Vulkan Q8 Conv2D: specialize shader on static parameters and tensor sizes #16036

morgolock · 2025-12-02T13:16:43Z

This change moves all fixed Conv2D parameters (kernel shape, stride, padding, dilation, groups) and the input/output tensor dimensions into Vulkan specialization constants. By making these values compile-time constants, the backend can generate more optimized pipelines, eliminate generic fallback paths, and reduce dynamic indexing overhead. This significantly improves performance across large and compute-intensive convolution workloads.

cc @SS-JIA @manuelcandales @digantdesai @cbilgin

pytorch-bot · 2025-12-02T13:16:50Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/16036

📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 1 New Failure, 1 Unrelated Failure

As of commit 010d843 with merge base 1cb85ef ():

NEW FAILURE - The following job has failed:

pull / test-openvino-linux / linux-job (gh)
RuntimeError: Command docker exec -t 5674f1eeccd237ec7127d7d7cb942b6c077fe0875a0cfd1d7922cf51e2e12cda /exec failed with exit code 1

UNSTABLE - The following job is marked as unstable, possibly due to flakiness on trunk:

pull / android / run-emulator (gh) (#16137)
Timeout waiting for emulator to boot.

This comment was automatically generated by Dr. CI and updates every 15 minutes.

meta-cla · 2025-12-02T13:16:50Z

Hi @morgolock!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at cla@meta.com. Thanks!

github-actions · 2025-12-02T13:17:25Z

This PR needs a `release notes:` label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

Copilot

Pull request overview

This pull request migrates Vulkan Q8 Conv2D operations from using Uniform Buffer Objects (UBOs) to Vulkan specialization constants for all fixed Conv2D parameters (kernel shape, stride, padding, dilation, groups) and tensor dimensions. This architectural change enables the Vulkan compiler to perform compile-time optimizations, eliminate generic fallback paths, and reduce dynamic indexing overhead in the shaders.

Key changes:

Added GenerateSpecConstants function in C++ to create specialization constants from Conv2D parameters and tensor sizes
Replaced UBO parameter buffers with specialization constants across all Conv2D shader variants
Updated GLSL shaders to use specialization constants with flattened naming (e.g., conv2d_params_stride_x instead of conv2d_params.stride.x)

Reviewed changes

Copilot reviewed 18 out of 18 changed files in this pull request and generated 17 comments.

Show a summary per file

File	Description
backends/vulkan/runtime/graph/ops/impl/QuantizedConvolution.cpp	Implements GenerateSpecConstants function; removes UBO parameter buffers from all conv2d dispatch nodes and replaces with spec constants
backends/vulkan/test/custom_ops/utils.cpp	Contains commented-out exception throw (debug code that should be removed)
backends/vulkan/test/custom_ops/q8ta_q8csw_q8to_conv2d.cpp	Contains preprocessor-guarded test configuration changes and commented-out validation (debug code)
backends/vulkan/runtime/graph/ops/glsl/quantize_and_pack_im2col.glsl	Migrates from UBO to specialization constants; reconstructs size vectors from spec constants
backends/vulkan/runtime/graph/ops/glsl/im2col_packed_int8_utils.glslh	Updates to use flattened spec constant names instead of struct member access
backends/vulkan/runtime/graph/ops/glsl/im2col_packed_int8.glsl	Migrates from UBO to specialization constants for Conv2D parameters and tensor sizes
backends/vulkan/runtime/graph/ops/glsl/im2col.glsl	Migrates from UBO to specialization constants; removes struct member access syntax
backends/vulkan/runtime/graph/ops/glsl/conv2d_q8ta_q8csw_q8to_linear_tiled.glsl	Migrates from UBO to specialization constants; contains commented-out input_sizes
backends/vulkan/runtime/graph/ops/glsl/conv2d_q8ta_q8csw_q8to.glsl	Migrates from UBO to specialization constants; updates all parameter references
backends/vulkan/runtime/graph/ops/glsl/conv2d_q8ta_q8csw_linear_tiled.glsl	Migrates from UBO to specialization constants for linear tiled convolution
backends/vulkan/runtime/graph/ops/glsl/conv2d_q8csw_linear_tiled.glsl	Migrates from UBO to specialization constants; updates loop bounds and calculations
backends/vulkan/runtime/graph/ops/glsl/conv2d_q8_utils.glslh	Updates utility functions to use flattened spec constant names
backends/vulkan/runtime/graph/ops/glsl/conv2d_pw_q8ta_q8csw_q8to_tiled.glsl	Migrates pointwise convolution from UBO to specialization constants
backends/vulkan/runtime/graph/ops/glsl/conv2d_fp_im2col_block_load.glslh	Updates block loading logic to use flattened spec constant names
backends/vulkan/runtime/graph/ops/glsl/conv2d_fp_im2col_block.glslh	Updates im2col indexing calculations to use flattened spec constant names
backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_q8ta_q8csw_q8to.glsl	Migrates depthwise convolution from UBO to specialization constants
backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_q8_utils.glslh	Updates depthwise convolution utilities to use flattened spec constant names
backends/vulkan/runtime/graph/ops/glsl/col2im.glsl	Migrates col2im operation from UBO to specialization constants

