From: Craig Topper <craig.topper@gmail.com>
Date: Tue, 31 Jan 2017 06:49:50 +0000 (+0000)
Subject: [AVX-512] Fix copy and paste bug in execution domain fixing tables so that we can... 
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=bde6e7c9dea7d564f3d8ebd3beb70fdb1298a980;p=llvm

[AVX-512] Fix copy and paste bug in execution domain fixing tables so that we can convert 256-bit movnt instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@293606 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index 856fbf78037..cd030630a23 100644
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -8669,7 +8669,7 @@ static const uint16_t ReplaceableInstrs[][3] = {
   // AVX512 support
   { X86::VMOVLPSZ128mr,  X86::VMOVLPDZ128mr,  X86::VMOVPQI2QIZmr  },
   { X86::VMOVNTPSZ128mr, X86::VMOVNTPDZ128mr, X86::VMOVNTDQZ128mr },
-  { X86::VMOVNTPSZ128mr, X86::VMOVNTPDZ128mr, X86::VMOVNTDQZ128mr },
+  { X86::VMOVNTPSZ256mr, X86::VMOVNTPDZ256mr, X86::VMOVNTDQZ256mr },
   { X86::VMOVNTPSZmr,    X86::VMOVNTPDZmr,    X86::VMOVNTDQZmr    },
   { X86::VMOVSDZmr,      X86::VMOVSDZmr,      X86::VMOVPQI2QIZmr  },
   { X86::VMOVSSZmr,      X86::VMOVSSZmr,      X86::VMOVPDI2DIZmr  },
diff --git a/test/CodeGen/X86/nontemporal-2.ll b/test/CodeGen/X86/nontemporal-2.ll
index 92a35436d90..3d280b4e6bf 100644
--- a/test/CodeGen/X86/nontemporal-2.ll
+++ b/test/CodeGen/X86/nontemporal-2.ll
@@ -777,7 +777,7 @@ define void @test_arg_v8i32(<8 x i32> %arg, <8 x i32>* %dst) {
 ;
 ; VLX-LABEL: test_arg_v8i32:
 ; VLX:       # BB#0:
-; VLX-NEXT:    vmovntdq %ymm0, (%rdi)
+; VLX-NEXT:    vmovntps %ymm0, (%rdi)
 ; VLX-NEXT:    retq
   store <8 x i32> %arg, <8 x i32>* %dst, align 32, !nontemporal !1
   ret void
@@ -798,7 +798,7 @@ define void @test_arg_v4f64(<4 x double> %arg, <4 x double>* %dst) {
 ;
 ; VLX-LABEL: test_arg_v4f64:
 ; VLX:       # BB#0:
-; VLX-NEXT:    vmovntpd %ymm0, (%rdi)
+; VLX-NEXT:    vmovntps %ymm0, (%rdi)
 ; VLX-NEXT:    retq
   store <4 x double> %arg, <4 x double>* %dst, align 32, !nontemporal !1
   ret void
@@ -819,7 +819,7 @@ define void @test_arg_v4i64(<4 x i64> %arg, <4 x i64>* %dst) {
 ;
 ; VLX-LABEL: test_arg_v4i64:
 ; VLX:       # BB#0:
-; VLX-NEXT:    vmovntdq %ymm0, (%rdi)
+; VLX-NEXT:    vmovntps %ymm0, (%rdi)
 ; VLX-NEXT:    retq
   store <4 x i64> %arg, <4 x i64>* %dst, align 32, !nontemporal !1
   ret void
@@ -840,7 +840,7 @@ define void @test_arg_v16i16(<16 x i16> %arg, <16 x i16>* %dst) {
 ;
 ; VLX-LABEL: test_arg_v16i16:
 ; VLX:       # BB#0:
-; VLX-NEXT:    vmovntdq %ymm0, (%rdi)
+; VLX-NEXT:    vmovntps %ymm0, (%rdi)
 ; VLX-NEXT:    retq
   store <16 x i16> %arg, <16 x i16>* %dst, align 32, !nontemporal !1
   ret void
@@ -861,7 +861,7 @@ define void @test_arg_v32i8(<32 x i8> %arg, <32 x i8>* %dst) {
 ;
 ; VLX-LABEL: test_arg_v32i8:
 ; VLX:       # BB#0:
-; VLX-NEXT:    vmovntdq %ymm0, (%rdi)
+; VLX-NEXT:    vmovntps %ymm0, (%rdi)
 ; VLX-NEXT:    retq
   store <32 x i8> %arg, <32 x i8>* %dst, align 32, !nontemporal !1
   ret void