Method do_setup

examples/cl_cache.cpp:45–96

    CLCacheExample() = default;

    bool do_setup(int argc, char **argv) override
    {
        std::cout << "Once the program has run and created the file cache.bin, rerun with --restore_cache."
                  << std::endl;
        CLScheduler::get().default_init();

        if (argc > 1)
        {
            std::string argv1 = argv[1];
            std::transform(argv1.begin(), argv1.end(), argv1.begin(), ::tolower);
            if (argv1 == "--restore_cache")
            {
                // Load the precompiled kernels from a file into the kernel library, in this way the next time they are needed
                // compilation won't be required.
                restore_program_cache_from_file();
            }
            else
            {
                std::cout << "Unkown option " << argv1 << std::endl;
            }
        }

        // Initialise shapes
        init_tensor(TensorShape(8U, 4U, 2U), tensor_nchw, DataType::U8, DataLayout::NCHW);
        init_tensor(TensorShape(2U, 8U, 4U), tensor_nhwc, DataType::U8, DataLayout::NHWC);
        init_tensor(TensorShape(8U, 4U, 2U), tensor_nchw_result, DataType::U8, DataLayout::NCHW);

        // Create the permutation vector to turn a NCHW tensor to NHWC.
        // The input tensor is NCHW, which means that the fastest changing coordinate is W=8U.
        // For permutation vectors the fastest changing coordinate is the one on the left too.
        // Each element in the permutation vector specifies a mapping from the source tensor to the destination one, thus if we
        // use 2U in the permutation vector's first element we are telling the function to move the channels to the fastest
        // changing coordinate in the destination tensor.

        const PermutationVector vector_nchw_to_nhwc(2U, 0U, 1U);
        permute_nhwc.configure(&tensor_nchw, &tensor_nhwc, vector_nchw_to_nhwc);

        // Allocate and fill tensors
        tensor_nhwc.allocator()->allocate();
        tensor_nchw.allocator()->allocate();
        fill_tensor(tensor_nchw);

        // Demostrate autoconfigure for the output tensor
        const PermutationVector vector_nhwc_to_nchw(1U, 2U, 0U);
        permute_nchw.configure(&tensor_nhwc, &tensor_nchw_result, vector_nhwc_to_nchw);
        tensor_nchw_result.allocator()->allocate();

        // Save the opencl kernels to a file
        save_program_cache_to_file();

        return true;
    }
    void do_run() override
    {
        permute_nhwc.run();