diff options
| author | Evan Cheng <evan.cheng@apple.com> | 2010-01-06 19:38:29 +0000 |
|---|---|---|
| committer | Evan Cheng <evan.cheng@apple.com> | 2010-01-06 19:38:29 +0000 |
| commit | d40d03e1bd1d51857fc2f9f9230e334c3a32b249 (patch) | |
| tree | cc28a8f045a22783624b225433ca137ab6a815d9 /test/CodeGen/SystemZ | |
| parent | b419a5cfe9fae934d6c05da173934673f5402c99 (diff) | |
| download | llvm-d40d03e1bd1d51857fc2f9f9230e334c3a32b249.tar.gz llvm-d40d03e1bd1d51857fc2f9f9230e334c3a32b249.tar.bz2 llvm-d40d03e1bd1d51857fc2f9f9230e334c3a32b249.tar.xz | |
Teach dag combine to fold the following transformation more aggressively:
(OP (trunc x), (trunc y)) -> (trunc (OP x, y))
Unfortunately this simple change causes dag combine to infinite looping. The problem is the shrink demanded ops optimization tend to canonicalize expressions in the opposite manner. That is badness. This patch disable those optimizations in dag combine but instead it is done as a late pass in sdisel.
This also exposes some deficiencies in dag combine and x86 setcc / brcond lowering. Teach them to look pass ISD::TRUNCATE in various places.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@92849 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test/CodeGen/SystemZ')
| -rw-r--r-- | test/CodeGen/SystemZ/2009-06-02-Rotate.ll | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/test/CodeGen/SystemZ/2009-06-02-Rotate.ll b/test/CodeGen/SystemZ/2009-06-02-Rotate.ll index 3317864c01..07a164d426 100644 --- a/test/CodeGen/SystemZ/2009-06-02-Rotate.ll +++ b/test/CodeGen/SystemZ/2009-06-02-Rotate.ll @@ -5,8 +5,8 @@ target triple = "s390x-linux" define i32 @rotl(i32 %x, i32 %y, i32 %z) nounwind readnone { entry: - %shl = shl i32 %x, 0 ; <i32> [#uses=1] - %sub = sub i32 32, 0 ; <i32> [#uses=1] + %shl = shl i32 %x, 1 ; <i32> [#uses=1] + %sub = sub i32 32, 1 ; <i32> [#uses=1] %shr = lshr i32 %x, %sub ; <i32> [#uses=1] %or = or i32 %shr, %shl ; <i32> [#uses=1] ret i32 %or |