derek.holloway
7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
25346 lines
1.1 MiB
Plaintext
25346 lines
1.1 MiB
Plaintext
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx600 < %s | FileCheck --check-prefixes=GFX6 %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx700 < %s | FileCheck --check-prefixes=GFX7 %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1010 < %s | FileCheck --check-prefixes=GFX10-WGP %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1010 -mattr=+cumode < %s | FileCheck --check-prefixes=GFX10-CU %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdpal -O0 -mcpu=gfx700 -amdgcn-skip-cache-invalidations < %s | FileCheck --check-prefixes=SKIP-CACHE-INV %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx90a < %s | FileCheck -check-prefixes=GFX90A-NOTTGSPLIT %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx90a -mattr=+tgsplit < %s | FileCheck -check-prefixes=GFX90A-TGSPLIT %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx942 < %s | FileCheck -check-prefixes=GFX942-NOTTGSPLIT %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx942 -mattr=+tgsplit < %s | FileCheck -check-prefixes=GFX942-TGSPLIT %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1100 < %s | FileCheck --check-prefixes=GFX11-WGP %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1100 -mattr=+cumode < %s | FileCheck --check-prefixes=GFX11-CU %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1200 < %s | FileCheck --check-prefixes=GFX12-WGP %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1200 -mattr=+cumode < %s | FileCheck --check-prefixes=GFX12-CU %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1250 < %s | FileCheck --check-prefixes=GFX1250 %s
|
|
|
|
define amdgpu_kernel void @local_workgroup_unordered_load(
|
|
; GFX6-LABEL: local_workgroup_unordered_load:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: ds_read_b32 v1, v0
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_unordered_load:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: ds_read_b32 v1, v0
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_unordered_load:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: ds_read_b32 v1, v0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_unordered_load:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: ds_read_b32 v1, v0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_unordered_load:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_read_b32 v1, v0
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_unordered_load:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_unordered_load:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_unordered_load:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_unordered_load:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_unordered_load:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: ds_load_b32 v1, v0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_unordered_load:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: ds_load_b32 v1, v0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_unordered_load:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: ds_load_b32 v1, v0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_unordered_load:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: ds_load_b32 v1, v0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_unordered_load:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: ds_load_b32 v1, v0
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %in, ptr addrspace(3) %out) {
|
|
entry:
|
|
%val = load atomic i32, ptr addrspace(3) %in syncscope("workgroup") unordered, align 4
|
|
store i32 %val, ptr addrspace(3) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_monotonic_load(
|
|
; GFX6-LABEL: local_workgroup_monotonic_load:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: ds_read_b32 v1, v0
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_monotonic_load:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: ds_read_b32 v1, v0
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_monotonic_load:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: ds_read_b32 v1, v0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_monotonic_load:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: ds_read_b32 v1, v0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_load:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_read_b32 v1, v0
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_monotonic_load:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_monotonic_load:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_monotonic_load:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_monotonic_load:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_monotonic_load:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: ds_load_b32 v1, v0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_monotonic_load:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: ds_load_b32 v1, v0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_monotonic_load:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: ds_load_b32 v1, v0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_monotonic_load:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: ds_load_b32 v1, v0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_monotonic_load:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: ds_load_b32 v1, v0
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %in, ptr addrspace(3) %out) {
|
|
entry:
|
|
%val = load atomic i32, ptr addrspace(3) %in syncscope("workgroup") monotonic, align 4
|
|
store i32 %val, ptr addrspace(3) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_acquire_load(
|
|
; GFX6-LABEL: local_workgroup_acquire_load:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: ds_read_b32 v1, v0
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_acquire_load:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: ds_read_b32 v1, v0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_acquire_load:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: ds_read_b32 v1, v0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_acquire_load:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: ds_read_b32 v1, v0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_load:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_read_b32 v1, v0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acquire_load:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_acquire_load:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_acquire_load:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_acquire_load:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_acquire_load:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: ds_load_b32 v1, v0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_acquire_load:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: ds_load_b32 v1, v0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_acquire_load:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: ds_load_b32 v1, v0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_acquire_load:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: ds_load_b32 v1, v0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_acquire_load:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: ds_load_b32 v1, v0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %in, ptr addrspace(3) %out) {
|
|
entry:
|
|
%val = load atomic i32, ptr addrspace(3) %in syncscope("workgroup") acquire, align 4
|
|
store i32 %val, ptr addrspace(3) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_seq_cst_load(
|
|
; GFX6-LABEL: local_workgroup_seq_cst_load:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_read_b32 v1, v0
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_seq_cst_load:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_read_b32 v1, v0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_seq_cst_load:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_read_b32 v1, v0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_seq_cst_load:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_read_b32 v1, v0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_load:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_read_b32 v1, v0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_load:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_seq_cst_load:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_load:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_seq_cst_load:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_seq_cst_load:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_load_b32 v1, v0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_seq_cst_load:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_load_b32 v1, v0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_seq_cst_load:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_load_b32 v1, v0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_seq_cst_load:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_load_b32 v1, v0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_seq_cst_load:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_load_b32 v1, v0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %in, ptr addrspace(3) %out) {
|
|
entry:
|
|
%val = load atomic i32, ptr addrspace(3) %in syncscope("workgroup") seq_cst, align 4
|
|
store i32 %val, ptr addrspace(3) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_unordered_store(
|
|
; GFX6-LABEL: local_workgroup_unordered_store:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_unordered_store:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_unordered_store:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_unordered_store:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_unordered_store:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_unordered_store:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_unordered_store:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_unordered_store:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_unordered_store:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_unordered_store:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_unordered_store:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_unordered_store:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_unordered_store:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_unordered_store:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
i32 %in, ptr addrspace(3) %out) {
|
|
entry:
|
|
store atomic i32 %in, ptr addrspace(3) %out syncscope("workgroup") unordered, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_monotonic_store(
|
|
; GFX6-LABEL: local_workgroup_monotonic_store:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_monotonic_store:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_monotonic_store:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_monotonic_store:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_store:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_monotonic_store:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_monotonic_store:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_monotonic_store:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_monotonic_store:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_monotonic_store:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_monotonic_store:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_monotonic_store:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_monotonic_store:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_monotonic_store:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
i32 %in, ptr addrspace(3) %out) {
|
|
entry:
|
|
store atomic i32 %in, ptr addrspace(3) %out syncscope("workgroup") monotonic, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_release_store(
|
|
; GFX6-LABEL: local_workgroup_release_store:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_release_store:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_release_store:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_release_store:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_release_store:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_release_store:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_release_store:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_release_store:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_release_store:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_release_store:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_release_store:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_release_store:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_release_store:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_release_store:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
