[PATCH] Fix PR18014

Michael Liao michael.liao at intel.com
Fri Nov 22 09:42:51 PST 2013


Hi All,

The attached patch fixes PR18014 which is caused by an over optimization
in x86 DAG combining. To generate efficient code for BLEND, we simplify
the mask generation by only ensure to generate the MSB. If that mask is
also consumed by other non-BLEND insns, the code is broken. This patch
checks whether that mask is only consumed by BLEND eventually and skip
if it's not true.

Thanks for review

Yours
- Michael
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-Fix-PR18014.patch
Type: text/x-patch
Size: 2195 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20131122/92792075/attachment.bin>


More information about the llvm-commits mailing list