windows - 使用cygwin在Linux和Windows中编译的MPI fortran代码

Question

的代码和目标

我有一个称为fortran的mpi elast3d_mpi.f代码，可以在 Windows 和 Linux 系统中进行编译。

预期的行为

在和Linux 中进行编译的方式如下

mpif90 -o elast3d_mpi elast3d_mpi.f

然后可以使用 mpirun命令并行执行该程序

mpirun -n 2 elast3d_mpi

终端输出允许观察到2个处理器正在运行，正如预期的那样

 There are 2  processors running this job.
 Rank# 0 d1 = 1   d2 = 64
 Rank# 1 d1 = 65  d2 = 128
 ...

如果程序在 linux 中没有 mpirun的情况下运行，那么它将正常运行且没有错误，也没有并行处理。

问题

为了在Windows中进行编译，使用了 cygwin环境。安装这些软件包后

Package                  Version            Status
_autorebase              001007-1           OK
alternatives             1.3.30c-10         OK
base-cygwin              3.8-1              OK
base-files               4.3-2              OK
bash                     4.4.12-3           OK
binutils                 2.29-1             OK
bzip2                    1.0.8-1            OK
ca-certificates          2.32-1             OK
coreutils                8.26-2             OK
crypto-policies          20190218-1         OK
cygutils                 1.4.16-2           OK
cygwin                   3.0.7-1            OK
cygwin-debuginfo         3.0.7-1            OK
cygwin-devel             3.0.7-1            OK
dash                     0.5.9.1-1          OK
diffutils                3.5-2              OK
editrights               1.03-1             OK
file                     5.32-1             OK
findutils                4.6.0-1            OK
gawk                     5.0.1-1            OK
gcc-core                 7.4.0-1            OK
gcc-fortran              7.4.0-1            OK
getent                   2.18.90-4          OK
grep                     3.0-2              OK
groff                    1.22.4-1           OK
gzip                     1.8-1              OK
hostname                 3.13-1             OK
info                     6.7-1              OK
ipc-utils                1.0-2              OK
less                     530-1              OK
libargp                  20110921-3         OK
libatomic1               7.4.0-1            OK
libattr1                 2.4.48-2           OK
libblkid1                2.33.1-1           OK
libbz2_1                 1.0.8-1            OK
libcrypt0                2.1-1              OK
libfdisk1                2.33.1-1           OK
libffi6                  3.2.1-2            OK
libgc1                   8.0.4-1            OK
libgcc1                  7.4.0-1            OK
libgdbm4                 1.13-1             OK
libgfortran3             6.4.0-5            OK
libgfortran4             7.4.0-1            OK
libgmp10                 6.1.2-1            OK
libgomp1                 7.4.0-1            OK
libguile17               1.8.8-3            OK
libguile2.0_22           2.0.14-3           OK
libiconv                 1.14-3             OK
libiconv2                1.14-3             OK
libintl8                 0.19.8.1-2         OK
libisl15                 0.16.1-1           OK
libltdl7                 2.4.6-7            OK
liblzma5                 5.2.4-1            OK
libmpc3                  1.1.0-1            OK
libmpfr6                 4.0.2-1            OK
libncursesw10            6.1-1.20190727     OK
libopenmpi-devel         3.1.3-1            OK
libopenmpi12             1.10.7-1           OK
libopenmpi40             3.1.3-1            OK
libopenmpicxx1           1.10.4-1           OK
libopenmpifh12           1.10.7-1           OK
libopenmpifh40           3.1.3-1            OK
libopenmpiusef08_40      3.1.3-1            OK
libopenmpiusetkr40       3.1.3-1            OK
libp11-kit0              0.23.15-1          OK
libpcre1                 8.43-1             OK
libpipeline1             1.5.1-1            OK
libpkgconf3              1.6.0-1            OK
libpopt-common           1.16-2             OK
libpopt0                 1.16-2             OK
libquadmath0             7.4.0-1            OK
libreadline7             7.0.3-3            OK
libsigsegv2              2.10-2             OK
libsmartcols1            2.33.1-1           OK
libssl1.1                1.1.1d-1           OK
libstdc++6               7.4.0-1            OK
libtasn1_6               4.14-1             OK
libunistring2            0.9.10-1           OK
libuuid1                 2.33.1-1           OK
login                    1.13-1             OK
make                     4.2.1-2            OK
man-db                   2.7.6.1-1          OK
mintty                   3.0.6-1            OK
ncurses                  6.1-1.20190727     OK
openmpi                  3.1.3-1            OK
openmpi-debuginfo        3.1.1-2            OK
openssl                  1.1.1d-1           OK
p11-kit                  0.23.15-1          OK
p11-kit-trust            0.23.15-1          OK
pkg-config               1.6.0-1            OK
pkgconf                  1.6.0-1            OK
rebase                   4.4.4-1            OK
run                      1.3.4-2            OK
sed                      4.4-1              OK
tar                      1.29-1             OK
terminfo                 6.1-1.20190727     OK
terminfo-extra           6.1-1.20190727     OK
tzcode                   2019c-1            OK
tzdata                   2019c-1            OK
util-linux               2.33.1-1           OK
vim-minimal              8.1.1772-1         OK
w32api-headers           5.0.4-1            OK
w32api-runtime           5.0.4-1            OK
which                    2.20-2             OK
windows-default-manifest 6.4-1              OK
xz                       5.2.4-1            OK
zlib0                    1.2.11-1           OK

