Files
TheRock/compiler/amd-llvm/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.iterative.ll
T
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
Flatten subprojects into a single base monorepo
2026-09-02 16:34:47 -07:00

905 lines
51 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=amdgcn -mcpu=gfx90a -misched-cluster=0 -misched=gcn-iterative-minreg < %s | FileCheck -check-prefix=GCN-MINREG %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx90a -misched-cluster=0 -misched=gcn-iterative-max-occupancy-experimental < %s | FileCheck -check-prefix=GCN-MAXOCC %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx90a -misched-cluster=0 -misched=gcn-iterative-ilp < %s | FileCheck -check-prefix=GCN-ILP %s
define amdgpu_kernel void @test_sched_group_barrier_pipeline_MFMA_interleave(ptr addrspace(3) noalias %in, ptr addrspace(3) noalias %out) #0 {
; GCN-MINREG-LABEL: test_sched_group_barrier_pipeline_MFMA_interleave:
; GCN-MINREG: ; %bb.0: ; %entry
; GCN-MINREG-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
; GCN-MINREG-NEXT: v_lshlrev_b32_e32 v0, 7, v0
; GCN-MINREG-NEXT: v_and_b32_e32 v0, 0x1ff80, v0
; GCN-MINREG-NEXT: v_mov_b32_e32 v34, 1.0
; GCN-MINREG-NEXT: v_mov_b32_e32 v33, 2.0
; GCN-MINREG-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MINREG-NEXT: v_add_u32_e32 v36, s0, v0
; GCN-MINREG-NEXT: v_add_u32_e32 v37, s1, v0
; GCN-MINREG-NEXT: ds_read_b128 v[28:31], v36 offset:112
; GCN-MINREG-NEXT: ds_read_b128 v[24:27], v36 offset:96
; GCN-MINREG-NEXT: ds_read_b128 v[20:23], v36 offset:80
; GCN-MINREG-NEXT: ds_read_b128 v[16:19], v36 offset:64
; GCN-MINREG-NEXT: ds_read_b128 v[0:3], v36
; GCN-MINREG-NEXT: ds_read_b128 v[4:7], v36 offset:16
; GCN-MINREG-NEXT: ds_read_b128 v[8:11], v36 offset:32
; GCN-MINREG-NEXT: ds_read_b128 v[12:15], v36 offset:48
; GCN-MINREG-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MINREG-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v34, v33, v[0:31]
; GCN-MINREG-NEXT: v_mov_b32_e32 v32, s1
; GCN-MINREG-NEXT: v_add_u32_e32 v35, 0x6000, v36
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MINREG-NEXT: s_nop 15
; GCN-MINREG-NEXT: s_nop 0
; GCN-MINREG-NEXT: ds_write_b128 v37, v[28:31] offset:112
; GCN-MINREG-NEXT: ds_write_b128 v37, v[24:27] offset:96
; GCN-MINREG-NEXT: ds_write_b128 v37, v[20:23] offset:80
; GCN-MINREG-NEXT: ds_write_b128 v37, v[16:19] offset:64
; GCN-MINREG-NEXT: ds_write_b128 v37, v[12:15] offset:48
; GCN-MINREG-NEXT: ds_write_b128 v37, v[8:11] offset:32
; GCN-MINREG-NEXT: ds_write_b128 v37, v[4:7] offset:16
; GCN-MINREG-NEXT: ds_write_b128 v37, v[0:3]
; GCN-MINREG-NEXT: ds_read_b128 v[28:31], v36 offset:8304
; GCN-MINREG-NEXT: ds_read_b128 v[24:27], v36 offset:8288
; GCN-MINREG-NEXT: ds_read_b128 v[20:23], v36 offset:8272
; GCN-MINREG-NEXT: ds_read_b128 v[16:19], v36 offset:8256
; GCN-MINREG-NEXT: ds_read_b128 v[12:15], v36 offset:8240
; GCN-MINREG-NEXT: ds_read_b128 v[8:11], v36 offset:8224
; GCN-MINREG-NEXT: ds_read_b128 v[4:7], v36 offset:8208
; GCN-MINREG-NEXT: ds_read_b128 v[0:3], v36 offset:8192
; GCN-MINREG-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MINREG-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v34, v33, v[0:31]
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MINREG-NEXT: s_nop 15
; GCN-MINREG-NEXT: s_nop 2
; GCN-MINREG-NEXT: ds_write_b128 v32, v[24:27] offset:8288
; GCN-MINREG-NEXT: ds_write_b128 v32, v[28:31] offset:8304
; GCN-MINREG-NEXT: ds_write_b128 v32, v[16:19] offset:8256
; GCN-MINREG-NEXT: ds_write_b128 v32, v[20:23] offset:8272
; GCN-MINREG-NEXT: ds_write_b128 v32, v[8:11] offset:8224
; GCN-MINREG-NEXT: ds_write_b128 v32, v[12:15] offset:8240
; GCN-MINREG-NEXT: ds_write_b128 v32, v[0:3] offset:8192
; GCN-MINREG-NEXT: ds_write_b128 v32, v[4:7] offset:8208
; GCN-MINREG-NEXT: ds_read_b128 v[28:31], v36 offset:24688
; GCN-MINREG-NEXT: ds_read_b128 v[24:27], v36 offset:24672
; GCN-MINREG-NEXT: ds_read_b128 v[20:23], v36 offset:24656
; GCN-MINREG-NEXT: ds_read_b128 v[16:19], v36 offset:24640
; GCN-MINREG-NEXT: ds_read_b128 v[12:15], v36 offset:24624
; GCN-MINREG-NEXT: ds_read_b128 v[8:11], v36 offset:24608
; GCN-MINREG-NEXT: ds_read_b128 v[4:7], v36 offset:24592
; GCN-MINREG-NEXT: ds_read_b128 v[0:3], v36 offset:24576
; GCN-MINREG-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MINREG-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v34, v33, v[0:31]
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MINREG-NEXT: s_nop 15
; GCN-MINREG-NEXT: s_nop 2
; GCN-MINREG-NEXT: ds_write_b128 v32, v[24:27] offset:16480
; GCN-MINREG-NEXT: ds_write_b128 v32, v[28:31] offset:16496
; GCN-MINREG-NEXT: ds_write_b128 v32, v[16:19] offset:16448
; GCN-MINREG-NEXT: ds_write_b128 v32, v[20:23] offset:16464
; GCN-MINREG-NEXT: ds_write_b128 v32, v[8:11] offset:16416
; GCN-MINREG-NEXT: ds_write_b128 v32, v[12:15] offset:16432
; GCN-MINREG-NEXT: ds_write_b128 v32, v[0:3] offset:16384
; GCN-MINREG-NEXT: ds_write_b128 v32, v[4:7] offset:16400
; GCN-MINREG-NEXT: ds_read_b128 v[28:31], v36 offset:49264
; GCN-MINREG-NEXT: ds_read_b128 v[24:27], v36 offset:49248
; GCN-MINREG-NEXT: ds_read_b128 v[20:23], v36 offset:49232
; GCN-MINREG-NEXT: ds_read_b128 v[16:19], v36 offset:49216
; GCN-MINREG-NEXT: ds_read_b128 v[12:15], v36 offset:49200
; GCN-MINREG-NEXT: ds_read_b128 v[8:11], v36 offset:49184
; GCN-MINREG-NEXT: ds_read_b128 v[4:7], v36 offset:49168
; GCN-MINREG-NEXT: ds_read_b128 v[0:3], v36 offset:49152
; GCN-MINREG-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MINREG-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v34, v33, v[0:31]
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MINREG-NEXT: s_nop 15
; GCN-MINREG-NEXT: s_nop 2
; GCN-MINREG-NEXT: ds_write_b128 v32, v[24:27] offset:24672
; GCN-MINREG-NEXT: ds_write_b128 v32, v[28:31] offset:24688
; GCN-MINREG-NEXT: ds_write_b128 v32, v[16:19] offset:24640
; GCN-MINREG-NEXT: ds_write_b128 v32, v[20:23] offset:24656
