[X86][FMA4] Add test cases to demonstrate missed folding opportunities for FMA4 scala...

llvm-svn: 288008

diff --git a/llvm/test/CodeGen/X86/fma-scalar-memfold.ll b/llvm/test/CodeGen/X86/fma-scalar-memfold.ll
index 23baeaf..5326d7f 100644 (file)
--- a/llvm/test/CodeGen/X86/fma-scalar-memfold.ll
+++ b/llvm/test/CodeGen/X86/fma-scalar-memfold.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mcpu=core-avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=AVX2
 ; RUN: llc < %s -mcpu=skx | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512
+; RUN: llc < %s -mattr=fma4 | FileCheck %s --check-prefix=FMA4
 
 target triple = "x86_64-unknown-unknown"
 
@@ -21,6 +22,14 @@ define void @fmadd_aab_ss(float* %a, float* %b) {
 ; CHECK-NEXT:    vfmadd213ss (%rsi), %xmm0, %xmm0
 ; CHECK-NEXT:    vmovss %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
+;
+; FMA4-LABEL: fmadd_aab_ss:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; FMA4-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; FMA4-NEXT:    vfmaddss %xmm1, %xmm0, %xmm0, %xmm0
+; FMA4-NEXT:    vmovss %xmm0, (%rdi)
+; FMA4-NEXT:    retq
   %a.val = load float, float* %a
   %av0 = insertelement <4 x float> undef, float %a.val, i32 0
   %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1
@@ -47,6 +56,14 @@ define void @fmadd_aba_ss(float* %a, float* %b) {
 ; CHECK-NEXT:    vfmadd132ss (%rsi), %xmm0, %xmm0
 ; CHECK-NEXT:    vmovss %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
+;
+; FMA4-LABEL: fmadd_aba_ss:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; FMA4-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; FMA4-NEXT:    vfmaddss %xmm0, %xmm1, %xmm0, %xmm0
+; FMA4-NEXT:    vmovss %xmm0, (%rdi)
+; FMA4-NEXT:    retq
   %a.val = load float, float* %a
   %av0 = insertelement <4 x float> undef, float %a.val, i32 0
   %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1
@@ -73,6 +90,14 @@ define void @fmsub_aab_ss(float* %a, float* %b) {
 ; CHECK-NEXT:    vfmsub213ss (%rsi), %xmm0, %xmm0
 ; CHECK-NEXT:    vmovss %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
+;
+; FMA4-LABEL: fmsub_aab_ss:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; FMA4-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; FMA4-NEXT:    vfmsubss %xmm1, %xmm0, %xmm0, %xmm0
+; FMA4-NEXT:    vmovss %xmm0, (%rdi)
+; FMA4-NEXT:    retq
   %a.val = load float, float* %a
   %av0 = insertelement <4 x float> undef, float %a.val, i32 0
   %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1
@@ -99,6 +124,14 @@ define void @fmsub_aba_ss(float* %a, float* %b) {
 ; CHECK-NEXT:    vfmsub132ss (%rsi), %xmm0, %xmm0
 ; CHECK-NEXT:    vmovss %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
+;
+; FMA4-LABEL: fmsub_aba_ss:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; FMA4-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; FMA4-NEXT:    vfmsubss %xmm0, %xmm1, %xmm0, %xmm0
+; FMA4-NEXT:    vmovss %xmm0, (%rdi)
+; FMA4-NEXT:    retq
   %a.val = load float, float* %a
   %av0 = insertelement <4 x float> undef, float %a.val, i32 0
   %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1
@@ -125,6 +158,14 @@ define void @fnmadd_aab_ss(float* %a, float* %b) {
 ; CHECK-NEXT:    vfnmadd213ss (%rsi), %xmm0, %xmm0
 ; CHECK-NEXT:    vmovss %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
+;
+; FMA4-LABEL: fnmadd_aab_ss:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; FMA4-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; FMA4-NEXT:    vfnmaddss %xmm1, %xmm0, %xmm0, %xmm0
+; FMA4-NEXT:    vmovss %xmm0, (%rdi)
+; FMA4-NEXT:    retq
