[ExecuTorch][Vulkan] Add et_vk.fused_ce kernels (fused cross-entropy loss + dlogits)#20941
Merged
Conversation
pytorch-bot
Bot
added
the
module: vulkan
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20941
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New Failure, 1 Cancelled Job, 1 Unrelated FailureAs of commit f5d76df with merge base 21554e5 ( NEW FAILURE - The following job has failed:
CANCELLED JOB - The following job was cancelled. Please retry:
FLAKY - The following job failed but was likely due to flakiness present on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This was referenced Jul 14, 2026
This PR needs a
|
This was referenced Jul 14, 2026
Merged
[ExecuTorch][WebGPU] Add et_vk.adamw_step (on-GPU AdamW optimizer step) to the WebGPU backend
#20935
Merged
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
JCNTH
added a commit
that referenced
this pull request
Jul 21, 2026
…loss + dlogits) Pull Request resolved: #20941 Vulkan GLSL kernels + handler for the training custom op `et_vk.fused_ce` (fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured like `glsl/reduce_per_row_buffer.glsl` (`NWORKERS=64`, one workgroup/row, barrier tree-combine); two dispatch nodes share `loss_partial`, ordered by the runtime per-`vTensor` barrier (same pattern as `impl/SDPA.cpp` QK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner. ghstack-source-id: 405059103 @exported-using-ghexport Differential Revision: [D111761780](https://our.internmc.facebook.com/intern/diff/D111761780/)
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):
Vulkan GLSL kernels + handler for the training custom op
et_vk.fused_ce(fused CE: per-row online-softmax loss + dlogits, then an N->1 loss sum). Per-row shared-memory reduction structured likeglsl/reduce_per_row_buffer.glsl(NWORKERS=64, one workgroup/row, barrier tree-combine); two dispatch nodes shareloss_partial, ordered by the runtime per-vTensorbarrier (same pattern asimpl/SDPA.cppQK->softmax->AV). AOT registration already exists under the shared Vulkan partitioner.@exported-using-ghexport
Differential Revision: D111761780
Differential Revision: D111761780