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
535 lines
22 KiB
LLVM
535 lines
22 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc -global-isel -new-reg-bank-select -amdgpu-codegenprepare-disable-idiv-expansion=1 -mtriple=amdgcn-amd-amdpal < %s | FileCheck -check-prefixes=CHECK,GISEL %s
|
|
; RUN: llc -global-isel -new-reg-bank-select -amdgpu-codegenprepare-disable-idiv-expansion=0 -mtriple=amdgcn-amd-amdpal < %s | FileCheck -check-prefixes=CHECK,CGP %s
|
|
|
|
; The same 32-bit expansion is implemented in the legalizer and in AMDGPUCodeGenPrepare.
|
|
|
|
define i32 @v_udiv_i32(i32 %num, i32 %den) {
|
|
; GISEL-LABEL: v_udiv_i32:
|
|
; GISEL: ; %bb.0:
|
|
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GISEL-NEXT: v_cvt_f32_u32_e32 v2, v1
|
|
; GISEL-NEXT: v_sub_i32_e32 v3, vcc, 0, v1
|
|
; GISEL-NEXT: v_rcp_iflag_f32_e32 v2, v2
|
|
; GISEL-NEXT: v_mul_f32_e32 v2, 0x4f7ffffe, v2
|
|
; GISEL-NEXT: v_cvt_u32_f32_e32 v2, v2
|
|
; GISEL-NEXT: v_mul_lo_u32 v3, v3, v2
|
|
; GISEL-NEXT: v_mul_hi_u32 v3, v2, v3
|
|
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v3
|
|
; GISEL-NEXT: v_mul_hi_u32 v2, v0, v2
|
|
; GISEL-NEXT: v_mul_lo_u32 v3, v2, v1
|
|
; GISEL-NEXT: v_add_i32_e32 v4, vcc, 1, v2
|
|
; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v0, v3
|
|
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v1
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc
|
|
; GISEL-NEXT: v_sub_i32_e64 v3, s[4:5], v0, v1
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v0, v0, v3, vcc
|
|
; GISEL-NEXT: v_add_i32_e32 v3, vcc, 1, v2
|
|
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v1
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc
|
|
; GISEL-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CGP-LABEL: v_udiv_i32:
|
|
; CGP: ; %bb.0:
|
|
; CGP-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CGP-NEXT: v_cvt_f32_u32_e32 v2, v1
|
|
; CGP-NEXT: v_sub_i32_e32 v3, vcc, 0, v1
|
|
; CGP-NEXT: v_rcp_f32_e32 v2, v2
|
|
; CGP-NEXT: v_mul_f32_e32 v2, 0x4f7ffffe, v2
|
|
; CGP-NEXT: v_cvt_u32_f32_e32 v2, v2
|
|
; CGP-NEXT: v_mul_lo_u32 v3, v3, v2
|
|
; CGP-NEXT: v_mul_hi_u32 v3, v2, v3
|
|
; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v3
|
|
; CGP-NEXT: v_mul_hi_u32 v2, v0, v2
|
|
; CGP-NEXT: v_mul_lo_u32 v3, v2, v1
|
|
; CGP-NEXT: v_add_i32_e32 v4, vcc, 1, v2
|
|
; CGP-NEXT: v_sub_i32_e32 v0, vcc, v0, v3
|
|
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v0, v1
|
|
; CGP-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc
|
|
; CGP-NEXT: v_sub_i32_e64 v3, s[4:5], v0, v1
|
|
; CGP-NEXT: v_cndmask_b32_e32 v0, v0, v3, vcc
|
|
; CGP-NEXT: v_add_i32_e32 v3, vcc, 1, v2
|
|
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v0, v1
|
|
; CGP-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc
|
|
; CGP-NEXT: s_setpc_b64 s[30:31]
|
|
%result = udiv i32 %num, %den
|
|
ret i32 %result
|
|
}
|
|
|
|
; FIXME: This is a workaround for not handling uniform VGPR case.