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2025-12-02T13:25:25Z

backends/vulkan/test/custom_ops/q8ta_q8csw_q8to_conv2d.cpp

+#if 0
      OutInChannels(16, 8), // channels (out, in)
      InputSize2D(21, 17), // input_size (h, w)
      KernelSize(3, 3), // kernel
      Stride(1, 1), // stride
      Padding(1, 1), // padding
      Dilation(1, 1), // dilation
      2, // groups
+#else
+      OutInChannels(128, 128),
+      InputSize2D(128, 128),
+      KernelSize(5, 5),
+      Stride(2, 2),
+      Padding(2, 2),
+      Dilation(1, 1),
+      1,
+#endif


The test configuration has been changed using #if 0 preprocessor directives, leaving the old configuration commented out. This appears to be temporary debug code. Either remove the old configuration entirely, or if both configurations are needed, use a more maintainable approach such as creating separate test functions or configuration parameters.

Copilot · 2025-12-02T13:25:25Z

backends/vulkan/test/custom_ops/q8ta_q8csw_q8to_conv2d.cpp

+//    throw std::invalid_argument(
+//        "One or more dimensions exceed the allowed limit for reference implementation.");
+      std::cout << "Reference implementation: computation may take some time for large tensors..." << std::endl;


The exception throw for dimension limit validation has been commented out and replaced with a warning message. This removes an important safety check that prevents extremely slow reference implementation execution. This appears to be debug code that should not be committed. The exception should be restored to prevent unintended long-running tests.

Suggested change

// throw std::invalid_argument(

// "One or more dimensions exceed the allowed limit for reference implementation.");

std::cout << "Reference implementation: computation may take some time for large tensors..." << std::endl;

throw std::invalid_argument(

"One or more dimensions exceed the allowed limit for reference implementation.");

Copilot · 2025-12-02T13:25:26Z

backends/vulkan/test/custom_ops/q8ta_q8csw_q8to_conv2d.cpp

  // Execute test cases using the new framework with custom FLOP calculator
  auto results = execute_test_cases(
+#if 1
      generate_quantized_conv2d_test_cases,
+#else
+      generate_quantized_conv2d_easy_cases,
+#endif


The test case generator selection is controlled by #if 1 preprocessor directive with the alternative commented out. This appears to be temporary debug code. Either remove the unused alternative or use a more maintainable approach like a command-line flag or configuration parameter.

Suggested change

// Execute test cases using the new framework with custom FLOP calculator

auto results = execute_test_cases(

#if 1

generate_quantized_conv2d_test_cases,

#else

generate_quantized_conv2d_easy_cases,

#endif

// Select test case generator based on command-line argument

// Usage: ./q8ta_q8csw_q8to_conv2d [easy]

auto test_case_generator = generate_quantized_conv2d_test_cases;

if (argc > 1 && std::string(argv[1]) == "easy") {

test_case_generator = generate_quantized_conv2d_easy_cases;

}

// Execute test cases using the new framework with custom FLOP calculator

auto results = execute_test_cases(

test_case_generator,

Copilot · 2025-12-02T13:25:26Z

backends/vulkan/runtime/graph/ops/glsl/im2col_packed_int8.glsl

+//${layout_declare_ubo(B, "ivec4", "output_sizes")}
 // Sizes of the input image
-${layout_declare_ubo(B, "ivec4", "input_sizes")}
-
-${layout_declare_ubo(B, "Conv2DParams", "conv2d_params")}
+//${layout_declare_ubo(B, "ivec4", "input_sizes")}
+
+//${layout_declare_ubo(B, "Conv2DParams", "conv2d_params")}


The old UBO declarations are commented out but not removed. Since these are being replaced by specialization constants, the commented-out code should be completely removed to keep the codebase clean and avoid confusion.

