[RISCV][NFCI] Use isADDLike helper for or_is_add PatFrag

[asb]

This should be equivalent.

[llvmbot]

@llvm/pr-subscribers-backend-risc-v

Author: Alex Bradbury (asb)

Changes

This should be equivalent.

---

Full diff: https://github.com/llvm/llvm-project/pull/81137.diff

1 Files Affected:

• (modified) llvm/lib/Target/RISCV/RISCVInstrInfo.td (+1-5)

diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfo.td b/llvm/lib/Target/RISCV/RISCVInstrInfo.td
index 518982441e7c0..aa84a4553ec1f 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfo.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfo.td
@@ -1249,11 +1249,7 @@ def : PatGprUimmLog2XLen<sra, SRAI>;
 // Select 'or' as ADDI if the immediate bits are known to be 0 in $rs1. This
 // can improve compressibility.
 def or_is_add : PatFrag<(ops node:$lhs, node:$rhs), (or node:$lhs, node:$rhs),[{
-  if (N->getFlags().hasDisjoint())
-    return true;
-  KnownBits Known0 = CurDAG->computeKnownBits(N->getOperand(0), 0);
-  KnownBits Known1 = CurDAG->computeKnownBits(N->getOperand(1), 0);
-  return KnownBits::haveNoCommonBitsSet(Known0, Known1);
+  return CurDAG->isADDLike(SDValue(N, 0));
 }]>;
 def : PatGprSimm12<or_is_add, ADDI>;
 

[preames]

LGTM

[topperc]

Is it equivalent? Aren't there some special cases handled in isAddLIke? See haveNoCommonBitsSetCommutative in SelectionDAG.cpp.

[asb]

Is it equivalent? Aren't there some special cases handled in isAddLIke? See haveNoCommonBitsSetCommutative in SelectionDAG.cpp.

Good point, I'd assumed that SelectionDAG::haveNoCommonBitsSet was a trivial wrapper over KnownBits;:HaveNoCommonBitsSet. So there may be some functional change. Are you pointing out that the NFCI tag is likely incorrect and we probably want tests for any new patterns picked up, or do you have correctness concerns? I haven't stepped through the haveNoCommonBitsSetCommutative logic, but isADDLike would be buggy (per its documented semantics) if the or couldn't freely be replaced with an add when it returns true.

[topperc]

Is it equivalent? Aren't there some special cases handled in isAddLIke? See haveNoCommonBitsSetCommutative in SelectionDAG.cpp.

Good point, I'd assumed that SelectionDAG::haveNoCommonBitsSet was a trivial wrapper over KnownBits;:HaveNoCommonBitsSet. So there may be some functional change. Are you pointing out that the NFCI tag is likely incorrect and we probably want tests for any new patterns picked up, or do you have correctness concerns? I haven't stepped through the haveNoCommonBitsSetCommutative logic, but isADDLike would be buggy (per its documented semantics) if the or couldn't freely be replaced with an add when it returns true.

I don't have correctness concerns, just that we should have new tests.