Default construct QueryPool with nonzero values #3721
Closed
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
Summary: **We must bugfix this first to unblock either dumping querypool directly after inference or SDK development** Currently, if you try to init QueryPool during inference, it will crash at `vkCreateQueryPool`. Some debugging revealed in the test binary, it will crash only on VulkanComputeGraphTest but not VulkanComputeAPITest. It is because the `Context` instance is initialized in two different ways and then `QueryPoolConfig` is then initialized in two different ways. 1. Static singleton https://fburl.com/code/iqo8xcg2 with above nonzero max query count and the compile time flag to change the query pool. We don't use this for model inference. 2. Explicit constructor https://fburl.com/code/m1xga4ra, invoked in `VulkanBackend::Init()` https://fburl.com/code/4j8y3pzu which goes through default construction of `GraphConfig`, default construction of `QueryPoolConfig`, leading to zero max query count and crash. We use this for model inference. 3. This should be probably be unified, as it was extremely confusing to have the compile time flag work in non inference cases but not in inference case, and explicit const struct construction happens in the former vs completely default struct construction in the latter. 4. A quick solution to unblock. `QueryPoolConfig` has default init values of the compile time flags. Reviewed By: SS-JIA Differential Revision: D57735873
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/3721
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 579c8bf with merge base 3c43fd6 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D57735873 |
SS-JIA
approved these changes
May 23, 2024
|
This pull request has been merged in 1991346. |
kirklandsign
pushed a commit
to kirklandsign/executorch
that referenced
this pull request
May 24, 2024
Summary: Pull Request resolved: pytorch#3721 **We must bugfix this first to unblock either dumping querypool directly after inference or SDK development** Currently, if you try to init QueryPool during inference, it will crash at `vkCreateQueryPool`. Some debugging revealed in the test binary, it will crash only on VulkanComputeGraphTest but not VulkanComputeAPITest. It is because the `Context` instance is initialized in two different ways and then `QueryPoolConfig` is then initialized in two different ways. 1. Static singleton https://fburl.com/code/iqo8xcg2 with above nonzero max query count and the compile time flag to change the query pool. We don't use this for model inference. 2. Explicit constructor https://fburl.com/code/m1xga4ra, invoked in `VulkanBackend::Init()` https://fburl.com/code/4j8y3pzu which goes through default construction of `GraphConfig`, default construction of `QueryPoolConfig`, leading to zero max query count and crash. We use this for model inference. 3. This should be probably be unified, as it was extremely confusing to have the compile time flag work in non inference cases but not in inference case, and explicit const struct construction happens in the former vs completely default struct construction in the latter. 4. A quick solution to unblock. `QueryPoolConfig` has default init values of the compile time flags. Reviewed By: SS-JIA Differential Revision: D57735873 fbshipit-source-id: e04325f34f68576db462b3003ddfffefee7fc7f0
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.
Summary:
We must bugfix this first to unblock either dumping querypool directly after inference or SDK development
Currently, if you try to init QueryPool during inference, it will crash at
vkCreateQueryPool. Some debugging revealed in the test binary, it will crash only on VulkanComputeGraphTest but not VulkanComputeAPITest.It is because the
Contextinstance is initialized in two different ways and thenQueryPoolConfigis then initialized in two different ways.Static singleton https://fburl.com/code/iqo8xcg2 with above nonzero max query count and the compile time flag to change the query pool. We don't use this for model inference.
Explicit constructor https://fburl.com/code/m1xga4ra, invoked in
VulkanBackend::Init()https://fburl.com/code/4j8y3pzu which goes through default construction ofGraphConfig, default construction ofQueryPoolConfig, leading to zero max query count and crash. We use this for model inference.This should be probably be unified, as it was extremely confusing to have the compile time flag work in non inference cases but not in inference case, and explicit const struct construction happens in the former vs completely default struct construction in the latter.
A quick solution to unblock.
QueryPoolConfighas default init values of the compile time flags.Reviewed By: SS-JIA
Differential Revision: D57735873