tvm_tirx.backend.cuda.layout.mma_atom_layout#
- tvm_tirx.backend.cuda.layout.mma_atom_layout(dtype, swizzle_mode)#
Generate the MMA-compatible shared-memory atom layout.
- Parameters:
dtype (str)
swizzle_mode (SwizzleMode | int)
- Return type:
Generate the MMA-compatible shared-memory atom layout.
dtype (str)
swizzle_mode (SwizzleMode | int)