X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=test%2FCodeGen%2FX86%2Ffma_patterns_wide.ll;h=5541e05196e526d6b752954b32a8f6998b8a3c85;hp=5d608c91844803795dce23a477cb40188807bc33;hb=d87d4991ce23d1e6e300b054fbea3bedb6cf8f4b;hpb=c8dc70b58428df59c331f68543bc025d56d368f6

diff --git a/test/CodeGen/X86/fma_patterns_wide.ll b/test/CodeGen/X86/fma_patterns_wide.ll
index 5d608c91844..5541e05196e 100644
--- a/test/CodeGen/X86/fma_patterns_wide.ll
+++ b/test/CodeGen/X86/fma_patterns_wide.ll
@@ -1,71 +1,92 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; 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
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+fma -fp-contract=fast | FileCheck %s --check-prefix=FMA
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+fma4,+fma -fp-contract=fast | FileCheck %s --check-prefix=FMA
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+fma4 -fp-contract=fast | FileCheck %s --check-prefix=FMA4
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f -fp-contract=fast | FileCheck %s --check-prefix=AVX512
 
 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
+; FMA-LABEL: test_x86_fmadd_ps_y_wide:
+; FMA:       # BB#0:
+; FMA-NEXT:    vfmadd213ps %ymm4, %ymm2, %ymm0
+; FMA-NEXT:    vfmadd213ps %ymm5, %ymm3, %ymm1
+; FMA-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
+; FMA4-LABEL: test_x86_fmadd_ps_y_wide:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vfmaddps %ymm4, %ymm2, %ymm0, %ymm0
+; FMA4-NEXT:    vfmaddps %ymm5, %ymm3, %ymm1, %ymm1
+; FMA4-NEXT:    retq
+;
+; AVX512-LABEL: test_x86_fmadd_ps_y_wide:
+; AVX512:       # BB#0:
+; AVX512-NEXT:    vfmadd213ps %zmm2, %zmm1, %zmm0
+; AVX512-NEXT:    retq
   %x = fmul <16 x float> %a0, %a1
   %res = fadd <16 x float> %x, %a2
   ret <16 x float> %res
 }
 
 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
+; FMA-LABEL: test_x86_fmsub_ps_y_wide:
+; FMA:       # BB#0:
+; FMA-NEXT:    vfmsub213ps %ymm4, %ymm2, %ymm0
+; FMA-NEXT:    vfmsub213ps %ymm5, %ymm3, %ymm1
+; FMA-NEXT:    retq
+;
+; FMA4-LABEL: test_x86_fmsub_ps_y_wide:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vfmsubps %ymm4, %ymm2, %ymm0, %ymm0
+; FMA4-NEXT:    vfmsubps %ymm5, %ymm3, %ymm1, %ymm1
+; FMA4-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
+; AVX512-LABEL: test_x86_fmsub_ps_y_wide:
+; AVX512:       # BB#0:
+; AVX512-NEXT:    vfmsub213ps %zmm2, %zmm1, %zmm0
+; AVX512-NEXT:    retq
   %x = fmul <16 x float> %a0, %a1
   %res = fsub <16 x float> %x, %a2
   ret <16 x float> %res
 }
 
 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
+; FMA-LABEL: test_x86_fnmadd_ps_y_wide:
+; FMA:       # BB#0:
+; FMA-NEXT:    vfnmadd213ps %ymm4, %ymm2, %ymm0
+; FMA-NEXT:    vfnmadd213ps %ymm5, %ymm3, %ymm1
+; FMA-NEXT:    retq
+;
+; FMA4-LABEL: test_x86_fnmadd_ps_y_wide:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vfnmaddps %ymm4, %ymm2, %ymm0, %ymm0
+; FMA4-NEXT:    vfnmaddps %ymm5, %ymm3, %ymm1, %ymm1
+; FMA4-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
+; AVX512-LABEL: test_x86_fnmadd_ps_y_wide:
+; AVX512:       # BB#0:
+; AVX512-NEXT:    vfnmadd213ps %zmm2, %zmm1, %zmm0
+; AVX512-NEXT:    retq
   %x = fmul <16 x float> %a0, %a1
   %res = fsub <16 x float> %a2, %x
   ret <16 x float> %res
 }
 
 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
+; FMA-LABEL: test_x86_fnmsub_ps_y_wide:
+; FMA:       # BB#0:
+; FMA-NEXT:    vfnmsub213ps %ymm4, %ymm2, %ymm0
+; FMA-NEXT:    vfnmsub213ps %ymm5, %ymm3, %ymm1
+; FMA-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
+; FMA4-LABEL: test_x86_fnmsub_ps_y_wide:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vfnmsubps %ymm4, %ymm2, %ymm0, %ymm0
+; FMA4-NEXT:    vfnmsubps %ymm5, %ymm3, %ymm1, %ymm1
+; FMA4-NEXT:    retq
+;
+; AVX512-LABEL: test_x86_fnmsub_ps_y_wide:
+; AVX512:       # BB#0:
+; AVX512-NEXT:    vfnmsub213ps %zmm2, %zmm1, %zmm0
+; AVX512-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
@@ -73,34 +94,44 @@ define <16 x float> @test_x86_fnmsub_ps_y_wide(<16 x float> %a0, <16 x float> %a
 }
 
 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
+; FMA-LABEL: test_x86_fmadd_pd_y_wide:
+; FMA:       # BB#0:
+; FMA-NEXT:    vfmadd213pd %ymm4, %ymm2, %ymm0
+; FMA-NEXT:    vfmadd213pd %ymm5, %ymm3, %ymm1
+; FMA-NEXT:    retq
+;
+; FMA4-LABEL: test_x86_fmadd_pd_y_wide:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vfmaddpd %ymm4, %ymm2, %ymm0, %ymm0
+; FMA4-NEXT:    vfmaddpd %ymm5, %ymm3, %ymm1, %ymm1
+; FMA4-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
+; AVX512-LABEL: test_x86_fmadd_pd_y_wide:
+; AVX512:       # BB#0:
+; AVX512-NEXT:    vfmadd213pd %zmm2, %zmm1, %zmm0
+; AVX512-NEXT:    retq
   %x = fmul <8 x double> %a0, %a1
   %res = fadd <8 x double> %x, %a2
   ret <8 x double> %res
 }
 
 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
+; FMA-LABEL: test_x86_fmsub_pd_y_wide:
+; FMA:       # BB#0:
+; FMA-NEXT:    vfmsub213pd %ymm4, %ymm2, %ymm0
+; FMA-NEXT:    vfmsub213pd %ymm5, %ymm3, %ymm1
+; FMA-NEXT:    retq
+;
+; FMA4-LABEL: test_x86_fmsub_pd_y_wide:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vfmsubpd %ymm4, %ymm2, %ymm0, %ymm0
+; FMA4-NEXT:    vfmsubpd %ymm5, %ymm3, %ymm1, %ymm1
+; FMA4-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
+; AVX512-LABEL: test_x86_fmsub_pd_y_wide:
+; AVX512:       # BB#0:
+; AVX512-NEXT:    vfmsub213pd %zmm2, %zmm1, %zmm0
+; AVX512-NEXT:    retq
   %x = fmul <8 x double> %a0, %a1
   %res = fsub <8 x double> %x, %a2
   ret <8 x double> %res