i32 %in, ptr addrspace(3) %out) {
|
|
entry:
|
|
store atomic i32 %in, ptr addrspace(3) %out syncscope("workgroup") release, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_seq_cst_store(
|
|
; GFX6-LABEL: local_workgroup_seq_cst_store:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_seq_cst_store:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_seq_cst_store:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_seq_cst_store:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_store:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_store:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_seq_cst_store:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_store:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_seq_cst_store:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_seq_cst_store:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_seq_cst_store:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_seq_cst_store:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_seq_cst_store:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_seq_cst_store:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
i32 %in, ptr addrspace(3) %out) {
|
|
entry:
|
|
store atomic i32 %in, ptr addrspace(3) %out syncscope("workgroup") seq_cst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_monotonic_atomicrmw(
|
|
; GFX6-LABEL: local_workgroup_monotonic_atomicrmw:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_monotonic_atomicrmw:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_monotonic_atomicrmw:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_monotonic_atomicrmw:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_atomicrmw:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_monotonic_atomicrmw:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_monotonic_atomicrmw:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_monotonic_atomicrmw:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_monotonic_atomicrmw:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_monotonic_atomicrmw:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_monotonic_atomicrmw:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_monotonic_atomicrmw:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_monotonic_atomicrmw:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_monotonic_atomicrmw:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in) {
|
|
entry:
|
|
%val = atomicrmw volatile xchg ptr addrspace(3) %out, i32 %in syncscope("workgroup") monotonic
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_acquire_atomicrmw(
|
|
; GFX6-LABEL: local_workgroup_acquire_atomicrmw:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_acquire_atomicrmw:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_acquire_atomicrmw:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_acquire_atomicrmw:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_atomicrmw:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acquire_atomicrmw:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_acquire_atomicrmw:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_acquire_atomicrmw:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_acquire_atomicrmw:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_acquire_atomicrmw:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_acquire_atomicrmw:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_acquire_atomicrmw:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_acquire_atomicrmw:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_acquire_atomicrmw:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in) {
|
|
entry:
|
|
%val = atomicrmw volatile xchg ptr addrspace(3) %out, i32 %in syncscope("workgroup") acquire
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_release_atomicrmw(
|
|
; GFX6-LABEL: local_workgroup_release_atomicrmw:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_release_atomicrmw:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_release_atomicrmw:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_release_atomicrmw:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_release_atomicrmw:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_release_atomicrmw:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_release_atomicrmw:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_release_atomicrmw:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_release_atomicrmw:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_release_atomicrmw:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_release_atomicrmw:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_release_atomicrmw:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_release_atomicrmw:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_release_atomicrmw:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in) {
|
|
entry:
|
|
%val = atomicrmw volatile xchg ptr addrspace(3) %out, i32 %in syncscope("workgroup") release
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_acq_rel_atomicrmw(
|
|
; GFX6-LABEL: local_workgroup_acq_rel_atomicrmw:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_acq_rel_atomicrmw:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_acq_rel_atomicrmw:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_acq_rel_atomicrmw:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_acq_rel_atomicrmw:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_atomicrmw:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_acq_rel_atomicrmw:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_atomicrmw:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_acq_rel_atomicrmw:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_acq_rel_atomicrmw:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_acq_rel_atomicrmw:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_acq_rel_atomicrmw:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_acq_rel_atomicrmw:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_acq_rel_atomicrmw:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in) {
|
|
entry:
|
|
%val = atomicrmw volatile xchg ptr addrspace(3) %out, i32 %in syncscope("workgroup") acq_rel
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_seq_cst_atomicrmw(
|
|
; GFX6-LABEL: local_workgroup_seq_cst_atomicrmw:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_seq_cst_atomicrmw:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_seq_cst_atomicrmw:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_seq_cst_atomicrmw:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_atomicrmw:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_atomicrmw:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_seq_cst_atomicrmw:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_atomicrmw:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_seq_cst_atomicrmw:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_seq_cst_atomicrmw:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_seq_cst_atomicrmw:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_seq_cst_atomicrmw:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_seq_cst_atomicrmw:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_seq_cst_atomicrmw:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in) {
|
|
entry:
|
|
%val = atomicrmw volatile xchg ptr addrspace(3) %out, i32 %in syncscope("workgroup") seq_cst
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_acquire_ret_atomicrmw(
|
|
; GFX6-LABEL: local_workgroup_acquire_ret_atomicrmw:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_acquire_ret_atomicrmw:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_acquire_ret_atomicrmw:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_acquire_ret_atomicrmw:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_ret_atomicrmw:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acquire_ret_atomicrmw:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_acquire_ret_atomicrmw:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_acquire_ret_atomicrmw:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_acquire_ret_atomicrmw:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_acquire_ret_atomicrmw:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_acquire_ret_atomicrmw:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_acquire_ret_atomicrmw:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_acquire_ret_atomicrmw:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_acquire_ret_atomicrmw:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in) {
|
|
entry:
|
|
%val = atomicrmw volatile xchg ptr addrspace(3) %out, i32 %in syncscope("workgroup") acquire
|
|
store i32 %val, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_acq_rel_ret_atomicrmw(
|
|
; GFX6-LABEL: local_workgroup_acq_rel_ret_atomicrmw:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_acq_rel_ret_atomicrmw:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_acq_rel_ret_atomicrmw:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_acq_rel_ret_atomicrmw:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_acq_rel_ret_atomicrmw:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_ret_atomicrmw:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_acq_rel_ret_atomicrmw:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_ret_atomicrmw:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_acq_rel_ret_atomicrmw:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_acq_rel_ret_atomicrmw:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_acq_rel_ret_atomicrmw:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_acq_rel_ret_atomicrmw:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_acq_rel_ret_atomicrmw:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_acq_rel_ret_atomicrmw:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in) {
|
|
entry:
|
|
%val = atomicrmw volatile xchg ptr addrspace(3) %out, i32 %in syncscope("workgroup") acq_rel
|
|
store i32 %val, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_seq_cst_ret_atomicrmw(
|
|
; GFX6-LABEL: local_workgroup_seq_cst_ret_atomicrmw:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_seq_cst_ret_atomicrmw:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_seq_cst_ret_atomicrmw:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_seq_cst_ret_atomicrmw:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_ret_atomicrmw:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_ret_atomicrmw:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_seq_cst_ret_atomicrmw:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_ret_atomicrmw:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_seq_cst_ret_atomicrmw:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_seq_cst_ret_atomicrmw:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_seq_cst_ret_atomicrmw:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_seq_cst_ret_atomicrmw:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_seq_cst_ret_atomicrmw:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_seq_cst_ret_atomicrmw:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in) {
|
|
entry:
|
|
%val = atomicrmw volatile xchg ptr addrspace(3) %out, i32 %in syncscope("workgroup") seq_cst
|
|
store i32 %val, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_monotonic_monotonic_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_monotonic_monotonic_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_monotonic_monotonic_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_monotonic_monotonic_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_monotonic_monotonic_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_monotonic_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_monotonic_monotonic_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_monotonic_monotonic_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_monotonic_monotonic_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_monotonic_monotonic_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_monotonic_monotonic_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_monotonic_monotonic_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_monotonic_monotonic_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_monotonic_monotonic_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_monotonic_monotonic_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") monotonic monotonic
