[ET-VK] Consolidate shader compilation into one vkCreateComputePipelines call #11345
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
…nes call We target the QC Adreno driver implementation of Vulkan. The Vulkan API does not enforce how QC actually uses the cache. As the plural naming of `vkCreateComputePipelines` suggests, we observed that the `createInfoCount`, `pCreateInfos` and `pPipelines` arguments above allow construction of multiple compute pipelines in one invocation. We refactor ET-VK to accumulate metadata necessary for pipeline construction and invoke vkCreateComputePipelines only once. QC's implementation maximizes the cache if we create the same number of compute pipelines in fewer invocations of vkCreateComputePipelines. This decreases model load for a sample model from 1.7s to 1.0s, and down to 300ms once @SSJia removes the noop shader. Differential Revision: [D75763660](https://our.internmc.facebook.com/intern/diff/D75763660/) [ghstack-poisoned]
jorgep31415
added a commit
that referenced
this pull request
Jun 3, 2025
…nes call We target the QC Adreno driver implementation of Vulkan. The Vulkan API does not enforce how QC actually uses the cache. As the plural naming of `vkCreateComputePipelines` suggests, we observed that the `createInfoCount`, `pCreateInfos` and `pPipelines` arguments above allow construction of multiple compute pipelines in one invocation. We refactor ET-VK to accumulate metadata necessary for pipeline construction and invoke vkCreateComputePipelines only once. QC's implementation maximizes the cache if we create the same number of compute pipelines in fewer invocations of vkCreateComputePipelines. This decreases model load for a sample model from 1.7s to 1.0s, and down to 300ms once ssjia removes the noop shader. Differential Revision: [D75763660](https://our.internmc.facebook.com/intern/diff/D75763660/) ghstack-source-id: 287485414 Pull Request resolved: #11345
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/11345
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 528b571 with merge base af0a246 ( 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: D75763660 |
jorgep31415
added
the
release notes: vulkan
…mputePipelines call" We target the QC Adreno driver implementation of Vulkan. The Vulkan API does not enforce how QC actually uses the cache. As the plural naming of `vkCreateComputePipelines` suggests, we observed that the `createInfoCount`, `pCreateInfos` and `pPipelines` arguments above allow construction of multiple compute pipelines in one invocation. We refactor ET-VK to accumulate metadata necessary for pipeline construction and invoke vkCreateComputePipelines only once. QC's implementation maximizes the cache if we create the same number of compute pipelines in fewer invocations of vkCreateComputePipelines. This decreases model load for a sample model from 1.7s to 1.0s, and down to 300ms once ssjia removes the noop shader. Differential Revision: [D75763660](https://our.internmc.facebook.com/intern/diff/D75763660/) [ghstack-poisoned]
|
This pull request was exported from Phabricator. Differential Revision: D75763660 |
…mputePipelines call" We target the QC Adreno driver implementation of Vulkan. The Vulkan API does not enforce how QC actually uses the cache. As the plural naming of `vkCreateComputePipelines` suggests, we observed that the `createInfoCount`, `pCreateInfos` and `pPipelines` arguments above allow construction of multiple compute pipelines in one invocation. We refactor ET-VK to accumulate metadata necessary for pipeline construction and invoke vkCreateComputePipelines only once. QC's implementation maximizes the cache if we create the same number of compute pipelines in fewer invocations of vkCreateComputePipelines. This decreases model load for a sample model from 1.7s to 200ms. Differential Revision: [D75763660](https://our.internmc.facebook.com/intern/diff/D75763660/) [ghstack-poisoned]
|
This pull request was exported from Phabricator. Differential Revision: D75763660 |
SS-JIA
approved these changes
Jun 4, 2025
…mputePipelines call" We target the QC Adreno driver implementation of Vulkan. The Vulkan API does not enforce how QC actually uses the cache. As the plural naming of `vkCreateComputePipelines` suggests, we observed that the `createInfoCount`, `pCreateInfos` and `pPipelines` arguments above allow construction of multiple compute pipelines in one invocation. We refactor ET-VK to accumulate metadata necessary for pipeline construction and invoke vkCreateComputePipelines only once. QC's implementation maximizes the cache if we create the same number of compute pipelines in fewer invocations of vkCreateComputePipelines. This decreases model load for a sample model from 1.7s to 200ms. Differential Revision: [D75763660](https://our.internmc.facebook.com/intern/diff/D75763660/) [ghstack-poisoned]
|
This pull request was exported from Phabricator. Differential Revision: D75763660 |
facebook-github-bot
temporarily deployed
to
cherry-pick-bot
GitHub Actions
Inactive
lucylq
pushed a commit
that referenced
this pull request
Jun 4, 2025
…nes call (#11381) This PR was created by the merge bot to help merge the original PR into the main branch. ghstack PR number: #11345 by @jorgep31415 ^ Please use this as the source of truth for the PR details, comments, and reviews ghstack PR base: https://github.com/pytorch/executorch/tree/gh/jorgep31415/135/base ghstack PR head: https://github.com/pytorch/executorch/tree/gh/jorgep31415/135/head Merge bot PR base: https://github.com/pytorch/executorch/tree/main Merge bot PR head: https://github.com/pytorch/executorch/tree/gh/jorgep31415/135/orig @diff-train-skip-merge Co-authored-by: jorgep31415 <jorgepineda140@gmail.com>
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):
We target the QC Adreno driver implementation of Vulkan. The Vulkan API does not enforce how QC actually uses the cache. As the plural naming of
vkCreateComputePipelinessuggests, we observed that thecreateInfoCount,pCreateInfosandpPipelinesarguments above allow construction of multiple compute pipelines in one invocation. We refactor ET-VK to accumulate metadata necessary for pipeline construction and invoke vkCreateComputePipelines only once. QC's implementation maximizes the cache if we create the same number of compute pipelines in fewer invocations of vkCreateComputePipelines. This decreases model load for a sample model from 1.7s to 300ms.Differential Revision: D75763660