[ET-VK] 5/n Split dispatches between multiple command buffers. Track previously submitted command buffers in context and add function to execute all previous command buffers. #12527
+81
−7
Conversation
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
…previously submitted command buffers in context and add function to execute all previous command buffers. The diff adds changes to store command buffers submitted with final_use set to false. Storing these buffers is necessary for `execute()` function. Since, `encode_execute()` function is typically called once but `execute()` can be called multiple times, `submit_all_non_final_cmds` function is added so all recorded command buffers with `final_use = False` can be called multiple times in `execute()`. #### Key Changes * Added a flag `execute_pending_first_submission` to the `ComputeGraph` class to track whether execute nodes have been freshly encoded and need to be submitted first. * Added a new function `submit_all_non_final_cmds` to the `Context` class, which submits all non-final command buffers to the GPU. * Modified the `submit_cmd_to_gpu` function to add the submitted command buffer to the `non_final_cmds_` list if it's not marked as final use. * Updated the `execute` function in `ComputeGraph` to submit all non-final command buffers before executing the graph. Differential Revision: [D78360038](https://our.internmc.facebook.com/intern/diff/D78360038/) [ghstack-poisoned]
This was referenced Jul 15, 2025
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/12527
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (4 Unrelated Failures)As of commit 25626da with merge base b77e412 ( BROKEN TRUNK - The following jobs failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
trivedivivek
added a commit
that referenced
this pull request
Jul 15, 2025
…previously submitted command buffers in context and add function to execute all previous command buffers. The diff adds changes to store command buffers submitted with final_use set to false. Storing these buffers is necessary for `execute()` function. Since, `encode_execute()` function is typically called once but `execute()` can be called multiple times, `submit_all_non_final_cmds` function is added so all recorded command buffers with `final_use = False` can be called multiple times in `execute()`. #### Key Changes * Added a flag `execute_pending_first_submission` to the `ComputeGraph` class to track whether execute nodes have been freshly encoded and need to be submitted first. * Added a new function `submit_all_non_final_cmds` to the `Context` class, which submits all non-final command buffers to the GPU. * Modified the `submit_cmd_to_gpu` function to add the submitted command buffer to the `non_final_cmds_` list if it's not marked as final use. * Updated the `execute` function in `ComputeGraph` to submit all non-final command buffers before executing the graph. Differential Revision: [D78360038](https://our.internmc.facebook.com/intern/diff/D78360038/) ghstack-source-id: 296432758 Pull Request resolved: #12527
|
This pull request was exported from Phabricator. Differential Revision: D78360038 |
trivedivivek
added
the
release notes: vulkan
…ers. Track previously submitted command buffers in context and add function to execute all previous command buffers." The diff adds changes to store command buffers submitted with final_use set to false. Storing these buffers is necessary for `execute()` function. Since, `encode_execute()` function is typically called once but `execute()` can be called multiple times, `submit_all_non_final_cmds` function is added so all recorded command buffers with `final_use = False` can be called multiple times in `execute()`. #### Key Changes * Added a flag `execute_pending_first_submission` to the `ComputeGraph` class to track whether execute nodes have been freshly encoded and need to be submitted first. * Added a new function `submit_all_non_final_cmds` to the `Context` class, which submits all non-final command buffers to the GPU. * Modified the `submit_cmd_to_gpu` function to add the submitted command buffer to the `non_final_cmds_` list if it's not marked as final use. * Updated the `execute` function in `ComputeGraph` to submit all non-final command buffers before executing the graph. Differential Revision: [D78360038](https://our.internmc.facebook.com/intern/diff/D78360038/) [ghstack-poisoned]
|
This pull request was exported from Phabricator. Differential Revision: D78360038 |
…ers. Track previously submitted command buffers in context and add function to execute all previous command buffers." The diff adds changes to store command buffers submitted with final_use set to false. Storing these buffers is necessary for `execute()` function. Since, `encode_execute()` function is typically called once but `execute()` can be called multiple times, `submit_all_non_final_cmds` function is added so all recorded command buffers with `final_use = False` can be called multiple times in `execute()`. #### Key Changes * Added a flag `execute_pending_first_submission` to the `ComputeGraph` class to track whether execute nodes have been freshly encoded and need to be submitted first. * Added a new function `submit_all_non_final_cmds` to the `Context` class, which submits all non-final command buffers to the GPU. * Modified the `submit_cmd_to_gpu` function to add the submitted command buffer to the `non_final_cmds_` list if it's not marked as final use. * Updated the `execute` function in `ComputeGraph` to submit all non-final command buffers before executing the graph. Differential Revision: [D78360038](https://our.internmc.facebook.com/intern/diff/D78360038/) [ghstack-poisoned]
|
