[PATCH] D112117: [llvm][LV] Drop poison-generating flags from FP instructions

Sanjay Patel via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Thu Oct 28 06:49:56 PDT 2021


spatel added a comment.

In D112117#3092834 <https://reviews.llvm.org/D112117#3092834>, @dcaballe wrote:

> In D112117#3090902 <https://reviews.llvm.org/D112117#3090902>, @spatel wrote:
>
>> I didn't go through all of the tests, but the first reduction expansion diff seems unintended to me (see inline comment). 
>> We could enable dropping FMF only with an optional bool parameter? That would allow incremental updates. 
>> Or we handle that as a one-off in the motivating LV caller?
>
> Good catch! I like the optional bool parameter. Honestly, I don't feel very comfortable with dropping all the FMF flags by default since I don't understand in detail all the users of this utility. Let's go with that.

Sounds good...I was probably too optimistic when I wrote that it would just be minor regressions (perf) if we got this wrong. :)
We need to audit the callers to decide if/when FMF can be propagated.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112117/new/

https://reviews.llvm.org/D112117



More information about the llvm-commits mailing list