; GCN-MINREG-NEXT: ds_write_b128 v32, v[8:11] offset:24608
; GCN-MINREG-NEXT: ds_write_b128 v32, v[12:15] offset:24624
; GCN-MINREG-NEXT: ds_write_b128 v32, v[0:3] offset:24576
; GCN-MINREG-NEXT: ds_write_b128 v32, v[4:7] offset:24592
; GCN-MINREG-NEXT: ds_read_b128 v[28:31], v35 offset:57456
; GCN-MINREG-NEXT: ds_read_b128 v[24:27], v35 offset:57440
; GCN-MINREG-NEXT: ds_read_b128 v[20:23], v35 offset:57424
; GCN-MINREG-NEXT: ds_read_b128 v[16:19], v35 offset:57408
; GCN-MINREG-NEXT: ds_read_b128 v[0:3], v35 offset:57344
; GCN-MINREG-NEXT: ds_read_b128 v[4:7], v35 offset:57360
; GCN-MINREG-NEXT: ds_read_b128 v[8:11], v35 offset:57376
; GCN-MINREG-NEXT: ds_read_b128 v[12:15], v35 offset:57392
; GCN-MINREG-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MINREG-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v34, v33, v[0:31]
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MINREG-NEXT: s_nop 15
; GCN-MINREG-NEXT: s_nop 2
; GCN-MINREG-NEXT: ds_write_b128 v32, v[24:27] offset:32864
; GCN-MINREG-NEXT: ds_write_b128 v32, v[28:31] offset:32880
; GCN-MINREG-NEXT: ds_write_b128 v32, v[16:19] offset:32832
; GCN-MINREG-NEXT: ds_write_b128 v32, v[20:23] offset:32848
; GCN-MINREG-NEXT: ds_write_b128 v32, v[8:11] offset:32800
; GCN-MINREG-NEXT: ds_write_b128 v32, v[12:15] offset:32816
; GCN-MINREG-NEXT: ds_write_b128 v32, v[0:3] offset:32768
; GCN-MINREG-NEXT: ds_write_b128 v32, v[4:7] offset:32784
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MINREG-NEXT: s_endpgm
;
; GCN-MAXOCC-LABEL: test_sched_group_barrier_pipeline_MFMA_interleave:
; GCN-MAXOCC: ; %bb.0: ; %entry
; GCN-MAXOCC-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
; GCN-MAXOCC-NEXT: v_lshlrev_b32_e32 v0, 7, v0
; GCN-MAXOCC-NEXT: v_and_b32_e32 v33, 0x1ff80, v0
; GCN-MAXOCC-NEXT: v_mov_b32_e32 v34, 1.0
; GCN-MAXOCC-NEXT: v_mov_b32_e32 v35, 2.0
; GCN-MAXOCC-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MAXOCC-NEXT: v_add_u32_e32 v32, s0, v33
; GCN-MAXOCC-NEXT: ds_read_b128 v[28:31], v32 offset:112
; GCN-MAXOCC-NEXT: ds_read_b128 v[24:27], v32 offset:96
; GCN-MAXOCC-NEXT: ds_read_b128 v[20:23], v32 offset:80
; GCN-MAXOCC-NEXT: ds_read_b128 v[16:19], v32 offset:64
; GCN-MAXOCC-NEXT: ds_read_b128 v[0:3], v32
; GCN-MAXOCC-NEXT: ds_read_b128 v[4:7], v32 offset:16
; GCN-MAXOCC-NEXT: ds_read_b128 v[8:11], v32 offset:32
; GCN-MAXOCC-NEXT: ds_read_b128 v[12:15], v32 offset:48
; GCN-MAXOCC-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MAXOCC-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v34, v35, v[0:31]
; GCN-MAXOCC-NEXT: v_add_u32_e32 v33, s1, v33
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MAXOCC-NEXT: s_nop 15
; GCN-MAXOCC-NEXT: s_nop 1
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[28:31] offset:112
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[24:27] offset:96
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[20:23] offset:80
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[16:19] offset:64
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[12:15] offset:48
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[8:11] offset:32
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[4:7] offset:16
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[0:3]
; GCN-MAXOCC-NEXT: ds_read_b128 v[28:31], v32 offset:8304
; GCN-MAXOCC-NEXT: ds_read_b128 v[24:27], v32 offset:8288
; GCN-MAXOCC-NEXT: ds_read_b128 v[20:23], v32 offset:8272
; GCN-MAXOCC-NEXT: ds_read_b128 v[16:19], v32 offset:8256
; GCN-MAXOCC-NEXT: ds_read_b128 v[12:15], v32 offset:8240
; GCN-MAXOCC-NEXT: ds_read_b128 v[8:11], v32 offset:8224
; GCN-MAXOCC-NEXT: ds_read_b128 v[4:7], v32 offset:8208
; GCN-MAXOCC-NEXT: ds_read_b128 v[0:3], v32 offset:8192
; GCN-MAXOCC-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MAXOCC-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v34, v35, v[0:31]
; GCN-MAXOCC-NEXT: v_mov_b32_e32 v33, s1
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MAXOCC-NEXT: s_nop 15
; GCN-MAXOCC-NEXT: s_nop 1
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[24:27] offset:8288
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[28:31] offset:8304
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[16:19] offset:8256
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[20:23] offset:8272
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[8:11] offset:8224
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[12:15] offset:8240
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[0:3] offset:8192
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[4:7] offset:8208
; GCN-MAXOCC-NEXT: ds_read_b128 v[28:31], v32 offset:24688
; GCN-MAXOCC-NEXT: ds_read_b128 v[24:27], v32 offset:24672
; GCN-MAXOCC-NEXT: ds_read_b128 v[20:23], v32 offset:24656
; GCN-MAXOCC-NEXT: ds_read_b128 v[16:19], v32 offset:24640
; GCN-MAXOCC-NEXT: ds_read_b128 v[12:15], v32 offset:24624
; GCN-MAXOCC-NEXT: ds_read_b128 v[8:11], v32 offset:24608
; GCN-MAXOCC-NEXT: ds_read_b128 v[4:7], v32 offset:24592
; GCN-MAXOCC-NEXT: ds_read_b128 v[0:3], v32 offset:24576
; GCN-MAXOCC-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MAXOCC-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v34, v35, v[0:31]
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MAXOCC-NEXT: s_nop 15
; GCN-MAXOCC-NEXT: s_nop 2
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[24:27] offset:16480
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[28:31] offset:16496
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[16:19] offset:16448
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[20:23] offset:16464
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[8:11] offset:16416
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[12:15] offset:16432
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[0:3] offset:16384
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[4:7] offset:16400