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_acquire_monotonic_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_acquire_monotonic_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_acquire_monotonic_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_acquire_monotonic_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_acquire_monotonic_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_monotonic_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acquire_monotonic_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_acquire_monotonic_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_acquire_monotonic_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_acquire_monotonic_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_acquire_monotonic_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_acquire_monotonic_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_acquire_monotonic_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_acquire_monotonic_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_acquire_monotonic_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") acquire monotonic
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_release_monotonic_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_release_monotonic_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_release_monotonic_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_release_monotonic_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_release_monotonic_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_release_monotonic_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_release_monotonic_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_release_monotonic_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_release_monotonic_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_release_monotonic_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_release_monotonic_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_release_monotonic_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_release_monotonic_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_release_monotonic_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_release_monotonic_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") release monotonic
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_acq_rel_monotonic_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_acq_rel_monotonic_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_acq_rel_monotonic_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_acq_rel_monotonic_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_acq_rel_monotonic_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_acq_rel_monotonic_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_monotonic_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_acq_rel_monotonic_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_monotonic_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_acq_rel_monotonic_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_acq_rel_monotonic_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_acq_rel_monotonic_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_acq_rel_monotonic_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_acq_rel_monotonic_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_acq_rel_monotonic_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") acq_rel monotonic
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_seq_cst_monotonic_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_seq_cst_monotonic_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_seq_cst_monotonic_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_seq_cst_monotonic_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_seq_cst_monotonic_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_monotonic_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_monotonic_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_seq_cst_monotonic_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_monotonic_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_seq_cst_monotonic_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_seq_cst_monotonic_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_seq_cst_monotonic_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_seq_cst_monotonic_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_seq_cst_monotonic_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_seq_cst_monotonic_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") seq_cst monotonic
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_monotonic_acquire_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_monotonic_acquire_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_monotonic_acquire_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_monotonic_acquire_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_monotonic_acquire_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_acquire_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_monotonic_acquire_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_monotonic_acquire_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_monotonic_acquire_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_monotonic_acquire_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_monotonic_acquire_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_monotonic_acquire_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_monotonic_acquire_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_monotonic_acquire_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_monotonic_acquire_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") monotonic acquire
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_acquire_acquire_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_acquire_acquire_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_acquire_acquire_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_acquire_acquire_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_acquire_acquire_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_acquire_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acquire_acquire_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_acquire_acquire_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_acquire_acquire_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_acquire_acquire_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_acquire_acquire_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_acquire_acquire_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_acquire_acquire_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_acquire_acquire_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_acquire_acquire_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") acquire acquire
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_release_acquire_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_release_acquire_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_release_acquire_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_release_acquire_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_release_acquire_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_release_acquire_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_release_acquire_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_release_acquire_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_release_acquire_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_release_acquire_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_release_acquire_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_release_acquire_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_release_acquire_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_release_acquire_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_release_acquire_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") release acquire
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_acq_rel_acquire_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_acq_rel_acquire_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_acq_rel_acquire_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_acq_rel_acquire_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_acq_rel_acquire_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_acq_rel_acquire_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_acquire_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_acq_rel_acquire_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_acquire_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_acq_rel_acquire_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_acq_rel_acquire_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_acq_rel_acquire_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_acq_rel_acquire_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_acq_rel_acquire_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_acq_rel_acquire_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") acq_rel acquire
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_seq_cst_acquire_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_seq_cst_acquire_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_seq_cst_acquire_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_seq_cst_acquire_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_seq_cst_acquire_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_acquire_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_acquire_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_seq_cst_acquire_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_acquire_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_seq_cst_acquire_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_seq_cst_acquire_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_seq_cst_acquire_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_seq_cst_acquire_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_seq_cst_acquire_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_seq_cst_acquire_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") seq_cst acquire
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_monotonic_seq_cst_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_monotonic_seq_cst_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_monotonic_seq_cst_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_monotonic_seq_cst_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_monotonic_seq_cst_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_seq_cst_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_monotonic_seq_cst_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_monotonic_seq_cst_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_monotonic_seq_cst_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_monotonic_seq_cst_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_monotonic_seq_cst_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_monotonic_seq_cst_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_monotonic_seq_cst_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_monotonic_seq_cst_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_monotonic_seq_cst_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") monotonic seq_cst
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_acquire_seq_cst_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_acquire_seq_cst_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_acquire_seq_cst_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_acquire_seq_cst_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_acquire_seq_cst_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_seq_cst_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acquire_seq_cst_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_acquire_seq_cst_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_acquire_seq_cst_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_acquire_seq_cst_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_acquire_seq_cst_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_acquire_seq_cst_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_acquire_seq_cst_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_acquire_seq_cst_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_acquire_seq_cst_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") acquire seq_cst
