Conversation
facebook-github-bot
added
the
CLA Signed
danthe3rd
changed the title
Codecov Report
@@ Coverage Diff @@
## pipeline #388 +/- ##
=========================================
Coverage 91.11% 91.11%
=========================================
Files 75 75
Lines 4345 4345
=========================================
Hits 3959 3959
Misses 386 386
Flags with carried forward coverage won't be shown. Click here to find out more. Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here. |
fmassa
approved these changes
Sep 13, 2022
fmassa
left a comment
fmassa
left a comment
Contributor
There was a problem hiding this comment.
The speedups looks great! (And compilation time didn't seem affected on CI).
There seems to be some regressions for V100 for large batch sizes and K (like f16 fwd_gen B=256, M=1024, K=256), but the improvements on A100 are pretty amazing so let's get this in.
bertmaher
pushed a commit
to bertmaher/xformers
that referenced
this pull request
Dec 20, 2024
…ch#388) Co-authored-by: danthe3rd <danthe3rd>
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
4 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
When we iterate on
key_start, we do the second matmul with V and stored the result into gmem after rescaling the current value. This did: gmem read/write + shmem read/write (due to the epilogue).This PR keeps everything in registers, and we only write to global memory at the end of the kernel.
BENCHMARKS
A100
P100/V100 perf