【发布时间】:2017-01-22 15:48:19
【问题描述】:
给定一个 10x10 大小的网格,如下所示:
0,1,2,3,4, 5,6,7,8,9,
10,11,12,13,14, 15,16,17,18,19,
20,21,22,23,24, 25,26,27,28,29,
30,31,32,33,34, 35,36,37,38,39,
40,41,42,43,44, 45,46,47,48,49,
50,51,52,53,54, 55,56,57,58,59,
60,61,62,63,64, 65,66,67,68,69,
70,71,72,73,74, 75,76,77,78,79,
80,81,82,83,84, 85,86,87,88,89,
90,91,92,93,94, 95,96,97,98,99
如何将视觉上分离的块分散到四个进程?
我已经尝试过使用 Scatterv 描述的here(甚至here)的方式,并且它有效,但是我记得有一个项目有完全相同的问题并且不使用调整大小或 Scatterv 来解决它。
这是我所拥有的最小代码示例:
#include <stdio.h>
#include <mpi.h>
#include <stdlib.h>
#include <assert.h>
#include <memory.h>
#include <unistd.h>
void print_array(int* arr, int width, int height)
{
int i;
for (i = 0; i< width * height;i++)
{
if((i != 0) && (i % width == 0))
printf("\n");
printf("%4d ", arr[i]);
}
putchar('\n');
}
int main() {
int board[100] = {
0,1,2,3,4, 5,6,7,8,9,
10,11,12,13,14, 15,16,17,18,19,
20,21,22,23,24, 25,26,27,28,29,
30,31,32,33,34, 35,36,37,38,39,
40,41,42,43,44, 45,46,47,48,49,
50,51,52,53,54, 55,56,57,58,59,
60,61,62,63,64, 65,66,67,68,69,
70,71,72,73,74, 75,76,77,78,79,
80,81,82,83,84, 85,86,87,88,89,
90,91,92,93,94, 95,96,97,98,99
};
int numprocs, rank;
MPI_Init(NULL, NULL);
MPI_Comm_size(MPI_COMM_WORLD, &numprocs);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
int* block = calloc(25, sizeof(int));
assert(block != NULL);
MPI_Datatype sent_block_t, resized_sent_block_t;
MPI_Type_vector(5, 5, 10, MPI_INT, &sent_block_t);
MPI_Type_create_resized(sent_block_t, 0, 5*sizeof(int), &resized_sent_block_t);
MPI_Type_commit(&sent_block_t);
MPI_Type_commit(&resized_sent_block_t);
if (rank == 0) {
print_array(board, 10, 10);
MPI_Scatter(&(board[0]), 1, resized_sent_block_t,
&(block[0]), 25, MPI_INT,
0, MPI_COMM_WORLD);
}
else {
MPI_Scatter(NULL, 0, resized_sent_block_t,
&(block[0]), 25, MPI_INT,
0, MPI_COMM_WORLD);
}
for (int i = 0; i < numprocs; i++) {
MPI_Barrier(MPI_COMM_WORLD);
sleep(1);
if (i == rank) {
printf("\nRank: %d\n", rank);
print_array(block, 5, 5);
}
}
MPI_Finalize();
free(block);
}
用 4 个进程运行它,我得到了这个:
0 1 2 3 4 5 6 7 8 9
10 11 12 13 14 15 16 17 18 19
20 21 22 23 24 25 26 27 28 29
30 31 32 33 34 35 36 37 38 39
40 41 42 43 44 45 46 47 48 49
50 51 52 53 54 55 56 57 58 59
60 61 62 63 64 65 66 67 68 69
70 71 72 73 74 75 76 77 78 79
80 81 82 83 84 85 86 87 88 89
90 91 92 93 94 95 96 97 98 99
Rank: 0
0 1 2 3 4
10 11 12 13 14
20 21 22 23 24
30 31 32 33 34
40 41 42 43 44
Rank: 1
5 6 7 8 9
15 16 17 18 19
25 26 27 28 29
35 36 37 38 39
45 46 47 48 49
Rank: 2
10 11 12 13 14
20 21 22 23 24
30 31 32 33 34
40 41 42 43 44
50 51 52 53 54
Rank: 3
15 16 17 18 19
25 26 27 28 29
35 36 37 38 39
45 46 47 48 49
55 56 57 58 59
这个散射是错误的,注意rank 2和3的内容。
正确的是:
Rank 0: Rank 1:
0,1,2,3,4, 5,6,7,8,9,
10,11,12,13,14, 15,16,17,18,19,
20,21,22,23,24, 25,26,27,28,29,
30,31,32,33,34, 35,36,37,38,39,
40,41,42,43,44, 45,46,47,48,49,
Rank 2: Rank 3:
50,51,52,53,54, 55,56,57,58,59,
60,61,62,63,64, 65,66,67,68,69,
70,71,72,73,74, 75,76,77,78,79,
80,81,82,83,84, 85,86,87,88,89,
90,91,92,93,94, 95,96,97,98,99
问题
有没有什么方法可以在不使用 Scatterv 的情况下分散相同大小的网格块?
【问题讨论】:
-
所以你说你的代码有效并且你理解它很好。您的实际问题是什么?
-
@Zulan 不,我发布的结果不是正确的输出。请参阅帖子开头的给定板,它在视觉上分为四个块。我想要左上角,去进程 0,右上角到进程 1,左下角到进程 2,左下角到进程 3。(我已经添加了预期的输出。)
-
抱歉,我只是简要浏览了输出。有一个very good extensive explanation,关于如何正确地做你想做的事,使用
MPI_Scatterv。我无法想象只使用MPI_Scatter的正确实现,这不是不必要的hacky。