kupl_shm_request_start
执行request中的集合通信操作。
接口定义
int kupl_shm_request_start(kupl_shm_request_h request);
参数
参数名 |
类型 |
描述 |
输入/输出 |
---|---|---|---|
request |
kupl_shm_request_h |
需要执行集合通信操作的request |
输入 |
返回值
- 成功:返回KUPL_OK
- 失败:返回KUPL_ERROR
示例
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 | #include <stdio.h> #include <mpi.h> #include <unistd.h> #include "kupl.h" // 创建 kupl 通信域需要的回调函数 1 static int oob_barrier_callback(void *group) { return MPI_Barrier((MPI_Comm)group); } // 创建 kupl 通信域需要的回调函数 2 static int oob_allgather_callback(const void *sendbuf, void *recvbuf, int size, void *group, kupl_shm_datatype_t datatype) { switch (datatype) { case KUPL_SHM_DATATYPE_CHAR: return MPI_Allgather(sendbuf, size, MPI_CHAR, recvbuf, size, MPI_CHAR, (MPI_Comm)group); default: fprintf(stderr, "not support datatype"); return KUPL_ERROR; } } int main(int argc, char *argv[]) { // 初始化 MPI 环境 MPI_Init(&argc, &argv); MPI_Comm comm = MPI_COMM_WORLD; // 获取 MPI 通信域大小 int world_size; MPI_Comm_size(comm, &world_size); // 获取进程 rank 号 int world_rank; MPI_Comm_rank(comm, &world_rank); // 获取进程 pid 号 int pid = getpid(); // 创建 kupl 通信域 kupl_shm_oob_cb_t oob_cbs; kupl_shm_oob_cb_h oob_cbs_h = &oob_cbs; oob_cbs_h->oob_allgather = oob_allgather_callback; oob_cbs_h->oob_barrier = oob_barrier_callback; kupl_shm_comm_h kupl_comm; int ret = kupl_shm_comm_create(world_size, world_rank, pid, oob_cbs_h, (void *)comm, &kupl_comm); if (ret != KUPL_OK || kupl_comm == NULL) { fprintf(stderr, "kupl shm comm create failed: %d\n", ret); return -1; } kupl_shm_win_h win_send; kupl_shm_win_h win_recv; kupl_shm_win_h ompi_win_recv; void *sendbuf = NULL; void *recvbuf = NULL; void *ompibuf = NULL; int count = 16000; size_t buf_size = sizeof(int) * count; ret = kupl_shm_win_alloc(buf_size, kupl_comm, &sendbuf, &win_send); if (ret != KUPL_OK) { fprintf(stderr, "kupl sendbuf alloc failed: %d\n", ret); return -1; } ret = kupl_shm_win_alloc(buf_size, kupl_comm, &recvbuf, &win_recv); if (ret != KUPL_OK) { fprintf(stderr, "kupl recvbuf alloc failed: %d\n", ret); kupl_shm_win_free(win_send); win_send = NULL; return -1; } ret = kupl_shm_win_alloc(buf_size, kupl_comm, &ompibuf, &ompi_win_recv); if (ret != KUPL_OK) { fprintf(stderr, "kupl ompibuf alloc failed: %d\n", ret); kupl_shm_win_free(win_send); kupl_shm_win_free(win_recv); win_send = NULL; win_recv = NULL; return -1; } auto *t_sendbuf = static_cast<int *>(sendbuf); auto *t_recvbuf = static_cast<int *>(recvbuf); auto *t_ompibuf = static_cast<int *>(ompibuf); for (int i = 0; i < count; i++) { t_sendbuf[i] = world_rank + i; } kupl_shm_request_h request; MPI_Barrier(comm); kupl_shm_allreduce_init(sendbuf, recvbuf, count, KUPL_SHM_DATATYPE_INT, KUPL_SHM_REDUCE_OP_SUM, kupl_comm, &request); kupl_shm_request_start(request); kupl_shm_request_wait(request); kupl_shm_request_free(request); MPI_Barrier(comm); MPI_Allreduce(sendbuf, ompibuf, count, MPI_INT, MPI_SUM, comm); int check = 0; for (int i = 0; i < count; i++) { if (t_recvbuf[i] != t_ompibuf[i]) { check = 1; break; } } int result = 0; MPI_Reduce(&check, &result, 1, MPI_INT, MPI_SUM, 0, comm); if (result == 0) { printf("check success\n"); } else { printf("check failed\n"); } kupl_shm_win_free(win_send); kupl_shm_win_free(win_recv); kupl_shm_win_free(ompi_win_recv); win_send = NULL; win_recv = NULL; ompi_win_recv = NULL; kupl_shm_comm_destroy(kupl_comm); MPI_Finalize(); return 0; } |
运行结果如下。
check success

上述示例演示了kupl allreduce的流程。kupl_shm_request_start函数执行request。
父主题: 共享内存通信函数