[TT-TO-TTGWARP] Detect and handle flash attention with causal masking #2013
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
Signed-off-by: Julian Oppermann <[email protected]>
jopperm
requested review from
whitneywhtsang,
etiotto,
Dewei-Wang-sh and
a team
jopperm
changed the title
etiotto
reviewed
Aug 27, 2024
| @@ -383,6 +385,51 @@ class ConvertTritonToTritonGPUWarp | |||
| } | |||
| return WalkResult::advance(); | |||
| }); | |||
|
|
|||
| if (loops.size() == 2 && workloads.front() == Workload::Attention && | |||
There was a problem hiding this comment.
why the loop size has to be 2 ?
There was a problem hiding this comment.
flashAttention with causal mask has 2 loops.
third_party/intel/lib/TritonToTritonGPUWarp/TritonToTritonGPUWarpPass.cpp
Outdated
Show resolved
Hide resolved
third_party/intel/lib/TritonToTritonGPUWarp/TritonToTritonGPUWarpPass.cpp
Outdated
Show resolved
Hide resolved
| return; | ||
|
|
||
| if (auto cst = dyn_cast<arith::ConstantOp>(op)) { | ||
| transformArithConstantOp(cst, blockLayout); |
There was a problem hiding this comment.
here just assume the type w/o encoding has blockLayout, right?
I have a local change that aims to cover the full propagation, making most of the case go to L405 early return.
There was a problem hiding this comment.
Yes, this IR walk just patches previously unhandled ops, but is completely specific to causal flash attention. I added a FIXME above to make that clearer. So great if we could drop this workaround soon.
Signed-off-by: Julian Oppermann <[email protected]>
victor-eds
approved these changes
Sep 2, 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Outdated suggestions cannot be applied.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.
Detect and handle flash attention with causal masking in
-convert-triton-to-tritongpu-warpby supportingtt.make_rangeand two dependent attention-for-loops.See #1947 for more context / the complete PoC.