Copilot · 2025-12-02T13:25:26Z

backends/vulkan/runtime/graph/ops/glsl/conv2d_q8ta_q8csw_q8to_linear_tiled.glsl

+//${layout_declare_ubo(B, "ivec4", "output_sizes")}
 ${layout_declare_ubo(B, "ivec4", "im2col_sizes")}
-${layout_declare_ubo(B, "Conv2DParams", "conv2d_params")}
+//${layout_declare_ubo(B, "Conv2DParams", "conv2d_params")}


The old UBO declarations are commented out but not removed. Since these are being replaced by specialization constants, the commented-out code should be completely removed to keep the codebase clean and avoid confusion.

Suggested change

//${layout_declare_ubo(B, "Conv2DParams", "conv2d_params")}

Copilot · 2025-12-02T13:25:29Z

backends/vulkan/runtime/graph/ops/impl/QuantizedConvolution.cpp

 * LICENSE file in the root directory of this source tree.
 */

+#include <iostream>


The #include <iostream> is only used within #if 0 debug code (lines 642-653). Since that debug code should be removed, this include should also be removed to avoid unnecessary dependencies in production code.

Suggested change

#include <iostream>

Copilot · 2025-12-02T13:25:29Z

backends/vulkan/test/custom_ops/utils.cpp

          print_valuespec_data(output_spec, "ref output", true);

-          throw std::runtime_error("Correctness validation failed");
+//          throw std::runtime_error("Correctness validation failed");


This exception throw has been commented out, which means correctness validation failures will no longer stop test execution with an error. This appears to be debug/development code that should not be committed. The proper behavior is to throw an exception when correctness validation fails.

Suggested change

// throw std::runtime_error("Correctness validation failed");

throw std::runtime_error("Correctness validation failed");

Likely needs to be undone

Copilot · 2025-12-02T13:25:30Z

backends/vulkan/runtime/graph/ops/glsl/im2col.glsl

+//${layout_declare_ubo(B, "ivec4", "input_sizes")}
 // Sizes of the output image
-${layout_declare_ubo(B, "ivec4", "output_sizes")}
+//${layout_declare_ubo(B, "ivec4", "output_sizes")}
+
+//${layout_declare_ubo(B, "Conv2DParams", "conv2d_params")}


The old UBO declarations are commented out but not removed. Since these are being replaced by specialization constants, the commented-out code should be completely removed to keep the codebase clean and avoid confusion.

Copilot · 2025-12-02T13:25:30Z

backends/vulkan/runtime/graph/ops/glsl/conv2d_q8ta_q8csw_q8to.glsl

+//${layout_declare_ubo(B, "ivec4", "output_sizes")}
+//${layout_declare_ubo(B, "ivec4", "input_sizes")}
+//${layout_declare_ubo(B, "Conv2DParams", "conv2d_params")}


The old UBO declarations are commented out but not removed. Since these are being replaced by specialization constants, the commented-out code should be completely removed to keep the codebase clean and avoid confusion.

Suggested change

//${layout_declare_ubo(B, "ivec4", "output_sizes")}

//${layout_declare_ubo(B, "ivec4", "input_sizes")}

//${layout_declare_ubo(B, "Conv2DParams", "conv2d_params")}

Copilot · 2025-12-02T13:25:30Z

backends/vulkan/runtime/graph/ops/glsl/conv2d_pw_q8ta_q8csw_q8to_tiled.glsl

+//${layout_declare_ubo(B, "ivec4", "output_sizes")}
+//${layout_declare_ubo(B, "ivec4", "input_sizes")}
+//${layout_declare_ubo(B, "Conv2DParams", "conv2d_params")}



The old UBO declarations are commented out but not removed. Since these are being replaced by specialization constants, the commented-out code should be completely removed to keep the codebase clean and avoid confusion.

