摘要
Co-Array Fortran(CAF)已经成为Fortran语言标准的一部分,在科学计算领域逐渐被接受。基于软件共享存储实现了一个CAF编译器,其通过直接的数组赋值实现Co-array数据通信,利用数据垫塞技术提高数据局部性,减少伪共享,优化CAF程序性能。典型科学计算程序测试表明,CAF能够获得和MPI相当的性能。
Co-Array Fortran(CAF) has become part of Fortran Programming Language Standard,and has been widely accepted in scientific computing community.The paper presented the design and implementation of a CAF Compiler based on software shared memory.Several techniques were presented to improve the performance of CAF programs,including Co-array data communication by direct array assignment,Data Padding for raising locality and reducing false share.The benchmark testing shows that the performance of CAF programs is similar to that of MPI programs,with our CAF compiler.
出处
《计算机科学》
CSCD
北大核心
2012年第1期287-289,304,共4页
Computer Science
基金
"核高基"国产CPU编译系统及工具链项目(2009ZX01036-001-003)资助