triton.experimental.gluon.language.nvidia.blackwell.tma.async_atomic_max

triton.experimental.gluon.language.nvidia.blackwell.tma.async_atomic_max(tensor_desc, coord, src, _semantic=None)

Atomically compute the maximum of shared memory data and global memory using TMA.

Parameters:

• tensor_desc (tensor_descriptor) – Tensor descriptor (tiled).

• coord (Sequence[int | ttgl.constexpr | ttgl.tensor]) – Coordinates in the destination tensor.

• src (ttgl.shared_memory_descriptor) – Source memory descriptor.