[Mlir-commits] [mlir] [mlir][Linalg] Fix foldFillPackIntoFillOp to work for general cases (PR #74148)

[lorenzo chelini]

chelini wrote:

> > Hi @yzhang93 thanks for improving this. I have two comments/questions:
> > 
> > 1. I need help to understand what this PR fixes from the commit message. Could you please comment on what the "general case" is?
> 
> > 2. The added test is a bit overloading; can it be simplified?
> 
> Hi @chelini, I have a use case as shown in the added test. The definingOp of the pack op is not directly tensor.empty. The original codes didn't work for folding the fill op into the pack op.

Ok, thanks a lot. Next time please provide a more detailed commit message. I was unclear what "general" was referring to in this context.

https://github.com/llvm/llvm-project/pull/74148