xla

xla copied to clipboard

Make StreamExecutorGpuTopologyDescription compile without error

copybara-service[bot]

When computing the set of reduced times, keep just the latest one if multiple primitives enter the liveness profile consecutively.

copybara-service[bot]

[xla:spmd] Fix a problem in generating code for concat that requires padding. When an internal concat operand requires padding, previously, we don't zero out the padded elements and the generated...

copybara-service[bot]

[ROCM] GPU performance cost model specs for MI300

1

@ddunl @xla-rotation : would you take a look please?

Ruturaj4

[GPU] Multinode support for multihost_hlo_runner

2

This PR adds multinode support to the multihost_hlo_runner. To use, there is a new command line argument `address` for the address of the coordinator/root node. Example usage with SLURM: ```...

trevor-m

Disable some tests that time out on certain platforms.

copybara-service[bot]

[XLA:CPU][oneDNN] Enable oneDNN MatMul + BiasAdd + Sigmoid fusion.

2

This PR enables oneDNN MatMul+BiasAdd+Sigmoid fusion for FP32, BF16 and F16.

nhatleSummer22

Improve the accuracy of asin(x) and asinh(x) for complex x using modified Hull et al algorithm.

15

As in the title. ~Fixes https://github.com/openxla/xla/issues/8553~ - PR https://github.com/openxla/xla/pull/9802 disabled the fix. Update: the fix to https://github.com/openxla/xla/issues/8553 will be available via https://github.com/openxla/stablehlo/pull/2357

pearu

Hello, I'm very new to the XLA project, so pardon my ignorance here. I'm trying to learn more about the project, optimizations, and their interactions with the hardware. The code...

abhaskumarsinha

Allow overriding some methods in all_gather_decomposer.

copybara-service[bot]