For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).
Mojo function
dispatch_im2col_matmul_conv2d
def dispatch_im2col_matmul_conv2d[input_type: DType, filter_type: DType, output_type: DType, //, filter_is_fcrs: Bool = False, maybe_epilogue_func: Optional[def[dtype: DType, rank: Int, width: Int, alignment: Int = 1](IndexList[rank], SIMD[dtype, width]) capturing -> None] = None, m_tile_byte_budget: Int = 268435456](input: TileTensor[input_type, address_space=input.address_space, linear_idx_type=input.linear_idx_type, element_size=input.element_size], filter: TileTensor[filter_type, address_space=filter.address_space, linear_idx_type=filter.linear_idx_type, element_size=filter.element_size], output: TileTensor[output_type, address_space=output.address_space, linear_idx_type=output.linear_idx_type, element_size=output.element_size], stride: IndexList[2], dilation: IndexList[2], symmetric_padding: IndexList[2], num_groups: Int, ctx: DeviceContext) -> Bool
Try to dispatch a 2-D conv as explicit im2col + generic matmul.
Returns True if the conv was handled; False if the caller should fall back to another implementation (naive Mojo kernel, cuDNN, etc.).
Skips on: non-bf16 dtype, grouped conv, dilation != 1, kernel size 1x1 (the vectorized naive kernel wins on tiny shapes), and K too small for the matmul fast path.
Returns:
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!