我有一组用户定义的类型。两个处理器中的每一个都唯一地写入阵列的一部分，最后，我想要组合阵列。为什么我的代码会产生分段错误？
program type_derived_recv_send
use nodeinfo
implicit none
include 'mpif.h'
integer(4) :: ierr
integer(4) :: istatus1(MPI_STATUS_SIZE)
integer(4) :: istatus2(MPI_STATUS_SIZE)
integer(4) :: i,j,k,l
integer(4) :: iblock(8)
integer(4) :: idisp(8)
integer(4) :: itype(8)
integer(4) :: inewtype
integer(4) :: iextent
integer(4) :: itag, isend, irecv
integer(4) :: n,m,p,q
TYPE :: newXYZ
real :: x, u
real :: y, v
real :: z, w
real,dimension(3) :: uvw
END TYPE
TYPE (newXYZ) :: xyz, xyz2
TYPE (newXYZ), allocatable :: xyzArray(:,:,:,:)
n = 1
m = 1
p = 1
q = 2
allocate( xyzArray(m,n,p,q) )
call MPI_INIT(ierr)
icomm = MPI_COMM_WORLD
call MPI_COMM_SIZE(icomm,nproc,ierr)
call MPI_COMM_RANK(icomm,iproc,ierr)
do l=1,q
do k=1,p
do j=1,n
do i=1,m
xyzArray(i,j,k,m)%x = 0
xyzArray(i,j,k,m)%y = 0
xyzArray(i,j,k,m)%z = 0
xyzArray(i,j,k,m)%u = 0
xyzArray(i,j,k,m)%v = 0
xyzArray(i,j,k,m)%w = 0
xyzArray(i,j,k,m)%uvw = (/0,0,0/)
end do
end do
end do
end do
if (iproc == 1) then
xyzArray(1,1,1,2)%x = 1.1
xyzArray(1,1,1,2)%y = 2.1
xyzArray(1,1,1,2)%z = 3.1
xyzArray(1,1,1,2)%u = 4.1
xyzArray(1,1,1,2)%v = 5.1
xyzArray(1,1,1,2)%w = 6.1
xyzArray(1,1,1,2)%uvw = (/10.1,10.1,10.1/)
else
xyzArray(1,1,1,1)%x = 0.1
xyzArray(1,1,1,1)%y = 0.1
xyzArray(1,1,1,1)%z = 0.1
xyzArray(1,1,1,1)%u = 0.1
xyzArray(1,1,1,1)%v = 0.1
xyzArray(1,1,1,1)%w = 0.1
xyzArray(1,1,1,1)%uvw = (/0.1,0.1,0.1/)
endif
itag = 1
xyz = xyzArray(1,1,1,2)
xyz2 = xyzArray(1,1,1,1)
if ( iproc == 1) call MPI_ISEND(xyzArray(1,1,1,2),36,MPI_INTEGER,0,1,icomm,isend,ierr)
if ( iproc == 0) call MPI_IRECV(xyzArray(1,1,1,2),36,MPI_INTEGER,1,1,icomm,irecv,ierr)
call MPI_WAIT(isend,istatus1,ierr)
call MPI_WAIT(irecv,istatus2,ierr)
if ( iproc == 0 )write(6,*)'iproc = ',iproc
if ( iproc == 0 ) write(6,*)xyzArray
if ( iproc == 1 )write(6,*)'iproc = ',iproc
if ( iproc == 1 ) write(6,*)xyzArray
call MPI_FINALIZE(ierr)
end program type_derived_recv_send
输出：
--------------------------------------------------------------------------
newtype extent = 36
xyz = 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
xyz2 = 0.1000000 0.1000000 0.1000000 0.1000000 0.1000000 0.1000000 0.1000000 0.1000000 0.1000000
[west0334:09950] *** Process received signal ***
[west0334:09950] Signal: Segmentation fault (11)
[west0334:09950] Signal code: Address not mapped (1)
[west0334:09950] Failing at address: 0x40
[west0334:09950] [ 0] /lib64/libpthread.so.0 [0x3e00e0eb70]
[west0334:09950] [ 1] /n/sw/openmpi-1.2.5-gcc-4.1.2/lib64/libmpi.so.0(ompi_request_wait+0x17) [0x2b33f8628607]
[west0334:09950] [ 2] /n/sw/openmpi-1.2.5-gcc-4.1.2/lib64/libmpi.so.0(PMPI_Wait+0x88) [0x2b33f86566e8]
[west0334:09950] [ 3] /n/sw/openmpi-1.2.5-gcc-4.1.2/lib64/libmpi_f77.so.0(PMPI_WAIT+0x3a) [0x2b33f83eac8a]
