[PATCH] D38521: [InstCombine] Improve support for ashr in foldICmpAndShift

[Sanjay Patel via Phabricator via llvm-commits]

spatel accepted this revision.
spatel added a comment.
This revision is now accepted and ready to land.

LGTM.



================
Comment at: lib/Transforms/InstCombine/InstCombineCompares.cpp:1540
       // obvious, but we can prove that they are correct using an SMT solver.
-      if (!Cmp.isSigned() ||
-          (!C2.shl(*C3).isNegative() && !C1.shl(*C3).isNegative()))
-        CanFold = true;
+      // For an arithmetic shift right we can do the same, if we we ensure
+      // the And doesn't use any bits being shifted in. Normally these would
----------------
extra 'we'


================
Comment at: lib/Transforms/InstCombine/InstCombineCompares.cpp:1542
+      // the And doesn't use any bits being shifted in. Normally these would
+      // turned in lshr by SimplifyDemandedBits, but not if there is an
+      // additional user.
----------------
turned in -> be turned into


https://reviews.llvm.org/D38521