perf: avoid redundant permute+cont in non-interleaved RoPE - #2102
Open
daniandtheweb wants to merge 1 commit into
Open
daniandtheweb wants to merge 1 commit into
daniandtheweb wants to merge 1 commit into
Conversation
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
This PR reduces the number of full ggml_cont passes in apply rope for the non-interleaved RoPE layout from 3 copies to 1.
After the single required H/L reorder, the two contiguous halves of the head dim (i, i + d_head/2) are read as views and the rotation is assembled with one ggml_concat. This eliminates 2 of the 3 full copies. The interleaved path is unchanged.
The output is identical to the previous implementation.
As for performance goes, on my RX 7800XT an Anima generation at 1024x1024, cfg 5 goes from 2.21 s/it to 2.15s/it.
This performance optimization possibility was found by a Pi agent running Qwen 3.8 27B locally.
The optimization was then made by me and the AI helped with the review.
Checklist