Hermetic Toolchains for ML

Warning

This project is under active development and is not yet ready for production use.

This project provides Bazel rules for ML project to achieve hermetic and cross-platform builds.

C++ and CUDA hermetic builds benefits:

• Reproducibility: Every build produces identical results regardless of the developer's machine environment.
• Consistency: Eliminates "works on my machine" issues, ensuring builds are consistent across different development environments.
• Isolation: Builds are isolated from the host system, minimizing unexpected dependencies and side effects.

C++ cross-platform builds benefits:

• Single Source of Truth: Develop and maintain a single codebase that can be built for various target platforms (e.g., Linux, macOS).
• Efficiency: Streamlines the build and release process for multiple platforms.

Configure C++ toolchains

How to configure toolchains for ML project

Add below code before CUDA initialization in WORKSPACE file

http_archive(
    name = "rules_ml_toolchain",
    sha256 = "562e0517f4e833afe0de7bb8da49f9adafcbca30a8259f118a65b4adf533b51f",
    strip_prefix = "rules_ml_toolchain-4995c0be587c6e173fe8cf8dc614f92011f7913d",
    urls = [
        "https://github.com/google-ml-infra/rules_ml_toolchain/archive/4995c0be587c6e173fe8cf8dc614f92011f7913d.zip",
    ],
)

load(
    "@rules_ml_toolchain//cc/deps:cc_toolchain_deps.bzl",
    "cc_toolchain_deps",
)

cc_toolchain_deps()

register_toolchains("@rules_ml_toolchain//cc:linux_x86_64_linux_x86_64")
register_toolchains("@rules_ml_toolchain//cc:linux_x86_64_linux_x86_64_cuda")

It must be ensured that builds for Linux x86_64 are run without the --noincompatible_enable_cc_toolchain_resolution flag. Furthermore, reliance on environment variables like CLANG_COMPILER_PATH, BAZEL_COMPILER, CC, or CXX must be avoided.

For diagnosing the utility set being used during build or test execution, the --subcommands flag should be appended to the Bazel command. This will facilitate checking that the compiler or linker are not being used from your machine.

How to run this project tests

CPU Hermetic tests

Project supports CPU hermetic builds on:

• Linux x86_64
• macOS aarch64

The command allows you to run hermetic build tests:

bazel test //cc/tests/cpu:all

If project doesn't support cross-platform builds for specified platform, it will use host utilities and host sysroot for running such build.

GPU Hermetic test

Project supports GPU hermetic builds on Linux x86_64 and requires machine with NVIDIA GPU

You could run hermetic tests with help of command:

Build by Clang

bazel test //cc/tests/gpu:all --config=build_cuda_with_clang --config=cuda --config=cuda_libraries_from_stubs

Build by NVCC

bazel test //cc/tests/gpu:all --config=build_cuda_with_nvcc --config=cuda --config=cuda_libraries_from_stubs

Non-hermetic builds

When executor and a target are the same, you still can run non-hermetic build. Command should look like:

bazel build //cc/tests/cpu:all --//common:enable_hermetic_cc=False

Cross-platform builds

Project supports cross-platform builds only on Linux x86_64 executor and allows build for such targets:

• Linux aarch64
• macOS aarch64

Build for Linux aarch64

bazel build //cc/tests/cpu/... --platforms=//common:linux_aarch64

Build for macOS aarch64

Prepare SDK before run the following command.

bazel build //cc/tests/cpu/... --platforms=//common:macos_aarch64