[Mlir-commits] [mlir] [mlir][tosa] Fix unranked tosa canonicalizations crashes (PR #188188)

[Luke Hutton]

================
@@ -1736,8 +1757,10 @@ OpFoldResult ResizeOp::fold(FoldAdaptor adaptor) {
   }
 
   auto input = getInput();
-  auto inputTy = llvm::cast<RankedTensorType>(input.getType());
-  auto resultTy = llvm::cast<RankedTensorType>(getType());
+  auto inputTy = llvm::dyn_cast<RankedTensorType>(input.getType());
+  auto resultTy = llvm::dyn_cast<RankedTensorType>(getType());
+  if (!inputTy || !resultTy)
----------------
lhutton1 wrote:

I wonder if this is possible given the input should be constant and therefore have a static type, could we add a test for `resize`?

Assuming what I said above is correct, we shouldn't need to cast to a `RankedTensorType` here and can just have:
```
return foldToInputIfTypeMatches(*this, input.getType())
```

https://github.com/llvm/llvm-project/pull/188188