For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).

Mojo function

accumulate_wo_tile_2d

def accumulate_wo_tile_2d[micro_kernel_height: Int, micro_kernel_width: Int, simd_size: Int, partial_load_filter: Bool, effected_by_padding: Bool, input_dt: DType, filter_dt: DType](c_tile_size: Int, RS: IndexList[Int(2)], mut acc: _Accumulator, input: Pointer[Scalar[input_dt], address_space=input.address_space], input_stride: Int, input_stride_to_nbr: IndexList[Int(2)], filter: Pointer[Scalar[filter_dt], address_space=filter.address_space], filter_stride: Int, filter_stride_to_nbr: IndexList[Int(2)], partial_load_filter_size: Int, hw: IndexList[Int(2)], HW: IndexList[Int(2)], dilation: IndexList[Int(2)])

Accumulates one output row tile for a 2D convolution by iterating over the R and S filter-window dimensions and delegating each row to accumulate_wo_tile_1d.

Parameters:

• ​micro_kernel_height (Int): Number of input rows covered by the micro tile in register tiling.
• ​micro_kernel_width (Int): Number of SIMD registers assigned to the F dimension per row.
• ​simd_size (Int): Number of elements in a SIMD register.
• ​partial_load_filter (Bool): True when the final filter segment is smaller than a full SIMD vector and must be partially loaded.
• ​effected_by_padding (Bool): True when the tile may touch padded input regions, requiring per-point bounds checks.
• ​input_dt (DType): Element type of the input tensor.
• ​filter_dt (DType): Element type of the filter tensor.

Args:

• ​c_tile_size (Int): Number of input channels in the current C tile.
• ​RS (IndexList[Int(2)]): Filter window extents as (R, S) with R the height and S the width.
• ​acc (_Accumulator): Register-tile accumulator updated in place with the convolution products.
• ​input (Pointer[Scalar[input_dt], address_space=input.address_space]): Pointer to the first input element of the WO tile.
• ​input_stride (Int): Stride between consecutive output points along WO in the input, equal to C * stride_w in NHWC layout.
• ​input_stride_to_nbr (IndexList[Int(2)]): Strides to the input neighbor for each spatial axis (R, S), i.e. (stride_to_R_neighbor, stride_to_S_neighbor).
• ​filter (Pointer[Scalar[filter_dt], address_space=filter.address_space]): Pointer to the first filter coefficient in the filter window.
• ​filter_stride (Int): Stride between consecutive filter segments of size micro_kernel_width * simd_size along the F dimension.
• ​filter_stride_to_nbr (IndexList[Int(2)]): Strides to the filter neighbor for each spatial axis (R, S).
• ​partial_load_filter_size (Int): Number of valid elements in the final partial filter SIMD vector when F is not a multiple of simd_size.
• ​hw (IndexList[Int(2)]): Input spatial coordinate (h, w) of the tile's first output point before padding adjustment.
• ​HW (IndexList[Int(2)]): Input spatial extents (H, W) used for padding bounds checks.
• ​dilation (IndexList[Int(2)]): Dilation factors (dilation_h, dilation_w) applied to the filter window.