RPCSX/llvm
1
0
Fork 0
mirror of https://github.com/RPCSX/llvm.git synced 2024-12-03 01:12:59 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[X86][FMA] Refreshed fma tests

Browse Source 
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@247508 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Simon Pilgrim 2015-09-12 15:33:05 +00:00
parent 69f08e598c
commit 21e0b1f45b
2 changed files with 227 additions and 151 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

262
test/CodeGen/X86/fma_patterns.ll
View File

@ -1,209 +1,263 @@
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=core-avx2 -mattr=avx2,+fma -fp-contract=fast | FileCheck %s
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -mattr=-fma4 -fp-contract=fast | FileCheck %s
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=bdver1 -fp-contract=fast | FileCheck %s --check-prefix=CHECK_FMA4
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+fma -fp-contract=fast | FileCheck %s
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+fma4,+fma -fp-contract=fast | FileCheck %s
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+fma4 -fp-contract=fast | FileCheck %s --check-prefix=CHECK_FMA4
; CHECK: test_x86_fmadd_ps
; CHECK: vfmadd213ps %xmm2, %xmm1, %xmm0
; CHECK: ret
; CHECK_FMA4: test_x86_fmadd_ps
; CHECK_FMA4: vfmaddps %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4: ret
define <4 x float> @test_x86_fmadd_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) {
; CHECK-LABEL: test_x86_fmadd_ps:
; CHECK: # BB#0:
; CHECK-NEXT: vfmadd213ps %xmm2, %xmm1, %xmm0
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fmadd_ps:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfmaddps %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4-NEXT: retq
%x = fmul <4 x float> %a0, %a1
%res = fadd <4 x float> %x, %a2
ret <4 x float> %res
}
; CHECK: test_x86_fmsub_ps
; CHECK: fmsub213ps %xmm2, %xmm1, %xmm0
; CHECK: ret
; CHECK_FMA4: test_x86_fmsub_ps
; CHECK_FMA4: vfmsubps %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4: ret
define <4 x float> @test_x86_fmsub_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) {
; CHECK-LABEL: test_x86_fmsub_ps:
; CHECK: # BB#0:
; CHECK-NEXT: vfmsub213ps %xmm2, %xmm1, %xmm0
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fmsub_ps:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfmsubps %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4-NEXT: retq
%x = fmul <4 x float> %a0, %a1
%res = fsub <4 x float> %x, %a2
ret <4 x float> %res
}
; CHECK: test_x86_fnmadd_ps
; CHECK: fnmadd213ps %xmm2, %xmm1, %xmm0
; CHECK: ret
; CHECK_FMA4: test_x86_fnmadd_ps
; CHECK_FMA4: vfnmaddps %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4: ret
define <4 x float> @test_x86_fnmadd_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) {
; CHECK-LABEL: test_x86_fnmadd_ps:
; CHECK: # BB#0:
; CHECK-NEXT: vfnmadd213ps %xmm2, %xmm1, %xmm0
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fnmadd_ps:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfnmaddps %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4-NEXT: retq
%x = fmul <4 x float> %a0, %a1
%res = fsub <4 x float> %a2, %x
ret <4 x float> %res
}
; CHECK: test_x86_fnmsub_ps
; CHECK: fnmsub213ps %xmm2, %xmm1, %xmm0
; CHECK: ret
; CHECK_FMA4: test_x86_fnmsub_ps
; CHECK_FMA4: fnmsubps %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4: ret
define <4 x float> @test_x86_fnmsub_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) {
; CHECK-LABEL: test_x86_fnmsub_ps:
; CHECK: # BB#0:
; CHECK-NEXT: vfnmsub213ps %xmm2, %xmm1, %xmm0
