tvm_tirx.backend.cuda.op.mma_fill

tvm_tirx.backend.cuda.op.mma_fill(local_size, local_ptr, offset, *, ty)

Zero-initialize an MMA accumulation register.