|
|
declare i32 @llvm.amdgcn.readfirstlane(i32)
|
|
|
|
define amdgpu_ps i32 @s_udiv_i32(i32 inreg %num, i32 inreg %den) {
|
|
; GISEL-LABEL: s_udiv_i32:
|
|
; GISEL: ; %bb.0:
|
|
; GISEL-NEXT: v_cvt_f32_u32_e32 v0, s1
|
|
; GISEL-NEXT: s_sub_i32 s2, 0, s1
|
|
; GISEL-NEXT: v_rcp_iflag_f32_e32 v0, v0
|
|
; GISEL-NEXT: v_mul_f32_e32 v0, 0x4f7ffffe, v0
|
|
; GISEL-NEXT: v_cvt_u32_f32_e32 v0, v0
|
|
; GISEL-NEXT: v_readfirstlane_b32 s3, v0
|
|
; GISEL-NEXT: s_mul_i32 s2, s2, s3
|
|
; GISEL-NEXT: v_mul_hi_u32 v0, v0, s2
|
|
; GISEL-NEXT: v_readfirstlane_b32 s2, v0
|
|
; GISEL-NEXT: s_add_i32 s2, s3, s2
|
|
; GISEL-NEXT: v_mov_b32_e32 v0, s2
|
|
; GISEL-NEXT: v_mul_hi_u32 v0, s0, v0
|
|
; GISEL-NEXT: v_readfirstlane_b32 s2, v0
|
|
; GISEL-NEXT: s_mul_i32 s3, s2, s1
|
|
; GISEL-NEXT: s_sub_i32 s0, s0, s3
|
|
; GISEL-NEXT: s_cmp_ge_u32 s0, s1
|
|
; GISEL-NEXT: s_cselect_b32 s3, 1, 0
|
|
; GISEL-NEXT: s_add_i32 s4, s2, 1
|
|
; GISEL-NEXT: s_cmp_lg_u32 s3, 0
|
|
; GISEL-NEXT: s_cselect_b32 s2, s4, s2
|
|
; GISEL-NEXT: s_sub_i32 s4, s0, s1
|
|
; GISEL-NEXT: s_cmp_lg_u32 s3, 0
|
|
; GISEL-NEXT: s_cselect_b32 s0, s4, s0
|
|
; GISEL-NEXT: s_cmp_ge_u32 s0, s1
|
|
; GISEL-NEXT: s_cselect_b32 s0, 1, 0
|
|
; GISEL-NEXT: s_add_i32 s1, s2, 1
|
|
; GISEL-NEXT: s_cmp_lg_u32 s0, 0
|
|
; GISEL-NEXT: s_cselect_b32 s0, s1, s2
|
|
; GISEL-NEXT: ; return to shader part epilog
|
|
;
|
|
; CGP-LABEL: s_udiv_i32:
|
|
; CGP: ; %bb.0:
|
|
; CGP-NEXT: v_cvt_f32_u32_e32 v0, s1
|
|
; CGP-NEXT: s_sub_i32 s2, 0, s1
|
|
; CGP-NEXT: v_rcp_f32_e32 v0, v0
|
|
; CGP-NEXT: v_mul_f32_e32 v0, 0x4f7ffffe, v0
|
|
; CGP-NEXT: v_cvt_u32_f32_e32 v0, v0
|
|
; CGP-NEXT: v_readfirstlane_b32 s3, v0
|
|
; CGP-NEXT: s_mul_i32 s2, s2, s3
|
|
; CGP-NEXT: v_mul_hi_u32 v0, v0, s2
|
|
; CGP-NEXT: v_readfirstlane_b32 s2, v0
|
|
; CGP-NEXT: s_add_i32 s2, s3, s2
|
|
; CGP-NEXT: v_mov_b32_e32 v0, s2
|
|
; CGP-NEXT: v_mul_hi_u32 v0, s0, v0
|
|
; CGP-NEXT: v_readfirstlane_b32 s2, v0
|
|
; CGP-NEXT: s_mul_i32 s3, s2, s1
|
|
; CGP-NEXT: s_sub_i32 s0, s0, s3
|
|
; CGP-NEXT: s_cmp_ge_u32 s0, s1
|
|
; CGP-NEXT: s_cselect_b32 s3, 1, 0
|
|
; CGP-NEXT: s_add_i32 s4, s2, 1
|
|
; CGP-NEXT: s_cmp_lg_u32 s3, 0
|
|
; CGP-NEXT: s_cselect_b32 s2, s4, s2
|
|
; CGP-NEXT: s_sub_i32 s4, s0, s1
|
|
; CGP-NEXT: s_cmp_lg_u32 s3, 0
|
|
; CGP-NEXT: s_cselect_b32 s0, s4, s0
|
|
; CGP-NEXT: s_cmp_ge_u32 s0, s1
|
|
; CGP-NEXT: s_cselect_b32 s0, 1, 0
|
|
; CGP-NEXT: s_add_i32 s1, s2, 1
|
|
; CGP-NEXT: s_cmp_lg_u32 s0, 0
|
|
; CGP-NEXT: s_cselect_b32 s0, s1, s2
|
|
; CGP-NEXT: ; return to shader part epilog
|
|
%result = udiv i32 %num, %den
|
|
%readlane = call i32 @llvm.amdgcn.readfirstlane(i32 %result)
|
|
ret i32 %readlane
|
|
}
|
|
|
|
define <2 x i32> @v_udiv_v2i32(<2 x i32> %num, <2 x i32> %den) {
|
|
; GISEL-LABEL: v_udiv_v2i32:
|
|
; GISEL: ; %bb.0:
|
|
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GISEL-NEXT: v_cvt_f32_u32_e32 v4, v2
|
|
; GISEL-NEXT: v_sub_i32_e32 v5, vcc, 0, v2
|
|
; GISEL-NEXT: v_cvt_f32_u32_e32 v6, v3