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fnmsub_ps:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfnmsubps %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4-NEXT: retq
%x = fmul <4 x float> %a0, %a1
%y = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x
%res = fsub <4 x float> %y, %a2
ret <4 x float> %res
}
; CHECK: test_x86_fmadd_ps_y
; CHECK: vfmadd213ps %ymm2, %ymm1, %ymm0
; CHECK: ret
; CHECK_FMA4: test_x86_fmadd_ps_y
; CHECK_FMA4: vfmaddps %ymm2, %ymm1, %ymm0, %ymm0
; CHECK_FMA4: ret
define <8 x float> @test_x86_fmadd_ps_y(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) {
; CHECK-LABEL: test_x86_fmadd_ps_y:
; CHECK: # BB#0:
; CHECK-NEXT: vfmadd213ps %ymm2, %ymm1, %ymm0
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fmadd_ps_y:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfmaddps %ymm2, %ymm1, %ymm0, %ymm0
; CHECK_FMA4-NEXT: retq
%x = fmul <8 x float> %a0, %a1
%res = fadd <8 x float> %x, %a2
ret <8 x float> %res
}
; CHECK: test_x86_fmsub_ps_y
; CHECK: vfmsub213ps %ymm2, %ymm1, %ymm0
; CHECK: ret
; CHECK_FMA4: test_x86_fmsub_ps_y
; CHECK_FMA4: vfmsubps %ymm2, %ymm1, %ymm0, %ymm0
; CHECK_FMA4: ret
define <8 x float> @test_x86_fmsub_ps_y(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) {
; CHECK-LABEL: test_x86_fmsub_ps_y:
; CHECK: # BB#0:
; CHECK-NEXT: vfmsub213ps %ymm2, %ymm1, %ymm0
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fmsub_ps_y:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfmsubps %ymm2, %ymm1, %ymm0, %ymm0
; CHECK_FMA4-NEXT: retq
%x = fmul <8 x float> %a0, %a1
%res = fsub <8 x float> %x, %a2
ret <8 x float> %res
}
; CHECK: test_x86_fnmadd_ps_y
; CHECK: vfnmadd213ps %ymm2, %ymm1, %ymm0
; CHECK: ret
; CHECK_FMA4: test_x86_fnmadd_ps_y
; CHECK_FMA4: vfnmaddps %ymm2, %ymm1, %ymm0, %ymm0
; CHECK_FMA4: ret
define <8 x float> @test_x86_fnmadd_ps_y(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) {
; CHECK-LABEL: test_x86_fnmadd_ps_y:
; CHECK: # BB#0:
; CHECK-NEXT: vfnmadd213ps %ymm2, %ymm1, %ymm0
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fnmadd_ps_y:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfnmaddps %ymm2, %ymm1, %ymm0, %ymm0
; CHECK_FMA4-NEXT: retq
%x = fmul <8 x float> %a0, %a1
%res = fsub <8 x float> %a2, %x
ret <8 x float> %res
}
; CHECK: test_x86_fnmsub_ps_y
; CHECK: vfnmsub213ps %ymm2, %ymm1, %ymm0
; CHECK: ret
define <8 x float> @test_x86_fnmsub_ps_y(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) {
; CHECK-LABEL: test_x86_fnmsub_ps_y:
; CHECK: # BB#0:
; CHECK-NEXT: vfnmsub213ps %ymm2, %ymm1, %ymm0
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fnmsub_ps_y:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfnmsubps %ymm2, %ymm1, %ymm0, %ymm0
; CHECK_FMA4-NEXT: retq
%x = fmul <8 x float> %a0, %a1
%y = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x
%res = fsub <8 x float> %y, %a2
ret <8 x float> %res
}
; CHECK: test_x86_fmadd_pd_y
; CHECK: vfmadd213pd %ymm2, %ymm1, %ymm0
; CHECK: ret
; CHECK_FMA4: test_x86_fmadd_pd_y
; CHECK_FMA4: vfmaddpd %ymm2, %ymm1, %ymm0, %ymm0
; CHECK_FMA4: ret
define <4 x double> @test_x86_fmadd_pd_y(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) {
; CHECK-LABEL: test_x86_fmadd_pd_y:
; CHECK: # BB#0:
; CHECK-NEXT: vfmadd213pd %ymm2, %ymm1, %ymm0
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fmadd_pd_y:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfmaddpd %ymm2, %ymm1, %ymm0, %ymm0
; CHECK_FMA4-NEXT: retq
%x = fmul <4 x double> %a0, %a1
%res = fadd <4 x double> %x, %a2
ret <4 x double> %res
}
; CHECK: test_x86_fmsub_pd_y
