Jeremy Kun

> This is necessary for targeting RLWE HW accelerators Explicitly, this means it has no control flow?

> Oh very true, I expect we'd probably create blocks of separate computation partitioning the program and then apply the analysis to specific blocks in this case. See my issue...

Scale and batch size will be needed for sure, but I think batch size will probably already be present in the IR attributes from how we handle packing. The mul...

CKKS is well supported now, and we have separate issues for sub-tasks related to scale, etc. raised in this issue. Closing.

Great overview! I might add, because it seems implicit here, that the memory access path (non-SIMD) reduces to the loop+if conversions, so I would expect you can lower as affine.for,...

I agreed to scope the work for this.

# Summary of https://eprint.iacr.org/2023/1445.pdf ## Section II Key operations for switching between arithmetic and boolean ciphertexts: - SampleExtract: given an RLWE ciphertext and an index i, extract the ith element...

# Summary of https://github.com/heir-compiler/heir ## Dialects There appears to actually be only one dialect, "HEIR", which has all the ops and the separation above is logical, not dialect...ical. The authors...

Here's a possible sketch of what we could do to get what appears to be feature parity with that project: ## Get test inputs Use polygeist to convert some of...

The biggest part of the proposal above that I don't like is the raising from lower-level pack/unpack ops to higher-level OpenFHE API calls. I would prefer if we had a...