Allow parallelism scaling of layernorm below N/SIMD = 12.#1534
Open
preusser-amd wants to merge 4 commits intodevfrom
Open
Allow parallelism scaling of layernorm below N/SIMD = 12.#1534preusser-amd wants to merge 4 commits intodevfrom
preusser-amd wants to merge 4 commits intodevfrom
Conversation
…newly supported cases.
STFleming
approved these changes
Mar 20, 2026
Collaborator
STFleming
left a comment
STFleming
left a comment
There was a problem hiding this comment.
Looks great. I removed the assertions from the compiler catching this and added some additional test coverage for the new capabilities, all tests are passing on my end.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The
layernormimplementation was not able to scale to a SIMD parallelism so large that it would push the N/SIMD ratio below 12 due to a corresponding II limitation in thersqrtimplementation. This PR removes this constraint with a significant overhaul of thesqrtimplementation introducing compute interleaving and DSP unrolling, which are leveraged as needed to meet the requested II requirement.