tvm_tirx.backend.cuda.op.ptx_legacy_mma#

tvm_tirx.backend.cuda.op.ptx_legacy_mma(shape, a_layout, b_layout, a_dtype, b_dtype, c_dtype, a_ptr, a_offset, b_ptr, b_offset, acc_ptr, c_offset, saturate, operator=None, *, ty=None, loc=UNKNOWN_LOC)#

Construct legacy MMA with element offsets and an optional bit operator.