diff options
| author | Arnold Schwaighofer <aschwaighofer@apple.com> | 2013-10-29 01:33:57 +0000 |
|---|---|---|
| committer | Arnold Schwaighofer <aschwaighofer@apple.com> | 2013-10-29 01:33:57 +0000 |
| commit | 89ae217422b6af9665bd6da578d69be0afa44da5 (patch) | |
| tree | 33c4059374daa0c896f23bfae61d7083f16b343b /llvm/test/Transforms/SLPVectorizer | |
| parent | 77af0f6e82ca671ecc72f295d88d6123d5028bc4 (diff) | |
| download | bcm5719-llvm-89ae217422b6af9665bd6da578d69be0afa44da5.tar.gz bcm5719-llvm-89ae217422b6af9665bd6da578d69be0afa44da5.zip | |
ARM cost model: Unaligned vectorized double stores are expensive
Updated a test case that assumed that <2 x double> would vectorize to use
<4 x float>.
radar://15338229
llvm-svn: 193574
Diffstat (limited to 'llvm/test/Transforms/SLPVectorizer')
| -rw-r--r-- | llvm/test/Transforms/SLPVectorizer/ARM/memory.ll | 20 |
1 files changed, 20 insertions, 0 deletions
diff --git a/llvm/test/Transforms/SLPVectorizer/ARM/memory.ll b/llvm/test/Transforms/SLPVectorizer/ARM/memory.ll new file mode 100644 index 00000000000..383c808d21c --- /dev/null +++ b/llvm/test/Transforms/SLPVectorizer/ARM/memory.ll @@ -0,0 +1,20 @@ +; RUN: opt < %s -basicaa -slp-vectorizer -S -mtriple=thumbv7-apple-ios3.0.0 -mcpu=swift | FileCheck %s + +target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:32:64-v128:32:128-a0:0:32-n32-S32" + +; On swift unaligned <2 x double> stores need 4uops and it is there for cheaper +; to do this scalar. + +; CHECK-LABEL: expensive_double_store +; CHECK-NOT: load <2 x double> +; CHECK-NOT: store <2 x double> +define void @expensive_double_store(double* noalias %dst, double* noalias %src, i64 %count) { +entry: + %0 = load double* %src, align 8 + store double %0, double* %dst, align 8 + %arrayidx2 = getelementptr inbounds double* %src, i64 1 + %1 = load double* %arrayidx2, align 8 + %arrayidx3 = getelementptr inbounds double* %dst, i64 1 + store double %1, double* %arrayidx3, align 8 + ret void +} |
