[Review]: Sink multiple instructions per iteration in InstCombine

Chandler Carruth chandlerc at google.com
Thu Jun 26 17:54:21 PDT 2014


I don't immediately understand why this is such a performance improvement
here (and not elsewhere in instcombine)... Do you have any insight into
that? Is it a locality problem that we don't re-visit the other sinkable
instructions in the right order? It feels pretty weird to have a
worklist-inside-a-worklist, so I'm just trynig to understand it better...

Also, the patch has a bunch of typos in comments and uses C-style
comments...


On Mon, Jun 23, 2014 at 10:57 PM, Aditya Nandakumar <
aditya_nandakumar at apple.com> wrote:

> Hi All
>
> I want to make a small optimization for instcombine. While sinking
> instructions (with single use) to the use block, also check if any other
> instructions which are used in the current instruction (operands) can be
> sunk.
>
> This speeds up InstCombine time in several cases as previously only one
> instruction would be sunk per iteration.
>
> Please review
>
> Thanks
> Aditya
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140627/51ef6e53/attachment.html>


More information about the llvm-commits mailing list