; GCN-MAXOCC-NEXT: ds_read_b128 v[28:31], v32 offset:49264
; GCN-MAXOCC-NEXT: ds_read_b128 v[24:27], v32 offset:49248
; GCN-MAXOCC-NEXT: ds_read_b128 v[20:23], v32 offset:49232
; GCN-MAXOCC-NEXT: ds_read_b128 v[16:19], v32 offset:49216
; GCN-MAXOCC-NEXT: ds_read_b128 v[12:15], v32 offset:49200
; GCN-MAXOCC-NEXT: ds_read_b128 v[8:11], v32 offset:49184
; GCN-MAXOCC-NEXT: ds_read_b128 v[4:7], v32 offset:49168
; GCN-MAXOCC-NEXT: ds_read_b128 v[0:3], v32 offset:49152
; GCN-MAXOCC-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MAXOCC-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v34, v35, v[0:31]
; GCN-MAXOCC-NEXT: v_add_u32_e32 v32, 0x6000, v32
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MAXOCC-NEXT: s_nop 15
; GCN-MAXOCC-NEXT: s_nop 1
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[24:27] offset:24672
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[28:31] offset:24688
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[16:19] offset:24640
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[20:23] offset:24656
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[8:11] offset:24608
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[12:15] offset:24624
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[0:3] offset:24576
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[4:7] offset:24592
; GCN-MAXOCC-NEXT: ds_read_b128 v[28:31], v32 offset:57456
; GCN-MAXOCC-NEXT: ds_read_b128 v[24:27], v32 offset:57440
; GCN-MAXOCC-NEXT: ds_read_b128 v[20:23], v32 offset:57424
; GCN-MAXOCC-NEXT: ds_read_b128 v[16:19], v32 offset:57408
; GCN-MAXOCC-NEXT: ds_read_b128 v[0:3], v32 offset:57344
; GCN-MAXOCC-NEXT: ds_read_b128 v[4:7], v32 offset:57360
; GCN-MAXOCC-NEXT: ds_read_b128 v[8:11], v32 offset:57376
; GCN-MAXOCC-NEXT: ds_read_b128 v[12:15], v32 offset:57392
; GCN-MAXOCC-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MAXOCC-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v34, v35, v[0:31]
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MAXOCC-NEXT: s_nop 15
; GCN-MAXOCC-NEXT: s_nop 2
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[24:27] offset:32864
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[28:31] offset:32880
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[16:19] offset:32832
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[20:23] offset:32848
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[8:11] offset:32800
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[12:15] offset:32816
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[0:3] offset:32768
; GCN-MAXOCC-NEXT: ds_write_b128 v33, v[4:7] offset:32784
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: s_endpgm
;
; GCN-ILP-LABEL: test_sched_group_barrier_pipeline_MFMA_interleave:
; GCN-ILP: ; %bb.0: ; %entry
; GCN-ILP-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
; GCN-ILP-NEXT: v_lshlrev_b32_e32 v0, 7, v0
; GCN-ILP-NEXT: v_and_b32_e32 v32, 0x1ff80, v0
; GCN-ILP-NEXT: v_mov_b32_e32 v33, 1.0
; GCN-ILP-NEXT: v_mov_b32_e32 v34, 2.0
; GCN-ILP-NEXT: s_waitcnt lgkmcnt(0)
; GCN-ILP-NEXT: v_add_u32_e32 v35, s0, v32
; GCN-ILP-NEXT: ds_read_b128 v[12:15], v35 offset:48
; GCN-ILP-NEXT: ds_read_b128 v[8:11], v35 offset:32
; GCN-ILP-NEXT: ds_read_b128 v[4:7], v35 offset:16
; GCN-ILP-NEXT: ds_read_b128 v[0:3], v35
; GCN-ILP-NEXT: ds_read_b128 v[16:19], v35 offset:64
; GCN-ILP-NEXT: ds_read_b128 v[20:23], v35 offset:80
; GCN-ILP-NEXT: ds_read_b128 v[24:27], v35 offset:96
; GCN-ILP-NEXT: ds_read_b128 v[28:31], v35 offset:112
; GCN-ILP-NEXT: s_waitcnt lgkmcnt(0)
; GCN-ILP-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v33, v34, v[0:31]
; GCN-ILP-NEXT: v_add_u32_e32 v32, s1, v32
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-ILP-NEXT: s_nop 15
; GCN-ILP-NEXT: s_nop 1
; GCN-ILP-NEXT: ds_write_b128 v32, v[28:31] offset:112
; GCN-ILP-NEXT: ds_write_b128 v32, v[24:27] offset:96
; GCN-ILP-NEXT: ds_write_b128 v32, v[20:23] offset:80
; GCN-ILP-NEXT: ds_write_b128 v32, v[16:19] offset:64
; GCN-ILP-NEXT: ds_write_b128 v32, v[12:15] offset:48
; GCN-ILP-NEXT: ds_write_b128 v32, v[8:11] offset:32
; GCN-ILP-NEXT: ds_write_b128 v32, v[4:7] offset:16
; GCN-ILP-NEXT: ds_write_b128 v32, v[0:3]
; GCN-ILP-NEXT: ds_read_b128 v[0:3], v35 offset:8192
; GCN-ILP-NEXT: ds_read_b128 v[4:7], v35 offset:8208
; GCN-ILP-NEXT: ds_read_b128 v[8:11], v35 offset:8224
; GCN-ILP-NEXT: ds_read_b128 v[12:15], v35 offset:8240
; GCN-ILP-NEXT: ds_read_b128 v[16:19], v35 offset:8256
; GCN-ILP-NEXT: ds_read_b128 v[20:23], v35 offset:8272
; GCN-ILP-NEXT: ds_read_b128 v[24:27], v35 offset:8288
; GCN-ILP-NEXT: ds_read_b128 v[28:31], v35 offset:8304
; GCN-ILP-NEXT: s_waitcnt lgkmcnt(0)
; GCN-ILP-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v33, v34, v[0:31]
; GCN-ILP-NEXT: v_mov_b32_e32 v32, s1
; GCN-ILP-NEXT: s_nop 15
; GCN-ILP-NEXT: s_nop 1
; GCN-ILP-NEXT: ds_write_b128 v32, v[24:27] offset:8288
; GCN-ILP-NEXT: ds_write_b128 v32, v[28:31] offset:8304
; GCN-ILP-NEXT: ds_write_b128 v32, v[16:19] offset:8256
; GCN-ILP-NEXT: ds_write_b128 v32, v[20:23] offset:8272
; GCN-ILP-NEXT: ds_write_b128 v32, v[8:11] offset:8224
; GCN-ILP-NEXT: ds_write_b128 v32, v[12:15] offset:8240
; GCN-ILP-NEXT: ds_write_b128 v32, v[0:3] offset:8192
; GCN-ILP-NEXT: ds_write_b128 v32, v[4:7] offset:8208
; GCN-ILP-NEXT: ds_read_b128 v[0:3], v35 offset:24576
; GCN-ILP-NEXT: ds_read_b128 v[4:7], v35 offset:24592
; GCN-ILP-NEXT: ds_read_b128 v[8:11], v35 offset:24608
; GCN-ILP-NEXT: ds_read_b128 v[12:15], v35 offset:24624
; GCN-ILP-NEXT: ds_read_b128 v[16:19], v35 offset:24640
; GCN-ILP-NEXT: ds_read_b128 v[20:23], v35 offset:24656
; GCN-ILP-NEXT: ds_read_b128 v[24:27], v35 offset:24672
; GCN-ILP-NEXT: ds_read_b128 v[28:31], v35 offset:24688
; GCN-ILP-NEXT: s_waitcnt lgkmcnt(0)
; GCN-ILP-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v33, v34, v[0:31]
; GCN-ILP-NEXT: s_nop 15
; GCN-ILP-NEXT: s_nop 2
; GCN-ILP-NEXT: ds_write_b128 v32, v[4:7] offset:16400
