Jeremy Kun

Reopening for cleanup TODOs

This is done in the sense that we went ahead and implemented Vos-Vos-Erkin's algorithm, but we don't have a baseline. Do we still want a dumb baseline? I don't think...

This is an internal Googler (not on our team) who is making changes to both MLIR and HEIR at the same time. I don't think it will end up being...

Some minor notes to add. > By diagonalizing the matrix and performing rotation operations where appropriate Is _diagonalize_ really the term of art? I think we should avoid overloading that...

In https://www.shoup.net/papers/helib.pdf the authors describe it as > the parallel “systolic” multiplication algorithm But I don't quite understand how this is related to systolic arrays...

That's true, and it would break the lowerings from linalg. And now that I read the Fhelipe paper, I think long term we'd want a custom attribute to represent the...

Do arithmetic FHE approaches to ML use quantization? We did the quantization in those passes for CGGI.

It should be blocking LLVM integrate, but we can't stop a force submit in general. I think it will just be an occasional hazard to deal with. On Thu, Jul...

An internal dev decided to force-push a broken change. Sigh.

It was fixed upstream in a few patches https://github.com/llvm/llvm-project/commits/main/utils/bazel/llvm-project-overlay/mlir/BUILD.bazel. We could just wait for the next integrate CL.