   %a.val = load float, float* %a
   %av0 = insertelement <4 x float> undef, float %a.val, i32 0
   %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1
@@ -151,6 +192,14 @@ define void @fnmadd_aba_ss(float* %a, float* %b) {
 ; CHECK-NEXT:    vfnmadd132ss (%rsi), %xmm0, %xmm0
 ; CHECK-NEXT:    vmovss %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
+;
+; FMA4-LABEL: fnmadd_aba_ss:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; FMA4-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; FMA4-NEXT:    vfnmaddss %xmm0, %xmm1, %xmm0, %xmm0
+; FMA4-NEXT:    vmovss %xmm0, (%rdi)
+; FMA4-NEXT:    retq
   %a.val = load float, float* %a
   %av0 = insertelement <4 x float> undef, float %a.val, i32 0
   %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1
@@ -177,6 +226,14 @@ define void @fnmsub_aab_ss(float* %a, float* %b) {
 ; CHECK-NEXT:    vfnmsub213ss (%rsi), %xmm0, %xmm0
 ; CHECK-NEXT:    vmovss %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
+;
+; FMA4-LABEL: fnmsub_aab_ss:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; FMA4-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; FMA4-NEXT:    vfnmsubss %xmm1, %xmm0, %xmm0, %xmm0
+; FMA4-NEXT:    vmovss %xmm0, (%rdi)
+; FMA4-NEXT:    retq
   %a.val = load float, float* %a
   %av0 = insertelement <4 x float> undef, float %a.val, i32 0
   %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1
@@ -203,6 +260,14 @@ define void @fnmsub_aba_ss(float* %a, float* %b) {
 ; CHECK-NEXT:    vfnmsub132ss (%rsi), %xmm0, %xmm0
 ; CHECK-NEXT:    vmovss %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
+;
+; FMA4-LABEL: fnmsub_aba_ss:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; FMA4-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; FMA4-NEXT:    vfnmsubss %xmm0, %xmm1, %xmm0, %xmm0
+; FMA4-NEXT:    vmovss %xmm0, (%rdi)
+; FMA4-NEXT:    retq
   %a.val = load float, float* %a
   %av0 = insertelement <4 x float> undef, float %a.val, i32 0
   %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1
@@ -229,6 +294,14 @@ define void @fmadd_aab_sd(double* %a, double* %b) {
 ; CHECK-NEXT:    vfmadd213sd (%rsi), %xmm0, %xmm0
 ; CHECK-NEXT:    vmovlpd %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
+;
+; FMA4-LABEL: fmadd_aab_sd:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; FMA4-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; FMA4-NEXT:    vfmaddsd %xmm1, %xmm0, %xmm0, %xmm0
+; FMA4-NEXT:    vmovlpd %xmm0, (%rdi)
+; FMA4-NEXT:    retq
   %a.val = load double, double* %a
   %av0 = insertelement <2 x double> undef, double %a.val, i32 0
   %av  = insertelement <2 x double> %av0, double 0.000000e+00, i32 1
@@ -251,6 +324,14 @@ define void @fmadd_aba_sd(double* %a, double* %b) {
 ; CHECK-NEXT:    vfmadd132sd (%rsi), %xmm0, %xmm0
 ; CHECK-NEXT:    vmovlpd %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
+;
+; FMA4-LABEL: fmadd_aba_sd:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; FMA4-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; FMA4-NEXT:    vfmaddsd %xmm0, %xmm1, %xmm0, %xmm0
+; FMA4-NEXT:    vmovlpd %xmm0, (%rdi)
+; FMA4-NEXT:    retq
   %a.val = load double, double* %a
   %av0 = insertelement <2 x double> undef, double %a.val, i32 0
   %av  = insertelement <2 x double> %av0, double 0.000000e+00, i32 1
@@ -273,6 +354,14 @@ define void @fmsub_aab_sd(double* %a, double* %b) {
 ; CHECK-NEXT:    vfmsub213sd (%rsi), %xmm0, %xmm0
 ; CHECK-NEXT:    vmovlpd %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
+;