; GCN-ILP-NEXT: ds_read_b128 v[4:7], v35 offset:49168
; GCN-ILP-NEXT: ds_write_b128 v32, v[0:3] offset:16384
; GCN-ILP-NEXT: ds_read_b128 v[0:3], v35 offset:49152
; GCN-ILP-NEXT: ds_write_b128 v32, v[12:15] offset:16432
; GCN-ILP-NEXT: ds_read_b128 v[12:15], v35 offset:49200
; GCN-ILP-NEXT: ds_write_b128 v32, v[8:11] offset:16416
; GCN-ILP-NEXT: ds_read_b128 v[8:11], v35 offset:49184
; GCN-ILP-NEXT: ds_write_b128 v32, v[20:23] offset:16464
; GCN-ILP-NEXT: ds_read_b128 v[20:23], v35 offset:49232
; GCN-ILP-NEXT: ds_write_b128 v32, v[16:19] offset:16448
; GCN-ILP-NEXT: ds_read_b128 v[16:19], v35 offset:49216
; GCN-ILP-NEXT: ds_write_b128 v32, v[28:31] offset:16496
; GCN-ILP-NEXT: ds_read_b128 v[28:31], v35 offset:49264
; GCN-ILP-NEXT: ds_write_b128 v32, v[24:27] offset:16480
; GCN-ILP-NEXT: ds_read_b128 v[24:27], v35 offset:49248
; GCN-ILP-NEXT: s_waitcnt lgkmcnt(0)
; GCN-ILP-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v33, v34, v[0:31]
; GCN-ILP-NEXT: v_add_u32_e32 v35, 0x6000, v35
; GCN-ILP-NEXT: s_nop 15
; GCN-ILP-NEXT: s_nop 1
; GCN-ILP-NEXT: ds_write_b128 v32, v[4:7] offset:24592
; GCN-ILP-NEXT: ds_read_b128 v[4:7], v35 offset:57360
; GCN-ILP-NEXT: ds_write_b128 v32, v[0:3] offset:24576
; GCN-ILP-NEXT: ds_read_b128 v[0:3], v35 offset:57344
; GCN-ILP-NEXT: ds_write_b128 v32, v[12:15] offset:24624
; GCN-ILP-NEXT: ds_read_b128 v[12:15], v35 offset:57392
; GCN-ILP-NEXT: ds_write_b128 v32, v[8:11] offset:24608
; GCN-ILP-NEXT: ds_read_b128 v[8:11], v35 offset:57376
; GCN-ILP-NEXT: ds_write_b128 v32, v[20:23] offset:24656
; GCN-ILP-NEXT: ds_read_b128 v[20:23], v35 offset:57424
; GCN-ILP-NEXT: ds_write_b128 v32, v[16:19] offset:24640
; GCN-ILP-NEXT: ds_read_b128 v[16:19], v35 offset:57408
; GCN-ILP-NEXT: ds_write_b128 v32, v[28:31] offset:24688
; GCN-ILP-NEXT: ds_read_b128 v[28:31], v35 offset:57456
; GCN-ILP-NEXT: ds_write_b128 v32, v[24:27] offset:24672
; GCN-ILP-NEXT: ds_read_b128 v[24:27], v35 offset:57440
; GCN-ILP-NEXT: s_waitcnt lgkmcnt(0)
; GCN-ILP-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v33, v34, v[0:31]
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-ILP-NEXT: s_nop 15
; GCN-ILP-NEXT: s_nop 2
; GCN-ILP-NEXT: ds_write_b128 v32, v[24:27] offset:32864
; GCN-ILP-NEXT: ds_write_b128 v32, v[28:31] offset:32880
; GCN-ILP-NEXT: ds_write_b128 v32, v[16:19] offset:32832
; GCN-ILP-NEXT: ds_write_b128 v32, v[20:23] offset:32848
; GCN-ILP-NEXT: ds_write_b128 v32, v[8:11] offset:32800
; GCN-ILP-NEXT: ds_write_b128 v32, v[12:15] offset:32816
; GCN-ILP-NEXT: ds_write_b128 v32, v[0:3] offset:32768
; GCN-ILP-NEXT: ds_write_b128 v32, v[4:7] offset:32784
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-ILP-NEXT: s_endpgm
entry:
%idx = call i32 @llvm.amdgcn.workitem.id.x()
%load.0.addr = getelementptr <32 x float>, ptr addrspace(3) %in, i32 %idx
%load.0 = load <32 x float>, ptr addrspace(3) %load.0.addr
%load.1.addr = getelementptr <32 x float>, ptr addrspace(3) %load.0.addr, i32 64
%load.1 = load <32 x float>, ptr addrspace(3) %load.1.addr
%load.2.addr = getelementptr <32 x float>, ptr addrspace(3) %load.1.addr, i32 128
%load.2 = load <32 x float>, ptr addrspace(3) %load.2.addr
%load.3.addr = getelementptr <32 x float>, ptr addrspace(3) %load.2.addr, i32 192
%load.3 = load <32 x float>, ptr addrspace(3) %load.3.addr
%load.4.addr = getelementptr <32 x float>, ptr addrspace(3) %load.3.addr, i32 256
%load.4 = load <32 x float>, ptr addrspace(3) %load.4.addr
%mai.0 = tail call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float 1.0, float 2.0, <32 x float> %load.0, i32 0, i32 0, i32 0)
%mai.1 = tail call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float 1.0, float 2.0, <32 x float> %load.1, i32 0, i32 0, i32 0)
%mai.2 = tail call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float 1.0, float 2.0, <32 x float> %load.2, i32 0, i32 0, i32 0)
%mai.3 = tail call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float 1.0, float 2.0, <32 x float> %load.3, i32 0, i32 0, i32 0)
%mai.4 = tail call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float 1.0, float 2.0, <32 x float> %load.4, i32 0, i32 0, i32 0)
%store.0.addr = getelementptr <32 x float>, ptr addrspace(3) %out, i32 %idx
store <32 x float> %mai.0, ptr addrspace(3) %store.0.addr
%store.1.addr = getelementptr <32 x float>, ptr addrspace(3) %out, i32 64
store <32 x float> %mai.1, ptr addrspace(3) %store.1.addr
%store.2.addr = getelementptr <32 x float>, ptr addrspace(3) %out, i32 128
store <32 x float> %mai.2, ptr addrspace(3) %store.2.addr
%store.3.addr = getelementptr <32 x float>, ptr addrspace(3) %out, i32 192
store <32 x float> %mai.3, ptr addrspace(3) %store.3.addr
%store.4.addr = getelementptr <32 x float>, ptr addrspace(3) %out, i32 256
store <32 x float> %mai.4, ptr addrspace(3) %store.4.addr
; 8 DS read
call void @llvm.amdgcn.sched.group.barrier(i32 256, i32 8, i32 0)
; 1 MFMA
call void @llvm.amdgcn.sched.group.barrier(i32 8, i32 1, i32 0)
; 8 DS write
call void @llvm.amdgcn.sched.group.barrier(i32 512, i32 8, i32 0)
; 8 DS read
call void @llvm.amdgcn.sched.group.barrier(i32 256, i32 8, i32 0)
; 1 MFMA
call void @llvm.amdgcn.sched.group.barrier(i32 8, i32 1, i32 0)
; 8 DS write
call void @llvm.amdgcn.sched.group.barrier(i32 512, i32 8, i32 0)
; 8 DS read
call void @llvm.amdgcn.sched.group.barrier(i32 256, i32 8, i32 0)
; 1 MFMA
call void @llvm.amdgcn.sched.group.barrier(i32 8, i32 1, i32 0)
; 8 DS write
call void @llvm.amdgcn.sched.group.barrier(i32 512, i32 8, i32 0)
; 8 DS read
call void @llvm.amdgcn.sched.group.barrier(i32 256, i32 8, i32 0)
; 1 MFMA
call void @llvm.amdgcn.sched.group.barrier(i32 8, i32 1, i32 0)
; 8 DS write
call void @llvm.amdgcn.sched.group.barrier(i32 512, i32 8, i32 0)
; 8 DS read
call void @llvm.amdgcn.sched.group.barrier(i32 256, i32 8, i32 0)
; 1 MFMA
call void @llvm.amdgcn.sched.group.barrier(i32 8, i32 1, i32 0)
; 8 DS write
