MPI并行程序编程-China-VO.ppt

创建新运算 归约 func是用户提供的用于完成运算的外部函数 commute用来指明所定义的运算是否满足交换律 一个运算创建后和MPI预定义的运算一样使用 归约 用户自定义函数 invec与inoutvec分别指出要被归约的数据所在缓冲的首地址 函数返回时，运算结果储存在inoutvec中 datatype指出归约对象的数据类型 len给出了invec与inoutvec中包含的元素个数（相当于归约函数中的count） 归约 Sample - Fortran PROGRAM UserOP C Run with 8 processes INCLUDE mpif.h INTEGER err, rank, size integer source, reslt EXTERNAL digit LOGICAL commute INTEGER myop CALL MPI_INIT(err) CALL MPI_COMM_RANK(MPI_WORLD_COMM,rank,err) CALL MPI_COMM_SIZE(MPI_WORLD_COMM,size,err) commute= true 满足交换律 call MPI_OP_CREATE(digit,commute,myop,err) source=(rank+1)**2 call MPI_BARRIER(MPI_COM_WORLD,err) call MPI_SCAN(source,reslt,1,MPI_INTEGER,myop,MPI_COMM_WORLD,err) print *,P:,rank, my result is ,reslt CALL MPI_OP_FREE(myop,err) CALL MPI_FINALIZE(err) END integer function digit(in,inout,len,type) integer len,type integer in(len),inout(len) do i=1,len inout(i)=mod((in(i)+inout(i)),10) end do digit=5 end Program Output P:6 my result is 0 P:5 my result is 1 P:7 my result is 4 P:1 my result is 5 P:3 my result is 0 P:2 my result is 4 P:4 my result is 5 P:0 my result is 1 source={ 1,4,9,16,25,36,49,64} MPI truly is a portable standard Source code portability between machines running MPI this is the “beauty” of having a standard Point-to-Point improve performance Derived Datatypes send a structure in C send a record in F90 send a collection of non-uniform data types Collective communication group of processes that do something together Both math and communications functions Virtual topologies method of making a collection of processes act like they are in a particular shap * * The arguments in the MPI_SEND call are “handles” briefly introduce arguments here - more detail will be