[ET-VK] Persistently map staging buffers #5021
Merged
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
added 6 commits
August 30, 2024 15:08
1. Tensors accessible to both CPU and GPU use `StorageBuffer`. 2. Tensors accessible to GPU only use `vTensorStorage`. Since `StorageBuffer` is only used for [staging buffers](https://vulkan-tutorial.com/Vertex_buffers/Staging_buffer), i.e., buffers accessible to both CPU and GPU, `StagingBuffer` is a fitting name for (1) that decreases confusion between (1) and (2). Differential Revision: [D62049779](https://our.internmc.facebook.com/intern/diff/D62049779/) [ghstack-poisoned]
TSIA Differential Revision: [D62049777](https://our.internmc.facebook.com/intern/diff/D62049777/) [ghstack-poisoned]
There are ~4 lines of duplicated code but IMO, the distinct function names make the code much clearer. Differential Revision: [D62049778](https://our.internmc.facebook.com/intern/diff/D62049778/) [ghstack-poisoned]
…on methods" There are ~4 lines of duplicated code but IMO, the distinct function names make the code much clearer. Differential Revision: [D62049778](https://our.internmc.facebook.com/intern/diff/D62049778/) [ghstack-poisoned]
`StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/) [ghstack-poisoned]
There are ~4 lines of duplicated code but IMO, the distinct function names make the code much clearer. Differential Revision: [D62049778](https://our.internmc.facebook.com/intern/diff/D62049778/) [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/5021
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 0bd4b96 with merge base 3cf9237 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
This was referenced Aug 30, 2024
|
This pull request was exported from Phabricator. Differential Revision: D59706627 |
junpi3
pushed a commit
that referenced
this pull request
Aug 30, 2024
`StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/) ghstack-source-id: 233620788 Pull Request resolved: #5021
`StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/) [ghstack-poisoned]
|
This pull request was exported from Phabricator. Differential Revision: D59706627 |
junpi3
pushed a commit
that referenced
this pull request
Sep 3, 2024
Pull Request resolved: #5021 `StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. ghstack-source-id: 240798121 @exported-using-ghexport Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/)
added 2 commits
September 3, 2024 10:19
`StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/) [ghstack-poisoned]
`StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/) [ghstack-poisoned]
|
This pull request was exported from Phabricator. Differential Revision: D59706627 |
junpi3
pushed a commit
that referenced
this pull request
Sep 3, 2024
Pull Request resolved: #5021 `StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. ghstack-source-id: 240813745 @exported-using-ghexport Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/)
junpi3
changed the base branch from
gh/jorgep31415/109/base
to
gh/jorgep31415/108/head
added 2 commits
September 3, 2024 10:36
`StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/) [ghstack-poisoned]
`StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/) [ghstack-poisoned]
|
This pull request was exported from Phabricator. Differential Revision: D59706627 |
junpi3
pushed a commit
that referenced
this pull request
Sep 3, 2024
Pull Request resolved: #5021 `StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. ghstack-source-id: 240814116 @exported-using-ghexport Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/)
added 2 commits
September 3, 2024 10:40
`StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/) [ghstack-poisoned]
`StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/) [ghstack-poisoned]
|
This pull request was exported from Phabricator. Differential Revision: D59706627 |
junpi3
pushed a commit
that referenced
this pull request
Sep 3, 2024
Pull Request resolved: #5021 `StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. ghstack-source-id: 240818434 @exported-using-ghexport Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/)
added 2 commits
September 3, 2024 11:30
`StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/) [ghstack-poisoned]
`StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/) [ghstack-poisoned]
junpi3
pushed a commit
that referenced
this pull request
Sep 3, 2024
Pull Request resolved: #5021 `StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. ghstack-source-id: 240831447 @exported-using-ghexport Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/)
|
This pull request was exported from Phabricator. Differential Revision: D59706627 |
added 2 commits
September 4, 2024 12:20
`StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/) [ghstack-poisoned]
`StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/) [ghstack-poisoned]
|
This pull request was exported from Phabricator. Differential Revision: D59706627 |
junpi3
pushed a commit
that referenced
this pull request
Sep 4, 2024
Pull Request resolved: #5021 `StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. ghstack-source-id: 241025872 @exported-using-ghexport Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/)
SS-JIA
approved these changes
Sep 4, 2024
added 2 commits
September 4, 2024 13:52
`StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/) [ghstack-poisoned]
`StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/) [ghstack-poisoned]
junpi3
pushed a commit
that referenced
this pull request
Sep 4, 2024
Pull Request resolved: #5021 `StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. ghstack-source-id: 241047807 @exported-using-ghexport Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/)
|
This pull request was exported from Phabricator. Differential Revision: D59706627 |
added 2 commits
September 4, 2024 13:57
`StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/) [ghstack-poisoned]
`StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/) [ghstack-poisoned]
junpi3
pushed a commit
that referenced
this pull request
Sep 4, 2024
Pull Request resolved: #5021 `StagingBuffer`s are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, [persistently map the memory](https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/memory_mapping.html) once on model load. ghstack-source-id: 241050420 @exported-using-ghexport Differential Revision: [D59706627](https://our.internmc.facebook.com/intern/diff/D59706627/)
|
This pull request was exported from Phabricator. Differential Revision: D59706627 |
junpi3
pushed a commit
that referenced
this pull request
Sep 9, 2024
I missed migration of this callsite in #5021 Differential Revision: [D62403691](https://our.internmc.facebook.com/intern/diff/D62403691/) [ghstack-poisoned]
Merged
junpi3
pushed a commit
that referenced
this pull request
Sep 9, 2024
I missed migration of this callsite in #5021 Differential Revision: [D62403691](https://our.internmc.facebook.com/intern/diff/D62403691/) ghstack-source-id: 241692821 Pull Request resolved: #5202
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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):
StagingBuffers are used for input, output, and constant tensors. For input and output, they are mapped every inference which is unnecessary. Instead, persistently map the memory once on model load.Differential Revision: D59706627