[PATCH] D60660: [InstCombine] Prune fshl/fshr with masked operands

[Nikita Popov via Phabricator via llvm-commits]

nikic added a comment.

The idea here looks reasonable to me, but this doesn't seem like the right place to make the transform. This is ultimately a demanded bits problem and as such should be handled in either BDCE or InstCombineSimplifyDemanded.



================
Comment at: lib/Transforms/InstCombine/InstCombineCalls.cpp:2090
+      if (match(Op0, m_c_And(m_Value(), m_APInt(B)))) {
+        APInt D = dyn_cast<ConstantInt>(ShAmtC)->getValue();
+        uint64_t DLV = D.getLimitedValue();
----------------
You cannot assume that the ShAmtC is a ConstantInt here. It could also be a vector (something that should be covered in tests).


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D60660/new/

https://reviews.llvm.org/D60660