Method Run

dali/kernels/imgproc/paste/paste_gpu.h:285–305

  void Run(
      KernelContext &context,
      const OutListGPU<OutputType, ndims> &out,
      const span<InListGPU<InputType, ndims>> &ins) {
    paste::FillPointers(make_span(sample_descriptors_), make_span(patch_descriptors_), out, ins);

    SampleDesc *samples_gpu;
    PatchDesc  *patches_gpu;
    BlockDesc *blocks_gpu;

    std::tie(samples_gpu, patches_gpu, blocks_gpu) = context.scratchpad->ToContiguousGPU(
        context.gpu.stream, sample_descriptors_, patch_descriptors_, block_setup_.Blocks());

    dim3 patch_dim = block_setup_.GridDim();
    dim3 block_dim = block_setup_.BlockDim();
    auto stream = context.gpu.stream;

    paste::PasteKernel<<<patch_dim, block_dim, 0, stream>>>(
        samples_gpu, patches_gpu, blocks_gpu);
    CUDA_CALL(cudaGetLastError());
  }
};

}  // namespace kernels