|
|
; GISEL-NEXT: v_sub_i32_e32 v7, vcc, 0, v3
|
|
; GISEL-NEXT: v_rcp_iflag_f32_e32 v4, v4
|
|
; GISEL-NEXT: v_rcp_iflag_f32_e32 v6, v6
|
|
; GISEL-NEXT: v_mul_f32_e32 v4, 0x4f7ffffe, v4
|
|
; GISEL-NEXT: v_mul_f32_e32 v6, 0x4f7ffffe, v6
|
|
; GISEL-NEXT: v_cvt_u32_f32_e32 v4, v4
|
|
; GISEL-NEXT: v_cvt_u32_f32_e32 v6, v6
|
|
; GISEL-NEXT: v_mul_lo_u32 v5, v5, v4
|
|
; GISEL-NEXT: v_mul_lo_u32 v7, v7, v6
|
|
; GISEL-NEXT: v_mul_hi_u32 v5, v4, v5
|
|
; GISEL-NEXT: v_mul_hi_u32 v7, v6, v7
|
|
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5
|
|
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v6, v7
|
|
; GISEL-NEXT: v_mul_hi_u32 v4, v0, v4
|
|
; GISEL-NEXT: v_mul_hi_u32 v5, v1, v5
|
|
; GISEL-NEXT: v_mul_lo_u32 v6, v4, v2
|
|
; GISEL-NEXT: v_add_i32_e32 v7, vcc, 1, v4
|
|
; GISEL-NEXT: v_mul_lo_u32 v8, v5, v3
|
|
; GISEL-NEXT: v_add_i32_e32 v9, vcc, 1, v5
|
|
; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v0, v6
|
|
; GISEL-NEXT: v_sub_i32_e32 v1, vcc, v1, v8
|
|
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v2
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v4, v4, v7, vcc
|
|
; GISEL-NEXT: v_sub_i32_e64 v6, s[4:5], v0, v2
|
|
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v1, v3
|
|
; GISEL-NEXT: v_cndmask_b32_e64 v5, v5, v9, s[4:5]
|
|
; GISEL-NEXT: v_sub_i32_e64 v7, s[6:7], v1, v3
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v0, v0, v6, vcc
|
|
; GISEL-NEXT: v_add_i32_e32 v6, vcc, 1, v4
|
|
; GISEL-NEXT: v_cndmask_b32_e64 v1, v1, v7, s[4:5]
|
|
; GISEL-NEXT: v_add_i32_e32 v7, vcc, 1, v5
|
|
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v2
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v0, v4, v6, vcc
|
|
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v1, v3
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v1, v5, v7, vcc
|
|
; GISEL-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CGP-LABEL: v_udiv_v2i32:
|
|
; CGP: ; %bb.0:
|
|
; CGP-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CGP-NEXT: v_cvt_f32_u32_e32 v4, v2
|
|
; CGP-NEXT: v_sub_i32_e32 v5, vcc, 0, v2
|
|
; CGP-NEXT: v_cvt_f32_u32_e32 v6, v3
|
|
; CGP-NEXT: v_sub_i32_e32 v7, vcc, 0, v3
|
|
; CGP-NEXT: v_rcp_f32_e32 v4, v4
|
|
; CGP-NEXT: v_rcp_f32_e32 v6, v6
|
|
; CGP-NEXT: v_mul_f32_e32 v4, 0x4f7ffffe, v4
|
|
; CGP-NEXT: v_mul_f32_e32 v6, 0x4f7ffffe, v6
|
|
; CGP-NEXT: v_cvt_u32_f32_e32 v4, v4
|
|
; CGP-NEXT: v_cvt_u32_f32_e32 v6, v6
|
|
; CGP-NEXT: v_mul_lo_u32 v5, v5, v4
|
|
; CGP-NEXT: v_mul_lo_u32 v7, v7, v6
|
|
; CGP-NEXT: v_mul_hi_u32 v5, v4, v5
|
|
; CGP-NEXT: v_mul_hi_u32 v7, v6, v7
|
|
; CGP-NEXT: v_add_i32_e32 v4, vcc, v4, v5
|
|
; CGP-NEXT: v_add_i32_e32 v5, vcc, v6, v7
|
|
; CGP-NEXT: v_mul_hi_u32 v4, v0, v4
|
|
; CGP-NEXT: v_mul_hi_u32 v5, v1, v5
|
|
; CGP-NEXT: v_mul_lo_u32 v6, v4, v2
|
|
; CGP-NEXT: v_add_i32_e32 v7, vcc, 1, v4
|
|
; CGP-NEXT: v_mul_lo_u32 v8, v5, v3
|
|
; CGP-NEXT: v_add_i32_e32 v9, vcc, 1, v5
|
|
; CGP-NEXT: v_sub_i32_e32 v0, vcc, v0, v6
|
|
