Method run_op

src/gpu/cl/kernels/ClWeightsReshapeKernel.cpp:141–188

}

void ClWeightsReshapeKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
{
    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
    ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS(ICLKernel::window(), window);

    auto src    = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
    auto biases = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_BIAS));
    auto dst    = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));

    Window out_window;
    out_window.use_tensor_dimensions(dst->info()->tensor_shape());

    Window in_slice  = window.first_slice_window_3D();
    Window out_slice = out_window.first_slice_window_2D();

    Window biases_window;
    Window biases_slice;

    unsigned int idx = num_arguments_per_3D_tensor() + num_arguments_per_2D_tensor();
    idx += (biases != nullptr) ? num_arguments_per_1D_tensor() : 0;
    _kernel.setArg<cl_uint>(idx++, src->info()->dimension(0));
    _kernel.setArg<cl_uint>(idx++, src->info()->dimension(1));
    _kernel.setArg<cl_uint>(idx++, src->info()->dimension(2));
    _kernel.setArg<cl_uint>(idx++, src->info()->dimension(3));
    _kernel.setArg<cl_uint>(idx++, dst->info()->strides_in_bytes().z());

    if (biases != nullptr)
    {
        biases_window.use_tensor_dimensions(biases->info()->tensor_shape());
        biases_slice = biases_window.first_slice_window_1D();
    }

    do
    {
        // Set arguments
        unsigned idx = 0;
        add_3D_tensor_argument(idx, src, in_slice);
        add_2D_tensor_argument(idx, dst, out_slice);
        if (biases != nullptr)
        {
            add_1D_tensor_argument(idx, biases, biases_slice);
            ARM_COMPUTE_UNUSED(biases_window.slide_window_slice_1D(biases_slice));
        }

        // Run kernel
        enqueue(queue, *this, in_slice, lws_hint());
    } while (window.slide_window_slice_4D(in_slice) && out_window.slide_window_slice_2D(out_slice));
}
} // namespace kernels
} // namespace opencl
} // namespace arm_compute