Suggested change

//${layout_declare_ubo(B, "ivec4", "output_sizes")}

//${layout_declare_ubo(B, "ivec4", "input_sizes")}

//${layout_declare_ubo(B, "Conv2DParams", "conv2d_params")}

Copilot

Pull request overview

Copilot reviewed 18 out of 18 changed files in this pull request and generated 4 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2025-12-09T08:43:04Z

backends/vulkan/test/custom_ops/q8ta_q8csw_q8to_conv2d.cpp

+#if 0
      OutInChannels(16, 8), // channels (out, in)
      InputSize2D(21, 17), // input_size (h, w)
      KernelSize(3, 3), // kernel
      Stride(1, 1), // stride
      Padding(1, 1), // padding
      Dilation(1, 1), // dilation
      2, // groups
+#else
+      OutInChannels(128, 128),
+      InputSize2D(128, 128),
+      KernelSize(5, 5),
+      Stride(2, 2),
+      Padding(2, 2),
+      Dilation(1, 1),
+      1,
+#endif


The #if 0 preprocessor directive disables the original test configuration. This appears to be temporary debugging code that should not be committed. Please remove the #if 0 ... #else ... #endif block and restore the appropriate test configuration.

Removed #if 0

Copilot · 2025-12-09T08:43:05Z

backends/vulkan/runtime/graph/ops/impl/QuantizedConvolution.cpp

    apply_bias = 0;
  }

+  vkapi::SpecVarList spec_constants = GenerateSpecConstants(graph, conv_params, groups, output_image);


Incorrect parameter passed to GenerateSpecConstants. The fourth parameter should be apply_bias, but output_image (a ValueRef) is being passed instead. This will not compile or will produce incorrect specialization constants. Should be: GenerateSpecConstants(graph, conv_params, groups, apply_bias)

Suggested change

vkapi::SpecVarList spec_constants = GenerateSpecConstants(graph, conv_params, groups, output_image);

vkapi::SpecVarList spec_constants = GenerateSpecConstants(graph, conv_params, groups, apply_bias);

Valid comment here as well

Made changes as suggested

Copilot · 2025-12-09T08:43:05Z

backends/vulkan/runtime/graph/ops/impl/QuantizedConvolution.cpp

+      conv2d_params_dilation_x, conv2d_params_dilation_y,
+      conv2d_params_kernel_size_x, conv2d_params_kernel_size_y,
+      in_channels_per_group, out_channels_per_group,
+      K4_per_group, K4, K_per_group, logical_K, logical_K_per_group, groups


Incorrect type for groups in the specialization constants list. The groups parameter is a ValueRef, but it should be converted to a uint32_t using graph.get_int(groups) before being added to the spec_constants list. This will cause a type mismatch. Should be: K4_per_group, K4, K_per_group, logical_K, logical_K_per_group, graph.get_int(groups)

Suggested change

K4_per_group, K4, K_per_group, logical_K, logical_K_per_group, groups

K4_per_group, K4, K_per_group, logical_K, logical_K_per_group, graph.get_int(groups)

valid comment - need to first extract the value of groups i.e.

uint32_t groups_val = graph.get_int(groups);

As suggested using uint32_t groups_val = graph.get_int(groups);

Copilot · 2025-12-09T08:43:05Z

backends/vulkan/runtime/graph/ops/impl/QuantizedConvolution.cpp

 * LICENSE file in the root directory of this source tree.
 */

+#include <iostream>


The #include <iostream> directive is added but doesn't appear to be used anywhere in this file. Consider removing this unused include.

Suggested change

#include <iostream>

SS-JIA

Some valid comments by the AI reviewer

SS-JIA · 2025-12-11T15:48:00Z

backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_q8ta_q8csw_q8to.glsl

 ${layout_declare_ubo(B, "ivec4", "output_sizes")}
 ${layout_declare_ubo(B, "ivec4", "input_sizes")}
-${layout_declare_ubo(B, "Conv2DParams", "conv2d_params")}
+//${layout_declare_ubo(B, "Conv2DParams", "conv2d_params")}


Remove commented section?

SS-JIA · 2025-12-11T15:49:37Z

backends/vulkan/runtime/graph/ops/impl/QuantizedConvolution.cpp

+      conv2d_params_dilation_x, conv2d_params_dilation_y,
+      conv2d_params_kernel_size_x, conv2d_params_kernel_size_y,
+      in_channels_per_group, out_channels_per_group,
+      K4_per_group, K4, K_per_group, logical_K, logical_K_per_group, groups


valid comment - need to first extract the value of groups i.e.

uint32_t groups_val = graph.get_int(groups);

SS-JIA · 2025-12-11T15:50:38Z

backends/vulkan/runtime/graph/ops/impl/QuantizedConvolution.cpp

    apply_bias = 0;
  }

+  vkapi::SpecVarList spec_constants = GenerateSpecConstants(graph, conv_params, groups, output_image);