[west0334:09950] [ 4] ./type_derived_recv_send.x(MAIN__+0x730) [0x401cd0]
newtype extent = 36
xyz = 1.100000 4.100000 2.100000 5.100000 3.100000 6.100000 10.10000 10.10000 10.10000
xyz2 = 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
[west0334:09951] *** Process received signal ***
[west0334:09951] Signal: Segmentation fault (11)
[west0334:09951] Signal code: Address not mapped (1)
[west0334:09951] Failing at address: 0x40
[west0334:09950] [ 5] ./type_derived_recv_send.x(main+0xe) [0x4049ae]
[west0334:09950] [ 6] /lib64/libc.so.6(__libc_start_main+0xf4) [0x3e0021d994]
[west0334:09950] [ 7] ./type_derived_recv_send.x [0x4014e9]
[west0334:09950] *** End of error message ***
[west0334:09951] [ 0] /lib64/libpthread.so.0 [0x3e00e0eb70]
[west0334:09951] [ 1] /n/sw/openmpi-1.2.5-gcc-4.1.2/lib64/libmpi.so.0(ompi_request_wait+0x17) [0x2b73c1aca607]
[west0334:09951] [ 2] /n/sw/openmpi-1.2.5-gcc-4.1.2/lib64/libmpi.so.0(PMPI_Wait+0x88) [0x2b73c1af86e8]
[west0334:09951] [ 3] /n/sw/openmpi-1.2.5-gcc-4.1.2/lib64/libmpi_f77.so.0(PMPI_WAIT+0x3a) [0x2b73c188cc8a]
[west0334:09951] [ 4] ./type_derived_recv_send.x(MAIN__+0x748) [0x401ce8]
[west0334:09951] [ 5] ./type_derived_recv_send.x(main+0xe) [0x4049ae]
[west0334:09951] [ 6] /lib64/libc.so.6(__libc_start_main+0xf4) [0x3e0021d994]
[west0334:09951] [ 7] ./type_derived_recv_send.x [0x4014e9]
[west0334:09951] *** End of error message ***
Nov 16 20:17:14 2012 9939 4 7.06 handleTSRegisterTerm(): TS reports task <0> pid <9950> on host<west0334> killed or core dumped
Job /lsf/7.0/linux2.6-glibc2.3-x86_64/bin/openmpi_wrapper ./type_derived_recv_send.x
答案 0 :(得分：4)
错误在于：
if ( iproc == 1) call MPI_ISEND(xyzArray(1,1,1,2),36,MPI_INTEGER,0,1,icomm,isend,ierr)
if ( iproc == 0) call MPI_IRECV(xyzArray(1,1,1,2),36,MPI_INTEGER,1,1,icomm,irecv,ierr)
call MPI_WAIT(isend,istatus1,ierr)
call MPI_WAIT(irecv,istatus2,ierr)
irecv请求句柄在排名0中保持未初始化状态，而isend请求句柄仍保留在排名1中。这里有两个选项 - 将MPI_WAIT调用与其相应的非阻塞MPI操作分组，或者在此之前将isend和irecv初始化为MPI_REQUEST_NULL。
解决方案1：
if ( iproc == 1) then
call MPI_ISEND(xyzArray(1,1,1,2),36,MPI_INTEGER,0,1,icomm,isend,ierr)
call MPI_WAIT(isend,istatus1,ierr)
end if
if ( iproc == 0) then
call MPI_IRECV(xyzArray(1,1,1,2),36,MPI_INTEGER,1,1,icomm,irecv,ierr)
call MPI_WAIT(irecv,istatus2,ierr)
end if
解决方案2（简化）：
ireq = MPI_REQUEST_NULL
if ( iproc == 1) call MPI_ISEND(xyzArray(1,1,1,2),36,MPI_INTEGER,0,1,icomm,ireq,ierr)
if ( iproc == 0) call MPI_IRECV(xyzArray(1,1,1,2),36,MPI_INTEGER,1,1,icomm,ireq,ierr)
call MPI_WAIT(ireq,istatus,ierr)
您实际上不需要两个单独的请求句柄以及两个单独的状态数组。使用空句柄初始化ireq时，此代码也可以使用超过2个MPI进程正常运行。
请注意，为了便于移植，您应该使用MPI_TYPE_CREATE_STRUCT构建派生结构数据类型，并使用它来发送和接收TYPE(newXYZ)的元素。
答案 1 :(得分：0)
为什么要发送一个带有整数MPI数据类型的9 real的类型？为什么不发送9 MPI_REAL或36 MPI_BYTE