diff options
| author | Michael J. Spencer <bigcheesegs@gmail.com> | 2014-07-18 00:15:50 +0000 |
|---|---|---|
| committer | Michael J. Spencer <bigcheesegs@gmail.com> | 2014-07-18 00:15:50 +0000 |
| commit | 1eb023013e5e513162ece799f5b94eadfb625131 (patch) | |
| tree | 89c6068880ed82921e6ffc5b5ecb5477a9d62ca4 /llvm/test | |
| parent | 722b1df869685778662a09e92f47a5d7fda7905a (diff) | |
| download | bcm5719-llvm-1eb023013e5e513162ece799f5b94eadfb625131.tar.gz bcm5719-llvm-1eb023013e5e513162ece799f5b94eadfb625131.zip | |
Revert "[x86] Fold extract_vector_elt of a load into the Load's address computation."
There's a bug where this can create cycles in the DAG. It will take a bit
to fix, so I'm backing it out for now.
llvm-svn: 213339
Diffstat (limited to 'llvm/test')
| -rw-r--r-- | llvm/test/CodeGen/X86/vec_splat.ll | 20 |
1 files changed, 1 insertions, 19 deletions
diff --git a/llvm/test/CodeGen/X86/vec_splat.ll b/llvm/test/CodeGen/X86/vec_splat.ll index 28f2a9074cb..a02e3836078 100644 --- a/llvm/test/CodeGen/X86/vec_splat.ll +++ b/llvm/test/CodeGen/X86/vec_splat.ll @@ -1,6 +1,5 @@ ; RUN: llc < %s -march=x86 -mcpu=pentium4 -mattr=+sse2 | FileCheck %s -check-prefix=SSE2 ; RUN: llc < %s -march=x86 -mcpu=pentium4 -mattr=+sse3 | FileCheck %s -check-prefix=SSE3 -; RUN: llc < %s -march=x86-64 -mattr=+avx | FileCheck %s -check-prefix=AVX define void @test_v4sf(<4 x float>* %P, <4 x float>* %Q, float %X) nounwind { %tmp = insertelement <4 x float> zeroinitializer, float %X, i32 0 ; <<4 x float>> [#uses=1] @@ -38,23 +37,6 @@ define void @test_v2sd(<2 x double>* %P, <2 x double>* %Q, double %X) nounwind { define <4 x float> @load_extract_splat(<4 x float>* nocapture readonly %ptr, i64 %i, i64 %j) nounwind { %1 = getelementptr inbounds <4 x float>* %ptr, i64 %i %2 = load <4 x float>* %1, align 16 - %3 = trunc i64 %j to i32 - %4 = extractelement <4 x float> %2, i32 %3 - %5 = insertelement <4 x float> undef, float %4, i32 0 - %6 = insertelement <4 x float> %5, float %4, i32 1 - %7 = insertelement <4 x float> %6, float %4, i32 2 - %8 = insertelement <4 x float> %7, float %4, i32 3 - ret <4 x float> %8 - -; AVX-LABEL: load_extract_splat -; AVX-NOT: rsp -; AVX: vbroadcastss -} - -; Fold extract of a load into the load's address computation. This avoids spilling to the stack. -define <4 x float> @load_extract_splat1(<4 x float>* nocapture readonly %ptr, i64 %i, i64 %j) nounwind { - %1 = getelementptr inbounds <4 x float>* %ptr, i64 %i - %2 = load <4 x float>* %1, align 16 %3 = extractelement <4 x float> %2, i64 %j %4 = insertelement <4 x float> undef, float %3, i32 0 %5 = insertelement <4 x float> %4, float %3, i32 1 @@ -62,7 +44,7 @@ define <4 x float> @load_extract_splat1(<4 x float>* nocapture readonly %ptr, i6 %7 = insertelement <4 x float> %6, float %3, i32 3 ret <4 x float> %7 -; AVX-LABEL: load_extract_splat1 +; AVX-LABEL: load_extract_splat ; AVX-NOT: movs ; AVX: vbroadcastss } |

