[ET-VK] Add nchw_to_int8x4_buffer shader for prepacking int8 staging data #17392
+227
−4
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
…data This adds a GLSL compute shader and supporting C++ dispatch logic to transfer int8 tensor data from a staging buffer (in NCHW contiguous order) to a GPU buffer in any PackedInt8 layout (4W, 4C, 4W4C, 4H4W, 4C1W). Previously there was no prepack path for kInt8x4 tensors, so constant int8 tensors (TensorRef inputs) could not be transferred to GPU buffers. This is needed to support constant quantized weights in q8ta operators. The shader uses texel-level dispatch where each thread writes one texel (one int32 = 4 packed int8 values). It decomposes the texel index into block-space coordinates using BufferMetadata strides and hashed_layout dim_order, then reads the corresponding int8 bytes from the staging buffer (interpreted as int32 for device compatibility, avoiding the need for 8-bit buffer support). New files: - nchw_to_int8x4_buffer.glsl: Compute shader handling all PackedInt8 layouts - nchw_to_int8x4_buffer.yaml: Shader variant config - Q8taStaging.h/cpp: C++ dispatch function creating the PrepackNode Modified files: - Staging.cpp: Routes kInt8x4 tensors in prepack_op() to the new function - TestQ8taBinary.cpp: Prepacks TensorRef inputs before quantization - test_q8ta_binary.cpp: Adds const_b test cases for constant tensor B inputs This diff was authored with Claude. Differential Revision: [D93000169](https://our.internmc.facebook.com/intern/diff/D93000169/) [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/17392
Note: Links to docs will display an error until the docs builds have been completed. ❌ 3 New Failures, 1 Unrelated FailureAs of commit b58d3cb with merge base 964c565 ( NEW FAILURES - The following jobs have failed:
BROKEN TRUNK - The following job failed but were 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 Feb 11, 2026
This PR needs a
|
meta-cla
bot
added
the
CLA Signed
…t8 staging data" This adds a GLSL compute shader and supporting C++ dispatch logic to transfer int8 tensor data from a staging buffer (in NCHW contiguous order) to a GPU buffer in any PackedInt8 layout (4W, 4C, 4W4C, 4H4W, 4C1W). Previously there was no prepack path for kInt8x4 tensors, so constant int8 tensors (TensorRef inputs) could not be transferred to GPU buffers. This is needed to support constant quantized weights in q8ta operators. The shader uses texel-level dispatch where each thread writes one texel (one int32 = 4 packed int8 values). It decomposes the texel index into block-space coordinates using BufferMetadata strides and hashed_layout dim_order, then reads the corresponding int8 bytes from the staging buffer (interpreted as int32 for device compatibility, avoiding the need for 8-bit buffer support). New files: - nchw_to_int8x4_buffer.glsl: Compute shader handling all PackedInt8 layouts - nchw_to_int8x4_buffer.yaml: Shader variant config - Q8taStaging.h/cpp: C++ dispatch function creating the PrepackNode Modified files: - Staging.cpp: Routes kInt8x4 tensors in prepack_op() to the new function - TestQ8taBinary.cpp: Prepacks TensorRef inputs before quantization - test_q8ta_binary.cpp: Adds const_b test cases for constant tensor B inputs This diff was authored with Claude. Differential Revision: [D93000169](https://our.internmc.facebook.com/intern/diff/D93000169/) [ghstack-poisoned]
…t8 staging data" This adds a GLSL compute shader and supporting C++ dispatch logic to transfer int8 tensor data from a staging buffer (in NCHW contiguous order) to a GPU buffer in any PackedInt8 layout (4W, 4C, 4W4C, 4H4W, 4C1W). Previously there was no prepack path for kInt8x4 tensors, so constant int8 tensors (TensorRef inputs) could not be transferred to GPU buffers. This is needed to support constant quantized weights in q8ta operators. The shader uses texel-level dispatch where each thread writes one texel (one int32 = 4 packed int8 values). It decomposes the texel index into block-space coordinates using BufferMetadata strides and hashed_layout dim_order, then reads the corresponding int8 bytes from the staging buffer (interpreted as int32 for device compatibility, avoiding the need for 8-bit buffer support). New files: - nchw_to_int8x4_buffer.glsl: Compute shader handling all PackedInt8 layouts - nchw_to_int8x4_buffer.yaml: Shader variant config - Q8taStaging.h/cpp: C++ dispatch function creating the PrepackNode Modified files: - Staging.cpp: Routes kInt8x4 tensors in prepack_op() to the new function - TestQ8taBinary.cpp: Prepacks TensorRef inputs before quantization - test_q8ta_binary.cpp: Adds const_b test cases for constant tensor B inputs This diff was authored with Claude. Differential Revision: [D93000169](https://our.internmc.facebook.com/intern/diff/D93000169/) [ghstack-poisoned]
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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 adds a GLSL compute shader and supporting C++ dispatch logic to transfer
int8 tensor data from a staging buffer (in NCHW contiguous order) to a GPU
buffer in any PackedInt8 layout (4W, 4C, 4W4C, 4H4W, 4C1W).
Previously there was no prepack path for kInt8x4 tensors, so constant int8
tensors (TensorRef inputs) could not be transferred to GPU buffers. This is
needed to support constant quantized weights in q8ta operators.
The shader uses texel-level dispatch where each thread writes one texel (one
int32 = 4 packed int8 values). It decomposes the texel index into block-space
coordinates using BufferMetadata strides and hashed_layout dim_order, then reads
the corresponding int8 bytes from the staging buffer (interpreted as int32 for
device compatibility, avoiding the need for 8-bit buffer support).
New files:
Modified files:
This diff was authored with Claude.
Differential Revision: D93000169