; CGP-NEXT: v_sub_i32_e32 v1, vcc, v1, v8
|
|
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v0, v2
|
|
; CGP-NEXT: v_cndmask_b32_e32 v4, v4, v7, vcc
|
|
; CGP-NEXT: v_sub_i32_e64 v6, s[4:5], v0, v2
|
|
; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v1, v3
|
|
; CGP-NEXT: v_cndmask_b32_e64 v5, v5, v9, s[4:5]
|
|
; CGP-NEXT: v_sub_i32_e64 v7, s[6:7], v1, v3
|
|
; CGP-NEXT: v_cndmask_b32_e32 v0, v0, v6, vcc
|
|
; CGP-NEXT: v_add_i32_e32 v6, vcc, 1, v4
|
|
; CGP-NEXT: v_cndmask_b32_e64 v1, v1, v7, s[4:5]
|
|
; CGP-NEXT: v_add_i32_e32 v7, vcc, 1, v5
|
|
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v0, v2
|
|
; CGP-NEXT: v_cndmask_b32_e32 v0, v4, v6, vcc
|
|
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v1, v3
|
|
; CGP-NEXT: v_cndmask_b32_e32 v1, v5, v7, vcc
|
|
; CGP-NEXT: s_setpc_b64 s[30:31]
|
|
%result = udiv <2 x i32> %num, %den
|
|
ret <2 x i32> %result
|
|
}
|
|
|
|
define i32 @v_udiv_i32_pow2k_denom(i32 %num) {
|
|
; CHECK-LABEL: v_udiv_i32_pow2k_denom:
|
|
; CHECK: ; %bb.0:
|
|
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CHECK-NEXT: v_lshrrev_b32_e32 v0, 12, v0
|
|
; CHECK-NEXT: s_setpc_b64 s[30:31]
|
|
%result = udiv i32 %num, 4096
|
|
ret i32 %result
|
|
}
|
|
|
|
define <2 x i32> @v_udiv_v2i32_pow2k_denom(<2 x i32> %num) {
|
|
; CHECK-LABEL: v_udiv_v2i32_pow2k_denom:
|
|
; CHECK: ; %bb.0:
|
|
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CHECK-NEXT: v_lshrrev_b32_e32 v0, 12, v0
|
|
; CHECK-NEXT: v_lshrrev_b32_e32 v1, 12, v1
|
|
; CHECK-NEXT: s_setpc_b64 s[30:31]
|
|
%result = udiv <2 x i32> %num, <i32 4096, i32 4096>
|
|
ret <2 x i32> %result
|
|
}
|
|
|
|
define i32 @v_udiv_i32_oddk_denom(i32 %num) {
|
|
; CHECK-LABEL: v_udiv_i32_oddk_denom:
|
|
; CHECK: ; %bb.0:
|
|
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CHECK-NEXT: v_mov_b32_e32 v1, 0xb2a50881
|
|
; CHECK-NEXT: v_mul_hi_u32 v1, v0, v1
|
|
; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
|
|
; CHECK-NEXT: v_lshrrev_b32_e32 v0, 1, v0
|
|
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v1
|
|
; CHECK-NEXT: v_lshrrev_b32_e32 v0, 20, v0
|
|
; CHECK-NEXT: s_setpc_b64 s[30:31]
|
|
%result = udiv i32 %num, 1235195
|
|
ret i32 %result
|
|
}
|
|
|
|
define <2 x i32> @v_udiv_v2i32_oddk_denom(<2 x i32> %num) {
|
|
; CHECK-LABEL: v_udiv_v2i32_oddk_denom:
|
|
; CHECK: ; %bb.0:
|
|
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CHECK-NEXT: v_mov_b32_e32 v2, 0xb2a50881
|
|
; CHECK-NEXT: v_mul_hi_u32 v3, v0, v2
|
|
; CHECK-NEXT: v_mul_hi_u32 v2, v1, v2
|
|
; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v0, v3
|
|
; CHECK-NEXT: v_sub_i32_e32 v1, vcc, v1, v2
|
|
; CHECK-NEXT: v_lshrrev_b32_e32 v0, 1, v0
|
|
; CHECK-NEXT: v_lshrrev_b32_e32 v1, 1, v1
|
|
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v3
|
|
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v2
|
|
; CHECK-NEXT: v_lshrrev_b32_e32 v0, 20, v0
|
|
; CHECK-NEXT: v_lshrrev_b32_e32 v1, 20, v1
|
|
; CHECK-NEXT: s_setpc_b64 s[30:31]
|
|
