Mojo function
q_out_tma
q_out_tma[dtype: DType, //, swizzle_mode: TensorMapSwizzle, *, BM: Int, depth: Int, padded_depth: Int, q_num_heads: Int, group: Int, decoding: Bool](ctx: DeviceContext, ptr: UnsafePointer[Scalar[dtype]], rows: Int, out res: TMATensorTile[dtype, tile_layout_k_major[dtype, max(group, 8) if decoding else BM, 64 if decoding else padded_depth, swizzle_mode](), _tma_desc_tile_layout[dtype, 2, IndexList[2, DType.int64](max(group, 8) if decoding else BM, 64 if decoding else padded_depth, Tuple[]()), swizzle_mode=swizzle_mode]()])
Returns:
TMATensorTile
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!