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_release_seq_cst_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_release_seq_cst_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_release_seq_cst_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_release_seq_cst_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_release_seq_cst_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_release_seq_cst_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_release_seq_cst_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_release_seq_cst_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_release_seq_cst_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_release_seq_cst_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_release_seq_cst_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_release_seq_cst_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_release_seq_cst_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_release_seq_cst_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_release_seq_cst_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") release seq_cst
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_acq_rel_seq_cst_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_acq_rel_seq_cst_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_acq_rel_seq_cst_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_acq_rel_seq_cst_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_acq_rel_seq_cst_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_acq_rel_seq_cst_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_seq_cst_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_acq_rel_seq_cst_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_seq_cst_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_acq_rel_seq_cst_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_acq_rel_seq_cst_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_acq_rel_seq_cst_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_acq_rel_seq_cst_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_acq_rel_seq_cst_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_acq_rel_seq_cst_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") acq_rel seq_cst
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_seq_cst_seq_cst_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_seq_cst_seq_cst_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_seq_cst_seq_cst_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_seq_cst_seq_cst_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_seq_cst_seq_cst_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_seq_cst_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_seq_cst_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_seq_cst_seq_cst_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_seq_cst_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_seq_cst_seq_cst_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_seq_cst_seq_cst_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_seq_cst_seq_cst_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_seq_cst_seq_cst_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_seq_cst_seq_cst_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_seq_cst_seq_cst_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") seq_cst seq_cst
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_monotonic_monotonic_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_monotonic_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") monotonic monotonic
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_acquire_monotonic_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_acquire_monotonic_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_acquire_monotonic_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_acquire_monotonic_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_acquire_monotonic_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_monotonic_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acquire_monotonic_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_acquire_monotonic_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_acquire_monotonic_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_acquire_monotonic_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_acquire_monotonic_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_acquire_monotonic_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_acquire_monotonic_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_acquire_monotonic_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_acquire_monotonic_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") acquire monotonic
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_release_monotonic_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_release_monotonic_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_release_monotonic_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_release_monotonic_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_release_monotonic_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_release_monotonic_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_release_monotonic_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_release_monotonic_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_release_monotonic_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_release_monotonic_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_release_monotonic_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_release_monotonic_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_release_monotonic_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_release_monotonic_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_release_monotonic_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") release monotonic
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_acq_rel_monotonic_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_acq_rel_monotonic_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") acq_rel monotonic
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_seq_cst_monotonic_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_monotonic_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") seq_cst monotonic
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_monotonic_acquire_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_monotonic_acquire_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_monotonic_acquire_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_monotonic_acquire_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_monotonic_acquire_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_acquire_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_monotonic_acquire_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_monotonic_acquire_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_monotonic_acquire_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_monotonic_acquire_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_monotonic_acquire_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_monotonic_acquire_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_monotonic_acquire_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_monotonic_acquire_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_monotonic_acquire_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") monotonic acquire
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_acquire_acquire_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_acquire_acquire_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_acquire_acquire_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_acquire_acquire_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_acquire_acquire_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_acquire_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acquire_acquire_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_acquire_acquire_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_acquire_acquire_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_acquire_acquire_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_acquire_acquire_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_acquire_acquire_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_acquire_acquire_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_acquire_acquire_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_acquire_acquire_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") acquire acquire
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_release_acquire_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_release_acquire_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_release_acquire_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_release_acquire_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_release_acquire_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_release_acquire_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_release_acquire_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_release_acquire_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_release_acquire_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_release_acquire_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_release_acquire_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_release_acquire_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_release_acquire_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_release_acquire_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_release_acquire_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") release acquire
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_acq_rel_acquire_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_acq_rel_acquire_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") acq_rel acquire
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_seq_cst_acquire_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_acquire_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") seq_cst acquire
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_monotonic_seq_cst_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_seq_cst_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") monotonic seq_cst
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_acquire_seq_cst_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_seq_cst_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") acquire seq_cst
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_release_seq_cst_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_release_seq_cst_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_release_seq_cst_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_release_seq_cst_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_release_seq_cst_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_release_seq_cst_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_release_seq_cst_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_release_seq_cst_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_release_seq_cst_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_release_seq_cst_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_release_seq_cst_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_release_seq_cst_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_release_seq_cst_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_release_seq_cst_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_release_seq_cst_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") release seq_cst
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_acq_rel_seq_cst_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_acq_rel_seq_cst_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") acq_rel seq_cst
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_seq_cst_seq_cst_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: buffer_gl0_inv
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_seq_cst_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: buffer_wbinvl1_vol
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: buffer_inv sc0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: buffer_gl0_inv
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_waitcnt_vscnt null, 0x0
