[PATCH] D61726: [Pass Pipeline] Run another round of reassociation after loop pipeline

[Sanjay Patel via Phabricator via llvm-commits]

spatel added a reviewer: efriedma.
spatel added a comment.

The motivation makes sense to me, but someone else should also review this patch in case there's a better way.
We also need to know if there's a compile-time impact. Get data for building test-suite, clang itself, SPEC, or some other benchmarks?



================
Comment at: test/Transforms/LoopVectorize/X86/masked_load_store.ll:2-4
 ; RUN: opt < %s  -O3 -mcpu=corei7-avx -S | FileCheck %s -check-prefix=AVX -check-prefix=AVX1
 ; RUN: opt < %s  -O3 -mcpu=core-avx2 -S | FileCheck %s -check-prefix=AVX -check-prefix=AVX2
 ; RUN: opt < %s  -O3 -mcpu=knl -S | FileCheck %s -check-prefix=AVX512
----------------
Regardless of anything else, this test file was over-reaching, so I fixed that problem:
rL360340

If you update/rebase, this should not wiggle with this patch now.


================
Comment at: test/Transforms/Reassociate/reassociate-after-unroll.ll:1
+; RUN: opt -O2 -S < %s | FileCheck %s
+target datalayout = "e-m:e-i64:64-n32:64"
----------------
This test file belongs in test/Transforms/PhaseOrdering. I prefer to have the baseline test with complete, auto-generated checks (utils/update_test_checks.py) committed as a preliminary step, so we can see the before/after diff in this review.

If you're updating the new pass manager in this patch, this test should have another RUN line to exercise/verify that path.


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D61726/new/

https://reviews.llvm.org/D61726