diff options
| author | Arnold Schwaighofer <aschwaighofer@apple.com> | 2014-01-19 03:18:31 +0000 |
|---|---|---|
| committer | Arnold Schwaighofer <aschwaighofer@apple.com> | 2014-01-19 03:18:31 +0000 |
| commit | cc742dd9e4e9b625733fd0de83476da461f9efe0 (patch) | |
| tree | 443927de52fa2a8410d6960720f074a7021522f8 /llvm/test/Transforms/LoopVectorize | |
| parent | 043949d4465f1c49557e4ac1b9a1646ed6c80312 (diff) | |
| download | bcm5719-llvm-cc742dd9e4e9b625733fd0de83476da461f9efe0.tar.gz bcm5719-llvm-cc742dd9e4e9b625733fd0de83476da461f9efe0.zip | |
LoopVectorizer: A reduction that has multiple uses of the reduction value is not
a reduction.
Really. Under certain circumstances (the use list of an instruction has to be
set up right - hence the extra pass in the test case) we would not recognize
when a value in a potential reduction cycle was used multiple times by the
reduction cycle.
Fixes PR18526.
radar://15851149
llvm-svn: 199570
Diffstat (limited to 'llvm/test/Transforms/LoopVectorize')
| -rw-r--r-- | llvm/test/Transforms/LoopVectorize/multi-use-reduction-bug.ll | 42 |
1 files changed, 42 insertions, 0 deletions
diff --git a/llvm/test/Transforms/LoopVectorize/multi-use-reduction-bug.ll b/llvm/test/Transforms/LoopVectorize/multi-use-reduction-bug.ll new file mode 100644 index 00000000000..88a29c50df5 --- /dev/null +++ b/llvm/test/Transforms/LoopVectorize/multi-use-reduction-bug.ll @@ -0,0 +1,42 @@ +; RUN: opt -indvars -loop-vectorize -force-vector-width=2 -force-vector-unroll=1 -S < %s | FileCheck %s + +target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" +target triple = "x86_64-apple-macosx10.9.0" + +; We must not vectorize this loop. %add55 is not reduction. Its value is used +; multiple times. + +; PR18526 + +; CHECK: multiple_use_of_value +; CHECK-NOT: <2 x i32> + +define void @multiple_use_of_value() { +entry: + %n = alloca i32, align 4 + %k7 = alloca i32, align 4 + %nf = alloca i32, align 4 + %0 = load i32* %k7, align 4 + %.neg1 = sub i32 0, %0 + %n.promoted = load i32* %n, align 4 + %nf.promoted = load i32* %nf, align 4 + br label %for.body + +for.body: + %inc107 = phi i32 [ undef, %entry ], [ %inc10, %for.body ] + %inc6 = phi i32 [ %nf.promoted, %entry ], [ undef, %for.body ] + %add55 = phi i32 [ %n.promoted, %entry ], [ %add5, %for.body ] + %.neg2 = sub i32 0, %inc6 + %add.neg = add i32 0, %add55 + %add4.neg = add i32 %add.neg, %.neg1 + %sub = add i32 %add4.neg, %.neg2 + %add5 = add i32 %sub, %add55 + %inc10 = add i32 %inc107, 1 + %cmp = icmp ult i32 %inc10, 61 + br i1 %cmp, label %for.body, label %for.end + +for.end: + %add5.lcssa = phi i32 [ %add5, %for.body ] + store i32 %add5.lcssa, i32* %n, align 4 + ret void +} |
