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.

Sign up for GitHub

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

Jump to bottom

Use buffer pool for prepacked matrices #128

Closed
robertknight opened this issue Apr 27, 2024 · 0 comments · Fixed by #130
Closed

Use buffer pool for prepacked matrices #128

robertknight opened this issue Apr 27, 2024 · 0 comments · Fixed by #130
Labels
performance Issues that affect model inference or loading performance

Comments

@robertknight
Copy link
Owner

A buffer pool was added in #108 to enable re-use of buffers for operator outputs and temporary tensors. This is now used by almost all operators and for most large temporary buffers, with one exception: prepacked matrix weights created by GemmExecutor::{prepack_a, prepack_b} are not allocated from the pool yet. They should be.
@robertknight robertknight added the performance Issues that affect model inference or loading performance label Apr 27, 2024
robertknight added a commit that referenced this issue Apr 27, 2024
@robertknight
Allocated prepacked matrices from pool
059e588 
Allocate and auto-return prepacked matrices to the shared pool. There is an
exception for the GRU op, because I'm expecting to revise its internals shortly
as part of #85.

Fixes #128
robertknight added a commit that referenced this issue Apr 27, 2024
@robertknight
Allocated prepacked matrices from pool
4ed15c1 
Allocate and auto-return prepacked matrices to the shared pool. There is an
exception for the GRU op, because I'm expecting to revise its internals shortly
as part of #85.

Fixes #128
robertknight added a commit that referenced this issue Apr 27, 2024
@robertknight
Allocated prepacked matrices from pool
f2445d0 
Allocate and auto-return prepacked matrices to the shared pool. There is an
exception for the GRU op, because I'm expecting to revise its internals shortly
as part of #85.

Fixes #128
@robertknight robertknight mentioned this issue Apr 27, 2024
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
performance Issues that affect model inference or loading performance
Projects
None yet
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

Allocate buffers for prepacked matrices from pool robertknight/rten
1 participant
@robertknight