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
203 lines
8.2 KiB
LLVM
203 lines
8.2 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+avx | FileCheck %s -check-prefixes=CHECK,NOFMA
|
|
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+fma | FileCheck %s -check-prefixes=CHECK,FMA3,FMA3_256
|
|
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+fma,+avx512f | FileCheck %s -check-prefixes=CHECK,FMA3,FMA3_512
|
|
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+fma4 | FileCheck %s -check-prefixes=CHECK,FMA4
|
|
|
|
; This test checks the fusing of MUL + SUB/ADD to FMSUBADD.
|
|
|
|
define <2 x double> @mul_subadd_pd128(<2 x double> %A, <2 x double> %B, <2 x double> %C) {
|
|
; NOFMA-LABEL: mul_subadd_pd128:
|
|
; NOFMA: # %bb.0: # %entry
|
|
; NOFMA-NEXT: vmulpd %xmm1, %xmm0, %xmm0
|
|
; NOFMA-NEXT: vsubpd %xmm2, %xmm0, %xmm1
|
|
; NOFMA-NEXT: vaddpd %xmm2, %xmm0, %xmm0
|
|
; NOFMA-NEXT: vmovsd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
|
|
; NOFMA-NEXT: retq
|
|
;
|
|
; FMA3-LABEL: mul_subadd_pd128:
|
|
; FMA3: # %bb.0: # %entry
|
|
; FMA3-NEXT: vfmsubadd213pd {{.*#+}} xmm0 = (xmm1 * xmm0) -/+ xmm2
|
|
; FMA3-NEXT: retq
|
|
;
|
|
; FMA4-LABEL: mul_subadd_pd128:
|
|
; FMA4: # %bb.0: # %entry
|
|
; FMA4-NEXT: vfmsubaddpd {{.*#+}} xmm0 = (xmm0 * xmm1) -/+ xmm2
|
|
; FMA4-NEXT: retq
|
|
entry:
|
|
%AB = fmul contract<2 x double> %A, %B
|
|
%Sub = fsub contract<2 x double> %AB, %C
|
|
%Add = fadd contract<2 x double> %AB, %C
|
|
%subadd = shufflevector <2 x double> %Add, <2 x double> %Sub, <2 x i32> <i32 0, i32 3>
|
|
ret <2 x double> %subadd
|
|
}
|
|
|
|
define <4 x float> @mul_subadd_ps128(<4 x float> %A, <4 x float> %B, <4 x float> %C) {
|
|
; NOFMA-LABEL: mul_subadd_ps128:
|
|
; NOFMA: # %bb.0: # %entry
|
|
; NOFMA-NEXT: vmulps %xmm1, %xmm0, %xmm0
|
|
; NOFMA-NEXT: vsubps %xmm2, %xmm0, %xmm1
|
|
; NOFMA-NEXT: vaddps %xmm2, %xmm0, %xmm0
|
|
; NOFMA-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
|
|
; NOFMA-NEXT: retq
|
|
;
|
|
; FMA3-LABEL: mul_subadd_ps128:
|
|
; FMA3: # %bb.0: # %entry
|
|
; FMA3-NEXT: vfmsubadd213ps {{.*#+}} xmm0 = (xmm1 * xmm0) -/+ xmm2
|
|
; FMA3-NEXT: retq
|
|
;
|
|
; FMA4-LABEL: mul_subadd_ps128:
|
|
; FMA4: # %bb.0: # %entry
|
|
; FMA4-NEXT: vfmsubaddps {{.*#+}} xmm0 = (xmm0 * xmm1) -/+ xmm2
|
|
; FMA4-NEXT: retq
|
|
entry:
|
|
%AB = fmul contract <4 x float> %A, %B
|
|
%Sub = fsub contract <4 x float> %AB, %C
|
|
%Add = fadd contract <4 x float> %AB, %C
|
|
%subadd = shufflevector <4 x float> %Add, <4 x float> %Sub, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
|
|
ret <4 x float> %subadd
|
|
}
|
|
|
|
define <4 x double> @mul_subadd_pd256(<4 x double> %A, <4 x double> %B, <4 x double> %C) {
|
|
; NOFMA-LABEL: mul_subadd_pd256:
|
|
; NOFMA: # %bb.0: # %entry
|
|
; NOFMA-NEXT: vmulpd %ymm1, %ymm0, %ymm0
|
|
; NOFMA-NEXT: vsubpd %ymm2, %ymm0, %ymm1
|
|
; NOFMA-NEXT: vaddpd %ymm2, %ymm0, %ymm0
|
|
; NOFMA-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3]
|
|
; NOFMA-NEXT: retq
|
|
;
|
|
; FMA3-LABEL: mul_subadd_pd256:
|
|
; FMA3: # %bb.0: # %entry
|
|
; FMA3-NEXT: vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2
|
|
; FMA3-NEXT: retq
|
|
;
|
|
; FMA4-LABEL: mul_subadd_pd256:
|
|
; FMA4: # %bb.0: # %entry
|
|
; FMA4-NEXT: vfmsubaddpd {{.*#+}} ymm0 = (ymm0 * ymm1) -/+ ymm2
|
|
; FMA4-NEXT: retq
|
|
entry:
|
|
%AB = fmul contract <4 x double> %A, %B
|
|
%Sub = fsub contract <4 x double> %AB, %C
|
|
%Add = fadd contract <4 x double> %AB, %C
|
|
%subadd = shufflevector <4 x double> %Add, <4 x double> %Sub, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
|
|
ret <4 x double> %subadd
|
|
}
|
|
|
|
define <8 x float> @mul_subadd_ps256(<8 x float> %A, <8 x float> %B, <8 x float> %C) {
|
|
; NOFMA-LABEL: mul_subadd_ps256:
|
|
; NOFMA: # %bb.0: # %entry
|
|
; NOFMA-NEXT: vmulps %ymm1, %ymm0, %ymm0
|
|
; NOFMA-NEXT: vsubps %ymm2, %ymm0, %ymm1
|
|
; NOFMA-NEXT: vaddps %ymm2, %ymm0, %ymm0
|
|
; NOFMA-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
|
|
; NOFMA-NEXT: retq
|
|
;
|
|
; FMA3-LABEL: mul_subadd_ps256:
|
|
; FMA3: # %bb.0: # %entry
|
|
; FMA3-NEXT: vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2
|
|
; FMA3-NEXT: retq
|
|
;
|
|
; FMA4-LABEL: mul_subadd_ps256:
|
|
; FMA4: # %bb.0: # %entry
|
|
; FMA4-NEXT: vfmsubaddps {{.*#+}} ymm0 = (ymm0 * ymm1) -/+ ymm2
|
|
; FMA4-NEXT: retq
|
|
entry:
|
|
%AB = fmul contract <8 x float> %A, %B
|
|
%Sub = fsub contract <8 x float> %AB, %C
|
|
%Add = fadd contract <8 x float> %AB, %C
|
|
%subadd = shufflevector <8 x float> %Add, <8 x float> %Sub, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
|
|
ret <8 x float> %subadd
|
|
}
|
|
|
|
define <8 x double> @mul_subadd_pd512(<8 x double> %A, <8 x double> %B, <8 x double> %C) {
|
|
; NOFMA-LABEL: mul_subadd_pd512:
|
|
; NOFMA: # %bb.0: # %entry
|
|
; NOFMA-NEXT: vmulpd %ymm2, %ymm0, %ymm0
|
|
; NOFMA-NEXT: vmulpd %ymm3, %ymm1, %ymm1
|
|
; NOFMA-NEXT: vsubpd %ymm5, %ymm1, %ymm2
|
|
; NOFMA-NEXT: vsubpd %ymm4, %ymm0, %ymm3
|
|
; NOFMA-NEXT: vaddpd %ymm5, %ymm1, %ymm1
|
|
; NOFMA-NEXT: vblendpd {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2],ymm2[3]
|
|
; NOFMA-NEXT: vaddpd %ymm4, %ymm0, %ymm0
|
|
; NOFMA-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm3[1],ymm0[2],ymm3[3]
|
|
; NOFMA-NEXT: retq
|
|
;
|
|
; FMA3_256-LABEL: mul_subadd_pd512:
|
|
; FMA3_256: # %bb.0: # %entry
|
|
; FMA3_256-NEXT: vfmsubadd213pd {{.*#+}} ymm0 = (ymm2 * ymm0) -/+ ymm4
|
|
; FMA3_256-NEXT: vfmsubadd213pd {{.*#+}} ymm1 = (ymm3 * ymm1) -/+ ymm5
|
|
; FMA3_256-NEXT: retq
|
|
;
|
|
; FMA3_512-LABEL: mul_subadd_pd512:
|
|
; FMA3_512: # %bb.0: # %entry
|
|
; FMA3_512-NEXT: vfmsubadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2
|
|
; FMA3_512-NEXT: retq
|
|
;
|
|
; FMA4-LABEL: mul_subadd_pd512:
|
|
; FMA4: # %bb.0: # %entry
|
|
