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_sm100_conv2d

def dispatch_sm100_conv2d[input_type: DType, filter_type: DType, output_type: DType, //, filter_is_fcrs: Bool = False, elementwise_lambda_fn: Optional[def[dtype: DType, width: Int, *, alignment: Int = 1](IndexList[2], SIMD[dtype, width]) capturing -> None] = None, has_residual: Bool = False](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], symmetric_padding: IndexList[2], ctx: DeviceContext, source_ptr: OptionalReg[UnsafePointer[Scalar[output_type], MutAnyOrigin]] = None, 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_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[def[dtype: DType, width: Int, *, alignment: Int = 1](IndexList[2], SIMD[dtype, width]) capturing -> None]): Optional void epilogue lambda applied after output write. Signature: def(IndexList[2], SIMD) -> None.
has_residual (Bool): If True, fuse residual add D = Conv(A,B) + beta*C.

Args:

input (TileTensor[input_type, address_space=input.address_space, linear_idx_type=input.linear_idx_type, element_size=input.element_size]): Input activation tensor in NHWC layout.
filter (TileTensor[filter_type, address_space=filter.address_space, linear_idx_type=filter.linear_idx_type, element_size=filter.element_size]): Filter weights tensor.
output (TileTensor[output_type, address_space=output.address_space, linear_idx_type=output.linear_idx_type, element_size=output.element_size]): Output tensor in NHWC layout.
symmetric_padding (IndexList[2]): Symmetric padding (pad_h, pad_w).
ctx (DeviceContext): Device context for kernel launch.
source_ptr (OptionalReg[UnsafePointer[Scalar[output_type], MutAnyOrigin]]): 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.