[PATCH] D113900: [PowerPC] Prevent the optimizer from producing wide vector types in IR.

Amy Kwan via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Tue Jan 25 06:45:43 PST 2022


amyk added inline comments.


================
Comment at: llvm/test/Transforms/PGOProfile/ppc-prevent-mma-types.ll:4
+; RUN: FileCheck %s
+; RUN: opt -passes='default<O3>' -mcpu=pwr10 -pgo-kind=pgo-instr-gen-pipeline \
+; RUN:   -mtriple=powerpc64le-unknown-linux-gnu -S < %s | FileCheck %s
----------------
fhahn wrote:
> Is there a reason this test checks the full pipeline compare to just the SLP vectorizer pass? The changes look completely unrelated to PGO/profiling.
Thanks for taking a look at this. I originally found this test case under the full PGO pipeline. Then, I realized the SLP vectorizer was also an important culprit in reproducing the issue, but it appears to be a bit more complex than that. It seems like there are other passes that come into play when trying to reproduce this issue.

I've attempted to find a minimal set of passes and I've posted a patch to hopefully be a bit more specific with the pass pipeline: https://reviews.llvm.org/D118142
I'd appreciate your opinion on the updated pass pipeline. 


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D113900/new/

https://reviews.llvm.org/D113900



More information about the llvm-commits mailing list