xla
xla copied to clipboard
openxla
A machine learning compiler for GPUs, CPUs, and ML accelerators
[xla] Add support for output target to DynamicParameterBinding In addition to specifying dynamic parameter binding for parameters (inputs), it should be possible to bind dynamic parameter to an output dimension.
![copybara-service[bot] avatar](https://avatars.githubusercontent.com/in/44061?v=4 "copybara-service[bot] avatar"){kind=avatar}
Do not choose bitcasts as fusion roots. This just adds indexing overhead without any benefit. Bitcasts outside of fusions are no-ops.
Select the first platform that is not kInterpreter by default when more than two platforms are available.
PR #59515: Cublaslt fp8 matmul restriction work-around Imported from GitHub PR https://github.com/tensorflow/tensorflow/pull/59515 As of 2/1/2023, cublasLt f8 matmul only support col major input(default to cublas). But calling from TF/XLA, input...
Change some callers of execute_sharded_on_local_devices to execute_sharded_on_local_devices_with_tokens.
[xla:pjrt]Tfrt Cpu PjRt client that enables JAX with dynamic shapes on top of XLA CPU pipeline
Add an option to configure layout assignment for transposes on GPU backend. By default, layout assignment tries to assign a layout to transposes that make them a bitcast. This layout...
Integrate LLVM at llvm/llvm-project@7c9b15fbaeb2 Updates LLVM usage to match [7c9b15fbaeb2](https://github.com/llvm/llvm-project/commit/7c9b15fbaeb2)
[jax] Add mhlo DynamicParameterBinding attribute to dynamically-shaped arguments and results
Add UnregisterFileSystem to allow unregister the existing filesystem of a given scheme.