[PATCH] D90431: [x86] add cost overrides for mul with overflow
Sanjay Patel via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Fri Oct 30 05:05:44 PDT 2020
spatel added a comment.
In D90431#2363926 <https://reviews.llvm.org/D90431#2363926>, @pengfei wrote:
> Do we need to update the vector version cost as well?
Yes, but it will be more complicated because we will need to map operations/sizes to different ISA attributes. I think it is also less likely to see the vector versions of the intrinsics. For example, we can create this intrinsic from instcombine ( InstCombinerImpl::foldUnsignedMultiplicationOverflowCheck() ). But creating a vector version would mean that we had also encountered/created vector integer division.
My underlying motivation for this is that I am cleaning up the basic class cost model implementation of getIntrinsicInstrCost(), and if we don't have this x86 scalar override, there is a potential regression.
So for those reasons, I prefer to defer the vector implementation to follow-up patch(es). I can add a 'TODO' comment here if that is ok.
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D90431/new/
https://reviews.llvm.org/D90431
More information about the llvm-commits
mailing list