%result = udiv <2 x i32> %num, <i32 1235195, i32 1235195>
|
|
ret <2 x i32> %result
|
|
}
|
|
|
|
define i32 @v_udiv_i32_pow2_shl_denom(i32 %x, i32 %y) {
|
|
; CHECK-LABEL: v_udiv_i32_pow2_shl_denom:
|
|
; CHECK: ; %bb.0:
|
|
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CHECK-NEXT: v_lshl_b32_e32 v1, 0x1000, v1
|
|
; CHECK-NEXT: v_cvt_f32_u32_e32 v2, v1
|
|
; CHECK-NEXT: v_sub_i32_e32 v3, vcc, 0, v1
|
|
; CHECK-NEXT: v_rcp_iflag_f32_e32 v2, v2
|
|
; CHECK-NEXT: v_mul_f32_e32 v2, 0x4f7ffffe, v2
|
|
; CHECK-NEXT: v_cvt_u32_f32_e32 v2, v2
|
|
; CHECK-NEXT: v_mul_lo_u32 v3, v3, v2
|
|
; CHECK-NEXT: v_mul_hi_u32 v3, v2, v3
|
|
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v3
|
|
; CHECK-NEXT: v_mul_hi_u32 v2, v0, v2
|
|
; CHECK-NEXT: v_mul_lo_u32 v3, v2, v1
|
|
; CHECK-NEXT: v_add_i32_e32 v4, vcc, 1, v2
|
|
; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v0, v3
|
|
; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v0, v1
|
|
; CHECK-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc
|
|
; CHECK-NEXT: v_sub_i32_e64 v3, s[4:5], v0, v1
|
|
; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v3, vcc
|
|
; CHECK-NEXT: v_add_i32_e32 v3, vcc, 1, v2
|
|
; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v0, v1
|
|
; CHECK-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc
|
|
; CHECK-NEXT: s_setpc_b64 s[30:31]
|
|
%shl.y = shl i32 4096, %y
|
|
%r = udiv i32 %x, %shl.y
|
|
ret i32 %r
|
|
}
|
|
|
|
define <2 x i32> @v_udiv_v2i32_pow2_shl_denom(<2 x i32> %x, <2 x i32> %y) {
|
|
; GISEL-LABEL: v_udiv_v2i32_pow2_shl_denom:
|
|
; GISEL: ; %bb.0:
|
|
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GISEL-NEXT: v_lshl_b32_e32 v2, 0x1000, v2
|
|
; GISEL-NEXT: v_lshl_b32_e32 v3, 0x1000, v3
|
|
; GISEL-NEXT: v_cvt_f32_u32_e32 v4, v2
|
|
; GISEL-NEXT: v_sub_i32_e32 v5, vcc, 0, v2
|
|
; GISEL-NEXT: v_cvt_f32_u32_e32 v6, v3
|
|
; GISEL-NEXT: v_sub_i32_e32 v7, vcc, 0, v3
|
|
; GISEL-NEXT: v_rcp_iflag_f32_e32 v4, v4
|
|
; GISEL-NEXT: v_rcp_iflag_f32_e32 v6, v6
|
|
; GISEL-NEXT: v_mul_f32_e32 v4, 0x4f7ffffe, v4
|
|
; GISEL-NEXT: v_mul_f32_e32 v6, 0x4f7ffffe, v6
|
|
; GISEL-NEXT: v_cvt_u32_f32_e32 v4, v4
|
|
; GISEL-NEXT: v_cvt_u32_f32_e32 v6, v6
|
|
; GISEL-NEXT: v_mul_lo_u32 v5, v5, v4
|
|
; GISEL-NEXT: v_mul_lo_u32 v7, v7, v6
|
|
; GISEL-NEXT: v_mul_hi_u32 v5, v4, v5
|
|
; GISEL-NEXT: v_mul_hi_u32 v7, v6, v7
|
|
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5
|
|
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v6, v7
|
|
; GISEL-NEXT: v_mul_hi_u32 v4, v0, v4
|
|
; GISEL-NEXT: v_mul_hi_u32 v5, v1, v5
|
|
; GISEL-NEXT: v_mul_lo_u32 v6, v4, v2
|
|
; GISEL-NEXT: v_add_i32_e32 v7, vcc, 1, v4
|
|
; GISEL-NEXT: v_mul_lo_u32 v8, v5, v3
|
|
; GISEL-NEXT: v_add_i32_e32 v9, vcc, 1, v5
|
|
; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v0, v6
|
|
; GISEL-NEXT: v_sub_i32_e32 v1, vcc, v1, v8
|
|
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v2
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v4, v4, v7, vcc
|
|
