[Mlir-commits] [mlir] [mlir][fold-memref-alias-ops] Add support for folding memref.expand_shape involving dynamic dims (PR #89093)

[llvmlistbot at llvm.org]

MaheshRavishankar wrote:

IIUC what you are trying to do is effectively this

```
SmallVector<OpFoldResult> sizes = .... 
AffineExpr s0, s1;
bindSymbol(builder.getContext(), s0, s1);
OpFoldResult product = builder.getIndexAttr(1);
SmallVector<OpFoldResult> suffixProduct(sizes.size());
int index = sizes.size() - 1;
suffixProduct[index--] = product;
for (auto size : llvm::reverse(sizes.drop_front())) {
  product = affine::makeComposedFoldedAffineApply(builder, loc, s0 * s1, {product, size});
  suffixProduct[index--] = product;
}
return suffixProduct.
```

The mixed static dynamic cases will all be handled automatically? and the `sizes` is what is returned by the `inferOutputShape`. I dont see why the logic needs to be duplicated.


https://github.com/llvm/llvm-project/pull/89093