Question

我是MPI的新手，如果满足条件，我试图从其中一个进程广播。问题是当我尝试将其打印出来时，我在其他进程上看不到广播的消息。当进程通过第二个if条件时，它会广播numpy数组。在我的代码中，进程0,1,3,5,7传递第二个if条件。这是代码：

import numpy as np
from mpi4py import MPI
comm = MPI.COMM_WORLD
rank = comm.Get_rank()

rand_num=np.empty(5,dtype='float')
bands  = [(0, 65), (66, 132), (133, 198), (199, 264), (265, 330), (331, 396), (397, 462), (463, 529)]
volts = {'map1':[478], 'map2':[346], 'map3':[213], 'map4':[81], 'map5':[34]}

for items in volts.keys():
    if(len(volts[items])!=0):
        neur_id = volts[items][0]
        if neur_id in range(bands[comm.rank][0],bands[comm.rank][1]):
            rand_num = np.array([comm.rank for i in range(0,5)],dtype='float')
            comm.Bcast(rand_num, root=comm.rank)
        comm.Barrier() #hoping to stop other processes from proceeding.
    print 'data and rank',rand_num,comm.rank

我期待输出看起来像：

data and rank [ 0.  0.  0.  0.  0.] 0           data and rank [ 1.  1.  1.  1.  1.] 0          data and rank [ 3.  3.  3.  3.  3.] 0          data and rank [ 5.  5.  5.  5.  5.] 0          data and rank [ 7.  7.  7.  7.  7.] 0  
data and rank [ 0.  0.  0.  0.  0.] 1           data and rank [ 1.  1.  1.  1.  1.] 1          data and rank [ 3.  3.  3.  3.  3.] 1          data and rank [ 5.  5.  5.  5.  5.] 1          data and rank [ 7.  7.  7.  7.  7.] 1   
data and rank [ 0.  0.  0.  0.  0.] 2           data and rank [ 1.  1.  1.  1.  1.] 2          data and rank [ 3.  3.  3.  3.  3.] 2           data and rank [ 5.  5.  5.  5.  5.] 2         data and rank [ 7.  7.  7.  7.  7.] 2
data and rank [ 0.  0.  0.  0.  0.] 3          data and rank [ 1.  1.  1.  1.  1.] 3           data and rank [ 3.  3.  3.  3.  3.] 3          data and rank [ 5.  5.  5.  5.  5.] 3          data and rank [ 7.  7.  7.  7.  7.] 3
data and rank [ 0.  0.  0.  0.  0.] 4          data and rank [ 1.  1.  1.  1.  1.] 4            data and rank [ 3.  3.  3.  3.  3.] 4          data and rank [ 5.  5.  5.  5.  5.] 4          data and rank [ 7.  7.  7.  7.  7.] 4
data and rank [ 0.  0.  0.  0.  0.] 5          data and rank [ 1.  1.  1.  1.  1.] 5           data and rank [ 3.  3.  3.  3.  3.] 5          data and rank [ 5.  5.  5.  5.  5.] 5           data and rank [ 7.  7.  7.  7.  7.] 5
data and rank [ 0.  0.  0.  0.  0.] 6          data and rank [ 1.  1.  1.  1.  1.] 6            data and rank [ 3.  3.  3.  3.  3.] 6         data and rank [ 5.  5.  5.  5.  5.] 6           data and rank [ 7.  7.  7.  7.  7.] 6
data and rank [ 0.  0.  0.  0.  0.] 7          data and rank [ 1.  1.  1.  1.  1.] 7             data and rank [ 3.  3.  3.  3.  3.] 7         data and rank [ 5.  5.  5.  5.  5.] 7            data and rank [ 7.  7.  7.  7.  7.] 7

然而，输出看起来完全不同。我的代码中没有任何逻辑错误。 PS：我注意到这个问题与this重复。我的进程都没有陷入僵局。我的问题是关于同步。我还阅读here如果所有流程都没有调用Bcast，则MPI不会同步Bcast，这就是我Barrier之后立即使用Bcast的原因。由于未在所有进程（排名）上调用bcast，因此将其标记为重复。如果在我的申请中，MPI bcast无法按要求使用，那么我应该使用哪个MPI命令进行＆＃34;广播＆＃34;？

因为这个问题被标记为重复，我自己无法回答。我发布了我的解决方案，我在for循环中使用了Send和Recv来解决我的问题：

import numpy as np
from mpi4py import MPI
from mpi4py.MPI import ANY_SOURCE
comm = MPI.COMM_WORLD
rank = comm.Get_rank()
size = comm.Get_size()

rand_num=np.empty(5,dtype='float')
bands  = [(0, 65), (66, 132), (133, 198), (199, 264), (265, 330), (331, 396), (397, 462), (463, 529)]
volts = {'map1':[478], 'map2':[346], 'map3':[213], 'map4':[81], 'map5':[34]}

for items in volts.keys():
    if(len(volts[items])!=0):
        neur_id = volts[items][0]
        if neur_id in range(bands[comm.rank][0],bands[comm.rank][1]):
            rand_num = np.array([comm.rank for i in range(0,5)],dtype='float')
            all_ranks = [i for i in range(0,size)]
            all_ranks.remove(rank)
            rest_ranks = all_ranks 
            for ranks in rest_ranks: 
                comm.Send(rand_num, dest= ranks)

        else:
            comm.Recv(rand_num, source = ANY_SOURCE)

    print 'data and rank',rand_num,comm.rank, items

如果在MPI中满足条件，则从进程广播

0 个答案: