Optimizing Data Movement

In the OpenCL™ programming model, all data are transferred from the host memory to the global memory on the device first and then from the global memory to the kernel for computation. The computation results are written back from the kernel to the global memory and lastly from the global memory to the host memory. How data can be efficiently moved around in this programming model is a key factor for determining strategies for kernel computation optimization, so it is recommended to optimize the data movement in your application before taking on optimizing the computation.

During data movement optimization, it is important to isolate data transfer code from computation code because inefficiency in computation may cause stalls in data movement. Xilinx recommends that you modify the host code and kernels with data transfer code only for this optimization step. The goal is to maximize the system level data throughput by maximizing PCIe bandwidth utilization and DDR bandwidth utilization. It usually takes multiple iterations of running CPU emulation, hardware emulation, as well as execution on FPGAs to achieve the goal.