[ET-VK] Update fused SDPA operator to support ViT attention#19114
Merged
meta-codesync[bot] merged 4 commits intogh/SS-JIA/523/basefrom Apr 27, 2026
Merged
[ET-VK] Update fused SDPA operator to support ViT attention#19114meta-codesync[bot] merged 4 commits intogh/SS-JIA/523/basefrom
meta-codesync[bot] merged 4 commits intogh/SS-JIA/523/basefrom
Conversation
This diff extends the ET-VK fused SDPA operator so it can be used for the ViT attention blocks in the EdgeTAM ViT-S encoder. The main correctness problem is that Q@K^T dot products in ViT attention can exceed the fp16 max (65504), so fp32 accumulation is required. **fp16 overflow fix**: The intermediate `attn_weights` buffer is now always fp32 regardless of input dtype. Previously the QK shader accumulated in fp32 but stored to an fp16 buffer, causing overflow. The softmax shader reads fp32 attention weights and writes fp16 softmax output (safe since values are in [0, 1]). **Texture support**: The QK and AV shaders support both buffer and texture3d storage for Q/K/V/output. The intermediate `attn_weights` and `attn_weights_softmax` tensors now inherit the storage type of the input/output (q_projected for the LLM path, out for the fused path), so the entire fused SDPA pipeline runs in a uniform storage type and no SDPA-internal layout transitions are needed. Differential Revision: [D102360200](https://our.internmc.facebook.com/intern/diff/D102360200/) [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19114
Note: Links to docs will display an error until the docs builds have been completed. ❗ 1 Active SEVsThere are 1 currently active SEVs. If your PR is affected, please view them below: ⏳ 2 Pending, 3 Unrelated FailuresAs of commit 55b382d with merge base bf64fa1 ( FLAKY - The following job failed but was likely due to flakiness present on trunk:
BROKEN TRUNK - The following jobs failed but was present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This was referenced Apr 24, 2026
meta-cla
Bot
added
the
CLA Signed
This PR needs a
|
added 3 commits
April 24, 2026 12:29
This diff extends the ET-VK fused SDPA operator so it can be used for the ViT attention blocks in the EdgeTAM ViT-S encoder. The main correctness problem is that QK^T dot products in ViT attention can exceed the fp16 max (65504), so fp32 accumulation is required. **fp16 overflow fix**: The intermediate `attn_weights` buffer is now always fp32 regardless of input dtype. Previously the QK shader accumulated in fp32 but stored to an fp16 buffer, causing overflow. The softmax shader reads fp32 attention weights and writes fp16 softmax output (safe since values are in [0, 1]). **Texture support**: The QK and AV shaders support both buffer and texture3d storage for Q/K/V/output. The intermediate `attn_weights` and `attn_weights_softmax` tensors now inherit the storage type of the input/output (q_projected for the LLM path, out for the fused path), so the entire fused SDPA pipeline runs in a uniform storage type and no SDPA-internal layout transitions are needed. Differential Revision: [D102360200](https://our.internmc.facebook.com/intern/diff/D102360200/) [ghstack-poisoned]
This diff extends the ET-VK fused SDPA operator so it can be used for the ViT attention blocks in the EdgeTAM ViT-S encoder. The main correctness problem is that QK^T dot products in ViT attention can exceed the fp16 max (65504), so fp32 accumulation is required. **fp16 overflow fix**: The intermediate `attn_weights` buffer is now always fp32 regardless of input dtype. Previously the QK shader accumulated in fp32 but stored to an fp16 buffer, causing overflow. The softmax shader reads fp32 attention weights and writes fp16 softmax output (safe since values are in [0, 1]). **Texture support**: The QK and AV shaders support both buffer and texture3d storage for Q/K/V/output. The intermediate `attn_weights` and `attn_weights_softmax` tensors now inherit the storage type of the input/output (q_projected for the LLM path, out for the fused path), so the entire fused SDPA pipeline runs in a uniform storage type and no SDPA-internal layout transitions are needed. Differential Revision: [D102360200](https://our.internmc.facebook.com/intern/diff/D102360200/) [ghstack-poisoned]