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-WGP-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: global_inv scope:SCOPE_SE
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: s_wait_bvhcnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_samplecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_storecnt 0x0
|
|
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: s_wait_storecnt 0x0
|
|
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup") seq_cst seq_cst
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_unordered_load(
|
|
; GFX6-LABEL: local_workgroup_one_as_unordered_load:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: ds_read_b32 v1, v0
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_unordered_load:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: ds_read_b32 v1, v0
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_unordered_load:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: ds_read_b32 v1, v0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_unordered_load:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: ds_read_b32 v1, v0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_unordered_load:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_read_b32 v1, v0
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_unordered_load:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_unordered_load:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_unordered_load:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_unordered_load:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_unordered_load:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: ds_load_b32 v1, v0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_unordered_load:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: ds_load_b32 v1, v0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_unordered_load:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: ds_load_b32 v1, v0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_unordered_load:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: ds_load_b32 v1, v0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_unordered_load:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: ds_load_b32 v1, v0
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %in, ptr addrspace(3) %out) {
|
|
entry:
|
|
%val = load atomic i32, ptr addrspace(3) %in syncscope("workgroup-one-as") unordered, align 4
|
|
store i32 %val, ptr addrspace(3) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_monotonic_load(
|
|
; GFX6-LABEL: local_workgroup_one_as_monotonic_load:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: ds_read_b32 v1, v0
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_monotonic_load:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: ds_read_b32 v1, v0
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_monotonic_load:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: ds_read_b32 v1, v0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_monotonic_load:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: ds_read_b32 v1, v0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_load:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_read_b32 v1, v0
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_load:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_load:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_load:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_load:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_monotonic_load:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: ds_load_b32 v1, v0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_monotonic_load:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: ds_load_b32 v1, v0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_monotonic_load:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: ds_load_b32 v1, v0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_monotonic_load:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: ds_load_b32 v1, v0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_monotonic_load:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: ds_load_b32 v1, v0
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %in, ptr addrspace(3) %out) {
|
|
entry:
|
|
%val = load atomic i32, ptr addrspace(3) %in syncscope("workgroup-one-as") monotonic, align 4
|
|
store i32 %val, ptr addrspace(3) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_acquire_load(
|
|
; GFX6-LABEL: local_workgroup_one_as_acquire_load:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: ds_read_b32 v1, v0
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_acquire_load:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: ds_read_b32 v1, v0
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_acquire_load:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: ds_read_b32 v1, v0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_acquire_load:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: ds_read_b32 v1, v0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_load:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_read_b32 v1, v0
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_load:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_acquire_load:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_load:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_acquire_load:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_acquire_load:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: ds_load_b32 v1, v0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_acquire_load:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: ds_load_b32 v1, v0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_acquire_load:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: ds_load_b32 v1, v0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_acquire_load:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: ds_load_b32 v1, v0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_acquire_load:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: ds_load_b32 v1, v0
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %in, ptr addrspace(3) %out) {
|
|
entry:
|
|
%val = load atomic i32, ptr addrspace(3) %in syncscope("workgroup-one-as") acquire, align 4
|
|
store i32 %val, ptr addrspace(3) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_seq_cst_load(
|
|
; GFX6-LABEL: local_workgroup_one_as_seq_cst_load:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: ds_read_b32 v1, v0
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_seq_cst_load:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: ds_read_b32 v1, v0
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_seq_cst_load:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: ds_read_b32 v1, v0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_seq_cst_load:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: ds_read_b32 v1, v0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_load:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_read_b32 v1, v0
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_load:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_load:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_load:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_load:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_read_b32 v1, v0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_seq_cst_load:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: ds_load_b32 v1, v0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_seq_cst_load:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: ds_load_b32 v1, v0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_seq_cst_load:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: ds_load_b32 v1, v0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_seq_cst_load:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: ds_load_b32 v1, v0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_seq_cst_load:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: ds_load_b32 v1, v0
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %in, ptr addrspace(3) %out) {
|
|
entry:
|
|
%val = load atomic i32, ptr addrspace(3) %in syncscope("workgroup-one-as") seq_cst, align 4
|
|
store i32 %val, ptr addrspace(3) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_unordered_store(
|
|
; GFX6-LABEL: local_workgroup_one_as_unordered_store:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_unordered_store:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_unordered_store:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_unordered_store:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_unordered_store:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_unordered_store:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_unordered_store:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_unordered_store:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_unordered_store:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_unordered_store:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_unordered_store:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_unordered_store:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_unordered_store:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_unordered_store:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
i32 %in, ptr addrspace(3) %out) {
|
|
entry:
|
|
store atomic i32 %in, ptr addrspace(3) %out syncscope("workgroup-one-as") unordered, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_monotonic_store(
|
|
; GFX6-LABEL: local_workgroup_one_as_monotonic_store:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_monotonic_store:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_monotonic_store:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_monotonic_store:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_store:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_store:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_store:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_store:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_store:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_monotonic_store:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_monotonic_store:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_monotonic_store:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_monotonic_store:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_monotonic_store:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
i32 %in, ptr addrspace(3) %out) {
|
|
entry:
|
|
store atomic i32 %in, ptr addrspace(3) %out syncscope("workgroup-one-as") monotonic, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_release_store(
|
|
; GFX6-LABEL: local_workgroup_one_as_release_store:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_release_store:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_release_store:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_release_store:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_release_store:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_store:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_release_store:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_store:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_release_store:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_release_store:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_release_store:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_release_store:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_release_store:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_release_store:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
i32 %in, ptr addrspace(3) %out) {
|
|
entry:
|
|
store atomic i32 %in, ptr addrspace(3) %out syncscope("workgroup-one-as") release, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_seq_cst_store(
|
|
; GFX6-LABEL: local_workgroup_one_as_seq_cst_store:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_seq_cst_store:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_seq_cst_store:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_seq_cst_store:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_store:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_store:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_store:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_store:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_store:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_seq_cst_store:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_seq_cst_store:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_seq_cst_store:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_seq_cst_store:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_seq_cst_store:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
