From e91c6128e6a243d405e5eddd46ea5f10d11215b9 Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper@gmail.com>
Date: Sun, 5 Feb 2017 22:25:40 +0000
Subject: [PATCH] [AVX-512] Add masked scalar FMA intrinsics to
 isNonFoldablePartialRegisterLoad to improve load folding of scalar loads.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@294151 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86InstrInfo.cpp       | 24 ++++++++++++++++++++++++
 test/CodeGen/X86/avx512-intrinsics.ll | 12 ++++--------
 2 files changed, 28 insertions(+), 8 deletions(-)

diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index 8750880ba95..ced83c140de 100644
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -7740,6 +7740,18 @@ static bool isNonFoldablePartialRegisterLoad(const MachineInstr &LoadMI,
     case X86::VFMSUB132SSZr_Int: case X86::VFNMSUB132SSZr_Int:
     case X86::VFMSUB213SSZr_Int: case X86::VFNMSUB213SSZr_Int:
     case X86::VFMSUB231SSZr_Int: case X86::VFNMSUB231SSZr_Int:
+    case X86::VFMADD132SSZr_Intk: case X86::VFNMADD132SSZr_Intk:
+    case X86::VFMADD213SSZr_Intk: case X86::VFNMADD213SSZr_Intk:
+    case X86::VFMADD231SSZr_Intk: case X86::VFNMADD231SSZr_Intk:
+    case X86::VFMSUB132SSZr_Intk: case X86::VFNMSUB132SSZr_Intk:
+    case X86::VFMSUB213SSZr_Intk: case X86::VFNMSUB213SSZr_Intk:
+    case X86::VFMSUB231SSZr_Intk: case X86::VFNMSUB231SSZr_Intk:
+    case X86::VFMADD132SSZr_Intkz: case X86::VFNMADD132SSZr_Intkz:
+    case X86::VFMADD213SSZr_Intkz: case X86::VFNMADD213SSZr_Intkz:
+    case X86::VFMADD231SSZr_Intkz: case X86::VFNMADD231SSZr_Intkz:
+    case X86::VFMSUB132SSZr_Intkz: case X86::VFNMSUB132SSZr_Intkz:
+    case X86::VFMSUB213SSZr_Intkz: case X86::VFNMSUB213SSZr_Intkz:
+    case X86::VFMSUB231SSZr_Intkz: case X86::VFNMSUB231SSZr_Intkz:
       return false;
     default:
       return true;
@@ -7773,6 +7785,18 @@ static bool isNonFoldablePartialRegisterLoad(const MachineInstr &LoadMI,
     case X86::VFMSUB132SDZr_Int: case X86::VFNMSUB132SDZr_Int:
     case X86::VFMSUB213SDZr_Int: case X86::VFNMSUB213SDZr_Int:
     case X86::VFMSUB231SDZr_Int: case X86::VFNMSUB231SDZr_Int:
+    case X86::VFMADD132SDZr_Intk: case X86::VFNMADD132SDZr_Intk:
+    case X86::VFMADD213SDZr_Intk: case X86::VFNMADD213SDZr_Intk:
+    case X86::VFMADD231SDZr_Intk: case X86::VFNMADD231SDZr_Intk:
+    case X86::VFMSUB132SDZr_Intk: case X86::VFNMSUB132SDZr_Intk:
+    case X86::VFMSUB213SDZr_Intk: case X86::VFNMSUB213SDZr_Intk:
+    case X86::VFMSUB231SDZr_Intk: case X86::VFNMSUB231SDZr_Intk:
+    case X86::VFMADD132SDZr_Intkz: case X86::VFNMADD132SDZr_Intkz:
+    case X86::VFMADD213SDZr_Intkz: case X86::VFNMADD213SDZr_Intkz:
+    case X86::VFMADD231SDZr_Intkz: case X86::VFNMADD231SDZr_Intkz:
+    case X86::VFMSUB132SDZr_Intkz: case X86::VFNMSUB132SDZr_Intkz:
+    case X86::VFMSUB213SDZr_Intkz: case X86::VFNMSUB213SDZr_Intkz:
+    case X86::VFMSUB231SDZr_Intkz: case X86::VFNMSUB231SDZr_Intkz:
       return false;
     default:
       return true;
diff --git a/test/CodeGen/X86/avx512-intrinsics.ll b/test/CodeGen/X86/avx512-intrinsics.ll
index 916fd3af51a..fbd49a3258e 100644
--- a/test/CodeGen/X86/avx512-intrinsics.ll
+++ b/test/CodeGen/X86/avx512-intrinsics.ll
@@ -4843,10 +4843,9 @@ define void @fmadd_ss_mask_memfold(float* %a, float* %b, i8 %c) {
 ; CHECK-LABEL: fmadd_ss_mask_memfold:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; CHECK-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
 ; CHECK-NEXT:    andl $1, %edx
 ; CHECK-NEXT:    kmovw %edx, %k1
-; CHECK-NEXT:    vfmadd213ss %xmm0, %xmm1, %xmm0 {%k1}
+; CHECK-NEXT:    vfmadd132ss (%rsi), %xmm0, %xmm0 {%k1}
 ; CHECK-NEXT:    vmovss %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
   %a.val = load float, float* %a
@@ -4872,10 +4871,9 @@ define void @fmadd_ss_maskz_memfold(float* %a, float* %b, i8 %c) {
 ; CHECK-LABEL: fmadd_ss_maskz_memfold:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; CHECK-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
 ; CHECK-NEXT:    andl $1, %edx
 ; CHECK-NEXT:    kmovw %edx, %k1
-; CHECK-NEXT:    vfmadd213ss %xmm0, %xmm1, %xmm0 {%k1} {z}
+; CHECK-NEXT:    vfmadd132ss (%rsi), %xmm0, %xmm0 {%k1} {z}
 ; CHECK-NEXT:    vmovss %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
   %a.val = load float, float* %a
@@ -4901,10 +4899,9 @@ define void @fmadd_sd_mask_memfold(double* %a, double* %b, i8 %c) {
 ; CHECK-LABEL: fmadd_sd_mask_memfold:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
-; CHECK-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
 ; CHECK-NEXT:    andl $1, %edx
 ; CHECK-NEXT:    kmovw %edx, %k1
-; CHECK-NEXT:    vfmadd213sd %xmm0, %xmm1, %xmm0 {%k1}
+; CHECK-NEXT:    vfmadd132sd (%rsi), %xmm0, %xmm0 {%k1}
 ; CHECK-NEXT:    vmovlps %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
   %a.val = load double, double* %a
@@ -4926,10 +4923,9 @@ define void @fmadd_sd_maskz_memfold(double* %a, double* %b, i8 %c) {
 ; CHECK-LABEL: fmadd_sd_maskz_memfold:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
-; CHECK-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
 ; CHECK-NEXT:    andl $1, %edx
 ; CHECK-NEXT:    kmovw %edx, %k1
-; CHECK-NEXT:    vfmadd213sd %xmm0, %xmm1, %xmm0 {%k1} {z}
+; CHECK-NEXT:    vfmadd132sd (%rsi), %xmm0, %xmm0 {%k1} {z}
 ; CHECK-NEXT:    vmovlps %xmm0, (%rdi)
 ; CHECK-NEXT:    retq
   %a.val = load double, double* %a
-- 
2.50.1