Currently in our code in tflite
we use the eigen library for our custom kernels for the Android devices. If we compile our library with EIGEN_USE_SYCL=1
will there be any speed increase using:
cl::sycl::gpu_selector selector;
Eigen::QueueInterface queue(selector);
Eigen::SyclDevice device(&queue);
instead of what we currently do:
const EigenForTFLite::ThreadPoolDevice* device =
eigen_support::GetThreadPoolDevice(context);