[Mlir-commits] [mlir] [mlir][tosa] Harden folds/canonicalizations for unranked and dynamic shapes (PR #188188)
Luke Hutton
llvmlistbot at llvm.org
Thu Mar 26 06:45:33 PDT 2026
================
@@ -1784,28 +1752,22 @@ OpFoldResult ResizeOp::fold(FoldAdaptor adaptor) {
return {};
}
- auto input = getInput();
- auto inputTy = llvm::cast<RankedTensorType>(input.getType());
- auto resultTy = llvm::cast<RankedTensorType>(getType());
- if (inputTy != resultTy)
- return {};
-
- return input;
+ return foldToInputIfTypeMatches(*this, getInput());
}
OpFoldResult ReverseOp::fold(FoldAdaptor adaptor) {
auto operand = getInput1();
auto operandTy = llvm::cast<ShapedType>(operand.getType());
auto axis = getAxis();
- auto operandAttr =
- llvm::dyn_cast_if_present<SplatElementsAttr>(adaptor.getInput1());
- if (operandAttr)
- return operandAttr;
+ bool noOpReverse =
+ llvm::isa_and_nonnull<SplatElementsAttr>(adaptor.getInput1());
+
+ // If the dim-length is 1, or reversing axis is unit-dim, also a no-op.
+ noOpReverse |= operandTy.hasRank() &&
+ (operandTy.getRank() == 0 || operandTy.getDimSize(axis) == 1);
- // If the dim-length is 1, tosa.reverse is a no-op.
- if (operandTy.hasRank() &&
- (operandTy.getRank() == 0 || operandTy.getDimSize(axis) == 1))
- return operand;
+ if (noOpReverse)
+ return foldToInputIfTypeMatches(*this, operand);
----------------
lhutton1 wrote:
Apologies, I thought I'd left this comment in the previous review, but it seems to have gone missing..
Could we convert this to? IMO it seems more readable. I've also removed the `operandTy.getRank() == 0` check since the tablegen definition enforces the input is rank >= 1 if it is ranked, so it's not needed here.
```suggestion
// If the dim-length is 1, or reversing axis is unit-dim, also a no-op.
const bool noOpReverse =
llvm::isa_and_nonnull<SplatElementsAttr>(adaptor.getInput1());
if (!noOpReverse || !operandTy.hasRank() || !operandTy.getDimSize(axis) == 0)
return {};
return foldToInputIfTypeMatches(*this, operand);
```
https://github.com/llvm/llvm-project/pull/188188
More information about the Mlir-commits
mailing list