From 19ff9cf62b09cf67255af6606b37c69bbe0edbef Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Tue, 6 Jun 2017 14:18:39 +0000
Subject: [PATCH] [X86][AVX1] Split 256-bit vector non-temporal FastISel loads
 to keep it non-temporal (PR32744)

Extension to D33728

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304798 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86FastISel.cpp            |  6 ++++
 test/CodeGen/X86/fast-isel-nontemporal.ll | 36 +++++++++++++++++++----
 2 files changed, 36 insertions(+), 6 deletions(-)

diff --git a/lib/Target/X86/X86FastISel.cpp b/lib/Target/X86/X86FastISel.cpp
index 3cfb924abd0..621505aaded 100644
--- a/lib/Target/X86/X86FastISel.cpp
+++ b/lib/Target/X86/X86FastISel.cpp
@@ -414,6 +414,8 @@ bool X86FastISel::X86FastEmitLoad(EVT VT, X86AddressMode &AM,
     assert(HasAVX);
     if (IsNonTemporal && Alignment >= 32 && HasAVX2)
       Opc = HasVLX ? X86::VMOVNTDQAZ256rm : X86::VMOVNTDQAYrm;
+    else if (IsNonTemporal && Alignment >= 16)
+      return false; // Force split for X86::VMOVNTDQArm
     else if (Alignment >= 32)
       Opc = HasVLX ? X86::VMOVAPSZ256rm : X86::VMOVAPSYrm;
     else
@@ -424,6 +426,8 @@ bool X86FastISel::X86FastEmitLoad(EVT VT, X86AddressMode &AM,
     assert(HasAVX);
     if (IsNonTemporal && Alignment >= 32 && HasAVX2)
       Opc = X86::VMOVNTDQAYrm;
+    else if (IsNonTemporal && Alignment >= 16)
+      return false; // Force split for X86::VMOVNTDQArm
     else if (Alignment >= 32)
       Opc = HasVLX ? X86::VMOVAPDZ256rm : X86::VMOVAPDYrm;
     else
@@ -437,6 +441,8 @@ bool X86FastISel::X86FastEmitLoad(EVT VT, X86AddressMode &AM,
     assert(HasAVX);
     if (IsNonTemporal && Alignment >= 32 && HasAVX2)
       Opc = X86::VMOVNTDQAYrm;
+    else if (IsNonTemporal && Alignment >= 16)
+      return false; // Force split for X86::VMOVNTDQArm
     else if (Alignment >= 32)
       Opc = HasVLX ? X86::VMOVDQA64Z256rm : X86::VMOVDQAYrm;
     else
diff --git a/test/CodeGen/X86/fast-isel-nontemporal.ll b/test/CodeGen/X86/fast-isel-nontemporal.ll
index 306012aa3bf..33d001cdc21 100644
--- a/test/CodeGen/X86/fast-isel-nontemporal.ll
+++ b/test/CodeGen/X86/fast-isel-nontemporal.ll
@@ -545,7 +545,11 @@ define <8 x float> @test_load_nt8xfloat(<8 x float>* nocapture %ptr) {
 ;
 ; AVX1-LABEL: test_load_nt8xfloat:
 ; AVX1:       # BB#0: # %entry
-; AVX1-NEXT:    vmovaps (%rdi), %ymm0
+; AVX1-NEXT:    vmovntdqa (%rdi), %xmm0
+; AVX1-NEXT:    # implicit-def: %YMM1
+; AVX1-NEXT:    vmovaps %xmm0, %xmm1
+; AVX1-NEXT:    vmovntdqa 16(%rdi), %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_load_nt8xfloat:
@@ -583,7 +587,11 @@ define <4 x double> @test_load_nt4xdouble(<4 x double>* nocapture %ptr) {
 ;
 ; AVX1-LABEL: test_load_nt4xdouble:
 ; AVX1:       # BB#0: # %entry
-; AVX1-NEXT:    vmovapd (%rdi), %ymm0
+; AVX1-NEXT:    vmovntdqa (%rdi), %xmm0
+; AVX1-NEXT:    # implicit-def: %YMM1
+; AVX1-NEXT:    vmovaps %xmm0, %xmm1
+; AVX1-NEXT:    vmovntdqa 16(%rdi), %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_load_nt4xdouble:
@@ -621,7 +629,11 @@ define <32 x i8> @test_load_nt32xi8(<32 x i8>* nocapture %ptr) {
 ;
 ; AVX1-LABEL: test_load_nt32xi8:
 ; AVX1:       # BB#0: # %entry
-; AVX1-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX1-NEXT:    vmovntdqa (%rdi), %xmm0
+; AVX1-NEXT:    # implicit-def: %YMM1
+; AVX1-NEXT:    vmovaps %xmm0, %xmm1
+; AVX1-NEXT:    vmovntdqa 16(%rdi), %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_load_nt32xi8:
@@ -659,7 +671,11 @@ define <16 x i16> @test_load_nt16xi16(<16 x i16>* nocapture %ptr) {
 ;
 ; AVX1-LABEL: test_load_nt16xi16:
 ; AVX1:       # BB#0: # %entry
-; AVX1-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX1-NEXT:    vmovntdqa (%rdi), %xmm0
+; AVX1-NEXT:    # implicit-def: %YMM1
+; AVX1-NEXT:    vmovaps %xmm0, %xmm1
+; AVX1-NEXT:    vmovntdqa 16(%rdi), %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_load_nt16xi16:
@@ -697,7 +713,11 @@ define <8 x i32> @test_load_nt8xi32(<8 x i32>* nocapture %ptr) {
 ;
 ; AVX1-LABEL: test_load_nt8xi32:
 ; AVX1:       # BB#0: # %entry
-; AVX1-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX1-NEXT:    vmovntdqa (%rdi), %xmm0
+; AVX1-NEXT:    # implicit-def: %YMM1
+; AVX1-NEXT:    vmovaps %xmm0, %xmm1
+; AVX1-NEXT:    vmovntdqa 16(%rdi), %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_load_nt8xi32:
@@ -735,7 +755,11 @@ define <4 x i64> @test_load_nt4xi64(<4 x i64>* nocapture %ptr) {
 ;
 ; AVX1-LABEL: test_load_nt4xi64:
 ; AVX1:       # BB#0: # %entry
-; AVX1-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX1-NEXT:    vmovntdqa (%rdi), %xmm0
+; AVX1-NEXT:    # implicit-def: %YMM1
+; AVX1-NEXT:    vmovaps %xmm0, %xmm1
+; AVX1-NEXT:    vmovntdqa 16(%rdi), %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_load_nt4xi64: