maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

R600: Remove unnecessary part of computeKnownBitsForTargetNode 

Zero-width BFEs are combined away already, so there's no point in
handling them.

llvm-svn: 219868
This commit is contained in:
Matt Arsenault 2014-10-15 23:37:49 +00:00
parent 6de7af4242
commit 20893b3611
1 changed files with 0 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
llvm/lib/Target/R600/AMDGPUISelLowering.cpp
View File

@ -2388,11 +2388,6 @@ void AMDGPUTargetLowering::computeKnownBitsForTargetNode(
unsigned BitWidth = 32;
uint32_t Width = CWidth->getZExtValue() & 0x1f;
if (Width == 0) {
KnownZero = APInt::getAllOnesValue(BitWidth);
KnownOne = APInt::getNullValue(BitWidth);
return;
}
// FIXME: This could do a lot more. If offset is 0, should be the same as
// sign_extend_inreg implementation, but that involves duplicating it.