Mojo function

dispatch_sm100_conv2d

dispatch_sm100_conv2d[input_layout: Layout, filter_layout: Layout, output_layout: Layout, input_type: DType, filter_type: DType, output_type: DType, filter_is_fcrs: Bool, elementwise_lambda_fn: Optional[elementwise_epilogue_type] = None, has_residual: Bool = False](input: LayoutTensor[input_type, input_layout, input.origin, address_space=input.address_space, element_layout=input.element_layout, layout_int_type=input.layout_int_type, linear_idx_type=input.linear_idx_type, masked=input.masked, alignment=input.alignment], filter: LayoutTensor[filter_type, filter_layout, filter.origin, address_space=filter.address_space, element_layout=filter.element_layout, layout_int_type=filter.layout_int_type, linear_idx_type=filter.linear_idx_type, masked=filter.masked, alignment=filter.alignment], output: LayoutTensor[output_type, output_layout, output.origin, address_space=output.address_space, element_layout=output.element_layout, layout_int_type=output.layout_int_type, linear_idx_type=output.linear_idx_type, masked=output.masked, alignment=output.alignment], symmetric_padding: IndexList[2], ctx: DeviceContext, source_ptr: UnsafePointer[Scalar[output_type], MutAnyOrigin] = UnsafePointer(), beta: Float32 = 0)

Dispatch to SM100 structured conv2d with filter transpose.

This function gates the SM100 kernel import behind @parameter if on dtype, so the kernel is never compiled for unsupported dtypes.

Parameters:

input_layout (Layout): Layout of the input activation tensor.
filter_layout (Layout): Layout of the filter weights tensor.
output_layout (Layout): Layout of the output tensor.
input_type (DType): Data type of the input activation tensor.
filter_type (DType): Data type of the filter weights tensor.
output_type (DType): Data type of the output tensor.
filter_is_fcrs (Bool): If True, filter is FCRS layout; otherwise RSCF.
elementwise_lambda_fn (Optional): Optional void epilogue lambda applied after output write. Signature: fn(IndexList[2], SIMD) -> None.
has_residual (Bool): If True, fuse residual add D = Conv(A,B) + beta*C.

Args:

input (LayoutTensor): Input activation tensor in NHWC layout.
filter (LayoutTensor): Filter weights tensor.
output (LayoutTensor): Output tensor in NHWC layout.
symmetric_padding (IndexList): Symmetric padding (pad_h, pad_w).
ctx (DeviceContext): Device context for kernel launch.
source_ptr (UnsafePointer): Pointer to residual source tensor C (NHWC, same shape as output). Only used when has_residual is True.
beta (Float32): Residual scale factor. D = Conv(A,B) + beta*C.

Raises:

Error if kernel launch fails.