Valid comment here as well

SS-JIA · 2025-12-11T15:51:03Z

backends/vulkan/test/custom_ops/utils.cpp

          print_valuespec_data(output_spec, "ref output", true);

-          throw std::runtime_error("Correctness validation failed");
+//          throw std::runtime_error("Correctness validation failed");


Likely needs to be undone

Copilot

Pull request overview

Copilot reviewed 16 out of 17 changed files in this pull request and generated 2 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2025-12-11T18:30:00Z

backends/vulkan/test/custom_ops/q8ta_q8csw_q8to_conv2d.cpp

+      throw std::invalid_argument(
        "One or more dimensions exceed the allowed limit for reference implementation.");
+      std::cout << "Reference implementation: computation may take some time for large tensors..." << std::endl;


This line of code is unreachable because it appears after a throw statement on line 459. The exception will be thrown and execution will not reach this cout statement. Either the cout should be moved before the throw, or the throw should be removed if this was intended to be a warning rather than an error.

Suggested change

throw std::invalid_argument(

"One or more dimensions exceed the allowed limit for reference implementation.");

std::cout << "Reference implementation: computation may take some time for large tensors..." << std::endl;

std::cout << "Reference implementation: computation may take some time for large tensors..." << std::endl;

throw std::invalid_argument(

"One or more dimensions exceed the allowed limit for reference implementation.");

Copilot · 2025-12-11T18:30:01Z

backends/vulkan/runtime/graph/ops/glsl/conv2d_q8ta_q8csw_q8to_linear_tiled.glsl

 ${layout_declare_tensor(B, "r", "t_bias", DTYPE, "buffer", is_scalar_array=False)}

-${layout_declare_ubo(B, "ivec4", "output_sizes")}
+//${layout_declare_ubo(B, "ivec4", "output_sizes")}


The output_sizes UBO is commented out but is still being used on line 91 where it's passed to make_block_extents(output_sizes). This will cause a compilation error as output_sizes is not defined. Either this UBO declaration should not be commented out, or the usage should be replaced with an alternative method to obtain output sizes.

Suggested change

//${layout_declare_ubo(B, "ivec4", "output_sizes")}

${layout_declare_ubo(B, "ivec4", "output_sizes")}

Copilot

Pull request overview

Copilot reviewed 16 out of 17 changed files in this pull request and generated 1 comment.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2025-12-15T10:15:44Z

backends/vulkan/test/custom_ops/q8ta_q8csw_q8to_conv2d.cpp

-    throw std::invalid_argument(
+      throw std::invalid_argument(
        "One or more dimensions exceed the allowed limit for reference implementation.");
+      std::cout << "Reference implementation: computation may take some time for large tensors..." << std::endl;


The std::cout statement on line 461 is unreachable code because the exception is thrown on lines 459-460 immediately before it. If the dimensions exceed the limit, the function will exit via the exception and never print the message. Either remove this line or move it before the throw statement if you want to log this information before throwing.

Suggested change

std::cout << "Reference implementation: computation may take some time for large tensors..." << std::endl;

SS-JIA

LGTM. Thank you for this optimization - it will be critical for deploying ET-VK effectively on Arm GPUs!

SS-JIA

Putting a temporary RC since I discovered that using specialization constants this heavily almost doubles model load time.

Let's discuss this further in tomorrow's meeting!

Copilot

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

SS-JIA · 2025-12-17T20:24:50Z

@morgolock it seems that you will need to run lintrunner before this PR can be merged to ensure that lint is clean 😅

To do so:

pip install lintrunner
pip install lintrunner_adapters

cd ~/executorch
lintrunner init
lintrunner lint --apply-patches --verbose -m main

…izes This change moves all fixed Conv2D parameters (kernel shape, stride, padding, dilation, groups) into Vulkan specialization constants. By making these values compile-time constants, the backend can generate more optimized pipelines, eliminate generic fallback paths, and reduce dynamic indexing overhead. This significantly improves performance across large and compute-intensive convolution workloads. Signed-off-by: Pablo Marquez Tello <pablo.tello@arm.com> Change-Id: I3efe3de80dece91341ae4111bef1254c6779a1db

meta-codesync · 2025-12-18T00:32:39Z

@SS-JIA has imported this pull request. If you are a Meta employee, you can view this in D89421309.

