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

Pack matmuls with small dimensions #963

Merged
merged 3 commits into from
Sep 6, 2024

Conversation

adam-smnk
Copy link
Collaborator

Extends pack matmul control function to automatically adapt blocking factors for small dimensions.

The new behavior is that the defined or default packing sizes define the upper bound on the tile size. Dimensions smaller than their block factors are still packed using their whole size.
This change allows processing matmuls on tall-and-skinny matrices and partially decouples the decision of what and how to pack from the mechanism itself. For simplicity, any small matmul is be packed now.

Long-term, packing driver should be augmented with a separate cost functions that governs operation selection, need for padding, smartly adjusts blocking factors etc.

@adam-smnk adam-smnk added the benchmark Triggers benchmark jobs label Sep 5, 2024
@adam-smnk adam-smnk merged commit 99cb584 into plaidml:main Sep 6, 2024
18 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
benchmark Triggers benchmark jobs
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants