opencl: enable the general fp mm for non-cont input and as a fallback for specialized kqv kernel for adreno #18970
+166
−13
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
github-actions
bot
added
ggml
max-krasnyansky
approved these changes
Jan 22, 2026
Collaborator
max-krasnyansky
left a comment
max-krasnyansky
left a comment
There was a problem hiding this comment.
Looks good.
Didn't get a chance to try it on my setup yet. I'm testing some Hexagon changes and will give this a go along with that.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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Outdated suggestions cannot be applied.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 specialized kqv kernel for Adreno uses
image1d_buffer_tto wrapdstto utilize texture processor. However,image1d_buffer_thas dimension limit. The limit can be reached when there are enough tokens in kv cache.For example, providing gpt-oss-20b with a prompt of about 4k tokens using
llama-completionwith-ub 1024will hit the limit, resulting in assert failure.This PR utilizes the general mm for floating point by copying to
src0andsrc1to contiguous if necessary. I believe this has been used in vulkan backend. A check is added to ensure that the arguments are within the limit ofimage1d_buffer_tfor the specialized kqv kernel.