[PATCH] D112754: X86: Fold masked merge pattern when and-not is not available

[Sanjay Patel via Phabricator via llvm-commits]

spatel added inline comments.


================
Comment at: llvm/lib/Target/X86/X86ISelLowering.cpp:46412
+  SDValue N0 = Node->getOperand(0);
+  if (N0->getOpcode() != ISD::AND)
+    return SDValue();
----------------
I suspect we want to avoid this transform if either operand has more than one use. Please add tests with something like this:

```
define i32 @masked_merge0_extra_use(i32 %a0, i32 %a1, i32 %a2, i32* %p1) {
  %and0 = and i32 %a0, %a1
  %not = xor i32 %a0, -1
  %and1 = and i32 %not, %a2
  store i32 %and1, i32* %p1 ; extra use of intermediate op
  %or = or i32 %and0, %and1
  ret i32 %or
}

```


================
Comment at: llvm/lib/Target/X86/X86ISelLowering.cpp:46527
 
+  // We should fold "masked merge" patterns when `andn` is not avilable.
+  if (!Subtarget.hasBMI() && VT.isScalarInteger() && VT != MVT::i1) {
----------------
typo: available


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112754/new/

https://reviews.llvm.org/D112754