[Mlir-commits] [mlir] [mlir][linalg] Support subtracting accumulation in partial reduction … (PR #214033)

[Renato Golin]

rengolin wrote:

> > Given the unusual length of the summary I feel that I should ask - did AI assist with this?
> 
> Oh definitely. There should be a "co-authored" trailer on the commit giving attribution. This is my first time looking at these files in LLVM. Claude identified the fix based on a regression I was seeing with the IREE compiler. I did my best to understand and simplify it (as you say, it tends to be overly verbose).

In addition, we also require all PRs with AI assistance to identify this in the PR description as well. Usually something like "Assisted by: Some model / framework".

You are also required to understand and _"be able to explain"_ every change AI made on your behalf, not just "best effort". To avoid concerns from the reviewers, make sure you add your own take on top of what AI did, you tested beyond what AI tested, and write the PR description yourself (or heavily modified by you), so that we know you know what's going on.

You are _required_ to validate the AI output and not let for reviewers to do so. Similar PRs have been rejected on those grounds, so please be careful when submitting AI generated PRs in such way.

https://github.com/llvm/llvm-project/pull/214033