Copilot AI review requested due to automatic review settings December 2, 2025 13:16

morgolock requested a review from SS-JIA as a code owner December 2, 2025 13:16

pytorch-bot bot added the module: vulkan Issues related to the Vulkan delegate and code under backends/vulkan/ label Dec 2, 2025

Copilot started reviewing on behalf of morgolock December 2, 2025 13:17 View session

Copilot finished reviewing on behalf of morgolock December 2, 2025 13:21

Copilot AI reviewed Dec 2, 2025

View reviewed changes

robell requested a review from perheld December 4, 2025 10:03

morgolock force-pushed the vk_spec_consts branch 2 times, most recently from d902b0f to 960ac20 Compare December 8, 2025 10:43

Copilot AI review requested due to automatic review settings December 9, 2025 08:37

morgolock force-pushed the vk_spec_consts branch from 960ac20 to b87778b Compare December 9, 2025 08:37

Copilot started reviewing on behalf of morgolock December 9, 2025 08:37 View session

Copilot AI reviewed Dec 9, 2025

View reviewed changes

morgolock force-pushed the vk_spec_consts branch from b87778b to 78622ab Compare December 9, 2025 08:53

SS-JIA requested changes Dec 11, 2025

View reviewed changes

Copilot AI review requested due to automatic review settings December 11, 2025 18:25

morgolock force-pushed the vk_spec_consts branch from 78622ab to a9228e7 Compare December 11, 2025 18:25

Copilot started reviewing on behalf of morgolock December 11, 2025 18:25 View session

Copilot AI reviewed Dec 11, 2025

View reviewed changes

morgolock force-pushed the vk_spec_consts branch from a9228e7 to a3f3d21 Compare December 12, 2025 09:53

Copilot AI review requested due to automatic review settings December 15, 2025 10:10

morgolock force-pushed the vk_spec_consts branch from a3f3d21 to c1067ec Compare December 15, 2025 10:10

Copilot started reviewing on behalf of morgolock December 15, 2025 10:11 View session

Copilot AI reviewed Dec 15, 2025

View reviewed changes

morgolock force-pushed the vk_spec_consts branch from c1067ec to 66d4f1a Compare December 16, 2025 10:09

SS-JIA approved these changes Dec 16, 2025

View reviewed changes

meta-cla bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Dec 16, 2025

SS-JIA requested changes Dec 17, 2025

View reviewed changes

SS-JIA approved these changes Dec 17, 2025

View reviewed changes

Copilot AI review requested due to automatic review settings December 17, 2025 18:00

Copilot AI reviewed Dec 17, 2025

View reviewed changes

morgolock force-pushed the vk_spec_consts branch from b624b7c to 010d843 Compare December 17, 2025 21:11

SS-JIA merged commit 73ae813 into pytorch:main Dec 22, 2025
138 of 140 checks passed

	// throw std::runtime_error("Correctness validation failed");
	throw std::runtime_error("Correctness validation failed");

	//${layout_declare_ubo(B, "ivec4", "output_sizes")}
	//${layout_declare_ubo(B, "ivec4", "input_sizes")}
	//${layout_declare_ubo(B, "Conv2DParams", "conv2d_params")}

	vkapi::SpecVarList spec_constants = GenerateSpecConstants(graph, conv_params, groups, output_image);
	vkapi::SpecVarList spec_constants = GenerateSpecConstants(graph, conv_params, groups, apply_bias);

	K4_per_group, K4, K_per_group, logical_K, logical_K_per_group, groups
	K4_per_group, K4, K_per_group, logical_K, logical_K_per_group, graph.get_int(groups)

	//${layout_declare_ubo(B, "ivec4", "output_sizes")}
	${layout_declare_ubo(B, "ivec4", "output_sizes")}

Vulkan Q8 Conv2D: specialize shader on static parameters and tensor sizes #16036

Vulkan Q8 Conv2D: specialize shader on static parameters and tensor sizes #16036

Uh oh!

Conversation

morgolock commented Dec 2, 2025 • edited by pytorch-bot bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot bot commented Dec 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/16036

❌ 1 New Failure, 1 Unrelated Failure

Uh oh!

meta-cla bot commented Dec 2, 2025

Action Required

Process

Uh oh!

github-actions bot commented Dec 2, 2025

This PR needs a release notes: label

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Dec 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 2, 2025

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Dec 9, 2025

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 9, 2025

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 9, 2025

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

morgolock commented Dec 2, 2025 •

edited by pytorch-bot bot

Loading

pytorch-bot bot commented Dec 2, 2025 •

edited

Loading

This PR needs a `release notes:` label