[PATCH] D48584: [InstCombine] Delay foldICmpUsingKnownBits until simple transforms are done

[Max Kazantsev via Phabricator via llvm-commits]

mkazantsev created this revision.
mkazantsev added reviewers: spatel, lebedev.ri, DaniilSuchkov, reames.

This patch changes order of transform in InstCombineCompares to avoid
performing transforms based on ranges which produce complex bit arithmetics
before more simple things (like folding with constants) are done. See PR37636
for the motivating example.


https://reviews.llvm.org/D48584

Files:
  lib/Transforms/InstCombine/InstCombineCompares.cpp
  test/Analysis/ValueTracking/non-negative-phi-bits.ll
  test/Transforms/InstCombine/icmp-shl-nsw.ll
  test/Transforms/InstCombine/icmp-shr-lt-gt.ll
  test/Transforms/InstCombine/icmp_sdiv_with_and_without_range.ll
  test/Transforms/LoopVectorize/X86/masked_load_store.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D48584.152835.patch
Type: text/x-patch
Size: 8628 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20180626/6540e018/attachment.bin>