[PATCH] D118142: [PowerPC] Update ppc-prevent-mma-types.ll with custom opt pipeline

[Amy Kwan via Phabricator via llvm-commits]

amyk created this revision.
amyk added a reviewer: fhahn.
amyk added a project: LLVM.
Herald added subscribers: wenlei, shchenz, kbarton, nemanjai.
amyk requested review of this revision.

The `ppc-prevent-mma-types.ll` test that was committed in D113900 <https://reviews.llvm.org/D113900> had an issue
where wide vectors were being produced during PGO, and the SLP vectorizer appeared
to be an important pass to reproduce the issue seen in the original revision. However,
the original test had  the `-pgo-kind=pgo-instr-gen-pipeline`, which does not explicitly
communicate that the SLP vectorizer affected to test case to produce the behaviour
that was seen.

This patch updates the opt pass pipeline to a minimal set of passes needed to reproduce the issue.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D118142

Files:
  llvm/test/Transforms/PGOProfile/ppc-prevent-mma-types.ll


Index: llvm/test/Transforms/PGOProfile/ppc-prevent-mma-types.ll
===================================================================
--- llvm/test/Transforms/PGOProfile/ppc-prevent-mma-types.ll
+++ llvm/test/Transforms/PGOProfile/ppc-prevent-mma-types.ll
@@ -1,8 +1,19 @@
-; RUN: opt --vec-extabi=true -passes='default<O3>' -mcpu=pwr10 \
-; RUN:   -pgo-kind=pgo-instr-gen-pipeline -mtriple=powerpc-ibm-aix -S < %s | \
-; 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
+; RUN: opt --vec-extabi=true -mcpu=pwr10 -mtriple=powerpc-ibm-aix \
+; RUN:   -do-counter-promotion -unroll-threshold=300 \
+; RUN:   -passes='function(sroa,early-cse<>),pgo-instr-gen,\
+; RUN:function(loop(loop-rotate)),instrprof,\
+; RUN:function(instcombine,loop-mssa(licm,simple-loop-unswitch<nontrivial>),\
+; RUN:loop(loop-unroll-full)),\
+; RUN:function(loop-vectorize<no-interleave-forced-only>,slp-vectorizer)' \
+; RUN:  -S < %s | FileCheck %s
+; RUN: opt -mcpu=pwr10 -mtriple=powerpc64le-unknown-linux-gnu \
+; RUN:   -do-counter-promotion -unroll-threshold=300 \
+; RUN:   -passes='function(sroa,early-cse<>),pgo-instr-gen,\
+; RUN:function(loop(loop-rotate)),instrprof,\
+; RUN:function(instcombine,loop-mssa(licm,simple-loop-unswitch<nontrivial>),\
+; RUN:loop(loop-unroll-full)),\
+; RUN:function(loop-vectorize<no-interleave-forced-only>,slp-vectorizer)' \
+; RUN:  -S < %s | FileCheck %s
 
 ; When running this test case under opt + PGO, the SLPVectorizer previously had
 ; an opportunity to produce wide vector types (such as <256 x i1>) within the


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D118142.402890.patch
Type: text/x-patch
Size: 1659 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220125/3d9ad979/attachment.bin>