在 Windows (7)中，以类似方式编译程序，但使用 cygwin终端

mpif90 -o elast_3d_mpi.exe elast3d_mpi.f

1-当我尝试在 mpirun终端中使用 cygwin运行它时，出现以下错误

$ mpirun -n 2 elast3d_mpi.exe
-----------------------------------------------------------------
Sorry!  You were supposed to get help about:
    agent-not-found
from the file:
    help-plm-rsh.txt
But I couldn't find that topic in the file.  Sorry!
-----------------------------------------------------------------
[gauss:00824] [[INVALID],INVALID] FORCE-TERMINATE AT Not found:-13 - error /cygdrive/d/cyg_pub/devel/openmpi/v3.1/openmpi-3.1.3-1.x86_64/src/openmpi-3.1.3/orte/mca/plm/rsh/plm_rsh_component.c(327)
[gauss:00824] *** Process received signal ***
[gauss:00824] Signal: Segmentation fault (11)
[gauss:00824] Signal code: Address not mapped (23)
[gauss:00824] Failing at address: 0x0
Unable to print stack trace!
[gauss:00824] *** End of error message ***

2-当我在 orterun终端中使用 cygwing的 cmd实现运行它时，出现此错误

C:\Users\io\Documents\elast-mpi>orterun.exe -np 2 elast3d_mpi
------------------------------------------------------------
Sorry!  You were supposed to get help about:
agent-not-found
from the file:
help-plm-rsh.txt
But I couldn't find that topic in the file.  Sorry!
------------------------------------------------------------------
[gauss:00827] [[INVALID],INVALID] FORCE-TERMINATE AT Not found:-13 - 
error /cygd
rive/d/cyg_pub/devel/openmpi/v3.1/openmpi-3.1.3-1.x86_64/src/openmpi-3.1.3/orte/
mca/plm/rsh/plm_rsh_component.c(327)
[gauss:00827] *** Process received signal ***
[gauss:00827] Signal: Segmentation fault (11)
[gauss:00827] Signal code: Address not mapped (23)
[gauss:00827] Failing at address: 0x0
Unable to print stack trace!
[gauss:00827] *** End of error message ***
1 [main] orterun 827 cygwin_exception::open_stackdumpfile: 
Dumping stack t
race to orterun.exe.stackdump

