[llvm] Re apply 130577 narrow math for and operand (PR #133896)
Juan Manuel Martinez CaamaƱo via llvm-commits
llvm-commits at lists.llvm.org
Thu Apr 10 23:54:13 PDT 2025
================
@@ -1559,6 +1559,75 @@ void AMDGPUCodeGenPrepareImpl::expandDivRem64(BinaryOperator &I) const {
llvm_unreachable("not a division");
}
+/*
+This will cause non-byte load in consistency, for example:
+```
+ %load = load i1, ptr addrspace(4) %arg, align 4
+ %zext = zext i1 %load to
+ i64 %add = add i64 %zext
+```
+Instead of creating `s_and_b32 s0, s0, 1`,
+it will create `s_and_b32 s0, s0, 0xff`.
+We accept this change since the non-byte load assumes the upper bits
+within the byte are all 0.
+*/
+static bool tryNarrowMathIfNoOverflow(Instruction *I,
+ const SITargetLowering *TLI,
+ const TargetTransformInfo &TTI,
+ const DataLayout &DL) {
+ unsigned Opc = I->getOpcode();
+ Type *OldType = I->getType();
+
+ if (Opc != Instruction::Add && Opc != Instruction::Mul)
+ return false;
+
+ unsigned OrigBit = OldType->getScalarSizeInBits();
+ unsigned MaxBitsNeeded = OrigBit;
+
+ if (Opc != Instruction::Add && Opc != Instruction::Mul)
+ llvm_unreachable("Unexpected opcode, only valid for Instruction::Add and "
+ "Instruction::Mul.");
+
+ MaxBitsNeeded = computeKnownBits(I, DL).countMaxActiveBits();
----------------
jmmartinez wrote:
NIT:
```suggestion
if (Opc != Instruction::Add && Opc != Instruction::Mul)
llvm_unreachable("Unexpected opcode, only valid for Instruction::Add and "
"Instruction::Mul.");
unsigned MaxBitsNeeded = computeKnownBits(I, DL).countMaxActiveBits();
```
https://github.com/llvm/llvm-project/pull/133896
More information about the llvm-commits
mailing list