i32 %in, ptr addrspace(3) %out) {
|
|
entry:
|
|
store atomic i32 %in, ptr addrspace(3) %out syncscope("workgroup-one-as") seq_cst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_monotonic_atomicrmw(
|
|
; GFX6-LABEL: local_workgroup_one_as_monotonic_atomicrmw:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_monotonic_atomicrmw:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_monotonic_atomicrmw:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_monotonic_atomicrmw:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_atomicrmw:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_atomicrmw:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_atomicrmw:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_atomicrmw:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_atomicrmw:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_monotonic_atomicrmw:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_monotonic_atomicrmw:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_monotonic_atomicrmw:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_monotonic_atomicrmw:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_monotonic_atomicrmw:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in) {
|
|
entry:
|
|
%val = atomicrmw volatile xchg ptr addrspace(3) %out, i32 %in syncscope("workgroup-one-as") monotonic
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_acquire_atomicrmw(
|
|
; GFX6-LABEL: local_workgroup_one_as_acquire_atomicrmw:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_acquire_atomicrmw:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_acquire_atomicrmw:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_acquire_atomicrmw:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_atomicrmw:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_atomicrmw:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_acquire_atomicrmw:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_atomicrmw:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_acquire_atomicrmw:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_acquire_atomicrmw:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_acquire_atomicrmw:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_acquire_atomicrmw:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_acquire_atomicrmw:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_acquire_atomicrmw:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in) {
|
|
entry:
|
|
%val = atomicrmw volatile xchg ptr addrspace(3) %out, i32 %in syncscope("workgroup-one-as") acquire
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_release_atomicrmw(
|
|
; GFX6-LABEL: local_workgroup_one_as_release_atomicrmw:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_release_atomicrmw:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_release_atomicrmw:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_release_atomicrmw:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_release_atomicrmw:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_atomicrmw:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_release_atomicrmw:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_atomicrmw:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_release_atomicrmw:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_release_atomicrmw:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_release_atomicrmw:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_release_atomicrmw:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_release_atomicrmw:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_release_atomicrmw:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in) {
|
|
entry:
|
|
%val = atomicrmw volatile xchg ptr addrspace(3) %out, i32 %in syncscope("workgroup-one-as") release
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_acq_rel_atomicrmw(
|
|
; GFX6-LABEL: local_workgroup_one_as_acq_rel_atomicrmw:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_acq_rel_atomicrmw:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_acq_rel_atomicrmw:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_acq_rel_atomicrmw:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acq_rel_atomicrmw:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_atomicrmw:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_acq_rel_atomicrmw:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_atomicrmw:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_acq_rel_atomicrmw:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_acq_rel_atomicrmw:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_acq_rel_atomicrmw:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_acq_rel_atomicrmw:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_acq_rel_atomicrmw:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_acq_rel_atomicrmw:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in) {
|
|
entry:
|
|
%val = atomicrmw volatile xchg ptr addrspace(3) %out, i32 %in syncscope("workgroup-one-as") acq_rel
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_seq_cst_atomicrmw(
|
|
; GFX6-LABEL: local_workgroup_one_as_seq_cst_atomicrmw:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX6-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_seq_cst_atomicrmw:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_seq_cst_atomicrmw:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-WGP-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_seq_cst_atomicrmw:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX10-CU-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_atomicrmw:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_atomicrmw:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_atomicrmw:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_atomicrmw:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_atomicrmw:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v0, v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_seq_cst_atomicrmw:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_seq_cst_atomicrmw:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX11-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_seq_cst_atomicrmw:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-WGP-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_seq_cst_atomicrmw:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX12-CU-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_seq_cst_atomicrmw:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX1250-NEXT: ds_storexchg_rtn_b32 v0, v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in) {
|
|
entry:
|
|
%val = atomicrmw volatile xchg ptr addrspace(3) %out, i32 %in syncscope("workgroup-one-as") seq_cst
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_acquire_ret_atomicrmw(
|
|
; GFX6-LABEL: local_workgroup_one_as_acquire_ret_atomicrmw:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_acquire_ret_atomicrmw:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_acquire_ret_atomicrmw:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_acquire_ret_atomicrmw:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_ret_atomicrmw:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_ret_atomicrmw:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_acquire_ret_atomicrmw:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_ret_atomicrmw:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_acquire_ret_atomicrmw:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_acquire_ret_atomicrmw:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_acquire_ret_atomicrmw:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_acquire_ret_atomicrmw:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_acquire_ret_atomicrmw:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_acquire_ret_atomicrmw:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in) {
|
|
entry:
|
|
%val = atomicrmw volatile xchg ptr addrspace(3) %out, i32 %in syncscope("workgroup-one-as") acquire
|
|
store i32 %val, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_acq_rel_ret_atomicrmw(
|
|
; GFX6-LABEL: local_workgroup_one_as_acq_rel_ret_atomicrmw:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_acq_rel_ret_atomicrmw:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_acq_rel_ret_atomicrmw:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_acq_rel_ret_atomicrmw:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acq_rel_ret_atomicrmw:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_ret_atomicrmw:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_acq_rel_ret_atomicrmw:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_ret_atomicrmw:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_acq_rel_ret_atomicrmw:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_acq_rel_ret_atomicrmw:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_acq_rel_ret_atomicrmw:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_acq_rel_ret_atomicrmw:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_acq_rel_ret_atomicrmw:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_acq_rel_ret_atomicrmw:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in) {
|
|
entry:
|
|
%val = atomicrmw volatile xchg ptr addrspace(3) %out, i32 %in syncscope("workgroup-one-as") acq_rel
|
|
store i32 %val, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_seq_cst_ret_atomicrmw(
|
|
; GFX6-LABEL: local_workgroup_one_as_seq_cst_ret_atomicrmw:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_seq_cst_ret_atomicrmw:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_seq_cst_ret_atomicrmw:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_seq_cst_ret_atomicrmw:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_ret_atomicrmw:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_ret_atomicrmw:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_ret_atomicrmw:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_ret_atomicrmw:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_ret_atomicrmw:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_wrxchg_rtn_b32 v1, v0, v1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_seq_cst_ret_atomicrmw:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_seq_cst_ret_atomicrmw:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_seq_cst_ret_atomicrmw:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_seq_cst_ret_atomicrmw:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_seq_cst_ret_atomicrmw:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: ds_storexchg_rtn_b32 v1, v0, v1
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in) {
|
|
entry:
|
|
%val = atomicrmw volatile xchg ptr addrspace(3) %out, i32 %in syncscope("workgroup-one-as") seq_cst
|
|
store i32 %val, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_monotonic_monotonic_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_monotonic_monotonic_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_monotonic_monotonic_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_monotonic_monotonic_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_monotonic_monotonic_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_monotonic_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_monotonic_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_monotonic_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_monotonic_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_monotonic_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_monotonic_monotonic_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_monotonic_monotonic_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_monotonic_monotonic_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_monotonic_monotonic_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_monotonic_monotonic_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") monotonic monotonic
