Repository navigation
Arm backend: Fuse command stream in VGF. - #24838
Merged
Merged
Conversation
This PR introduces fused Vulkan command-stream submission for the ExecuTorch VGF backend. The implementation allows application-provided preprocessing and postprocessing command buffers to be submitted together with the VGF inference command buffer using a single vkQueueSubmit() and fence wait, reducing CPU submission and synchronisation overhead. The existing VGF execution path remains unchanged when fusion is not enabled. Fusion was faster in 8 out of 10 repetitions, with a median paired improvement of 1.144 ms. Signed-off-by: Elena Zhelezina <elena.zhelezina@arm.com> Change-Id: I7d918442b578e30383f0c8dc4f8cfad9e9eecda3
wwwind
added
partner: arm
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/24838
Note: Links to docs will display an error until the docs builds have been completed. ⏳ No Failures, 1 PendingAs of commit 76b8dcc with merge base 4411b8d ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
oscarandersson8218
approved these changes
Nov 10, 2026
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This PR introduces fused Vulkan command-stream
submission for the ExecuTorch VGF backend.
The implementation allows application-provided
preprocessing and postprocessing command buffers
to be submitted together with the VGF inference
command buffer using a single vkQueueSubmit()
and fence wait, reducing CPU submission and
synchronisation overhead.
The existing VGF execution path remains unchanged
when fusion is not enabled.
Fusion was faster in 8 out of 10 repetitions,
with a median paired improvement of 1.144 ms.
cc @digantdesai @freddan80 @per @zingo @oscarandersson8218 @mansnils @Sebastian-Larsson @robell @rascani