Throughput-Optimized OpenCL-based FPGA Accelerator for Large-Scale Convolutional Neural Networks