一种针对异构计算平台的编译期优化方法

针对OpenCL(open computing language)编译时期的特有模式, 提出了一种新的针对异构计算平台的编译期优化方法。该方法根据设备端和主机端的各自特点, 将设备端的一些冗余操作提到主机端或者新的设备端kernel中去执行, 以达到降低存储器读写的目的。这种方法充分利用了异构计算平台的特点, 较传统优化方法相对灵活。大多数情况下能有效提高OpenCL的运行速度, 测试用例中在应用原有编译器优化的基础上最快提高了270%。

Compile-time optimization method for heterogeneous computing platforms

For the OpenCL(open computing language) peculiar mode, this paper proposed a new compile-time optimization methods for heterogeneous computing platform. According to the characteristics of each device-side and host-side, this method took the redundant operation of the device-end to the host-side or new device-side to operate, which reduced the times of read and write memory. This method took full advantage of the characteristics of the heterogeneous computing platforms, relatively flexible compared to traditional optimization methods. It can effectively improve the overall speed in many cases, the fastest increase of 270% in test samples.