[WinEH] Fix stack alignment in funclets and ParentFrameOffset calculation

[oota-llvm.git] / test / CodeGen / X86 / nontemporal-2.ll

diff --git a/test/CodeGen/X86/nontemporal-2.ll b/test/CodeGen/X86/nontemporal-2.ll
index 8c08b3c163c0249493ce15abecad80f1ca422c90..c9767f88488c9d7b0b0ecd4a4d0e7ff0f8470d04 100644 (file)
--- a/test/CodeGen/X86/nontemporal-2.ll
+++ b/test/CodeGen/X86/nontemporal-2.ll
@@ -1,6 +1,6 @@
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=corei7 | FileCheck %s -check-prefix=CHECK -check-prefix=SSE
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=corei7-avx | FileCheck %s -check-prefix=CHECK -check-prefix=AVX
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=core-avx2 | FileCheck %s -check-prefix=CHECK -check-prefix=AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s -check-prefix=CHECK -check-prefix=SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s -check-prefix=CHECK -check-prefix=AVX
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s -check-prefix=CHECK -check-prefix=AVX2
 
 ; Make sure that we generate non-temporal stores for the test cases below.
 ; We use xorps for zeroing, so domain information isn't available anymore.
@@ -300,4 +300,19 @@ define void @test_op_v32i8(<32 x i8> %a, <32 x i8> %b, <32 x i8>* %dst) {
   ret void
 }
 
+; 256-bit NT stores require 256-bit alignment.
+; FIXME: For AVX, we could lower this to 2x movntps %xmm. Taken further, we
+; could even scalarize to movnti when we have 1-alignment: nontemporal is
+; probably always worth even some 20 instruction scalarization.
+define void @test_unaligned_v8f32(<8 x float> %a, <8 x float> %b, <8 x float>* %dst) {
+; CHECK-LABEL: test_unaligned_v8f32:
+; SSE: movntps %xmm
+; SSE: movntps %xmm
+; AVX-NOT: movnt
+; AVX: vmovups %ymm
+  %r = fadd <8 x float> %a, %b
+  store <8 x float> %r, <8 x float>* %dst, align 16, !nontemporal !1
+  ret void
+}
+
 !1 = !{i32 1}