[ET-VK] Efficient tiled int8 matmul #9804
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
Pull Request resolved: #9765 ## Context The weight tensor of a linear layer is usually stored in a transposed manner, such that when computing the matrix multiplication, the reduction traverses along the rows of the weight tensor as opposed to the columns. This results in a better memory access pattern for CPUs. However, for GPUs, I have found that "un-transposing" the weight tensors result in better performance. This is likely due to the fact since GPUs can compute multiple output elements in parallel, reading along the columns allows for coalescing memory loads among threads in a work group. ## Changes * Introduce the ability to transpose height and weight dims when transferring tensor data to the GPU. * Prepackthe weight tensor "un-transposed" for the int8 quantized linear operator ghstack-source-id: 275180033 @exported-using-ghexport Differential Revision: [D72066588](https://our.internmc.facebook.com/intern/diff/D72066588/)
Pull Request resolved: #9766 ## Context Introduce a optimized tiled implementation for computing the weight int8-quantized linear operation. This implementation takes advantage of the following principles to squeeze out performance: * Compute an output tile with each thread, rather than a single output element. This allows for better memory re-use of loaded input tensor data. * Compute the output tile by iteratively loading tiles of the input matrices, caching them in registers, and then performing the `fma` accumulations to obtain a partial output. By splitting the data loading and computation into distinct steps, the GPU is able to perform latency hiding more effectively, i.e. switching to a warp that needs to perform compute when the current warp is waiting on data load * Use a work group size of `{N, 1, 1}`. This makes it so that all the threads in a work group load the same row of the input matrx, and consecutive columns of the weight matrix. This way, the row of the input is kept hot in the cache, and accesses to the weight matrix can be coalesced due to the previous diff un-transposing the weight matrix. Differential Revision: [D72066587](https://our.internmc.facebook.com/intern/diff/D72066587/) ghstack-source-id: 275180032
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/9804
Note: Links to docs will display an error until the docs builds have been completed. ⏳ No Failures, 17 PendingAs of commit 1907ae2 with merge base 2aa7748 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
SS-JIA
approved these changes
Apr 1, 2025
kirklandsign
pushed a commit
that referenced
this pull request
Apr 11, 2025
Pull Request resolved: #9766 ## Context Introduce a optimized tiled implementation for computing the weight int8-quantized linear operation. This implementation takes advantage of the following principles to squeeze out performance: * Compute an output tile with each thread, rather than a single output element. This allows for better memory re-use of loaded input tensor data. * Compute the output tile by iteratively loading tiles of the input matrices, caching them in registers, and then performing the `fma` accumulations to obtain a partial output. By splitting the data loading and computation into distinct steps, the GPU is able to perform latency hiding more effectively, i.e. switching to a warp that needs to perform compute when the current warp is waiting on data load * Use a work group size of `{N, 1, 1}`. This makes it so that all the threads in a work group load the same row of the input matrx, and consecutive columns of the weight matrix. This way, the row of the input is kept hot in the cache, and accesses to the weight matrix can be coalesced due to the previous diff un-transposing the weight matrix. Differential Revision: [D72066587](https://our.internmc.facebook.com/intern/diff/D72066587/)
This was referenced Apr 14, 2025
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.
This PR was created by the merge bot to help merge the original PR into the main branch.
ghstack PR number: #9766 by @SS-JIA
^ 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/SS-JIA/205/base
ghstack PR head: https://github.com/pytorch/executorch/tree/gh/SS-JIA/205/head
Merge bot PR base: https://github.com/pytorch/executorch/tree/gh/SS-JIA/204/orig
Merge bot PR head: https://github.com/pytorch/executorch/tree/gh/SS-JIA/205/orig
@diff-train-skip-merge