; CHECK: vfmsub213pd %ymm2, %ymm1, %ymm0
; CHECK: ret
; CHECK_FMA4: test_x86_fmsub_pd_y
; CHECK_FMA4: vfmsubpd %ymm2, %ymm1, %ymm0, %ymm0
; CHECK_FMA4: ret
define <4 x double> @test_x86_fmsub_pd_y(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) {
; CHECK-LABEL: test_x86_fmsub_pd_y:
; CHECK: # BB#0:
; CHECK-NEXT: vfmsub213pd %ymm2, %ymm1, %ymm0
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fmsub_pd_y:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfmsubpd %ymm2, %ymm1, %ymm0, %ymm0
; CHECK_FMA4-NEXT: retq
%x = fmul <4 x double> %a0, %a1
%res = fsub <4 x double> %x, %a2
ret <4 x double> %res
}
; CHECK: test_x86_fmsub_pd
; CHECK: vfmsub213pd %xmm2, %xmm1, %xmm0
; CHECK: ret
; CHECK_FMA4: test_x86_fmsub_pd
; CHECK_FMA4: vfmsubpd %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4: ret
define <2 x double> @test_x86_fmsub_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) {
; CHECK-LABEL: test_x86_fmsub_pd:
; CHECK: # BB#0:
; CHECK-NEXT: vfmsub213pd %xmm2, %xmm1, %xmm0
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fmsub_pd:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfmsubpd %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4-NEXT: retq
%x = fmul <2 x double> %a0, %a1
%res = fsub <2 x double> %x, %a2
ret <2 x double> %res
}
; CHECK: test_x86_fnmadd_ss
; CHECK: vfnmadd213ss %xmm2, %xmm1, %xmm0
; CHECK: ret
; CHECK_FMA4: test_x86_fnmadd_ss
; CHECK_FMA4: vfnmaddss %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4: ret
define float @test_x86_fnmadd_ss(float %a0, float %a1, float %a2) {
; CHECK-LABEL: test_x86_fnmadd_ss:
; CHECK: # BB#0:
; CHECK-NEXT: vfnmadd213ss %xmm2, %xmm1, %xmm0
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fnmadd_ss:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfnmaddss %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4-NEXT: retq
%x = fmul float %a0, %a1
%res = fsub float %a2, %x
ret float %res
}
; CHECK: test_x86_fnmadd_sd
; CHECK: vfnmadd213sd %xmm2, %xmm1, %xmm0
; CHECK: ret
; CHECK_FMA4: test_x86_fnmadd_sd
; CHECK_FMA4: vfnmaddsd %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4: ret
define double @test_x86_fnmadd_sd(double %a0, double %a1, double %a2) {
; CHECK-LABEL: test_x86_fnmadd_sd:
; CHECK: # BB#0:
; CHECK-NEXT: vfnmadd213sd %xmm2, %xmm1, %xmm0
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fnmadd_sd:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfnmaddsd %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4-NEXT: retq
%x = fmul double %a0, %a1
%res = fsub double %a2, %x
ret double %res
}
; CHECK: test_x86_fmsub_sd
; CHECK: vfmsub213sd %xmm2, %xmm1, %xmm0
; CHECK: ret
; CHECK_FMA4: test_x86_fmsub_sd
; CHECK_FMA4: vfmsubsd %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4: ret
define double @test_x86_fmsub_sd(double %a0, double %a1, double %a2) {
; CHECK-LABEL: test_x86_fmsub_sd:
; CHECK: # BB#0:
; CHECK-NEXT: vfmsub213sd %xmm2, %xmm1, %xmm0
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fmsub_sd:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfmsubsd %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4-NEXT: retq
%x = fmul double %a0, %a1
%res = fsub double %x, %a2
ret double %res
}
; CHECK: test_x86_fnmsub_ss
; CHECK: vfnmsub213ss %xmm2, %xmm1, %xmm0
; CHECK: ret
; CHECK_FMA4: test_x86_fnmsub_ss
; CHECK_FMA4: vfnmsubss %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4: ret
define float @test_x86_fnmsub_ss(float %a0, float %a1, float %a2) {
; CHECK-LABEL: test_x86_fnmsub_ss:
; CHECK: # BB#0:
; CHECK-NEXT: vfnmsub213ss %xmm2, %xmm1, %xmm0
