[PATCH] D89566: [LV] Epilogue Vectorization with Optimal Control Flow

Sat Nov 7 09:18:00 PST 2020

Prashanth added a comment.

In D89566#2379543 <https://reviews.llvm.org/D89566#2379543>, @bmahjour wrote:

> In D89566#2379500 <https://reviews.llvm.org/D89566#2379500>, @Prashanth wrote:
>
>> In D89566#2379432 <https://reviews.llvm.org/D89566#2379432>, @bmahjour wrote:
>>
>>>> Just checking/summarising: is there consensus that this is the patch that is going to do epilogue vectorisation, and not D88819 <https://reviews.llvm.org/D88819>? Or does that depend on perf numbers as Florian suggested?
>>>
>>> I think we should go with this, unless @fhahn finds that D88819 <https://reviews.llvm.org/D88819> can outperform it (unlikely). There are more loops that can be transformed with this patch and I've listed other advantages in the description. @mivnay do you agree?
>>
>> Are you blocking D88819 <https://reviews.llvm.org/D88819> without running benchmarks?  i.e. Without a comparison on any one of the architectures. For x264, D88819 <https://reviews.llvm.org/D88819> might give better numbers....
>
> Have you been following the conversations in D88819 <https://reviews.llvm.org/D88819> and this revision? Blocking work has certainly not been my intention. I just resigned from D88819 <https://reviews.llvm.org/D88819>, in case people would like to move ahead with it. In terms of testing I've verified that x264 opportunity can be caught on POWER, however I do not have access to other architectures, which is why I would appreciate help on testing on other platforms.

It would be great if you are reviewer of the other patch.  Let us run the benchmarks on Graviton/Thunderx2/X86/Power and see what happens..

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D89566/new/

https://reviews.llvm.org/D89566