This pull request was exported from Phabricator. Differential Revision: D78360038 |
…ers. Track previously submitted command buffers in context and add function to execute all previous command buffers." The diff adds changes to store command buffers submitted with final_use set to false. Storing these buffers is necessary for `execute()` function. Since, `encode_execute()` function is typically called once but `execute()` can be called multiple times, `submit_all_non_final_cmds` function is added so all recorded command buffers with `final_use = False` can be called multiple times in `execute()`. #### Key Changes * Added a flag `execute_pending_first_submission` to the `ComputeGraph` class to track whether execute nodes have been freshly encoded and need to be submitted first. * Added a new function `submit_all_non_final_cmds` to the `Context` class, which submits all non-final command buffers to the GPU. * Modified the `submit_cmd_to_gpu` function to add the submitted command buffer to the `non_final_cmds_` list if it's not marked as final use. * Updated the `execute` function in `ComputeGraph` to submit all non-final command buffers before executing the graph. Differential Revision: [D78360038](https://our.internmc.facebook.com/intern/diff/D78360038/) [ghstack-poisoned]
trivedivivek
added a commit
that referenced
this pull request
Jul 16, 2025
…pport to defer command buffers in encode_execute function and submit all deferred commands in execute function. Pull Request resolved: #12527 The diff adds changes to store command buffers submitted with final_use set to false. Storing these buffers is necessary for `execute()` function. Since, `encode_execute()` function is typically called once but `execute()` can be called multiple times, `submit_all_non_final_cmds` function is added so all recorded command buffers with `final_use = False` can be called multiple times in `execute()`. #### Key Changes * Added a flag `execute_pending_first_submission` to the `ComputeGraph` class to track whether execute nodes have been freshly encoded and need to be submitted first. * Added a new function `submit_all_non_final_cmds` to the `Context` class, which submits all non-final command buffers to the GPU. * Modified the `submit_cmd_to_gpu` function to add the submitted command buffer to the `non_final_cmds_` list if it's not marked as final use. * Updated the `execute` function in `ComputeGraph` to submit all non-final command buffers before executing the graph. ghstack-source-id: 296562130 @exported-using-ghexport Differential Revision: [D78360038](https://our.internmc.facebook.com/intern/diff/D78360038/)
|
This pull request was exported from Phabricator. Differential Revision: D78360038 |
SS-JIA
approved these changes
Jul 16, 2025
…ers. Track previously submitted command buffers in context and add function to execute all previous command buffers." The diff adds changes to store command buffers submitted with final_use set to false. Storing these buffers is necessary for `execute()` function. Since, `encode_execute()` function is typically called once but `execute()` can be called multiple times, `submit_all_non_final_cmds` function is added so all recorded command buffers with `final_use = False` can be called multiple times in `execute()`. #### Key Changes * Added a flag `execute_pending_first_submission` to the `ComputeGraph` class to track whether execute nodes have been freshly encoded and need to be submitted first. * Added a new function `submit_all_non_final_cmds` to the `Context` class, which submits all non-final command buffers to the GPU. * Modified the `submit_cmd_to_gpu` function to add the submitted command buffer to the `non_final_cmds_` list if it's not marked as final use. * Updated the `execute` function in `ComputeGraph` to submit all non-final command buffers before executing the graph. Differential Revision: [D78360038](https://our.internmc.facebook.com/intern/diff/D78360038/) [ghstack-poisoned]
trivedivivek
added a commit
that referenced
this pull request
Jul 17, 2025
…pport to defer command buffers in encode_execute function and submit all deferred commands in execute function. Pull Request resolved: #12527 The diff adds changes to store command buffers for deferred submission. Storing these buffers is necessary for `execute()` function. Since, `encode_execute()` function is typically called once but `execute()` can be called multiple times, `submit_deferred_cmds` function is added so all recorded command buffers can be called multiple times in `execute()`. ghstack-source-id: 296939582 @exported-using-ghexport Differential Revision: [D78360038](https://our.internmc.facebook.com/intern/diff/D78360038/)
|
This pull request was exported from Phabricator. Differential Revision: D78360038 |
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.
Stack from ghstack (oldest at bottom):
The diff adds changes to store command buffers submitted with final_use set to false.
Storing these buffers is necessary for
execute()function. Since,encode_execute()function is typically called once butexecute()can be called multiple times,submit_all_non_final_cmdsfunction is added so all recorded command buffers withfinal_use = Falsecan be called multiple times inexecute().Key Changes
execute_pending_first_submissionto theComputeGraphclass to track whether execute nodes have been freshly encoded and need to be submitted first.submit_all_non_final_cmdsto theContextclass, which submits all non-final command buffers to the GPU.submit_cmd_to_gpufunction to add the submitted command buffer to thenon_final_cmds_list if it's not marked as final use.executefunction inComputeGraphto submit all non-final command buffers before executing the graph.Differential Revision: D78360038