Make x86 fast-isel correctly choose between aligned and unaligned operations for vector stores. Fixes PR16640.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@186491 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 49 insertions, 19 deletions

diff --git a/test/CodeGen/X86/2011-10-18-FastISel-VectorParams.ll b/test/CodeGen/X86/2011-10-18-FastISel-VectorParams.ll
index 8c09d97f08..e7d1e194d9 100644
--- a/test/CodeGen/X86/2011-10-18-FastISel-VectorParams.ll
+++ b/test/CodeGen/X86/2011-10-18-FastISel-VectorParams.ll
@@ -20,7 +20,7 @@ entry:
   %2 = load <4 x float>* %p3, align 16
   %3 = load <4 x float>* %p4, align 16
   %4 = load <4 x float>* %p5, align 16
-; CHECK:      movaps {{%xmm[0-7]}}, (%esp)
+; CHECK:      movups {{%xmm[0-7]}}, (%esp)
 ; CHECK-NEXT: calll _dovectortest 
   call void @dovectortest(<4 x float> %0, <4 x float> %1, <4 x float> %2, <4 x float> %3, <4 x float> %4)
   ret void
diff --git a/test/CodeGen/X86/fast-isel-store.ll b/test/CodeGen/X86/fast-isel-store.ll
new file mode 100644
index 0000000000..06f5b6621a
--- /dev/null
+++ b/test/CodeGen/X86/fast-isel-store.ll
@@ -0,0 +1,48 @@
+; RUN: llc -mtriple=x86_64-none-linux -fast-isel -fast-isel-abort -mattr=+sse2 < %s | FileCheck %s
+; RUN: llc -mtriple=i686-none-linux -fast-isel -fast-isel-abort -mattr=+sse2 < %s | FileCheck %s
+
+define i32 @test_store_32(i32* nocapture %addr, i32 %value) {
+entry:
+  store i32 %value, i32* %addr, align 1
+  ret i32 %value
+}
+
+; CHECK: ret
+
+define i16 @test_store_16(i16* nocapture %addr, i16 %value) {
+entry:
+  store i16 %value, i16* %addr, align 1
+  ret i16 %value
+}
+
+; CHECK: ret
+
+define <4 x i32> @test_store_4xi32(<4 x i32>* nocapture %addr, <4 x i32> %value, <4 x i32> %value2) {
+; CHECK: movdqu
+; CHECK: ret
+  %foo = add <4 x i32> %value, %value2 ; to force integer type on store
+  store <4 x i32> %foo, <4 x i32>* %addr, align 1
+  ret <4 x i32> %foo
+}
+
+define <4 x i32> @test_store_4xi32_aligned(<4 x i32>* nocapture %addr, <4 x i32> %value, <4 x i32> %value2) {
+; CHECK: movdqa
+; CHECK: ret
+  %foo = add <4 x i32> %value, %value2 ; to force integer type on store
+  store <4 x i32> %foo, <4 x i32>* %addr, align 16
+  ret <4 x i32> %foo
+}
+
+define <4 x float> @test_store_4xf32(<4 x float>* nocapture %addr, <4 x float> %value) {
+; CHECK: movups
+; CHECK: ret
+  store <4 x float> %value, <4 x float>* %addr, align 1
+  ret <4 x float> %value
+}
+
+define <4 x float> @test_store_4xf32_aligned(<4 x float>* nocapture %addr, <4 x float> %value) {
+; CHECK: movaps
+; CHECK: ret
+  store <4 x float> %value, <4 x float>* %addr, align 16
+  ret <4 x float> %value
+}
diff --git a/test/CodeGen/X86/fast-isel-unaligned-store.ll b/test/CodeGen/X86/fast-isel-unaligned-store.ll
deleted file mode 100644
index 7ce7f676ad..0000000000
--- a/test/CodeGen/X86/fast-isel-unaligned-store.ll
+++ /dev/null
@@ -1,18 +0,0 @@
-; RUN: llc -mtriple=x86_64-none-linux -fast-isel -fast-isel-abort < %s | FileCheck %s
-; RUN: llc -mtriple=i686-none-linux -fast-isel -fast-isel-abort < %s | FileCheck %s
-
-define i32 @test_store_32(i32* nocapture %addr, i32 %value) {
-entry:
-  store i32 %value, i32* %addr, align 1
-  ret i32 %value
-}
-
-; CHECK: ret
-
-define i16 @test_store_16(i16* nocapture %addr, i16 %value) {
-entry:
-  store i16 %value, i16* %addr, align 1
-  ret i16 %value
-}
-
-; CHECK: ret