如何将MPI派生的数据类型用于3D阵列? [英] How to use MPI derived data type for 3D array?
问题描述
我想编写一个在3D矩阵上工作的并行代码,其中每个进程都有其自己的子矩阵,但是为了完成其工作,他们需要一些有关其相邻进程的子矩阵(只是边界平面)的信息.我通过点对点通信发送这些信息,但是我知道对于大型矩阵来说,这不是一个好主意,因此我决定使用派生数据类型进行通信.我对mpi_type_vector
有问题:例如,我有一个NX*NY*NZ
矩阵,我想将常数为NY
的平面发送到另一个进程,我为此编写了这些行:
I want to write a parallel code that works on a 3D matrix where each processes has it's own sub matrix but for doing their jobs they need some information about their neighbouring processes' sub matrix (just boundary planes). I send these informations with point to point communication but I know that for large matrix it is not a good idea so I decide to use derived data type for communication. I have problem with mpi_type_vector
: for example I have a NX*NY*NZ
matrix and I want to send plane with constant NY
to another process I write these lines for doing this:
MPI_Datatype sub;
MPI_Type_vector(NX, NZ, NY*NZ, MPI_DOUBLE, &sub);
MPI_Type_commit(&sub);
但是它不起作用(无法发送我想要的飞机).怎么了? 我的测试代码在这里:
but it doesn't work (can not send my desired plane). What is wrong? my test code is here:
#include <mpi.h>
#include <iostream>
using namespace std;
int main(int argc,char ** argv)
{
int const IE=100,JE=25,KE=100;
int size,rank;
MPI_Status status;
MPI_Init(&argc,&argv);
MPI_Comm_size(MPI_COMM_WORLD,&size);
MPI_Comm_rank(MPI_COMM_WORLD,&rank);
MPI_Datatype sub;
MPI_Type_vector(KE,IE,IE+(JE-1)*IE,MPI_DOUBLE,&sub);
MPI_Type_commit(&sub);
if (rank==0){
double*** a=new double**[IE];
for(int i=0;i<IE;i++){
a[i]=new double *[JE];
for(int j=0;j<JE;j++){
a[i][j]=new double [KE];
}
}
for(int i=0;i<IE;i++){
for(int j=0;j<JE;j++){
for(int k=0;k<KE;k++){
a[i][j][k]=2;
}}}
for(int i=0;i<IE;i++){
for(int j=0;j<JE;j++){
a[i][j][0]=2;
}}
MPI_Send(&a[0][0][0],1,sub,1,52,MPI_COMM_WORLD);
}
if (rank==1){
double*** b=new double**[IE];
for(int i=0;i<IE;i++){
b[i]=new double *[JE];
for(int j=0;j<JE;j++){
b[i][j]=new double [KE];
}
}
for(int i=0;i<IE;i++){
for(int j=0;j<JE;j++){
for(int k=0;k<KE;k++){
b[i][j][k]=0;
}}}
MPI_Recv(&b[0][0][0][0],1,sub,0,52,MPI_COMM_WORLD,&status);
for(int i=0;i<IE;i++){
for(int j=0;j<JE;j++){
for(int k=0;k<KE;k++){
if(b[i][j][k]>0){
cout<<"b["<<i<<"]["<<j<<"]["<<k<<"]="<<b[i][j][k]<<endl;
}}}}
}
MPI_Finalize();
}
推荐答案
对于3d矩阵,通常必须使用向量vector(因为涉及两个步幅)-可能,但是简单得多是使用 MPI_Type_create_subarray()您可以切出所需的多维数组的平板.
With a 3d matrix, in general you'd have to use a vector of vectors (because there are two strides involved) - which is possible, but much simpler is to use MPI_Type_create_subarray() which just lets you carve out the slab of a multidimensional array you want.
更新:上述代码中的一个问题是您分配的3d数组不连续;它是IE * JE分配的1d数组的集合,该数组可能彼此相距不远.因此,没有可靠的方法可以从中提取数据平面.
Update: One problem in the above code is that the 3d array you allocate isn't contiguous; it's a collection of IE*JE allocated 1d arrays which may or may not be anywhere near each other. So there's no reliable way of extracting a plane of data out of it.
您需要执行以下操作:
double ***alloc3d(int l, int m, int n) {
double *data = new double [l*m*n];
double ***array = new double **[l];
for (int i=0; i<l; i++) {
array[i] = new double *[m];
for (int j=0; j<m; j++) {
array[i][j] = &(data[(i*m+j)*n]);
}
}
return array;
}
然后,将数据放在一个大立方体中,就像您期望的那样,其中指向数组的指针指向其中.这-C没有真正的多维数组的事实-C + MPI一直存在.
Then the data is in one big cube, like you'd expect, with an array of pointers pointing into it. This - the fact that C doesn't have real multidimensional arrays - comes up all the time with C + MPI.
这篇关于如何将MPI派生的数据类型用于3D阵列?的文章就介绍到这了,希望我们推荐的答案对大家有所帮助,也希望大家多多支持IT屋!