Fix bug in DML EP ExecuteCommandList fast path and simplify design #18866
Conversation
This file contains 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
PatriceVignola
approved these changes
Dec 18, 2023
jeffbloo
deleted the
user/jeffbloo/CommandRecorderFixAndSimplification
branch
jeffbloo
added a commit
that referenced
this pull request
Jan 4, 2024
…18866) ### Description This addresses a bug in a fast path that was added for submission of re-used command lists of fused graph kernels in the DML EP, addressing a D3D debug layer error. ### Motivation and Context The fast path in DmlCommandRecorder::ExecuteCommandList enabled a current non-reused command list, if empty, to be used for commands following submission of the fused command list. The fix ensures the associated command allocator is only re-used after the next fence value is completed, which is higher due to submission of the other command list. The command recorder design was intended to support batching of provided command list execution, however it submits command lists immedately as an implementation detail to maximize CPU/GPU parallelism. If that heuristic was removed, it would expose additional issues in this same fast path. Because of this and complexity and inefficiency of the old batching mechanism, I also removed this.
jslap-ubi
pushed a commit
to jslap-ubi/onnxruntime
that referenced
this pull request
Apr 5, 2024
…icrosoft#18866) ### Description This addresses a bug in a fast path that was added for submission of re-used command lists of fused graph kernels in the DML EP, addressing a D3D debug layer error. ### Motivation and Context The fast path in DmlCommandRecorder::ExecuteCommandList enabled a current non-reused command list, if empty, to be used for commands following submission of the fused command list. The fix ensures the associated command allocator is only re-used after the next fence value is completed, which is higher due to submission of the other command list. The command recorder design was intended to support batching of provided command list execution, however it submits command lists immedately as an implementation detail to maximize CPU/GPU parallelism. If that heuristic was removed, it would expose additional issues in this same fast path. Because of this and complexity and inefficiency of the old batching mechanism, I also removed this.
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.
Description
This addresses a bug in a fast path that was added for submission of re-used command lists of fused graph kernels in the DML EP, addressing a D3D debug layer error.
Motivation and Context
The fast path in DmlCommandRecorder::ExecuteCommandList enabled a current non-reused command list, if empty, to be used for commands following submission of the fused command list. The fix ensures the associated command allocator is only re-used after the next fence value is completed, which is higher due to submission of the other command list.
The command recorder design was intended to support batching of provided command list execution, however it submits command lists immedately as an implementation detail to maximize CPU/GPU parallelism. If that heuristic was removed, it would expose additional issues in this same fast path. Because of this and complexity and inefficiency of the old batching mechanism, I also removed this.