[PATCH] D52548: Stop instcombining propagating wider shufflevector arguments to predecessors.

[Sanjay Patel via Phabricator via llvm-commits]

spatel added inline comments.


================
Comment at: lib/Transforms/InstCombine/InstCombineVectorOps.cpp:900-902
+    case Instruction::FMul:
+    case Instruction::FDiv:
     case Instruction::FRem:
----------------
Leave formatting changes out of this patch, so we're not distracting from the real diff.


================
Comment at: lib/Transforms/InstCombine/InstCombineVectorOps.cpp:922
       for (Value *Operand : I->operands()) {
-        if (!CanEvaluateShuffled(Operand, Mask, Depth-1))
+        if (!CanEvaluateShuffled(Operand, Mask, Depth - 1))
           return false;
----------------
Leave formatting changes out of this patch, so we're not distracting from the real diff.


================
Comment at: lib/Transforms/InstCombine/InstCombineVectorOps.cpp:942
       }
-      return CanEvaluateShuffled(I->getOperand(0), Mask, Depth-1);
+      return CanEvaluateShuffled(I->getOperand(0), Mask, Depth - 1);
     }
----------------
Leave formatting changes out of this patch, so we're not distracting from the real diff.


================
Comment at: lib/Transforms/InstCombine/InstCombineVectorOps.cpp:1468
+  if (isa<UndefValue>(RHS) && CanEvaluateShuffled(LHS, Mask) &&
+      !SVI.increasesLength()) {
     Value *V = EvaluateInDifferentElementOrder(LHS, Mask);
----------------
The length check should be before CanEvaluateShuffled() to be more efficient (CanEval... is recursive and potentially costly).


Repository:
  rL LLVM

https://reviews.llvm.org/D52548