Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Enable the SLPVectorizer on Triton side with no scheduling #2714

Closed
wants to merge 1 commit into from

Conversation

chengjunlu
Copy link
Contributor

The IGCVectorizer doesn't support the flash attention kernel so far.
To enable the SLPVectorizer on Triton side with no scheduling in basic block will get better performance for now.

@chengjunlu chengjunlu changed the title [WIP] Enable the SLPVectorizer on Triton side with no scheduling Enable the SLPVectorizer on Triton side with no scheduling Nov 15, 2024
@@ -335,6 +335,9 @@ def make_spv(src, metadata, options):
else:
metadata["build_flags"] = ""

if os.getenv("TRITON_INTEL_ENABLE_POST_PROCESS_LLIR", "1") == "1":
metadata["build_flags"] += " -igc_opts 'DisablePHIScalarization=1'"
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Check with IGC about how to use the IGC options to set the optimization flag instead of using environment variable.

…ve the SLPVectorizer once the IGCVectorizer works well.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

[Performance] Enable the SLPVectorizer to improve flash attention performance
1 participant