xla
xla copied to clipboard
openxla
A machine learning compiler for GPUs, CPUs, and ML accelerators
Integrate Triton up to [243522bd](https://github.com/openai/triton/commits/243522bd337fb2e3f4245d58452b0247629f0b58)
![copybara-service[bot] avatar](https://avatars.githubusercontent.com/in/44061?v=4 "copybara-service[bot] avatar"){kind=avatar}
[XLA] [NFC] Add more fine-grained tracing annotations to compilation stages Reverts 9e4f0ab3d004d1dd2948830d644e518ebab4fccd
PR #15417: Add while loop config options and optional pass pipeline immediately before unroll. Imported from GitHub PR https://github.com/openxla/xla/pull/15417 This PR adds the availability to configure while loop unroll thresholds....
Remove duplicate location of ml_dtypes - `third_party/py/ml_dtypes` as it is not used. The actively used ml_dtypes is located in `third_party/tsl/third_party/py/ml_dtypes`
See for details: https://github.com/google/jax/issues/18787
Integrate LLVM at llvm/llvm-project@0c25f85e5b88 Updates LLVM usage to match [0c25f85e5b88](https://github.com/llvm/llvm-project/commit/0c25f85e5b88)
Fix `GetExecutablePath(..)` for the case when a command flag is passed to the python executable. Currently `GetExecutablePath(..)` returns an empty string if the command `python -c ''` is executed. The...
Add `ClearEntryComputationLayout` in hlo_module_util.
Add Platform::FindExisting to find pre-existing StreamExecutors for the given Platform without creating new ones.
[RFC] [XLA][HostOffloader] Updates UsesBeforeValueDefinition to consider host offloaded computation usage Updates: the use at an async host offloaded call occurs before values that are defined in the async wrapped computation