diff options
Diffstat (limited to 'llvm/test')
| -rw-r--r-- | llvm/test/CodeGen/X86/avx-splat.ll | 9 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/widen_cast-4.ll | 25 |
2 files changed, 26 insertions, 8 deletions
diff --git a/llvm/test/CodeGen/X86/avx-splat.ll b/llvm/test/CodeGen/X86/avx-splat.ll index 5d0781531f4..b1b2f8b97a7 100644 --- a/llvm/test/CodeGen/X86/avx-splat.ll +++ b/llvm/test/CodeGen/X86/avx-splat.ll @@ -43,13 +43,10 @@ entry: ret <4 x double> %vecinit6.i } -; Test this simple opt: +; Test this turns into a broadcast: ; shuffle (scalar_to_vector (load (ptr + 4))), undef, <0, 0, 0, 0> -; To: -; shuffle (vload ptr)), undef, <1, 1, 1, 1> -; CHECK: vmovdqa -; CHECK-NEXT: vpshufd $-1 -; CHECK-NEXT: vinsertf128 $1 +; +; CHECK: vbroadcastss define <8 x float> @funcE() nounwind { allocas: %udx495 = alloca [18 x [18 x float]], align 32 diff --git a/llvm/test/CodeGen/X86/widen_cast-4.ll b/llvm/test/CodeGen/X86/widen_cast-4.ll index 1bc06a77cbf..19b84f19a4f 100644 --- a/llvm/test/CodeGen/X86/widen_cast-4.ll +++ b/llvm/test/CodeGen/X86/widen_cast-4.ll @@ -1,8 +1,9 @@ ; RUN: llc < %s -march=x86 -mattr=+sse4.2 | FileCheck %s -; CHECK: psraw -; CHECK: psraw +; RUN: llc < %s -march=x86 -mattr=+sse4.2 -x86-experimental-vector-widening-legalization | FileCheck %s --check-prefix=CHECK-WIDE define void @update(i64* %dst_i, i64* %src_i, i32 %n) nounwind { +; CHECK-LABEL: update: +; CHECK-WIDE-LABEL: update: entry: %dst_i.addr = alloca i64* ; <i64**> [#uses=2] %src_i.addr = alloca i64* ; <i64**> [#uses=2] @@ -44,6 +45,26 @@ forbody: ; preds = %forcond %shr = ashr <8 x i8> %add, < i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2 > ; <<8 x i8>> [#uses=1] store <8 x i8> %shr, <8 x i8>* %arrayidx10 br label %forinc +; CHECK: %forbody +; CHECK: pmovzxbw +; CHECK-NEXT: paddw +; CHECK-NEXT: psllw $8 +; CHECK-NEXT: psraw $8 +; CHECK-NEXT: psraw $2 +; CHECK-NEXT: pshufb +; CHECK-NEXT: movlpd +; +; FIXME: We shouldn't require both a movd and an insert. +; CHECK-WIDE: %forbody +; CHECK-WIDE: movd +; CHECK-WIDE-NEXT: pinsrd +; CHECK-WIDE-NEXT: paddb +; CHECK-WIDE-NEXT: psrlw $2 +; CHECK-WIDE-NEXT: pand +; CHECK-WIDE-NEXT: pxor +; CHECK-WIDE-NEXT: psubb +; CHECK-WIDE-NEXT: pextrd +; CHECK-WIDE-NEXT: movd forinc: ; preds = %forbody %tmp15 = load i32* %i ; <i32> [#uses=1] |
