基于OpenCL框架的并行编译系统设计

摘要

Aiming at PAAG (Polymorphic Array Architecture for Graphics and image processing),which is designed and developed by Xi'An University of Posts and Telecommunications,this paper discusses the parallel compilation problems on the machine.It uses OpenCL framework to solve the difficult problem that developing parallel software on the PAAG is too difficult at present.It chooses LLVM with PAAG backend as OpenCL compiler.It designed a PAAG task loader to map tasks to PAAG array and fix the instruction and deals with the unique communication instruction of PAAG.Optimize OpenCL built function,reduces the cost of function call.The experiments prove that this compilation system can work normally.%针对自主设计的面向图形图像应用的16核阵列并行处理器PAAG (polymorphie array architecture for graphics and image processing),讨论在PAAG机器上的并行编译问题.提出使用OpenCL框架解决目前在PAAG平台上开发并行软件困难的问题.选择LLVM作为OpenCL框架中的OpenCL编译器,为LLVM移植PAAG后端.设计了PAAG任务加载器进行任务映射,指令修正,处理PAAG独特的通信指令.对内置OpenCL函数进行了优化,减少了函数调用开销.经实验证明本编译系统可正常工作.