pub unsafe fn get_convolution_forward_workspace_size(
    handle: cudnnHandle_t,
    x: cudnnTensorDescriptor_t,
    w: cudnnFilterDescriptor_t,
    conv: cudnnConvolutionDescriptor_t,
    y: cudnnTensorDescriptor_t,
    algo: cudnnConvolutionFwdAlgo_t
) -> Result<usize, CudnnError>
Expand description

Returns size in bytes. See nvidia docs

Safety

  • All handles & descriptors must still be allocated.
  • The pointers must point to valid memory.