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

conv_miopen

def conv_miopen[conv_rank: Int, input_type: DType, filter_type: DType, output_type: DType, //, filter_is_fcrs: Bool = False](input: TileTensor[input_type, Storage=input.Storage, address_space=input.address_space, linear_idx_type=input.linear_idx_type], filter: TileTensor[filter_type, Storage=filter.Storage, address_space=filter.address_space, linear_idx_type=filter.linear_idx_type], output: TileTensor[output_type, Storage=output.Storage, address_space=output.address_space, linear_idx_type=output.linear_idx_type], stride: IndexList[conv_rank], dilation: IndexList[conv_rank], padding: IndexList[conv_rank], num_groups: Int, ctx: DeviceContext)

Runs a convolution via MIOpen on AMD GPUs, transposing the filter to FRSC physical layout and dispatching to the cached MIOpen forward convolution path.

Parameters:

• ​conv_rank (Int): Number of spatial dimensions in the convolution (1, 2, or 3) (inferred).
• ​input_type (DType): Element type of the input tensor (inferred).
• ​filter_type (DType): Element type of the filter tensor (inferred).
• ​output_type (DType): Element type of the output tensor (inferred).
• ​filter_is_fcrs (Bool): True when the filter uses FCRS layout, otherwise RSCF (defaults to False).

Args:

• ​input (TileTensor[input_type, Storage=input.Storage, address_space=input.address_space, linear_idx_type=input.linear_idx_type]): Input activation tensor in NHWC or NDHWC layout.
• ​filter (TileTensor[filter_type, Storage=filter.Storage, address_space=filter.address_space, linear_idx_type=filter.linear_idx_type]): Filter weights tensor in RSCF, FCRS, or QRSCF layout.
• ​output (TileTensor[output_type, Storage=output.Storage, address_space=output.address_space, linear_idx_type=output.linear_idx_type]): Output tensor in NHWC or NDHWC layout.
• ​stride (IndexList[conv_rank]): Stride along each spatial dimension.
• ​dilation (IndexList[conv_rank]): Dilation factor along each spatial dimension.
• ​padding (IndexList[conv_rank]): Symmetric padding applied to each spatial dimension.
• ​num_groups (Int): Number of convolution groups for grouped convolution.
• ​ctx (DeviceContext): Device context for kernel launch.