Experiment Subgroup 8 for older gpus#14
Draft
rillomas wants to merge 90 commits into
Draft
Conversation
This reverts commit edccd26.
was failing on MUL_MAT(type_a=q4_0,type_b=f32,m=1,n=2048,k=8192,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1)
|
Have you seen this ggml-org#24408? Is this going to be added to the profits you're going to get there? |
Owner
Author
I haven't tested with both so hard to say. Since ARL-H will get coopmat enabled the benefits on my changes (which are for non-coopmat kernels) may not add-up. |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
2df11d7 is passing all test-backend-ops with default (subgroup 32),
set GGML_VK_INTEL_DEFAULT_SUBGROUP_SIZE=8andset GGML_VK_INTEL_DEFAULT_SUBGROUP_SIZE=16when run on ARL-H U7-265H (Windows, GPU driver: 32.0.101.8801).f2cf16d passes test-backend-ops and show good gains on specific piplines though seeing regressions on others as well. We shouldn't be seeing regressions so need to check
b5b1ea9 looking pretty good on ARL-H and Arc A770 with only minor regressions. May promote this version as the actual PR