+; FMA4-LABEL: fmsub_aab_sd:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; FMA4-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; FMA4-NEXT:    vfmsubsd %xmm1, %xmm0, %xmm0, %xmm0
+; FMA4-NEXT:    vmovlpd %xmm0, (%rdi)
+; FMA4-NEXT:    retq
   %a.val = load double, double* %a
   %av0 = insertelement <2 x double> undef, double %a.val, i32 0
   %av  = insertelement <2 x double> %av0, double 0.000000e+00, i32 1
@@ -295,6 +384,14 @@ define void @fmsub_aba_sd(double* %a, double* %b) {
 ; CHECK-NEXT:    vfmsub132sd (%rsi), %xmm0, %xmm0
 ; CHECK-NEXT:    vmovlpd %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
+;
+; FMA4-LABEL: fmsub_aba_sd:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; FMA4-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; FMA4-NEXT:    vfmsubsd %xmm0, %xmm1, %xmm0, %xmm0
+; FMA4-NEXT:    vmovlpd %xmm0, (%rdi)
+; FMA4-NEXT:    retq
   %a.val = load double, double* %a
   %av0 = insertelement <2 x double> undef, double %a.val, i32 0
   %av  = insertelement <2 x double> %av0, double 0.000000e+00, i32 1
@@ -317,6 +414,14 @@ define void @fnmadd_aab_sd(double* %a, double* %b) {
 ; CHECK-NEXT:    vfnmadd213sd (%rsi), %xmm0, %xmm0
 ; CHECK-NEXT:    vmovlpd %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
+;
+; FMA4-LABEL: fnmadd_aab_sd:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; FMA4-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; FMA4-NEXT:    vfnmaddsd %xmm1, %xmm0, %xmm0, %xmm0
+; FMA4-NEXT:    vmovlpd %xmm0, (%rdi)
+; FMA4-NEXT:    retq
   %a.val = load double, double* %a
   %av0 = insertelement <2 x double> undef, double %a.val, i32 0
   %av  = insertelement <2 x double> %av0, double 0.000000e+00, i32 1
@@ -339,6 +444,14 @@ define void @fnmadd_aba_sd(double* %a, double* %b) {
 ; CHECK-NEXT:    vfnmadd132sd (%rsi), %xmm0, %xmm0
 ; CHECK-NEXT:    vmovlpd %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
+;
+; FMA4-LABEL: fnmadd_aba_sd:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; FMA4-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; FMA4-NEXT:    vfnmaddsd %xmm0, %xmm1, %xmm0, %xmm0
+; FMA4-NEXT:    vmovlpd %xmm0, (%rdi)
+; FMA4-NEXT:    retq
   %a.val = load double, double* %a
   %av0 = insertelement <2 x double> undef, double %a.val, i32 0
   %av  = insertelement <2 x double> %av0, double 0.000000e+00, i32 1
@@ -361,6 +474,14 @@ define void @fnmsub_aab_sd(double* %a, double* %b) {
 ; CHECK-NEXT:    vfnmsub213sd (%rsi), %xmm0, %xmm0
 ; CHECK-NEXT:    vmovlpd %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
+;
+; FMA4-LABEL: fnmsub_aab_sd:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; FMA4-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; FMA4-NEXT:    vfnmsubsd %xmm1, %xmm0, %xmm0, %xmm0
+; FMA4-NEXT:    vmovlpd %xmm0, (%rdi)
+; FMA4-NEXT:    retq
   %a.val = load double, double* %a
   %av0 = insertelement <2 x double> undef, double %a.val, i32 0
   %av  = insertelement <2 x double> %av0, double 0.000000e+00, i32 1
@@ -383,6 +504,14 @@ define void @fnmsub_aba_sd(double* %a, double* %b) {
 ; CHECK-NEXT:    vfnmsub132sd (%rsi), %xmm0, %xmm0
 ; CHECK-NEXT:    vmovlpd %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
+;
+; FMA4-LABEL: fnmsub_aba_sd:
+; FMA4:       # BB#0:
+; FMA4-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; FMA4-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; FMA4-NEXT:    vfnmsubsd %xmm0, %xmm1, %xmm0, %xmm0
+; FMA4-NEXT:    vmovlpd %xmm0, (%rdi)
+; FMA4-NEXT:    retq
   %a.val = load double, double* %a
   %av0 = insertelement <2 x double> undef, double %a.val, i32 0
   %av  = insertelement <2 x double> %av0, double 0.000000e+00, i32 1