call void @llvm.amdgcn.sched.group.barrier(i32 512, i32 8, i32 0)
ret void
}
define amdgpu_kernel void @test_sched_group_barrier_pipeline_MFMA_interleave_split_region(ptr addrspace(3) noalias %in, ptr addrspace(3) noalias %out) #0 {
; GCN-MINREG-LABEL: test_sched_group_barrier_pipeline_MFMA_interleave_split_region:
; GCN-MINREG: ; %bb.0: ; %entry
; GCN-MINREG-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
; GCN-MINREG-NEXT: v_lshlrev_b32_e32 v0, 7, v0
; GCN-MINREG-NEXT: v_and_b32_e32 v0, 0x1ff80, v0
; GCN-MINREG-NEXT: v_mov_b32_e32 v33, 1.0
; GCN-MINREG-NEXT: v_mov_b32_e32 v32, 2.0
; GCN-MINREG-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MINREG-NEXT: v_add_u32_e32 v35, s0, v0
; GCN-MINREG-NEXT: ds_read_b128 v[28:31], v35 offset:112
; GCN-MINREG-NEXT: v_add_u32_e32 v34, s1, v0
; GCN-MINREG-NEXT: ds_read_b128 v[24:27], v35 offset:96
; GCN-MINREG-NEXT: ds_read_b128 v[20:23], v35 offset:80
; GCN-MINREG-NEXT: ds_read_b128 v[16:19], v35 offset:64
; GCN-MINREG-NEXT: ds_read_b128 v[0:3], v35
; GCN-MINREG-NEXT: ds_read_b128 v[4:7], v35 offset:16
; GCN-MINREG-NEXT: ds_read_b128 v[8:11], v35 offset:32
; GCN-MINREG-NEXT: ds_read_b128 v[12:15], v35 offset:48
; GCN-MINREG-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MINREG-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v33, v32, v[0:31]
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MINREG-NEXT: s_nop 15
; GCN-MINREG-NEXT: s_nop 2
; GCN-MINREG-NEXT: ds_write_b128 v34, v[28:31] offset:112
; GCN-MINREG-NEXT: ds_write_b128 v34, v[24:27] offset:96
; GCN-MINREG-NEXT: ds_write_b128 v34, v[20:23] offset:80
; GCN-MINREG-NEXT: ds_write_b128 v34, v[16:19] offset:64
; GCN-MINREG-NEXT: ds_write_b128 v34, v[12:15] offset:48
; GCN-MINREG-NEXT: ds_write_b128 v34, v[8:11] offset:32
; GCN-MINREG-NEXT: ds_write_b128 v34, v[4:7] offset:16
; GCN-MINREG-NEXT: ds_write_b128 v34, v[0:3]
; GCN-MINREG-NEXT: ds_read_b128 v[28:31], v35 offset:8304
; GCN-MINREG-NEXT: ds_read_b128 v[24:27], v35 offset:8288
; GCN-MINREG-NEXT: ds_read_b128 v[20:23], v35 offset:8272
; GCN-MINREG-NEXT: ds_read_b128 v[16:19], v35 offset:8256
; GCN-MINREG-NEXT: ds_read_b128 v[12:15], v35 offset:8240
; GCN-MINREG-NEXT: ds_read_b128 v[8:11], v35 offset:8224
; GCN-MINREG-NEXT: ds_read_b128 v[4:7], v35 offset:8208
; GCN-MINREG-NEXT: ds_read_b128 v[0:3], v35 offset:8192
; GCN-MINREG-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MINREG-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v33, v32, v[0:31]
; GCN-MINREG-NEXT: v_mov_b32_e32 v34, s1
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MINREG-NEXT: s_nop 15
; GCN-MINREG-NEXT: s_nop 1
; GCN-MINREG-NEXT: ds_write_b128 v34, v[24:27] offset:8288
; GCN-MINREG-NEXT: ds_write_b128 v34, v[28:31] offset:8304
; GCN-MINREG-NEXT: ds_write_b128 v34, v[16:19] offset:8256
; GCN-MINREG-NEXT: ds_write_b128 v34, v[20:23] offset:8272
; GCN-MINREG-NEXT: ds_write_b128 v34, v[8:11] offset:8224
; GCN-MINREG-NEXT: ds_write_b128 v34, v[12:15] offset:8240
; GCN-MINREG-NEXT: ds_write_b128 v34, v[0:3] offset:8192
; GCN-MINREG-NEXT: ds_write_b128 v34, v[4:7] offset:8208
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_barrier mask(0x00000000)
; GCN-MINREG-NEXT: ds_read_b128 v[28:31], v35 offset:24688
; GCN-MINREG-NEXT: ds_read_b128 v[24:27], v35 offset:24672
; GCN-MINREG-NEXT: ds_read_b128 v[20:23], v35 offset:24656
; GCN-MINREG-NEXT: ds_read_b128 v[16:19], v35 offset:24640
; GCN-MINREG-NEXT: ds_read_b128 v[0:3], v35 offset:24576
; GCN-MINREG-NEXT: ds_read_b128 v[4:7], v35 offset:24592
; GCN-MINREG-NEXT: ds_read_b128 v[8:11], v35 offset:24608
; GCN-MINREG-NEXT: ds_read_b128 v[12:15], v35 offset:24624
; GCN-MINREG-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MINREG-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v33, v32, v[0:31]
; GCN-MINREG-NEXT: v_add_u32_e32 v36, 0x6000, v35
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MINREG-NEXT: s_nop 15
; GCN-MINREG-NEXT: s_nop 1
; GCN-MINREG-NEXT: ds_write_b128 v34, v[28:31] offset:16496
; GCN-MINREG-NEXT: ds_write_b128 v34, v[24:27] offset:16480
; GCN-MINREG-NEXT: ds_write_b128 v34, v[20:23] offset:16464
; GCN-MINREG-NEXT: ds_write_b128 v34, v[16:19] offset:16448
; GCN-MINREG-NEXT: ds_write_b128 v34, v[12:15] offset:16432
; GCN-MINREG-NEXT: ds_write_b128 v34, v[8:11] offset:16416
; GCN-MINREG-NEXT: ds_write_b128 v34, v[4:7] offset:16400
; GCN-MINREG-NEXT: ds_write_b128 v34, v[0:3] offset:16384
; GCN-MINREG-NEXT: ds_read_b128 v[28:31], v35 offset:49264
; GCN-MINREG-NEXT: ds_read_b128 v[24:27], v35 offset:49248
; GCN-MINREG-NEXT: ds_read_b128 v[20:23], v35 offset:49232
; GCN-MINREG-NEXT: ds_read_b128 v[16:19], v35 offset:49216
; GCN-MINREG-NEXT: ds_read_b128 v[12:15], v35 offset:49200
; GCN-MINREG-NEXT: ds_read_b128 v[8:11], v35 offset:49184
; GCN-MINREG-NEXT: ds_read_b128 v[4:7], v35 offset:49168
; GCN-MINREG-NEXT: ds_read_b128 v[0:3], v35 offset:49152
; GCN-MINREG-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MINREG-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v33, v32, v[0:31]
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MINREG-NEXT: s_nop 15
; GCN-MINREG-NEXT: s_nop 2
; GCN-MINREG-NEXT: ds_write_b128 v34, v[28:31] offset:24688
; GCN-MINREG-NEXT: ds_write_b128 v34, v[24:27] offset:24672
; GCN-MINREG-NEXT: ds_write_b128 v34, v[20:23] offset:24656
; GCN-MINREG-NEXT: ds_write_b128 v34, v[16:19] offset:24640
; GCN-MINREG-NEXT: ds_write_b128 v34, v[12:15] offset:24624
; GCN-MINREG-NEXT: ds_write_b128 v34, v[8:11] offset:24608
; GCN-MINREG-NEXT: ds_write_b128 v34, v[4:7] offset:24592
; GCN-MINREG-NEXT: ds_write_b128 v34, v[0:3] offset:24576
; GCN-MINREG-NEXT: ds_read_b128 v[28:31], v36 offset:57456
; GCN-MINREG-NEXT: ds_read_b128 v[24:27], v36 offset:57440
; GCN-MINREG-NEXT: ds_read_b128 v[20:23], v36 offset:57424
; GCN-MINREG-NEXT: ds_read_b128 v[16:19], v36 offset:57408
; GCN-MINREG-NEXT: ds_read_b128 v[0:3], v36 offset:57344
