Repository navigation
Use square dot tiles when B is packed per split of n and A doesn't fit in cache. - #11713
Open
copybara-service[bot] wants to merge 1 commit into
Open
copybara-service[bot] wants to merge 1 commit into
copybara-service[bot] wants to merge 1 commit into
Conversation
…t in cache. With a dynamic B, B is packed outside the loop over m, so A is read once per split of n. With AVX2 kernels (block_n = 16) and large k, `choose_split_factors` picked tall 128x32 tiles, so an A larger than the cache was streamed from memory n / 32 times. Square tiles are now used when one pass over A exceeds 32 MB, which halves the reads of A. AVX-512 kernels already get 64x64 tiles for k >= 512, so on those CPUs only tall dots with small k change. PiperOrigin-RevId: 1000733554
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Use square dot tiles when B is packed per split of n and A doesn't fit in cache.
With a dynamic B, B is packed outside the loop over m, so A is read once per
split of n. With AVX2 kernels (block_n = 16) and large k,
choose_split_factorspicked tall 128x32 tiles, so an A larger than the cachewas streamed from memory n / 32 times. Square tiles are now used when one pass
over A exceeds 32 MB, which halves the reads of A. AVX-512 kernels already get
64x64 tiles for k >= 512, so on those CPUs only tall dots with small k change.