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

conv_transposed_cudnn

def conv_transposed_cudnn[input_type: DType, filter_type: DType, output_type: DType](input: TileTensor[input_type, Storage=input.Storage, linear_idx_type=input.linear_idx_type], filter: TileTensor[filter_type, Storage=filter.Storage, linear_idx_type=filter.linear_idx_type], output: TileTensor[output_type, Storage=output.Storage, linear_idx_type=output.linear_idx_type], stride: IndexList[Int(2)], dilation: IndexList[Int(2)], padding: IndexList[Int(2)], ctx: DeviceContext)

Runs a 2D transposed convolution on the GPU using cuDNN backward-data.

Pushes the device context and delegates to the internal cuDNN backward-data helper that configures the tensor, filter, and convolution descriptors and launches the cuDNN kernel.

Parameters:

input_type (DType): Element type of the input tensor.
filter_type (DType): Element type of the filter tensor.
output_type (DType): Element type of the output tensor.

Args:

input (TileTensor[input_type, Storage=input.Storage, linear_idx_type=input.linear_idx_type]): Input tensor in NCHW layout.
filter (TileTensor[filter_type, Storage=filter.Storage, linear_idx_type=filter.linear_idx_type]): Filter tensor in CKRS layout.
output (TileTensor[output_type, Storage=output.Storage, linear_idx_type=output.linear_idx_type]): Output tensor in NCHW layout.
stride (IndexList[Int(2)]): Stride along the height and width axes.
dilation (IndexList[Int(2)]): Dilation along the height and width axes of the filter.
padding (IndexList[Int(2)]): Padding along the height and width axes.
ctx (DeviceContext): Device context used to bind the cuDNN handle to the current stream.