ML SDK for Vulkan®

The ML SDK for Vulkan® is an SDK to facilitate the development of projects using the ML extensions for Vulkan®.

Content

This repository contains scripts to facilitate building all the ML SDK for Vulkan® components, utilities, and documentation.

Cloning the repository

To clone the ML SDK for Vulkan® repository, you can use the Repo tool. The tool is here. Cloning using Repo tool ensures all components and dependencies are fetched and in a suitable default location on the file system. To initialize the repo structure for the entire ML SDK for Vulkan®, including all components, run:

repo init -u https://github.com/arm/ai-ml-sdk-manifest.git -g all

After the repository is initialized, fetch the contents with:

repo sync --no-clone-bundle

Cloning on Windows®

To ensure nested submodules do not exceed the maximum long path length, you must enable long paths on Windows®, and you must clone close to the root directory or use a symlink. Make sure to use Git for Windows.

Using PowerShell:

Set-ItemProperty -Path "HKLM:\SYSTEM\CurrentControlSet\Control\FileSystem" -Name "LongPathsEnabled" -Value 1
git config --global core.longpaths true
git --version # Ensure you are using Git for Windows, for example 2.50.1.windows.1
git clone <git-repo-tool-url>
python <path-to-git-repo>\git-repo\repo init -u <manifest-url> -g all
python <path-to-git-repo>\git-repo\repo sync --no-clone-bundle

Using Git Bash:

cmd.exe "/c reg.exe add \"HKLM\System\CurrentControlSet\Control\FileSystem"" /v LongPathsEnabled /t REG_DWORD /d 1 /f"
git config --global core.longpaths true
git --version # Ensure you are using the Git for Windows, for example 2.50.1.windows.1
git clone <git-repo-tool-url>
python <path-to-git-repo>/git-repo/repo init -u <manifest-url> -g all
python <path-to-git-repo>/git-repo/repo sync --no-clone-bundle

Due to a known issue in git-repo, nested submodules do not always update as part of repo sync and need to be manually updated, for example:

cd dependencies/tosa_mlir_translator
git submodule update --init --recursive

After the sync command completes successfully, you can find the individual components in <repo_root>/sw/. You can also find all the required dependencies in <repo_root>/dependencies/.

You can initialize the repository with only the components that you need. This also fetches only the dependencies required by the selected components. The -g all in the preceding initialization command is equivalent to -g model-converter vgf-lib scenario-runner emulation-layer. Some components may also force other components to be included. For example, scenario-runner automatically includes vgf-lib because vgf-lib is required.

Although we do not recommend this method, you can check out the components using git clone commands directly. This requires checking out each component and dependency manually. The custom location to each dependency needs to be configured explicitly by options to the build command for each component which results in a more complicated workflow.

Build all the ML SDK for Vulkan® components

The ML SDK for Vulkan® consists of four components:

• VGF Library
• Model Converter
• Scenario Runner
• Emulation Layer

These can be built from the ML SDK for Vulkan® root repository or individually from their respective repositories.

The build system must have:

• CMake 3.25 or later.
• C/C++ 17 compiler: GCC, or optionally Clang on Linux and MSVC on Windows®.
• Python 3.10 or later. Required python libraries for building are listed in tooling-requirements.txt.
• Flatbuffers flatc compiler.
• Doxygen 1.9.1 or later. (When building documentation)

The following dependencies are also needed:

• Argument Parser for Modern C++.
• LLVM.
• TOSA Serialization Library.
• TOSA MLIR Translator.
• JSON for Modern C++.
• pybind11.
• GoogleTest. Optional, for testing.
• glslang.
• SPIRV-Headers.
• SPIRV-Tools.
• SPIRV-Cross.
• Vulkan-Headers.

For building the ML SDK for Vulkan® components, run the following command:

./scripts/build.py

If the ML SDK for Vulkan® components are installed in custom locations, specify their paths by adding the following command line option:

./scripts/build.py --$COMPONENT_NAME $PATH_TO_COMPONENT

COMPONENT_NAME, with the respective default relative locations within paranthesis, can be:

• vgf-lib (sw/vgf-lib)
• model-converter (sw/model-converter)
• scenario-runner (sw/scenario-runner)
• emulation-layer (sw/emulation-layer)

Similarly, dependencies with custom install locations need to be specifed. For instance if glslang was installed at $GLSLANG_REPO_PATH, use the following:

./scripts/build.py --glslang-path $GLSLANG_REPO_PATH

Upon a sparse checkout, missing components will be skipped during building. Note that some components have strict dependencies on each other, with VGF Library being required by Scenario Runner and Model Converter.

The build artifacts can be installed into a specified location by passing the option --install with the required path.

To create an archive with the build artifacts, add the option --package. The archive will be stored in the provided location.

Build the documentation

To build the documentation, run the following command:

./scripts/build.py --doc

More command line options

For more command line options, consult the program help:

./scripts/build.py --help

License

Apache-2.0

Security

If you believe you have discovered a security issue please refer to the Security Section

Trademark notice

Arm® is a registered trademarks of Arm Limited (or its subsidiaries) in the US and/or elsewhere.

Khronos®, Vulkan® and SPIR-V™ are registered trademarks of the Khronos® Group.