This lets us avoid bundling a whole another copy of LLVM with JAX packages
and so we can finally start building Mosaic GPU by default.
PiperOrigin-RevId: 638569750
With this change, `python3 build/build.py --enable_cuda --build_gpu_plugin --gpu_plugin_cuda_version=12` will generate three wheels:
| |size|wheel name |
|----------------------|----|-------------------------------------------------------------------------|
|jaxlib w/o cuda kernels|76M |jaxlib-0.4.20.dev20231101-cp310-cp310-manylinux2014_x86_64.whl |
|cuda pjrt |73M|jax_cuda12_pjrt-0.4.20.dev20231101-py3-none-manylinux2014_x86_64.whl |
|cuda kernels |6.6M|jax_cuda12_plugin-0.4.20.dev20231101-cp310-cp310-manylinux2014_x86_64.whl|
The size of jaxlib with cuda kernels and pjrt is 119M.
The cuda kernel wheel contains all the cuda kernels. A plugin_setup.py and plugin_pyproject.toml are added for this new pacakge.
PiperOrigin-RevId: 579861480
Add a --verbose option that logs all shell() commands run by the script.
Remove some Python 2 backward compatibility logic related to urllib and shutil.
Enable debug logging on Windows wheel builds.
Also include setuptools in the build requirements and test for its presence in build.py.
- Add a py extension to call the custom call C API.
- Change the implementation of register_custom_call_target to store handlers for the custom call targets and delays the registration until the handler for a xla platform is registered.
- Change register_plugin to load PJRT plugin when register_pluin is called (instead of when a client is created), and let it return the PJRT_Api* loaded.
- Delay calling discover_pjrt_plugins() and register_pjrt_plugin_factories_from_env() until the first time backends() is called.
PiperOrigin-RevId: 568265745
Add a build wheel, pyproject.toml and setup.py.
The directory structure in jax repo is:
jax/
└── plugins/
└── cuda/
├── __init__.py
├── pyproject.toml
└── setup.py
Installed package structure is:
jax_plugins/
└── xla_cuda_cu12/
├── __init__.py
└── xla_cuda_plugin.so
The major cuda version will be part of the package name.
The plugin wheel can be built with command:
python3 build/build.py --enable_cuda --build_gpu_plugin --gpu_plugin_cuda_version=12 --bazel_options="--override_repository=xla=$HOME/xla"
PiperOrigin-RevId: 565187954
The jaxlib wheel build currently uses `python setup.py bdist_wheel` to construct the wheel. Change it to use `python -m build -w` instead.
To avoid Python getting confused between the directory named `build` in the bazel tree and the Python `build` module, move `build_wheel.py` into `jaxlib/tools`.
PiperOrigin-RevId: 548133811
--
57af5360a1ca1356dbf7760c76e241f7134ef6dd by Jake VanderPlas <jakevdp@google.com>:
[Rollback] Update required Python version to 3.9
PiperOrigin-RevId: 528905991
* Add a new --configure_only option to build.py to allow build.py to generate a .bazelrc without necessarily building jaxlib.
* Add a bazel flag that make the dependency of //jax on //jaxlib optional. If //jaxlib isn't built by bazel, then tests will implicitly use a preinstalled jaxlib.
See: https://github.com/tensorflow/tensorflow/pull/55613
For a CUDA build at head with the default compute capabilities, reduces wheel size from 141MB to 112MB.
Don't redundantly specify default compute capabilities in .bazelrc and in
build.py.
