[PATCH] D93793: [IR] Let IRBuilder's CreateVectorSplat/CreateShuffleVector use poison as placeholder

Fri Jan 14 11:52:51 PST 2022

bmahjour added inline comments.

================
Comment at: llvm/test/Transforms/LoopVectorize/PowerPC/optimal-epilog-vectorization.ll:273-276
-; VF-TWO-CHECK-DAG: [[LOOPID_MV]] = distinct !{[[LOOPID_MV]], [[LOOPID_DISABLE_VECT:!.*]]}
-; VF-TWO-CHECK-DAG: [[LOOPID_EV]] = distinct !{[[LOOPID_EV]], [[LOOPID_DISABLE_UNROLL:!.*]], [[LOOPID_DISABLE_VECT:!.*]]}
-; VF-TWO-CHECK-DAG: [[LOOPID_DISABLE_VECT]] = [[DISABLE_VECT_STR:!{!"llvm.loop.isvectorized".*}.*]]
-; VF-TWO-CHECK-DAG: [[LOOPID_DISABLE_UNROLL]] = [[DISABLE_UNROLL_STR:!{!"llvm.loop.unroll.runtime.disable"}.*]]
----------------
efriedma wrote:
> bmahjour wrote:
> > these lines should not have been removed!
> It looks like someone manually hacked on the CHECK lines in a test marked "; NOTE: Assertions have been autogenerated by utils/update_test_checks.py".  This is generally a bad idea... it makes it easy to miss lines like this.  If CHECKs are manually written or modified, the NOTE should be removed as a warning to anyone modifying the test in the future.
> 
> Not sure what the best solution is here... at minimum, I guess we can just drop the "autogenerated" note?
Yes, I should have removed that comment (lesson learned), although that wouldn't necessarily have prevented the oversight.

We cannot put those lines back until the regression I reported in https://reviews.llvm.org/D116479 is fixed. I can remove the autogen note at that point.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D93793/new/

https://reviews.llvm.org/D93793