; GISEL-NEXT: v_sub_i32_e64 v6, s[4:5], v0, v2
|
|
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v1, v3
|
|
; GISEL-NEXT: v_cndmask_b32_e64 v5, v5, v9, s[4:5]
|
|
; GISEL-NEXT: v_sub_i32_e64 v7, s[6:7], v1, v3
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v0, v0, v6, vcc
|
|
; GISEL-NEXT: v_add_i32_e32 v6, vcc, 1, v4
|
|
; GISEL-NEXT: v_cndmask_b32_e64 v1, v1, v7, s[4:5]
|
|
; GISEL-NEXT: v_add_i32_e32 v7, vcc, 1, v5
|
|
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v2
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v0, v4, v6, vcc
|
|
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v1, v3
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v1, v5, v7, vcc
|
|
; GISEL-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CGP-LABEL: v_udiv_v2i32_pow2_shl_denom:
|
|
; CGP: ; %bb.0:
|
|
; CGP-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CGP-NEXT: v_lshl_b32_e32 v2, 0x1000, v2
|
|
; CGP-NEXT: v_lshl_b32_e32 v3, 0x1000, v3
|
|
; CGP-NEXT: v_cvt_f32_u32_e32 v4, v2
|
|
; CGP-NEXT: v_sub_i32_e32 v5, vcc, 0, v2
|
|
; CGP-NEXT: v_cvt_f32_u32_e32 v6, v3
|
|
; CGP-NEXT: v_sub_i32_e32 v7, vcc, 0, v3
|
|
; CGP-NEXT: v_rcp_f32_e32 v4, v4
|
|
; CGP-NEXT: v_rcp_f32_e32 v6, v6
|
|
; CGP-NEXT: v_mul_f32_e32 v4, 0x4f7ffffe, v4
|
|
; CGP-NEXT: v_mul_f32_e32 v6, 0x4f7ffffe, v6
|
|
; CGP-NEXT: v_cvt_u32_f32_e32 v4, v4
|
|
; CGP-NEXT: v_cvt_u32_f32_e32 v6, v6
|
|
; CGP-NEXT: v_mul_lo_u32 v5, v5, v4
|
|
; CGP-NEXT: v_mul_lo_u32 v7, v7, v6
|
|
; CGP-NEXT: v_mul_hi_u32 v5, v4, v5
|
|
; CGP-NEXT: v_mul_hi_u32 v7, v6, v7
|
|
; CGP-NEXT: v_add_i32_e32 v4, vcc, v4, v5
|
|
; CGP-NEXT: v_add_i32_e32 v5, vcc, v6, v7
|
|
; CGP-NEXT: v_mul_hi_u32 v4, v0, v4
|
|
; CGP-NEXT: v_mul_hi_u32 v5, v1, v5
|
|
; CGP-NEXT: v_mul_lo_u32 v6, v4, v2
|
|
; CGP-NEXT: v_add_i32_e32 v7, vcc, 1, v4
|
|
; CGP-NEXT: v_mul_lo_u32 v8, v5, v3
|
|
; CGP-NEXT: v_add_i32_e32 v9, vcc, 1, v5
|
|
; CGP-NEXT: v_sub_i32_e32 v0, vcc, v0, v6
|
|
; CGP-NEXT: v_sub_i32_e32 v1, vcc, v1, v8
|
|
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v0, v2
|
|
; CGP-NEXT: v_cndmask_b32_e32 v4, v4, v7, vcc
|
|
; CGP-NEXT: v_sub_i32_e64 v6, s[4:5], v0, v2
|
|
; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v1, v3
|
|
; CGP-NEXT: v_cndmask_b32_e64 v5, v5, v9, s[4:5]
|
|
; CGP-NEXT: v_sub_i32_e64 v7, s[6:7], v1, v3
|
|
; CGP-NEXT: v_cndmask_b32_e32 v0, v0, v6, vcc
|
|
; CGP-NEXT: v_add_i32_e32 v6, vcc, 1, v4
|
|
; CGP-NEXT: v_cndmask_b32_e64 v1, v1, v7, s[4:5]
|
|
; CGP-NEXT: v_add_i32_e32 v7, vcc, 1, v5
|
|
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v0, v2
|
|
; CGP-NEXT: v_cndmask_b32_e32 v0, v4, v6, vcc
|
|
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v1, v3
|
|
; CGP-NEXT: v_cndmask_b32_e32 v1, v5, v7, vcc
|
|
; CGP-NEXT: s_setpc_b64 s[30:31]
|
|
%shl.y = shl <2 x i32> <i32 4096, i32 4096>, %y
|
|
%r = udiv <2 x i32> %x, %shl.y
|
|
ret <2 x i32> %r
|
|
}
|
|
|
|
define i32 @v_udiv_i32_24bit(i32 %num, i32 %den) {
|
|
; GISEL-LABEL: v_udiv_i32_24bit:
|
|
