Refactor custom gemm heuristics #56
Merged
Conversation
This file contains 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
…ns. Disabling the now obsolete LLMM1 path
mawong-amd
reviewed
Jun 19, 2024
| weights.shape[0], | ||
| dtype=inp_view.dtype, | ||
| device='cuda') | ||
| _custom_C.wvSpltK(weights, inp_view, out, n, self.cu_count) |
There was a problem hiding this comment.
Not something that needs to be changed right now but we probably want to refactor this eventually so that the MP core count is done at the C++ level: IMO not good decomposition to have it here.
mawong-amd
reviewed
Jun 20, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Moving custom skinny gemm heuristic before hipblas or rocblas solutions.
Disabling the now obsolete LLMM1 path which is fully covered by the new kernel