; GCN-MINREG-NEXT: ds_read_b128 v[4:7], v36 offset:57360
; GCN-MINREG-NEXT: ds_read_b128 v[8:11], v36 offset:57376
; GCN-MINREG-NEXT: ds_read_b128 v[12:15], v36 offset:57392
; GCN-MINREG-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MINREG-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v33, v32, v[0:31]
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MINREG-NEXT: s_nop 15
; GCN-MINREG-NEXT: s_nop 2
; GCN-MINREG-NEXT: ds_write_b128 v34, v[28:31] offset:32880
; GCN-MINREG-NEXT: ds_write_b128 v34, v[24:27] offset:32864
; GCN-MINREG-NEXT: ds_write_b128 v34, v[20:23] offset:32848
; GCN-MINREG-NEXT: ds_write_b128 v34, v[16:19] offset:32832
; GCN-MINREG-NEXT: ds_write_b128 v34, v[12:15] offset:32816
; GCN-MINREG-NEXT: ds_write_b128 v34, v[8:11] offset:32800
; GCN-MINREG-NEXT: ds_write_b128 v34, v[4:7] offset:32784
; GCN-MINREG-NEXT: ds_write_b128 v34, v[0:3] offset:32768
; GCN-MINREG-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MINREG-NEXT: s_endpgm
;
; GCN-MAXOCC-LABEL: test_sched_group_barrier_pipeline_MFMA_interleave_split_region:
; GCN-MAXOCC: ; %bb.0: ; %entry
; GCN-MAXOCC-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
; GCN-MAXOCC-NEXT: v_lshlrev_b32_e32 v0, 7, v0
; GCN-MAXOCC-NEXT: v_and_b32_e32 v35, 0x1ff80, v0
; GCN-MAXOCC-NEXT: v_mov_b32_e32 v33, 1.0
; GCN-MAXOCC-NEXT: v_mov_b32_e32 v34, 2.0
; GCN-MAXOCC-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MAXOCC-NEXT: v_add_u32_e32 v32, s0, v35
; GCN-MAXOCC-NEXT: ds_read_b128 v[28:31], v32 offset:112
; GCN-MAXOCC-NEXT: ds_read_b128 v[24:27], v32 offset:96
; GCN-MAXOCC-NEXT: ds_read_b128 v[20:23], v32 offset:80
; GCN-MAXOCC-NEXT: ds_read_b128 v[16:19], v32 offset:64
; GCN-MAXOCC-NEXT: ds_read_b128 v[0:3], v32
; GCN-MAXOCC-NEXT: ds_read_b128 v[4:7], v32 offset:16
; GCN-MAXOCC-NEXT: ds_read_b128 v[8:11], v32 offset:32
; GCN-MAXOCC-NEXT: ds_read_b128 v[12:15], v32 offset:48
; GCN-MAXOCC-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MAXOCC-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v33, v34, v[0:31]
; GCN-MAXOCC-NEXT: v_add_u32_e32 v35, s1, v35
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MAXOCC-NEXT: s_nop 15
; GCN-MAXOCC-NEXT: s_nop 1
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[28:31] offset:112
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[24:27] offset:96
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[20:23] offset:80
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[16:19] offset:64
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[12:15] offset:48
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[8:11] offset:32
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[4:7] offset:16
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[0:3]
; GCN-MAXOCC-NEXT: ds_read_b128 v[28:31], v32 offset:8304
; GCN-MAXOCC-NEXT: ds_read_b128 v[24:27], v32 offset:8288
; GCN-MAXOCC-NEXT: ds_read_b128 v[20:23], v32 offset:8272
; GCN-MAXOCC-NEXT: ds_read_b128 v[16:19], v32 offset:8256
; GCN-MAXOCC-NEXT: ds_read_b128 v[12:15], v32 offset:8240
; GCN-MAXOCC-NEXT: ds_read_b128 v[8:11], v32 offset:8224
; GCN-MAXOCC-NEXT: ds_read_b128 v[4:7], v32 offset:8208
; GCN-MAXOCC-NEXT: ds_read_b128 v[0:3], v32 offset:8192
; GCN-MAXOCC-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MAXOCC-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v33, v34, v[0:31]
; GCN-MAXOCC-NEXT: v_mov_b32_e32 v35, s1
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MAXOCC-NEXT: s_nop 15
; GCN-MAXOCC-NEXT: s_nop 1
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[24:27] offset:8288
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[28:31] offset:8304
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[16:19] offset:8256
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[20:23] offset:8272
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[8:11] offset:8224
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[12:15] offset:8240
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[0:3] offset:8192
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[4:7] offset:8208
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_barrier mask(0x00000000)
; GCN-MAXOCC-NEXT: ds_read_b128 v[28:31], v32 offset:24688
; GCN-MAXOCC-NEXT: ds_read_b128 v[24:27], v32 offset:24672
; GCN-MAXOCC-NEXT: ds_read_b128 v[20:23], v32 offset:24656
; GCN-MAXOCC-NEXT: ds_read_b128 v[16:19], v32 offset:24640
; GCN-MAXOCC-NEXT: ds_read_b128 v[0:3], v32 offset:24576
; GCN-MAXOCC-NEXT: ds_read_b128 v[4:7], v32 offset:24592
; GCN-MAXOCC-NEXT: ds_read_b128 v[8:11], v32 offset:24608
; GCN-MAXOCC-NEXT: ds_read_b128 v[12:15], v32 offset:24624
; GCN-MAXOCC-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MAXOCC-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v33, v34, v[0:31]
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MAXOCC-NEXT: s_nop 15
; GCN-MAXOCC-NEXT: s_nop 2
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[28:31] offset:16496
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[24:27] offset:16480
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[20:23] offset:16464
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[16:19] offset:16448
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[12:15] offset:16432
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[8:11] offset:16416
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[4:7] offset:16400
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[0:3] offset:16384
; GCN-MAXOCC-NEXT: ds_read_b128 v[28:31], v32 offset:49264
; GCN-MAXOCC-NEXT: ds_read_b128 v[24:27], v32 offset:49248
; GCN-MAXOCC-NEXT: ds_read_b128 v[20:23], v32 offset:49232
; GCN-MAXOCC-NEXT: ds_read_b128 v[16:19], v32 offset:49216
; GCN-MAXOCC-NEXT: ds_read_b128 v[12:15], v32 offset:49200
; GCN-MAXOCC-NEXT: ds_read_b128 v[8:11], v32 offset:49184
; GCN-MAXOCC-NEXT: ds_read_b128 v[4:7], v32 offset:49168
; GCN-MAXOCC-NEXT: ds_read_b128 v[0:3], v32 offset:49152
