[ET-VK][CI] Add macOS MoltenVK Vulkan model tests

[SS-JIA]

Summary:
Add a test-vulkan-models-macos job to the pull CI workflow that tests
the Vulkan backend on macOS Apple Silicon using MoltenVK. This mirrors
the existing test-vulkan-models-linux job (which uses SwiftShader) but
runs on a macOS M1 runner with the LunarG Vulkan SDK providing MoltenVK
as the Vulkan ICD.

The new setup-vulkan-macos-deps.sh script downloads the macOS Vulkan
SDK and configures VK_ICD_FILENAMES to point at MoltenVK.

Generated with Claude.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/18723

• 📄 Preview Python docs built from this PR

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

❌ 3 New Failures, 1 Cancelled Job, 3 Unrelated Failures

As of commit d036931 with merge base 3d2c853 ():

NEW FAILURES - The following jobs have failed:

• pull / test-vulkan-models-macos / macos-job (gh)
  RuntimeError: Command bash /Users/ec2-user/runner/_work/_temp/exec_script failed with exit code 1
• trunk / unittest-release / linux / linux-job (gh)
  [ FAILED ] KernelTempMemoryAllocatorIntegrationTest.UsingTempMemoryAllocator
• trunk / unittest-release / macos / macos-job (gh)
  [ FAILED ] KernelTempMemoryAllocatorIntegrationTest.UsingTempMemoryAllocator

CANCELLED JOB - The following job was cancelled. Please retry:

• trunk / test-custom-ops-macos (cmake) / macos-job (gh)
  ##[error]The operation was canceled.

FLAKY - The following jobs failed but were likely due to flakiness present on trunk:

• pull / unittest / windows / windows-job (gh) (matched win rule in flaky-rules.json)
  ##[error]The operation was canceled.
• trunk / unittest-release / windows / windows-job (gh) (matched win rule in flaky-rules.json)
  ##[error]The operation was canceled.

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• pull / unittest-editable / windows / windows-job (gh) (trunk failure)
  ##[error]The operation was canceled.

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

[github-actions]

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.

[SS-JIA]

@claude analyse the failing test-vulkan-models-macos job.

[claude]

Claude finished @SS-JIA's task in 7m 58s —— View job

---

Analysis of failing test-vulkan-models-macos job

• Read the changed files for context
• Fetch CI job logs
• Analyze the failure and provide diagnosis

---

Root Cause

The job fails during pip install . --no-build-isolation (the ExecuTorch wheel build) with a CMake error at backends/vulkan/cmake/ShaderLibrary.cmake:30:

CMake Error at backends/vulkan/cmake/ShaderLibrary.cmake:30 (message):
  glslc from the Vulkan SDK must be installed to build the Vulkan backend.
  Please install the Vulkan SDK 1.4.321.0 or newer from
  https://vulkan.lunarg.com/sdk/home and ensure that the glslc binary is in
  your PATH.

CMake's find_program(GLSLC_PATH glslc PATHS $ENV{PATH}) cannot locate glslc.

Analysis

The setup-vulkan-macos-deps.sh script:

1. Downloads vulkansdk-macos-1.4.321.0.zip from LunarG (succeeds — no curl/unzip errors)
2. Extracts to /tmp/vulkansdk/
3. Sets VULKAN_SDK=/tmp/vulkansdk/VulkanSDK/1.4.321.0/macOS
4. Adds ${VULKAN_SDK}/bin to PATH → /tmp/vulkansdk/VulkanSDK/1.4.321.0/macOS/bin

The logs confirm PATH is propagated correctly through the entire chain: outer shell → conda run → setup-macos.sh → install_executorch.sh → pip install . --no-build-isolation → CMake. At line 236988 in the log, inside conda run, PATH includes /tmp/vulkansdk/VulkanSDK/1.4.321.0/macOS/bin.

The problem is that glslc does not exist at /tmp/vulkansdk/VulkanSDK/1.4.321.0/macOS/bin/glslc.

The macOS Vulkan SDK zip likely has a different directory structure than what the script assumes. Comparing with the Linux setup script (.ci/scripts/setup-vulkan-linux-deps.sh), which uses ${_vulkan_sdk_dir}/${VULKAN_SDK_VERSION}/x86_64/bin/ — the macOS SDK zip may similarly place binaries under an architecture-specific path rather than macOS/bin/.

Suggested Fix

Add a diagnostic step to setup-vulkan-macos-deps.sh after extracting the zip to verify the actual directory layout and confirm glslc is present. For example:

# After unzip, verify the SDK structure
ls -la "${VULKAN_SDK}/bin/" || echo "WARNING: ${VULKAN_SDK}/bin/ does not exist"
which glslc || echo "WARNING: glslc not found in PATH"
# Also list the top-level extracted directories to discover actual structure
find /tmp/vulkansdk -name "glslc" -type f 2>/dev/null

The macOS Vulkan SDK zip may place glslc under a path like:

• /tmp/vulkansdk/VulkanSDK/1.4.321.0/macOS/Bin/glslc (capital B)
• /tmp/vulkansdk/VulkanSDK/1.4.321.0/arm64/bin/glslc
• /tmp/vulkansdk/VulkanSDK/1.4.321.0/bin/glslc

Once you identify where glslc actually lives in the extracted zip, update the PATH accordingly in setup-vulkan-macos-deps.sh.

Secondary Note

There is also a minor earlier error at log line 236561:

/Users/ec2-user/runner/_work/_temp/c2e4b4fd-da9c-4cd9-834e-3b0480d21c0e.sh: line 10: 
  /Users/ec2-user/runner/_work/_temp/github_env_24048813425: No such file or directory

This appears to be a non-fatal issue in the macos_job.yml reusable workflow infrastructure, not directly related to the build failure.

---