; GISEL: ; %bb.0:
|
|
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GISEL-NEXT: v_and_b32_e32 v0, 0xffffff, v0
|
|
; GISEL-NEXT: v_and_b32_e32 v1, 0xffffff, v1
|
|
; GISEL-NEXT: v_cvt_f32_u32_e32 v2, v1
|
|
; GISEL-NEXT: v_sub_i32_e32 v3, vcc, 0, v1
|
|
; GISEL-NEXT: v_rcp_iflag_f32_e32 v2, v2
|
|
; GISEL-NEXT: v_mul_f32_e32 v2, 0x4f7ffffe, v2
|
|
; GISEL-NEXT: v_cvt_u32_f32_e32 v2, v2
|
|
; GISEL-NEXT: v_mul_lo_u32 v3, v3, v2
|
|
; GISEL-NEXT: v_mul_hi_u32 v3, v2, v3
|
|
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v3
|
|
; GISEL-NEXT: v_mul_hi_u32 v2, v0, v2
|
|
; GISEL-NEXT: v_mul_lo_u32 v3, v2, v1
|
|
; GISEL-NEXT: v_add_i32_e32 v4, vcc, 1, v2
|
|
; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v0, v3
|
|
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v1
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc
|
|
; GISEL-NEXT: v_sub_i32_e64 v3, s[4:5], v0, v1
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v0, v0, v3, vcc
|
|
; GISEL-NEXT: v_add_i32_e32 v3, vcc, 1, v2
|
|
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v1
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc
|
|
; GISEL-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CGP-LABEL: v_udiv_i32_24bit:
|
|
; CGP: ; %bb.0:
|
|
; CGP-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CGP-NEXT: v_and_b32_e32 v0, 0xffffff, v0
|
|
; CGP-NEXT: v_and_b32_e32 v1, 0xffffff, v1
|
|
; CGP-NEXT: v_cvt_f32_u32_e32 v0, v0
|
|
; CGP-NEXT: v_cvt_f32_u32_e32 v1, v1
|
|
; CGP-NEXT: v_rcp_f32_e32 v2, v1
|
|
; CGP-NEXT: v_mul_f32_e32 v2, v0, v2
|
|
; CGP-NEXT: v_trunc_f32_e32 v2, v2
|
|
; CGP-NEXT: v_fma_f32 v0, -v2, v1, v0
|
|
; CGP-NEXT: v_cvt_u32_f32_e32 v2, v2
|
|
; CGP-NEXT: v_cmp_ge_f32_e64 s[4:5], |v0|, v1
|
|
; CGP-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[4:5]
|
|
; CGP-NEXT: v_add_i32_e32 v0, vcc, v2, v0
|
|
; CGP-NEXT: v_and_b32_e32 v0, 0xffffff, v0
|
|
; CGP-NEXT: s_setpc_b64 s[30:31]
|
|
%num.mask = and i32 %num, 16777215
|
|
%den.mask = and i32 %den, 16777215
|
|
%result = udiv i32 %num.mask, %den.mask
|
|
ret i32 %result
|
|
}
|
|
|
|
define <2 x i32> @v_udiv_v2i32_24bit(<2 x i32> %num, <2 x i32> %den) {
|
|
; GISEL-LABEL: v_udiv_v2i32_24bit:
|
|
; GISEL: ; %bb.0:
|
|
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GISEL-NEXT: v_and_b32_e32 v0, 0xffffff, v0
|
|
; GISEL-NEXT: v_and_b32_e32 v1, 0xffffff, v1
|
|
; GISEL-NEXT: v_and_b32_e32 v2, 0xffffff, v2
|
|
; GISEL-NEXT: v_and_b32_e32 v3, 0xffffff, v3
|
|
; GISEL-NEXT: v_cvt_f32_u32_e32 v4, v2
|
|
; GISEL-NEXT: v_sub_i32_e32 v5, vcc, 0, v2
|
|
; GISEL-NEXT: v_cvt_f32_u32_e32 v6, v3
|
|
; GISEL-NEXT: v_sub_i32_e32 v7, vcc, 0, v3
|
|
; GISEL-NEXT: v_rcp_iflag_f32_e32 v4, v4
|
|
; GISEL-NEXT: v_rcp_iflag_f32_e32 v6, v6
|
|
; GISEL-NEXT: v_mul_f32_e32 v4, 0x4f7ffffe, v4
|
|
; GISEL-NEXT: v_mul_f32_e32 v6, 0x4f7ffffe, v6
|
|
; GISEL-NEXT: v_cvt_u32_f32_e32 v4, v4
|
|
; GISEL-NEXT: v_cvt_u32_f32_e32 v6, v6
|
|
; GISEL-NEXT: v_mul_lo_u32 v5, v5, v4
|
|
; GISEL-NEXT: v_mul_lo_u32 v7, v7, v6
