1) Hermetic CUDA rules allow building wheels with GPU support on a machine without GPUs, as well as running Bazel GPU tests on a machine with only GPUs and NVIDIA driver installed. When `--config=cuda` is provided in Bazel options, Bazel will download CUDA, CUDNN and NCCL redistributions in the cache, and use them during build and test phases.
[Default location of CUNN redistributions](https://developer.download.nvidia.com/compute/cudnn/redist/)
[Default location of CUDA redistributions](https://developer.download.nvidia.com/compute/cuda/redist/)
[Default location of NCCL redistributions](https://pypi.org/project/nvidia-nccl-cu12/#history)
2) To include hermetic CUDA rules in your project, add the following in the WORKSPACE of the downstream project dependent on XLA.
Note: use `@local_tsl` instead of `@tsl` in Tensorflow project.
```
load(
"@tsl//third_party/gpus/cuda/hermetic:cuda_json_init_repository.bzl",
"cuda_json_init_repository",
)
cuda_json_init_repository()
load(
"@cuda_redist_json//:distributions.bzl",
"CUDA_REDISTRIBUTIONS",
"CUDNN_REDISTRIBUTIONS",
)
load(
"@tsl//third_party/gpus/cuda/hermetic:cuda_redist_init_repositories.bzl",
"cuda_redist_init_repositories",
"cudnn_redist_init_repository",
)
cuda_redist_init_repositories(
cuda_redistributions = CUDA_REDISTRIBUTIONS,
)
cudnn_redist_init_repository(
cudnn_redistributions = CUDNN_REDISTRIBUTIONS,
)
load(
"@tsl//third_party/gpus/cuda/hermetic:cuda_configure.bzl",
"cuda_configure",
)
cuda_configure(name = "local_config_cuda")
load(
"@tsl//third_party/nccl/hermetic:nccl_redist_init_repository.bzl",
"nccl_redist_init_repository",
)
nccl_redist_init_repository()
load(
"@tsl//third_party/nccl/hermetic:nccl_configure.bzl",
"nccl_configure",
)
nccl_configure(name = "local_config_nccl")
```
PiperOrigin-RevId: 662981325
This is needed for hermetic CUDA integration in Google ML projects since tensorRT is not distributed in the same free way as other CUDA/CUDNN distributives.
PiperOrigin-RevId: 662601190
The plugin is released and the flag is no longer needed.
Also set default value of enable_gpu to False. enable_gpu will be removed in the next change.
PiperOrigin-RevId: 660059432
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
This removes commands that set up Python and JAX's build-time Python dependencies. These are now handled automatically by Bazel with hermetic Python.
We also no longer need to force Genrules to run locally to be able to set up `local_config_python` correctly.
This change should also resolve build failures in the cross-compile builds.
PiperOrigin-RevId: 636030000
Add a currently undocumented jax[cuda11_pip] and jax[cuda12_pip] that depend on the pip CUDA wheels.
Add a currently undocumented jax[cuda11_local] and jax[cuda12_local] that avoid the CUDA wheel dependency.
It seems to me that jaxlib development must be mostly happening on CI, because some basics are pretty essential. Here are a few things I've been typing/carrying for a while in my flow:
* Add .bazelrc.user to .gitignore so it doesn't accidentally get checked in.
* Add configs for 'debug_symbols' and 'debug' that make some things minimally workable under a debugger (or to get backtraces, etc).
* Add `--force-reinstall` to the copy/paste command to update a built jaxlib wheel (without this, if you are iterating, it fairly quietly does nothing).
