目的：整个大矩阵从 [Nx, Ny, Nz] 转到 [Nz, Nx, Ny]

每个进程的输入：大矩阵的 [Nx / total_proc_num, Ny, Nz] 的部分

每个进程的输出：大矩阵的 [Nz / total_proc_num, Nx, Ny] 的部分

一开始我大概有一个想法，假设两个进程的话，就把整个大矩阵分成 4 * 4 的块，进行分配，但是我不知道怎么分，就算是把转置之前的数据分布和转置之后的数据分布写出来了，也似乎找不到规律。

我做的图

xyz

zxy

后面通过 chatgpt 做出来了

#include <algorithm>
#include <iostream>
#include <mpi.h>
#include <vector>void transpose_3d_block(int* local_A, int* local_A_T, int local_nx, int Ny, int Nz)
{for (int i = 0; i < local_nx; ++i){for (int j = 0; j < Ny; ++j){for (int k = 0; k < Nz; ++k){local_A_T[k * local_nx * Ny + i * Ny + j] = local_A[i * Ny * Nz + j * Nz + k];}}}
}int main(int argc, char** argv)
{MPI_Init(&argc, &argv);int rank, size;MPI_Comm_rank(MPI_COMM_WORLD, &rank);MPI_Comm_size(MPI_COMM_WORLD, &size);// 矩阵维度const int Nx = 4, Ny = 6, Nz = 8;const int local_nx     = Nx / size;const int new_local_nx = Nz / size;// 创建局部数据块std::vector<int> local_A(local_nx * Ny * Nz);std::vector<int> local_A_T(new_local_nx * Nx * Ny);// 初始化矩阵，仅在主进程上进行if (rank == 0){std::vector<int> A(Nx * Ny * Nz);for (int i = 0; i < Nx * Ny * Nz; ++i){A[i] = i;}// 分割矩阵到各个进程for (int i = 0; i < size; ++i){if (i == 0){std::copy(A.begin(), A.begin() + local_nx * Ny * Nz, local_A.begin());}else{MPI_Send(A.data() + i * local_nx * Ny * Nz, local_nx * Ny * Nz, MPI_INT, i, 0, MPI_COMM_WORLD);}}}else{MPI_Recv(local_A.data(), local_nx * Ny * Nz, MPI_INT, 0, 0, MPI_COMM_WORLD, MPI_STATUS_IGNORE);}// 局部转置 (local_nx, Ny, Nz) -> (Nz, local_nx, Ny)transpose_3d_block(local_A.data(), local_A_T.data(), local_nx, Ny, Nz);// 准备交换数据std::vector<int> send_data(local_A_T);std::vector<int> recv_data(send_data.size());// 交换数据MPI_Alltoall(send_data.data(),new_local_nx * local_nx * Ny,MPI_INT,recv_data.data(),new_local_nx * local_nx * Ny,MPI_INT,MPI_COMM_WORLD);// 重组数据到local_resultstd::vector<int> local_result(new_local_nx * Nx * Ny);for (int i = 0; i < size; ++i){for (int j = 0; j < new_local_nx; ++j){std::copy(recv_data.begin() + (i * new_local_nx + j) * local_nx * Ny,recv_data.begin() + (i * new_local_nx + j + 1) * local_nx * Ny,local_result.begin() + j * Nx * Ny + i * local_nx * Ny);}}// 输出转置结果，仅在主进程上进行检查if (rank == 0){std::vector<int> A_T(Nz * Nx * Ny);std::copy(local_result.begin(), local_result.begin() + new_local_nx * Nx * Ny, A_T.begin());for (int i = 1; i < size; ++i){MPI_Recv(A_T.data() + i * new_local_nx * Nx * Ny,new_local_nx * Nx * Ny,MPI_INT,i,0,MPI_COMM_WORLD,MPI_STATUS_IGNORE);}std::cout << "Transposed matrix A_T:" << std::endl;for (int i = 0; i < Nz; ++i){for (int j = 0; j < Nx; ++j){for (int k = 0; k < Ny; ++k){std::cout << A_T[i * Nx * Ny + j * Ny + k] << " ";}std::cout << std::endl;}std::cout << std::endl;}}else{MPI_Send(local_result.data(), new_local_nx * Nx * Ny, MPI_INT, 0, 0, MPI_COMM_WORLD);}MPI_Finalize();return 0;
}

输出：

Transposed matrix A_T:
0 8 16 24 32 40 
48 56 64 72 80 88 
96 104 112 120 128 136 
144 152 160 168 176 184 1 9 17 25 33 41 
49 57 65 73 81 89 
97 105 113 121 129 137 
145 153 161 169 177 185 2 10 18 26 34 42 
50 58 66 74 82 90 
98 106 114 122 130 138 
146 154 162 170 178 186 3 11 19 27 35 43 
51 59 67 75 83 91 
99 107 115 123 131 139 
147 155 163 171 179 187 4 12 20 28 36 44 
52 60 68 76 84 92 
100 108 116 124 132 140 
148 156 164 172 180 188 5 13 21 29 37 45 
53 61 69 77 85 93 
101 109 117 125 133 141 
149 157 165 173 181 189 6 14 22 30 38 46 
54 62 70 78 86 94 
102 110 118 126 134 142 
150 158 166 174 182 190 7 15 23 31 39 47 
55 63 71 79 87 95 
103 111 119 127 135 143 
151 159 167 175 183 191