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fnmsub_ss:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfnmsubss %xmm2, %xmm1, %xmm0, %xmm0
; CHECK_FMA4-NEXT: retq
%x = fsub float -0.000000e+00, %a0
%y = fmul float %x, %a1
%res = fsub float %y, %a2
ret float %res
}
; CHECK: test_x86_fmadd_ps_load
; CHECK: vmovaps (%rdi), %xmm2
; CHECK: vfmadd213ps %xmm1, %xmm2, %xmm0
; CHECK: ret
; CHECK_FMA4: test_x86_fmadd_ps_load
; CHECK_FMA4: vfmaddps %xmm1, (%rdi), %xmm0, %xmm0
; CHECK_FMA4: ret
define <4 x float> @test_x86_fmadd_ps_load(<4 x float>* %a0, <4 x float> %a1, <4 x float> %a2) {
; CHECK-LABEL: test_x86_fmadd_ps_load:
; CHECK: # BB#0:
; CHECK-NEXT: vmovaps (%rdi), %xmm2
; CHECK-NEXT: vfmadd213ps %xmm1, %xmm2, %xmm0
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fmadd_ps_load:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfmaddps %xmm1, (%rdi), %xmm0, %xmm0
; CHECK_FMA4-NEXT: retq
%x = load <4 x float>, <4 x float>* %a0
%y = fmul <4 x float> %x, %a1
%res = fadd <4 x float> %y, %a2
ret <4 x float> %res
}
; CHECK: test_x86_fmsub_ps_load
; CHECK: vmovaps (%rdi), %xmm2
; CHECK: fmsub213ps %xmm1, %xmm2, %xmm0
; CHECK: ret
; CHECK_FMA4: test_x86_fmsub_ps_load
; CHECK_FMA4: vfmsubps %xmm1, (%rdi), %xmm0, %xmm0
; CHECK_FMA4: ret
define <4 x float> @test_x86_fmsub_ps_load(<4 x float>* %a0, <4 x float> %a1, <4 x float> %a2) {
; CHECK-LABEL: test_x86_fmsub_ps_load:
; CHECK: # BB#0:
; CHECK-NEXT: vmovaps (%rdi), %xmm2
; CHECK-NEXT: vfmsub213ps %xmm1, %xmm2, %xmm0
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fmsub_ps_load:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfmsubps %xmm1, (%rdi), %xmm0, %xmm0
; CHECK_FMA4-NEXT: retq
%x = load <4 x float>, <4 x float>* %a0
%y = fmul <4 x float> %x, %a1
%res = fsub <4 x float> %y, %a2

116
test/CodeGen/X86/fma_patterns_wide.ll
View File

@ -1,83 +1,105 @@
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=core-avx2 -mattr=avx2,+fma -fp-contract=fast | FileCheck %s
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -mattr=-fma4 -fp-contract=fast | FileCheck %s
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=bdver1 -fp-contract=fast | FileCheck %s --check-prefix=CHECK_FMA4
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+fma -fp-contract=fast | FileCheck %s
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+fma4,+fma -fp-contract=fast | FileCheck %s
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+fma4 -fp-contract=fast | FileCheck %s --check-prefix=CHECK_FMA4
; CHECK-LABEL: test_x86_fmadd_ps_y_wide
; CHECK: vfmadd213ps
; CHECK: vfmadd213ps
; CHECK: ret
; CHECK_FMA4-LABEL: test_x86_fmadd_ps_y_wide
; CHECK_FMA4: vfmaddps
; CHECK_FMA4: vfmaddps
; CHECK_FMA4: ret
define <16 x float> @test_x86_fmadd_ps_y_wide(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
; CHECK-LABEL: test_x86_fmadd_ps_y_wide:
; CHECK: # BB#0:
; CHECK-NEXT: vfmadd213ps %ymm4, %ymm2, %ymm0
; CHECK-NEXT: vfmadd213ps %ymm5, %ymm3, %ymm1
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fmadd_ps_y_wide:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfmaddps %ymm4, %ymm2, %ymm0, %ymm0
; CHECK_FMA4-NEXT: vfmaddps %ymm5, %ymm3, %ymm1, %ymm1
; CHECK_FMA4-NEXT: retq
%x = fmul <16 x float> %a0, %a1
%res = fadd <16 x float> %x, %a2
ret <16 x float> %res
}
; CHECK-LABEL: test_x86_fmsub_ps_y_wide
; CHECK: vfmsub213ps
; CHECK: vfmsub213ps
; CHECK: ret
; CHECK_FMA4-LABEL: test_x86_fmsub_ps_y_wide
; CHECK_FMA4: vfmsubps
; CHECK_FMA4: vfmsubps
; CHECK_FMA4: ret
define <16 x float> @test_x86_fmsub_ps_y_wide(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