This diff extends the ET-VK fused SDPA operator so it can be used for the ViT attention blocks in the EdgeTAM ViT-S encoder. The main correctness problem is that QK^T dot products in ViT attention can exceed the fp16 max (65504), so fp32 accumulation is required. **fp16 overflow fix**: The intermediate `attn_weights` buffer is now always fp32 regardless of input dtype. Previously the QK shader accumulated in fp32 but stored to an fp16 buffer, causing overflow. The softmax shader reads fp32 attention weights and writes fp16 softmax output (safe since values are in [0, 1]). **Texture support**: The QK and AV shaders support both buffer and texture3d storage for Q/K/V/output. The intermediate `attn_weights` and `attn_weights_softmax` tensors now inherit the storage type of the input/output (q_projected for the LLM path, out for the fused path), so the entire fused SDPA pipeline runs in a uniform storage type and no SDPA-internal layout transitions are needed. Differential Revision: [D102360200](https://our.internmc.facebook.com/intern/diff/D102360200/) [ghstack-poisoned]
GregoryComer
approved these changes
Apr 27, 2026
meta-codesync
Bot
merged commit 226a7b3
into
gh/SS-JIA/523/base
169 of 173 checks passed
SS-JIA
pushed a commit
that referenced
this pull request
Apr 27, 2026
Pull Request resolved: #19114 This diff extends the ET-VK fused SDPA operator so it can be used for the ViT attention blocks in the EdgeTAM ViT-S encoder. The main correctness problem is that Q@K^T dot products in ViT attention can exceed the fp16 max (65504), so fp32 accumulation is required. **fp16 overflow fix**: The intermediate `attn_weights` buffer is now always fp32 regardless of input dtype. Previously the QK shader accumulated in fp32 but stored to an fp16 buffer, causing overflow. The softmax shader reads fp32 attention weights and writes fp16 softmax output (safe since values are in [0, 1]). **Texture support**: The QK and AV shaders support both buffer and texture3d storage for Q/K/V/output. The intermediate `attn_weights` and `attn_weights_softmax` tensors now inherit the storage type of the input/output (q_projected for the LLM path, out for the fused path), so the entire fused SDPA pipeline runs in a uniform storage type and no SDPA-internal layout transitions are needed. ghstack-source-id: 373258239 @exported-using-ghexport Differential Revision: [D102360200](https://our.internmc.facebook.com/intern/diff/D102360200/)
zeel2104
pushed a commit
to zeel2104/executorch
that referenced
this pull request
May 5, 2026
Pull Request resolved: pytorch#19114 This diff extends the ET-VK fused SDPA operator so it can be used for the ViT attention blocks in the EdgeTAM ViT-S encoder. The main correctness problem is that Q@K^T dot products in ViT attention can exceed the fp16 max (65504), so fp32 accumulation is required. **fp16 overflow fix**: The intermediate `attn_weights` buffer is now always fp32 regardless of input dtype. Previously the QK shader accumulated in fp32 but stored to an fp16 buffer, causing overflow. The softmax shader reads fp32 attention weights and writes fp16 softmax output (safe since values are in [0, 1]). **Texture support**: The QK and AV shaders support both buffer and texture3d storage for Q/K/V/output. The intermediate `attn_weights` and `attn_weights_softmax` tensors now inherit the storage type of the input/output (q_projected for the LLM path, out for the fused path), so the entire fused SDPA pipeline runs in a uniform storage type and no SDPA-internal layout transitions are needed. ghstack-source-id: 373258239 @exported-using-ghexport Differential Revision: [D102360200](https://our.internmc.facebook.com/intern/diff/D102360200/)
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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):
This diff extends the ET-VK fused SDPA operator so it can be used for the ViT attention blocks in the EdgeTAM ViT-S encoder. The main correctness problem is that Q@K^T dot products in ViT attention can exceed the fp16 max (65504), so fp32 accumulation is required.
fp16 overflow fix: The intermediate
attn_weightsbuffer is now always fp32 regardless of input dtype. Previously the QK shader accumulated in fp32 but stored to an fp16 buffer, causing overflow. The softmax shader reads fp32 attention weights and writes fp16 softmax output (safe since values are in [0, 1]).Texture support: The QK and AV shaders support both buffer and texture3d storage for Q/K/V/output. The intermediate
attn_weightsandattn_weights_softmaxtensors now inherit the storage type of the input/output (q_projected for the LLM path, out for the fused path), so the entire fused SDPA pipeline runs in a uniform storage type and no SDPA-internal layout transitions are needed.Differential Revision: D102360200