; FMA4-NEXT: vfmsubaddpd {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm4
|
|
; FMA4-NEXT: vfmsubaddpd {{.*#+}} ymm1 = (ymm1 * ymm3) -/+ ymm5
|
|
; FMA4-NEXT: retq
|
|
entry:
|
|
%AB = fmul contract <8 x double> %A, %B
|
|
%Sub = fsub contract <8 x double> %AB, %C
|
|
%Add = fadd contract <8 x double> %AB, %C
|
|
%subadd = shufflevector <8 x double> %Add, <8 x double> %Sub, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
|
|
ret <8 x double> %subadd
|
|
}
|
|
|
|
define <16 x float> @mul_subadd_ps512(<16 x float> %A, <16 x float> %B, <16 x float> %C) {
|
|
; NOFMA-LABEL: mul_subadd_ps512:
|
|
; NOFMA: # %bb.0: # %entry
|
|
; NOFMA-NEXT: vmulps %ymm2, %ymm0, %ymm0
|
|
; NOFMA-NEXT: vmulps %ymm3, %ymm1, %ymm1
|
|
; NOFMA-NEXT: vsubps %ymm5, %ymm1, %ymm2
|
|
; NOFMA-NEXT: vsubps %ymm4, %ymm0, %ymm3
|
|
; NOFMA-NEXT: vaddps %ymm5, %ymm1, %ymm1
|
|
; NOFMA-NEXT: vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2],ymm2[3],ymm1[4],ymm2[5],ymm1[6],ymm2[7]
|
|
; NOFMA-NEXT: vaddps %ymm4, %ymm0, %ymm0
|
|
; NOFMA-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm3[1],ymm0[2],ymm3[3],ymm0[4],ymm3[5],ymm0[6],ymm3[7]
|
|
; NOFMA-NEXT: retq
|
|
;
|
|
; FMA3_256-LABEL: mul_subadd_ps512:
|
|
; FMA3_256: # %bb.0: # %entry
|
|
; FMA3_256-NEXT: vfmsubadd213ps {{.*#+}} ymm0 = (ymm2 * ymm0) -/+ ymm4
|
|
; FMA3_256-NEXT: vfmsubadd213ps {{.*#+}} ymm1 = (ymm3 * ymm1) -/+ ymm5
|
|
; FMA3_256-NEXT: retq
|
|
;
|
|
; FMA3_512-LABEL: mul_subadd_ps512:
|
|
; FMA3_512: # %bb.0: # %entry
|
|
; FMA3_512-NEXT: vfmsubadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2
|
|
; FMA3_512-NEXT: retq
|
|
;
|
|
; FMA4-LABEL: mul_subadd_ps512:
|
|
; FMA4: # %bb.0: # %entry
|
|
; FMA4-NEXT: vfmsubaddps {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm4
|
|
; FMA4-NEXT: vfmsubaddps {{.*#+}} ymm1 = (ymm1 * ymm3) -/+ ymm5
|
|
; FMA4-NEXT: retq
|
|
entry:
|
|
%AB = fmul contract <16 x float> %A, %B
|
|
%Sub = fsub contract <16 x float> %AB, %C
|
|
%Add = fadd contract <16 x float> %AB, %C
|
|
%subadd = shufflevector <16 x float> %Add, <16 x float> %Sub, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
|
|
ret <16 x float> %subadd
|
|
}
|
|
|
|
; This should not be matched to fmsubadd because the mul is on the wrong side of the fsub.
|
|
define <2 x double> @mul_subadd_bad_commute(<2 x double> %A, <2 x double> %B, <2 x double> %C) {
|
|
; CHECK-LABEL: mul_subadd_bad_commute:
|
|
; CHECK: # %bb.0: # %entry
|
|
; CHECK-NEXT: vmulpd %xmm1, %xmm0, %xmm0
|
|
; CHECK-NEXT: vsubpd %xmm0, %xmm2, %xmm1
|
|
; CHECK-NEXT: vaddpd %xmm2, %xmm0, %xmm0
|
|
; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
|
|
; CHECK-NEXT: retq
|
|
entry:
|
|
%AB = fmul contract <2 x double> %A, %B
|
|
%Sub = fsub <2 x double> %C, %AB
|
|
%Add = fadd <2 x double> %AB, %C
|
|
%subadd = shufflevector <2 x double> %Add, <2 x double> %Sub, <2 x i32> <i32 0, i32 3>
|
|
ret <2 x double> %subadd
|
|
}
|