; CHECK-LABEL: test_x86_fmsub_ps_y_wide:
; CHECK: # BB#0:
; CHECK-NEXT: vfmsub213ps %ymm4, %ymm2, %ymm0
; CHECK-NEXT: vfmsub213ps %ymm5, %ymm3, %ymm1
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fmsub_ps_y_wide:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfmsubps %ymm4, %ymm2, %ymm0, %ymm0
; CHECK_FMA4-NEXT: vfmsubps %ymm5, %ymm3, %ymm1, %ymm1
; CHECK_FMA4-NEXT: retq
%x = fmul <16 x float> %a0, %a1
%res = fsub <16 x float> %x, %a2
ret <16 x float> %res
}
; CHECK-LABEL: test_x86_fnmadd_ps_y_wide
; CHECK: vfnmadd213ps
; CHECK: vfnmadd213ps
; CHECK: ret
; CHECK_FMA4-LABEL: test_x86_fnmadd_ps_y_wide
; CHECK_FMA4: vfnmaddps
; CHECK_FMA4: vfnmaddps
; CHECK_FMA4: ret
define <16 x float> @test_x86_fnmadd_ps_y_wide(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
; CHECK-LABEL: test_x86_fnmadd_ps_y_wide:
; CHECK: # BB#0:
; CHECK-NEXT: vfnmadd213ps %ymm4, %ymm2, %ymm0
; CHECK-NEXT: vfnmadd213ps %ymm5, %ymm3, %ymm1
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fnmadd_ps_y_wide:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfnmaddps %ymm4, %ymm2, %ymm0, %ymm0
; CHECK_FMA4-NEXT: vfnmaddps %ymm5, %ymm3, %ymm1, %ymm1
; CHECK_FMA4-NEXT: retq
%x = fmul <16 x float> %a0, %a1
%res = fsub <16 x float> %a2, %x
ret <16 x float> %res
}
; CHECK-LABEL: test_x86_fnmsub_ps_y_wide
; CHECK: vfnmsub213ps
; CHECK: vfnmsub213ps
; CHECK: ret
define <16 x float> @test_x86_fnmsub_ps_y_wide(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
; CHECK-LABEL: test_x86_fnmsub_ps_y_wide:
; CHECK: # BB#0:
; CHECK-NEXT: vfnmsub213ps %ymm4, %ymm2, %ymm0
; CHECK-NEXT: vfnmsub213ps %ymm5, %ymm3, %ymm1
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fnmsub_ps_y_wide:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfnmsubps %ymm4, %ymm2, %ymm0, %ymm0
; CHECK_FMA4-NEXT: vfnmsubps %ymm5, %ymm3, %ymm1, %ymm1
; CHECK_FMA4-NEXT: retq
%x = fmul <16 x float> %a0, %a1
%y = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x
%res = fsub <16 x float> %y, %a2
ret <16 x float> %res
}
; CHECK-LABEL: test_x86_fmadd_pd_y_wide
; CHECK: vfmadd213pd
; CHECK: vfmadd213pd
; CHECK: ret
; CHECK_FMA4-LABEL: test_x86_fmadd_pd_y_wide
; CHECK_FMA4: vfmaddpd
; CHECK_FMA4: vfmaddpd
; CHECK_FMA4: ret
define <8 x double> @test_x86_fmadd_pd_y_wide(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
; CHECK-LABEL: test_x86_fmadd_pd_y_wide:
; CHECK: # BB#0:
; CHECK-NEXT: vfmadd213pd %ymm4, %ymm2, %ymm0
; CHECK-NEXT: vfmadd213pd %ymm5, %ymm3, %ymm1
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fmadd_pd_y_wide:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfmaddpd %ymm4, %ymm2, %ymm0, %ymm0
; CHECK_FMA4-NEXT: vfmaddpd %ymm5, %ymm3, %ymm1, %ymm1
; CHECK_FMA4-NEXT: retq
%x = fmul <8 x double> %a0, %a1
%res = fadd <8 x double> %x, %a2
ret <8 x double> %res
}
; CHECK-LABEL: test_x86_fmsub_pd_y_wide
; CHECK: vfmsub213pd
; CHECK: vfmsub213pd
; CHECK: ret
; CHECK_FMA4-LABEL: test_x86_fmsub_pd_y_wide
; CHECK_FMA4: vfmsubpd
; CHECK_FMA4: vfmsubpd
; CHECK_FMA4: ret
define <8 x double> @test_x86_fmsub_pd_y_wide(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
; CHECK-LABEL: test_x86_fmsub_pd_y_wide:
; CHECK: # BB#0:
; CHECK-NEXT: vfmsub213pd %ymm4, %ymm2, %ymm0
; CHECK-NEXT: vfmsub213pd %ymm5, %ymm3, %ymm1
; CHECK-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fmsub_pd_y_wide:
; CHECK_FMA4: # BB#0:
; CHECK_FMA4-NEXT: vfmsubpd %ymm4, %ymm2, %ymm0, %ymm0
; CHECK_FMA4-NEXT: vfmsubpd %ymm5, %ymm3, %ymm1, %ymm1
; CHECK_FMA4-NEXT: retq
%x = fmul <8 x double> %a0, %a1
%res = fsub <8 x double> %x, %a2
ret <8 x double> %res