Sergei Lebedev 5e418f5ab2 Added argument validation to mosaic_gpu_init_tma_desc ...

This should help with understanding cuTensorMapEncodeTiled failures, since
CUDA doesn't provide any details beyond the error return code.

Note that this change also ensures that TMA descriptors are 64-byte aligned.

PiperOrigin-RevId: 656062820

2024-07-25 13:16:34 -07:00

..

integrations/c

Switch Mosaic GPU to a custom pass pipeline and improve the lowering of GPU launch

2024-04-24 03:27:45 -07:00

BUILD

Use constant memory to pass in TMA descriptors to the kernel

2024-06-17 05:31:26 -07:00

custom_call.cc

[Mosaic GPU] Pass in TMA descriptors through kernel parameters

2024-07-02 09:30:52 -07:00

launch_lowering.cc

[Mosaic GPU] Strengthen cluster-related tests by covering more cluster shapes

2024-07-24 13:43:52 -07:00

launch_lowering.h

Switch Mosaic GPU to a custom pass pipeline and improve the lowering of GPU launch

2024-04-24 03:27:45 -07:00

mosaic_gpu_ext.cc

[Mosaic GPU] Integrate the ExecutionEngine with the jaxlib GPU plugin

2024-05-30 01:46:23 -07:00

pass_boilerplate.h

[Mosaic GPU] Load LLVM lowering interfaces for all dialects

2024-06-10 05:55:01 -07:00

passes.cc

[Mosaic GPU] Pass in TMA descriptors through kernel parameters

2024-07-02 09:30:52 -07:00

passes.h

[Mosaic GPU] Pass in TMA descriptors through kernel parameters

2024-07-02 09:30:52 -07:00

runtime.cc

Added argument validation to mosaic_gpu_init_tma_desc

2024-07-25 13:16:34 -07:00