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_acquire_monotonic_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_acquire_monotonic_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_acquire_monotonic_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_acquire_monotonic_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_acquire_monotonic_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_monotonic_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_monotonic_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_acquire_monotonic_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_monotonic_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_acquire_monotonic_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_acquire_monotonic_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_acquire_monotonic_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_acquire_monotonic_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_acquire_monotonic_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_acquire_monotonic_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acquire monotonic
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_release_monotonic_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_release_monotonic_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_release_monotonic_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_release_monotonic_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_release_monotonic_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_release_monotonic_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_monotonic_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_release_monotonic_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_monotonic_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_release_monotonic_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_release_monotonic_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_release_monotonic_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_release_monotonic_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_release_monotonic_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_release_monotonic_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") release monotonic
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_acq_rel_monotonic_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_acq_rel_monotonic_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_acq_rel_monotonic_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_acq_rel_monotonic_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_acq_rel_monotonic_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acq_rel_monotonic_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_monotonic_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_acq_rel_monotonic_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_monotonic_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_acq_rel_monotonic_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_acq_rel_monotonic_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_acq_rel_monotonic_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_acq_rel_monotonic_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_acq_rel_monotonic_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_acq_rel_monotonic_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acq_rel monotonic
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_seq_cst_monotonic_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_seq_cst_monotonic_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_seq_cst_monotonic_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_seq_cst_monotonic_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_seq_cst_monotonic_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_monotonic_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_monotonic_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_monotonic_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_monotonic_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_monotonic_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_seq_cst_monotonic_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_seq_cst_monotonic_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_seq_cst_monotonic_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_seq_cst_monotonic_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_seq_cst_monotonic_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") seq_cst monotonic
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_monotonic_acquire_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_monotonic_acquire_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_monotonic_acquire_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_monotonic_acquire_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_monotonic_acquire_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_acquire_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_acquire_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_acquire_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_acquire_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_acquire_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_monotonic_acquire_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_monotonic_acquire_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_monotonic_acquire_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_monotonic_acquire_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_monotonic_acquire_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") monotonic acquire
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_acquire_acquire_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_acquire_acquire_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_acquire_acquire_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_acquire_acquire_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_acquire_acquire_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_acquire_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_acquire_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_acquire_acquire_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_acquire_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_acquire_acquire_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_acquire_acquire_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_acquire_acquire_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_acquire_acquire_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_acquire_acquire_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_acquire_acquire_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acquire acquire
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_release_acquire_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_release_acquire_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_release_acquire_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_release_acquire_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_release_acquire_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_release_acquire_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_acquire_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_release_acquire_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_acquire_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_release_acquire_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_release_acquire_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_release_acquire_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_release_acquire_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_release_acquire_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_release_acquire_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") release acquire
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_acq_rel_acquire_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_acq_rel_acquire_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_acq_rel_acquire_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_acq_rel_acquire_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_acq_rel_acquire_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acq_rel_acquire_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_acquire_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_acq_rel_acquire_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_acquire_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_acq_rel_acquire_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_acq_rel_acquire_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_acq_rel_acquire_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_acq_rel_acquire_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_acq_rel_acquire_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_acq_rel_acquire_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acq_rel acquire
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_seq_cst_acquire_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_seq_cst_acquire_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_seq_cst_acquire_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_seq_cst_acquire_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_seq_cst_acquire_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_acquire_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_acquire_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_acquire_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_acquire_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_acquire_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_seq_cst_acquire_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_seq_cst_acquire_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_seq_cst_acquire_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_seq_cst_acquire_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_seq_cst_acquire_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") seq_cst acquire
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_monotonic_seq_cst_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_monotonic_seq_cst_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_monotonic_seq_cst_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_monotonic_seq_cst_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_monotonic_seq_cst_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_seq_cst_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_seq_cst_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_seq_cst_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_seq_cst_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_seq_cst_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_monotonic_seq_cst_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_monotonic_seq_cst_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_monotonic_seq_cst_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_monotonic_seq_cst_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_monotonic_seq_cst_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") monotonic seq_cst
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_acquire_seq_cst_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_acquire_seq_cst_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_acquire_seq_cst_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_acquire_seq_cst_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_acquire_seq_cst_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_seq_cst_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_seq_cst_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_acquire_seq_cst_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_seq_cst_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_acquire_seq_cst_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_acquire_seq_cst_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_acquire_seq_cst_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_acquire_seq_cst_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_acquire_seq_cst_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_acquire_seq_cst_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acquire seq_cst
