[PATCH] [NaryReassociate] Run EarlyCSE/GVN after NaryReassociate

[Daniel Berlin]

FWIW: Rerunning GVN can be *really* expensive.
I can make EarlyCSE really fast by removing the scoped hash table if
that works for you instead?



On Fri, May 22, 2015 at 2:14 PM, Jingyue Wu <jingyue at google.com> wrote:
> Hi broune, meheff,
>
> This patch made two improvements to NaryReassociate and the NVPTX pipeline
>
> 1. Run EarlyCSE/GVN after NaryReassociate to get rid of redundant common
> expressions.
>
> 2. When adding an instruction to SeenExprs, maps both the SCEV before and after
> reassociation to that instruction.
>
> http://reviews.llvm.org/D9947
>
> Files:
>   lib/Target/NVPTX/NVPTXTargetMachine.cpp
>   lib/Transforms/Scalar/NaryReassociate.cpp
>   test/Transforms/NaryReassociate/NVPTX/nary-gep.ll
>
> EMAIL PREFERENCES
>   http://reviews.llvm.org/settings/panel/emailpreferences/
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>