diff options
| author | Owen Anderson <resistor@mac.com> | 2008-02-18 09:24:53 +0000 |
|---|---|---|
| committer | Owen Anderson <resistor@mac.com> | 2008-02-18 09:24:53 +0000 |
| commit | 5aa4f2a0857563b4ad9115c614afed9501aa58f4 (patch) | |
| tree | d1def04f316ef138fc6afe4cb5a91349cfe71054 /test | |
| parent | 874a892c9969fbbe50a5017868c2f82923632b29 (diff) | |
| download | external_llvm-5aa4f2a0857563b4ad9115c614afed9501aa58f4.zip external_llvm-5aa4f2a0857563b4ad9115c614afed9501aa58f4.tar.gz external_llvm-5aa4f2a0857563b4ad9115c614afed9501aa58f4.tar.bz2 | |
Add support to GVN for performing sret return slot optimization. This means that, if an sret function tail calls
another sret function, it should pass its own sret parameter to the tail callee, allowing it to fill in the correct
return value. llvm-gcc does not emit this by default. Instead, it allocates space in the caller for the sret of
the tail call and then uses memcpy to copy the result into the caller's sret parameter. This optimization detects
and optimizes that case.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47265 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test')
| -rw-r--r-- | test/Transforms/GVN/sret.ll | 28 |
1 files changed, 28 insertions, 0 deletions
diff --git a/test/Transforms/GVN/sret.ll b/test/Transforms/GVN/sret.ll new file mode 100644 index 0000000..9ae73ef --- /dev/null +++ b/test/Transforms/GVN/sret.ll @@ -0,0 +1,28 @@ +; RUN: llvm-as < %s | opt -gvn | llvm-dis | grep memcpy | count 1 + +target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128" +target triple = "i686-apple-darwin9" + +define void @ccosl({ x86_fp80, x86_fp80 }* noalias sret %agg.result, { x86_fp80, x86_fp80 }* byval %z) nounwind { +entry: + %iz = alloca { x86_fp80, x86_fp80 } ; <{ x86_fp80, x86_fp80 }*> [#uses=3] + %memtmp = alloca { x86_fp80, x86_fp80 }, align 16 ; <{ x86_fp80, x86_fp80 }*> [#uses=2] + %tmp1 = getelementptr { x86_fp80, x86_fp80 }* %z, i32 0, i32 1 ; <x86_fp80*> [#uses=1] + %tmp2 = load x86_fp80* %tmp1, align 16 ; <x86_fp80> [#uses=1] + %tmp3 = sub x86_fp80 0xK80000000000000000000, %tmp2 ; <x86_fp80> [#uses=1] + %tmp4 = getelementptr { x86_fp80, x86_fp80 }* %iz, i32 0, i32 1 ; <x86_fp80*> [#uses=1] + %real = getelementptr { x86_fp80, x86_fp80 }* %iz, i32 0, i32 0 ; <x86_fp80*> [#uses=1] + %tmp7 = getelementptr { x86_fp80, x86_fp80 }* %z, i32 0, i32 0 ; <x86_fp80*> [#uses=1] + %tmp8 = load x86_fp80* %tmp7, align 16 ; <x86_fp80> [#uses=1] + store x86_fp80 %tmp3, x86_fp80* %real, align 16 + store x86_fp80 %tmp8, x86_fp80* %tmp4, align 16 + call void @ccoshl( { x86_fp80, x86_fp80 }* noalias sret %memtmp, { x86_fp80, x86_fp80 }* byval %iz ) nounwind + %memtmp14 = bitcast { x86_fp80, x86_fp80 }* %memtmp to i8* ; <i8*> [#uses=1] + %agg.result15 = bitcast { x86_fp80, x86_fp80 }* %agg.result to i8* ; <i8*> [#uses=1] + call void @llvm.memcpy.i32( i8* %agg.result15, i8* %memtmp14, i32 32, i32 16 ) + ret void +} + +declare void @ccoshl({ x86_fp80, x86_fp80 }* noalias sret , { x86_fp80, x86_fp80 }* byval ) nounwind + +declare void @llvm.memcpy.i32(i8*, i8*, i32, i32) nounwind |