|
|
; GISEL-NEXT: v_mul_hi_u32 v5, v4, v5
|
|
; GISEL-NEXT: v_mul_hi_u32 v7, v6, v7
|
|
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5
|
|
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v6, v7
|
|
; GISEL-NEXT: v_mul_hi_u32 v4, v0, v4
|
|
; GISEL-NEXT: v_mul_hi_u32 v5, v1, v5
|
|
; GISEL-NEXT: v_mul_lo_u32 v6, v4, v2
|
|
; GISEL-NEXT: v_add_i32_e32 v7, vcc, 1, v4
|
|
; GISEL-NEXT: v_mul_lo_u32 v8, v5, v3
|
|
; GISEL-NEXT: v_add_i32_e32 v9, vcc, 1, v5
|
|
; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v0, v6
|
|
; GISEL-NEXT: v_sub_i32_e32 v1, vcc, v1, v8
|
|
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v2
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v4, v4, v7, vcc
|
|
; GISEL-NEXT: v_sub_i32_e64 v6, s[4:5], v0, v2
|
|
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v1, v3
|
|
; GISEL-NEXT: v_cndmask_b32_e64 v5, v5, v9, s[4:5]
|
|
; GISEL-NEXT: v_sub_i32_e64 v7, s[6:7], v1, v3
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v0, v0, v6, vcc
|
|
; GISEL-NEXT: v_add_i32_e32 v6, vcc, 1, v4
|
|
; GISEL-NEXT: v_cndmask_b32_e64 v1, v1, v7, s[4:5]
|
|
; GISEL-NEXT: v_add_i32_e32 v7, vcc, 1, v5
|
|
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v2
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v0, v4, v6, vcc
|
|
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v1, v3
|
|
; GISEL-NEXT: v_cndmask_b32_e32 v1, v5, v7, vcc
|
|
; GISEL-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CGP-LABEL: v_udiv_v2i32_24bit:
|
|
; CGP: ; %bb.0:
|
|
; CGP-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CGP-NEXT: v_and_b32_e32 v0, 0xffffff, v0
|
|
; CGP-NEXT: v_and_b32_e32 v1, 0xffffff, v1
|
|
; CGP-NEXT: v_and_b32_e32 v2, 0xffffff, v2
|
|
; CGP-NEXT: v_and_b32_e32 v3, 0xffffff, v3
|
|
; CGP-NEXT: v_cvt_f32_u32_e32 v0, v0
|
|
; CGP-NEXT: v_cvt_f32_u32_e32 v2, v2
|
|
; CGP-NEXT: v_cvt_f32_u32_e32 v1, v1
|
|
; CGP-NEXT: v_cvt_f32_u32_e32 v3, v3
|
|
; CGP-NEXT: v_rcp_f32_e32 v4, v2
|
|
; CGP-NEXT: v_rcp_f32_e32 v5, v3
|
|
; CGP-NEXT: v_mul_f32_e32 v4, v0, v4
|
|
; CGP-NEXT: v_mul_f32_e32 v5, v1, v5
|
|
; CGP-NEXT: v_trunc_f32_e32 v4, v4
|
|
; CGP-NEXT: v_trunc_f32_e32 v5, v5
|
|
; CGP-NEXT: v_fma_f32 v0, -v4, v2, v0
|
|
; CGP-NEXT: v_cvt_u32_f32_e32 v4, v4
|
|
; CGP-NEXT: v_fma_f32 v1, -v5, v3, v1
|
|
; CGP-NEXT: v_cvt_u32_f32_e32 v5, v5
|
|
; CGP-NEXT: v_cmp_ge_f32_e64 s[4:5], |v0|, v2
|
|
; CGP-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[4:5]
|
|
; CGP-NEXT: v_cmp_ge_f32_e64 s[4:5], |v1|, v3
|
|
; CGP-NEXT: v_cndmask_b32_e64 v1, 0, 1, s[4:5]
|
|
; CGP-NEXT: v_add_i32_e32 v0, vcc, v4, v0
|
|
; CGP-NEXT: v_add_i32_e32 v1, vcc, v5, v1
|
|
; CGP-NEXT: v_and_b32_e32 v0, 0xffffff, v0
|
|
; CGP-NEXT: v_and_b32_e32 v1, 0xffffff, v1
|
|
; CGP-NEXT: s_setpc_b64 s[30:31]
|
|
%num.mask = and <2 x i32> %num, <i32 16777215, i32 16777215>
|
|
%den.mask = and <2 x i32> %den, <i32 16777215, i32 16777215>
|
|
%result = udiv <2 x i32> %num.mask, %den.mask
|
|
ret <2 x i32> %result
|
|
}
|