; GCN-MAXOCC-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MAXOCC-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v33, v34, v[0:31]
; GCN-MAXOCC-NEXT: v_add_u32_e32 v32, 0x6000, v32
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MAXOCC-NEXT: s_nop 15
; GCN-MAXOCC-NEXT: s_nop 1
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[28:31] offset:24688
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[24:27] offset:24672
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[20:23] offset:24656
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[16:19] offset:24640
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[12:15] offset:24624
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[8:11] offset:24608
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[4:7] offset:24592
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[0:3] offset:24576
; GCN-MAXOCC-NEXT: ds_read_b128 v[28:31], v32 offset:57456
; GCN-MAXOCC-NEXT: ds_read_b128 v[24:27], v32 offset:57440
; GCN-MAXOCC-NEXT: ds_read_b128 v[20:23], v32 offset:57424
; GCN-MAXOCC-NEXT: ds_read_b128 v[16:19], v32 offset:57408
; GCN-MAXOCC-NEXT: ds_read_b128 v[0:3], v32 offset:57344
; GCN-MAXOCC-NEXT: ds_read_b128 v[4:7], v32 offset:57360
; GCN-MAXOCC-NEXT: ds_read_b128 v[8:11], v32 offset:57376
; GCN-MAXOCC-NEXT: ds_read_b128 v[12:15], v32 offset:57392
; GCN-MAXOCC-NEXT: s_waitcnt lgkmcnt(0)
; GCN-MAXOCC-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v33, v34, v[0:31]
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-MAXOCC-NEXT: s_nop 15
; GCN-MAXOCC-NEXT: s_nop 2
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[28:31] offset:32880
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[24:27] offset:32864
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[20:23] offset:32848
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[16:19] offset:32832
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[12:15] offset:32816
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[8:11] offset:32800
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[4:7] offset:32784
; GCN-MAXOCC-NEXT: ds_write_b128 v35, v[0:3] offset:32768
; GCN-MAXOCC-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-MAXOCC-NEXT: s_endpgm
;
; GCN-ILP-LABEL: test_sched_group_barrier_pipeline_MFMA_interleave_split_region:
; GCN-ILP: ; %bb.0: ; %entry
; GCN-ILP-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
; GCN-ILP-NEXT: v_lshlrev_b32_e32 v0, 7, v0
; GCN-ILP-NEXT: v_and_b32_e32 v34, 0x1ff80, v0
; GCN-ILP-NEXT: v_mov_b32_e32 v32, 1.0
; GCN-ILP-NEXT: v_mov_b32_e32 v33, 2.0
; GCN-ILP-NEXT: s_waitcnt lgkmcnt(0)
; GCN-ILP-NEXT: v_add_u32_e32 v35, s0, v34
; GCN-ILP-NEXT: ds_read_b128 v[12:15], v35 offset:48
; GCN-ILP-NEXT: ds_read_b128 v[8:11], v35 offset:32
; GCN-ILP-NEXT: ds_read_b128 v[4:7], v35 offset:16
; GCN-ILP-NEXT: ds_read_b128 v[0:3], v35
; GCN-ILP-NEXT: ds_read_b128 v[16:19], v35 offset:64
; GCN-ILP-NEXT: ds_read_b128 v[20:23], v35 offset:80
; GCN-ILP-NEXT: ds_read_b128 v[24:27], v35 offset:96
; GCN-ILP-NEXT: ds_read_b128 v[28:31], v35 offset:112
; GCN-ILP-NEXT: s_waitcnt lgkmcnt(0)
; GCN-ILP-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v32, v33, v[0:31]
; GCN-ILP-NEXT: v_add_u32_e32 v34, s1, v34
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-ILP-NEXT: s_nop 15
; GCN-ILP-NEXT: s_nop 1
; GCN-ILP-NEXT: ds_write_b128 v34, v[0:3]
; GCN-ILP-NEXT: ds_read_b128 v[0:3], v35 offset:8192
; GCN-ILP-NEXT: ds_write_b128 v34, v[4:7] offset:16
; GCN-ILP-NEXT: ds_read_b128 v[4:7], v35 offset:8208
; GCN-ILP-NEXT: ds_write_b128 v34, v[8:11] offset:32
; GCN-ILP-NEXT: ds_read_b128 v[8:11], v35 offset:8224
; GCN-ILP-NEXT: ds_write_b128 v34, v[12:15] offset:48
; GCN-ILP-NEXT: ds_read_b128 v[12:15], v35 offset:8240
; GCN-ILP-NEXT: ds_write_b128 v34, v[16:19] offset:64
; GCN-ILP-NEXT: ds_read_b128 v[16:19], v35 offset:8256
; GCN-ILP-NEXT: ds_write_b128 v34, v[20:23] offset:80
; GCN-ILP-NEXT: ds_read_b128 v[20:23], v35 offset:8272
; GCN-ILP-NEXT: ds_write_b128 v34, v[24:27] offset:96
; GCN-ILP-NEXT: ds_read_b128 v[24:27], v35 offset:8288
; GCN-ILP-NEXT: ds_write_b128 v34, v[28:31] offset:112
; GCN-ILP-NEXT: ds_read_b128 v[28:31], v35 offset:8304
; GCN-ILP-NEXT: s_waitcnt lgkmcnt(0)
; GCN-ILP-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v32, v33, v[0:31]
; GCN-ILP-NEXT: v_mov_b32_e32 v34, s1
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-ILP-NEXT: s_nop 15
; GCN-ILP-NEXT: s_nop 1
; GCN-ILP-NEXT: ds_write_b128 v34, v[24:27] offset:8288
; GCN-ILP-NEXT: ds_write_b128 v34, v[28:31] offset:8304
; GCN-ILP-NEXT: ds_write_b128 v34, v[16:19] offset:8256
; GCN-ILP-NEXT: ds_write_b128 v34, v[20:23] offset:8272
; GCN-ILP-NEXT: ds_write_b128 v34, v[8:11] offset:8224
; GCN-ILP-NEXT: ds_write_b128 v34, v[12:15] offset:8240
; GCN-ILP-NEXT: ds_write_b128 v34, v[0:3] offset:8192
; GCN-ILP-NEXT: ds_write_b128 v34, v[4:7] offset:8208
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-ILP-NEXT: ; sched_barrier mask(0x00000000)
; GCN-ILP-NEXT: ds_read_b128 v[12:15], v35 offset:24624
; GCN-ILP-NEXT: ds_read_b128 v[8:11], v35 offset:24608
; GCN-ILP-NEXT: ds_read_b128 v[4:7], v35 offset:24592
; GCN-ILP-NEXT: ds_read_b128 v[0:3], v35 offset:24576
; GCN-ILP-NEXT: ds_read_b128 v[16:19], v35 offset:24640
; GCN-ILP-NEXT: ds_read_b128 v[20:23], v35 offset:24656
; GCN-ILP-NEXT: ds_read_b128 v[24:27], v35 offset:24672
; GCN-ILP-NEXT: ds_read_b128 v[28:31], v35 offset:24688
; GCN-ILP-NEXT: s_waitcnt lgkmcnt(0)
; GCN-ILP-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v32, v33, v[0:31]
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-ILP-NEXT: s_nop 15
; GCN-ILP-NEXT: s_nop 2
; GCN-ILP-NEXT: ds_write_b128 v34, v[28:31] offset:16496
; GCN-ILP-NEXT: ds_write_b128 v34, v[24:27] offset:16480
; GCN-ILP-NEXT: ds_write_b128 v34, v[20:23] offset:16464
; GCN-ILP-NEXT: ds_write_b128 v34, v[16:19] offset:16448
; GCN-ILP-NEXT: ds_write_b128 v34, v[12:15] offset:16432
