diff options
| author | Jingyue Wu <jingyue@google.com> | 2015-06-24 19:28:40 +0000 |
|---|---|---|
| committer | Jingyue Wu <jingyue@google.com> | 2015-06-24 19:28:40 +0000 |
| commit | 6f72aed3ecda230814e1701d96f235d9a04b84af (patch) | |
| tree | fc22625fadcab9e965646a900c63c4f02fd3d664 /llvm/test/Transforms/LoopReroll | |
| parent | c1ffd69c3424dc8cdb3a24f18cd3e4901ce918aa (diff) | |
| download | bcm5719-llvm-6f72aed3ecda230814e1701d96f235d9a04b84af.tar.gz bcm5719-llvm-6f72aed3ecda230814e1701d96f235d9a04b84af.zip | |
[LSR] canonicalize Prod*(1<<C) to Prod<<C
Summary:
Because LSR happens at a late stage where mul of a power of 2 is
typically canonicalized to shl, this canonicalization emits code that
can be better CSE'ed.
Test Plan:
Transforms/LoopStrengthReduce/shl.ll shows how this change makes GVN more
powerful. Fixes some existing tests due to this change.
Reviewers: sanjoy, majnemer, atrick
Reviewed By: majnemer, atrick
Subscribers: majnemer, llvm-commits
Differential Revision: http://reviews.llvm.org/D10448
llvm-svn: 240573
Diffstat (limited to 'llvm/test/Transforms/LoopReroll')
| -rw-r--r-- | llvm/test/Transforms/LoopReroll/nonconst_lb.ll | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/llvm/test/Transforms/LoopReroll/nonconst_lb.ll b/llvm/test/Transforms/LoopReroll/nonconst_lb.ll index 5effa42d07c..96090e8dc0b 100644 --- a/llvm/test/Transforms/LoopReroll/nonconst_lb.ll +++ b/llvm/test/Transforms/LoopReroll/nonconst_lb.ll @@ -52,7 +52,7 @@ for.end: ; preds = %for.body, %entry ; CHECK: %0 = add i32 %n, -1 ; CHECK: %1 = sub i32 %0, %m ; CHECK: %2 = lshr i32 %1, 2 -; CHECK: %3 = mul i32 %2, 4 +; CHECK: %3 = shl i32 %2, 2 ; CHECK: %4 = add i32 %m, %3 ; CHECK: %5 = add i32 %4, 3 ; CHECK: br label %for.body @@ -132,7 +132,7 @@ for.end: ; preds = %for.body, %entry ; CHECK: %0 = add i32 %n, -1 ; CHECK: %1 = sub i32 %0, %rem ; CHECK: %2 = lshr i32 %1, 2 -; CHECK: %3 = mul i32 %2, 4 +; CHECK: %3 = shl i32 %2, 2 ; CHECK: %4 = add i32 %rem, %3 ; CHECK: %5 = add i32 %4, 3 ; CHECK: br label %for.body |
