[PATCH] D54876: [DemandedBits] Add support for funnel shifts

[Sanjay Patel via Phabricator via llvm-commits]

spatel added inline comments.


================
Comment at: lib/Analysis/DemandedBits.cpp:147
+      case Intrinsic::fshr:
+        if (auto *SA = dyn_cast<ConstantInt>(II->getOperand(2))) {
+          // Normalize to funnel shift left. APInt shifts of BitWidth are well-
----------------
Is this analysis completely ignoring vector integer instructions? 

It would still be good to use 'match(X, m_APInt())' in case we ever upgrade the whole thing to handle vectors. But if the answer to the above question is 'yes', then a TODO comment is ok for now.


================
Comment at: lib/Analysis/DemandedBits.cpp:154-156
+          if (OperandNo == 0)
+            AB = AOut.lshr(ShiftAmt);
+          else if (OperandNo == 1)
----------------
Can we do something like D54478 for operand 2 in a follow-up patch?


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D54876/new/

https://reviews.llvm.org/D54876