; GCN-ILP-NEXT: ds_write_b128 v34, v[8:11] offset:16416
; GCN-ILP-NEXT: ds_write_b128 v34, v[4:7] offset:16400
; GCN-ILP-NEXT: ds_write_b128 v34, v[0:3] offset:16384
; GCN-ILP-NEXT: ds_read_b128 v[0:3], v35 offset:49152
; GCN-ILP-NEXT: ds_read_b128 v[4:7], v35 offset:49168
; GCN-ILP-NEXT: ds_read_b128 v[8:11], v35 offset:49184
; GCN-ILP-NEXT: ds_read_b128 v[12:15], v35 offset:49200
; GCN-ILP-NEXT: ds_read_b128 v[16:19], v35 offset:49216
; GCN-ILP-NEXT: ds_read_b128 v[20:23], v35 offset:49232
; GCN-ILP-NEXT: ds_read_b128 v[24:27], v35 offset:49248
; GCN-ILP-NEXT: ds_read_b128 v[28:31], v35 offset:49264
; GCN-ILP-NEXT: s_waitcnt lgkmcnt(0)
; GCN-ILP-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v32, v33, v[0:31]
; GCN-ILP-NEXT: v_add_u32_e32 v35, 0x6000, v35
; GCN-ILP-NEXT: s_nop 15
; GCN-ILP-NEXT: s_nop 1
; GCN-ILP-NEXT: ds_write_b128 v34, v[0:3] offset:24576
; GCN-ILP-NEXT: ds_read_b128 v[0:3], v35 offset:57344
; GCN-ILP-NEXT: ds_write_b128 v34, v[4:7] offset:24592
; GCN-ILP-NEXT: ds_read_b128 v[4:7], v35 offset:57360
; GCN-ILP-NEXT: ds_write_b128 v34, v[8:11] offset:24608
; GCN-ILP-NEXT: ds_read_b128 v[8:11], v35 offset:57376
; GCN-ILP-NEXT: ds_write_b128 v34, v[12:15] offset:24624
; GCN-ILP-NEXT: ds_read_b128 v[12:15], v35 offset:57392
; GCN-ILP-NEXT: ds_write_b128 v34, v[16:19] offset:24640
; GCN-ILP-NEXT: ds_read_b128 v[16:19], v35 offset:57408
; GCN-ILP-NEXT: ds_write_b128 v34, v[20:23] offset:24656
; GCN-ILP-NEXT: ds_read_b128 v[20:23], v35 offset:57424
; GCN-ILP-NEXT: ds_write_b128 v34, v[24:27] offset:24672
; GCN-ILP-NEXT: ds_read_b128 v[24:27], v35 offset:57440
; GCN-ILP-NEXT: ds_write_b128 v34, v[28:31] offset:24688
; GCN-ILP-NEXT: ds_read_b128 v[28:31], v35 offset:57456
; GCN-ILP-NEXT: s_waitcnt lgkmcnt(0)
; GCN-ILP-NEXT: v_mfma_f32_32x32x1f32 v[0:31], v32, v33, v[0:31]
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000100) size(8) SyncID(0)
; GCN-ILP-NEXT: s_nop 15
; GCN-ILP-NEXT: s_nop 2
; GCN-ILP-NEXT: ds_write_b128 v34, v[28:31] offset:32880
; GCN-ILP-NEXT: ds_write_b128 v34, v[24:27] offset:32864
; GCN-ILP-NEXT: ds_write_b128 v34, v[20:23] offset:32848
; GCN-ILP-NEXT: ds_write_b128 v34, v[16:19] offset:32832
; GCN-ILP-NEXT: ds_write_b128 v34, v[12:15] offset:32816
; GCN-ILP-NEXT: ds_write_b128 v34, v[8:11] offset:32800
; GCN-ILP-NEXT: ds_write_b128 v34, v[4:7] offset:32784
; GCN-ILP-NEXT: ds_write_b128 v34, v[0:3] offset:32768
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-ILP-NEXT: ; sched_group_barrier mask(0x00000200) size(8) SyncID(0)
; GCN-ILP-NEXT: s_endpgm
entry:
%idx = call i32 @llvm.amdgcn.workitem.id.x()
%load.0.addr = getelementptr <32 x float>, ptr addrspace(3) %in, i32 %idx
%load.0 = load <32 x float>, ptr addrspace(3) %load.0.addr
%load.1.addr = getelementptr <32 x float>, ptr addrspace(3) %load.0.addr, i32 64
%load.1 = load <32 x float>, ptr addrspace(3) %load.1.addr
%mai.0 = tail call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float 1.0, float 2.0, <32 x float> %load.0, i32 0, i32 0, i32 0)
%mai.1 = tail call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float 1.0, float 2.0, <32 x float> %load.1, i32 0, i32 0, i32 0)
%store.0.addr = getelementptr <32 x float>, ptr addrspace(3) %out, i32 %idx
store <32 x float> %mai.0, ptr addrspace(3) %store.0.addr
%store.1.addr = getelementptr <32 x float>, ptr addrspace(3) %out, i32 64
store <32 x float> %mai.1, ptr addrspace(3) %store.1.addr
; 8 DS read
call void @llvm.amdgcn.sched.group.barrier(i32 256, i32 8, i32 0)
; 1 MFMA
call void @llvm.amdgcn.sched.group.barrier(i32 8, i32 1, i32 0)
; 8 DS write
call void @llvm.amdgcn.sched.group.barrier(i32 512, i32 8, i32 0)
; 8 DS read
call void @llvm.amdgcn.sched.group.barrier(i32 256, i32 8, i32 0)
; 1 MFMA
call void @llvm.amdgcn.sched.group.barrier(i32 8, i32 1, i32 0)
; 8 DS write
call void @llvm.amdgcn.sched.group.barrier(i32 512, i32 8, i32 0)
; parition the regions
call void @llvm.amdgcn.sched.barrier(i32 0)
%load.2.addr = getelementptr <32 x float>, ptr addrspace(3) %load.1.addr, i32 128
%load.2 = load <32 x float>, ptr addrspace(3) %load.2.addr
%load.3.addr = getelementptr <32 x float>, ptr addrspace(3) %load.2.addr, i32 192
%load.3 = load <32 x float>, ptr addrspace(3) %load.3.addr
%load.4.addr = getelementptr <32 x float>, ptr addrspace(3) %load.3.addr, i32 256
%load.4 = load <32 x float>, ptr addrspace(3) %load.4.addr
%mai.2 = tail call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float 1.0, float 2.0, <32 x float> %load.2, i32 0, i32 0, i32 0)
%mai.3 = tail call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float 1.0, float 2.0, <32 x float> %load.3, i32 0, i32 0, i32 0)
%mai.4 = tail call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float 1.0, float 2.0, <32 x float> %load.4, i32 0, i32 0, i32 0)
%store.2.addr = getelementptr <32 x float>, ptr addrspace(3) %out, i32 128
store <32 x float> %mai.2, ptr addrspace(3) %store.2.addr
%store.3.addr = getelementptr <32 x float>, ptr addrspace(3) %out, i32 192
store <32 x float> %mai.3, ptr addrspace(3) %store.3.addr
%store.4.addr = getelementptr <32 x float>, ptr addrspace(3) %out, i32 256
store <32 x float> %mai.4, ptr addrspace(3) %store.4.addr
; 8 DS read
call void @llvm.amdgcn.sched.group.barrier(i32 256, i32 8, i32 0)
; 1 MFMA
call void @llvm.amdgcn.sched.group.barrier(i32 8, i32 1, i32 0)
; 8 DS write
call void @llvm.amdgcn.sched.group.barrier(i32 512, i32 8, i32 0)
; 8 DS read
call void @llvm.amdgcn.sched.group.barrier(i32 256, i32 8, i32 0)
; 1 MFMA
call void @llvm.amdgcn.sched.group.barrier(i32 8, i32 1, i32 0)
; 8 DS write
call void @llvm.amdgcn.sched.group.barrier(i32 512, i32 8, i32 0)
; 8 DS read
call void @llvm.amdgcn.sched.group.barrier(i32 256, i32 8, i32 0)
; 1 MFMA
call void @llvm.amdgcn.sched.group.barrier(i32 8, i32 1, i32 0)
; 8 DS write
call void @llvm.amdgcn.sched.group.barrier(i32 512, i32 8, i32 0)
ret void
}