Use new split pipe API for FlashAttention#130
Draft
learning-chip wants to merge 17 commits into
Draft
Conversation
learning-chip
force-pushed
the
fa_split_pipe
branch
from
17a3917 to
51384f2
Compare
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
To run code in this PR, must strictly use the same ptoas + ptoisa dependency following the dockerfile changes in this PR
Goal: 100% match this cpp performance https://gitcode.com/csjlchen/pto-isa/blob/feature_subtile/kernels/manual/common/flash_atten/fa_performance_kernel.cpp
Action items:
EntryOffetanymore), to see if it still matches best perf in feat: add FA cpp ref impl, and benchmark script #125TileSplitAxis::TILE_UP_DOWNloweringhttps://github.com/hw-native-sys/PTOAS/blob/v0.35/test/lit/pto/tpush_tpop_globaltensor_split_half_slot_a3.pto