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_release_seq_cst_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_release_seq_cst_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_release_seq_cst_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_release_seq_cst_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_release_seq_cst_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_release_seq_cst_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_seq_cst_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_release_seq_cst_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_seq_cst_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_release_seq_cst_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_release_seq_cst_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_release_seq_cst_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_release_seq_cst_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_release_seq_cst_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_release_seq_cst_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") release seq_cst
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_acq_rel_seq_cst_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acq_rel seq_cst
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_seq_cst_seq_cst_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX6-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-WGP-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX10-CU-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_b32 v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX11-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-WGP-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX12-CU-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX1250-NEXT: ds_cmpstore_b32 v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") seq_cst seq_cst
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_monotonic_monotonic_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_monotonic_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_monotonic_monotonic_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") monotonic monotonic
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_acquire_monotonic_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acquire monotonic
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_release_monotonic_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_release_monotonic_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_release_monotonic_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_release_monotonic_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_release_monotonic_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_release_monotonic_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_monotonic_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_release_monotonic_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_monotonic_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_release_monotonic_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_release_monotonic_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_release_monotonic_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_release_monotonic_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_release_monotonic_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_release_monotonic_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") release monotonic
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acq_rel monotonic
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") seq_cst monotonic
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_monotonic_acquire_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") monotonic acquire
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_acquire_acquire_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_acquire_acquire_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_acquire_acquire_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_acquire_acquire_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_acquire_acquire_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_acquire_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_acquire_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_acquire_acquire_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_acquire_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_acquire_acquire_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_acquire_acquire_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_acquire_acquire_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_acquire_acquire_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_acquire_acquire_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_acquire_acquire_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acquire acquire
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_release_acquire_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_release_acquire_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_release_acquire_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_release_acquire_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_release_acquire_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_release_acquire_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_acquire_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_release_acquire_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_acquire_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_release_acquire_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_release_acquire_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_release_acquire_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_release_acquire_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_release_acquire_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_release_acquire_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") release acquire
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_acq_rel_acquire_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acq_rel acquire
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_seq_cst_acquire_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") seq_cst acquire
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") monotonic seq_cst
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_acquire_seq_cst_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acquire seq_cst
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_release_seq_cst_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_release_seq_cst_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_release_seq_cst_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_release_seq_cst_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_release_seq_cst_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_release_seq_cst_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_seq_cst_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_release_seq_cst_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_seq_cst_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_release_seq_cst_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_release_seq_cst_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_release_seq_cst_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_release_seq_cst_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_release_seq_cst_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_release_seq_cst_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") release seq_cst
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acq_rel seq_cst
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @local_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg(
|
|
; GFX6-LABEL: local_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX6: ; %bb.0: ; %entry
|
|
; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x2
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX6-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX6-NEXT: s_mov_b32 m0, -1
|
|
; GFX6-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX6-NEXT: ds_write_b32 v0, v1
|
|
; GFX6-NEXT: s_endpgm
|
|
;
|
|
; GFX7-LABEL: local_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1
|
|
; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX7-NEXT: s_mov_b32 m0, -1
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: ds_write_b32 v0, v1
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX10-WGP-LABEL: local_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX10-WGP: ; %bb.0: ; %entry
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-WGP-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-WGP-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX10-CU-LABEL: local_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX10-CU: ; %bb.0: ; %entry
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX10-CU-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-CU-NEXT: ds_write_b32 v0, v1
|
|
; GFX10-CU-NEXT: s_endpgm
|
|
;
|
|
; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
|
|
; SKIP-CACHE-INV: ; %bb.0: ; %entry
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1
|
|
; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s2
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1
|
|
; SKIP-CACHE-INV-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; SKIP-CACHE-INV-NEXT: s_mov_b32 m0, -1
|
|
; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0
|
|
; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SKIP-CACHE-INV-NEXT: ds_write_b32 v0, v1
|
|
; SKIP-CACHE-INV-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX90A-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX90A-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4
|
|
; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX90A-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX90A-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX90A-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-NOTTGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-NOTTGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX942-TGSPLIT-LABEL: local_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX942-TGSPLIT: ; %bb.0: ; %entry
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4
|
|
; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX942-TGSPLIT-NEXT: ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX942-TGSPLIT-NEXT: ds_write_b32 v0, v1
|
|
; GFX942-TGSPLIT-NEXT: s_endpgm
|
|
;
|
|
; GFX11-WGP-LABEL: local_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX11-WGP: ; %bb.0: ; %entry
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX11-CU-LABEL: local_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX11-CU: ; %bb.0: ; %entry
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX11-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX11-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX11-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX12-WGP-LABEL: local_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX12-WGP: ; %bb.0: ; %entry
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-WGP-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-WGP-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-WGP-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-WGP-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-WGP-NEXT: s_endpgm
|
|
;
|
|
; GFX12-CU-LABEL: local_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX12-CU: ; %bb.0: ; %entry
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
|
|
; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x4
|
|
; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x8
|
|
; GFX12-CU-NEXT: s_wait_kmcnt 0x0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX12-CU-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX12-CU-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX12-CU-NEXT: s_wait_dscnt 0x0
|
|
; GFX12-CU-NEXT: ds_store_b32 v0, v1
|
|
; GFX12-CU-NEXT: s_endpgm
|
|
;
|
|
; GFX1250-LABEL: local_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
|
|
; GFX1250: ; %bb.0: ; %entry
|
|
; GFX1250-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
|
|
; GFX1250-NEXT: v_nop
|
|
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 nv
|
|
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 nv
|
|
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 nv
|
|
; GFX1250-NEXT: s_wait_kmcnt 0x0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX1250-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX1250-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:16
|
|
; GFX1250-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX1250-NEXT: s_wait_dscnt 0x0
|
|
; GFX1250-NEXT: ds_store_b32 v0, v1
|
|
; GFX1250-NEXT: s_endpgm
|
|
ptr addrspace(3) %out, i32 %in, i32 %old) {
|
|
entry:
|
|
%gep = getelementptr i32, ptr addrspace(3) %out, i32 4
|
|
%val = cmpxchg volatile ptr addrspace(3) %gep, i32 %old, i32 %in syncscope("workgroup-one-as") seq_cst seq_cst
|
|
%val0 = extractvalue { i32, i1 } %val, 0
|
|
store i32 %val0, ptr addrspace(3) %out, align 4
|
|
ret void
|
|
}
|