Refactor/cmma generalize #94
Closed
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
…l into refactor/cmma_generalize
…l into refactor/cmma_generalize
…l into refactor/cmma_generalize
|
See #101 instead |
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Outdated suggestions cannot be applied.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.
Generalized the cmma algorithm to allow
I reduced the complexity of all the different parameters (block sizes, cube dims and other comptime info) to two degrees of liberty only: one must choose B_MN which is both block size for m and n, and B_K (block size for k), with the additional constraints that these parameters must be divisible by the tile size, and that B_MN must be divisible by B_K. This allows for only a few combinations of B_MN, B_K.
The combination (32, 32) weirdly adds 0s if tensors are not small, so I ignored it. I don't intend to fix it in the short term because it's not likely an important combination.
The combination (128, 16) creates a 128x128 shared memory at the write output stage. With the other shared memories this will bust most SMEMs. I will work on removing the need for a final shared memory, as mentioned in #15 .
In the short term I will also work on #13