3-在没有 ortermpi.exe的Windows中运行程序，程序将输出下一个错误

C:\Users\io\Documents\elast-mpi>elast3d_mpi
---------------------------------------------------------------------
Sorry!  You were supposed to get help about:
agent-not-found
from the file:
help-plm-rsh.txt
But I couldn't find that topic in the file.  Sorry!
---------------------------------------------------------------------
[gauss:00833] [[INVALID],INVALID] FORCE-TERMINATE AT Not found:-13 - error /cygd
rive/d/cyg_pub/devel/openmpi/v3.1/openmpi-3.1.3-1.x86_64/src/openmpi-3.1.3/orte/
mca/plm/rsh/plm_rsh_component.c(327)
[gauss:00833] Process received signal 
[gauss:00833] Signal: Segmentation fault (11)
[gauss:00833] Signal code: Address not mapped (23)
[gauss:00833] Failing at address: 0x0
Unable to print stack trace!
[gauss:00833] End of error message 
[gauss:00832] [[INVALID],INVALID] ORTE_ERROR_LOG: Unable to start a daemon on th
e local node in file /cygdrive/d/cyg_pub/devel/openmpi/v3.1/openmpi-3.1.3-1.x86_
64/src/openmpi-3.1.3/orte/mca/ess/singleton/ess_singleton_module.c at line 532
[gauss:00832] [[INVALID],INVALID] ORTE_ERROR_LOG: Unable to start a daemon on th
e local node in file /cygdrive/d/cyg_pub/devel/openmpi/v3.1/openmpi-3.1.3-1.x86_
64/src/openmpi-3.1.3/orte/mca/ess/singleton/ess_singleton_module.c at line 166
--------------------------------------------------------------------
It looks like orte_init failed for some reason; your parallel process is
likely to abort.  There are many reasons that a parallel process can
fail during orte_init; some of which are due to configuration or
environment problems.  This failure appears to be an internal failure;
here's some additional information (which may only be relevant to an
Open MPI developer):
orte_ess_init failed
--> Returned value Unable to start a daemon on the local node (-127) instead o
f ORTE_SUCCESS
---------------------------------------------------------------------
---------------------------------------------------------------------
It looks like MPI_INIT failed for some reason; your parallel process is
likely to abort.  There are many reasons that a parallel process can
fail during MPI_INIT; some of which are due to configuration or environment
problems.  This failure appears to be an internal failure; here's some
additional information (which may only be relevant to an Open MPI
developer):
ompi_mpi_init: ompi_rte_init failed
Returned "Unable to start a daemon on the local node" (-127) instead of "Success" (0)
--------------------------------------------------------------------
An error occurred in MPI_Init
on a NULL communicator
MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort, and potentially your MPI job)
[gauss:00832] Local abort before MPI_INIT completed completed successfully, but
am not able to aggregate error messages, and not able to guarantee that all other processes were killed!

观察和问题

如果程序在 linux 中没有mpirun的情况下运行，则它可以正常运行且没有错误，也无需并行处理。

在没有ortermpi.exe的Windows中运行该程序，程序将输出错误。

听起来像是一个正在运行的(环境)问题。

是在Windows中编译此程序的最佳方法吗？

我可以在Windows和Linux中编译相同的mpi fortran代码程序吗？

为了使程序在Windows系统中运行，我可以尝试编译什么程序？

Answer 1

如@payam_sbr所指出的，openssh软件包的安装允许在窗口中并行运行程序。

测试

安装openssh软件包后，可以在Windows中运行该程序:

1-使用 cygwin 终端

mpirun -n 2 ./elast3d_mpi.exe

2-或使用 cmd 终端

orterun -np 2 ./elast3d_mpi.exe

在这两种情况下，结果都与 中的结果相同linux

There are 2  processors running this job.
 Rank# 0 d1 = 1   d2 = 64
 Rank# 1 d1 = 65  d2 = 128
 ...

观测

openssh的安装解决了情况1)和2)的问题，因此解决了该问题(使用cygwin在Windows中并行运行代码)。

另一方面，就程序的分发而言，在所有其他机器上运行该程序的所有cygwing软件包的安装都不是很实际。

有没有一种方法可以编译我的所有代码，以便在Windows中运行，而无需在要运行代码的每台计算机上手动安装cygwin软件包？有什么选择？