RFE: System introspection for cmake generator selection under Msys2 in build_opencoarrays.sh

[jbmaggard]

Native Win64 CAF solution using OpenCoarrays-2.0.0/MSYS2/gfortran/Intel-MPI.

• OpenCoarrays Version: 2.0.0
• Fortran Compiler: GNU Fortran (Rev2, Built by MSYS2 project) 7.3.0
• C compiler used for building lib: gcc.exe (Rev2, Built by MSYS2 project) 7.3.0
• Installation method: install.sh (with one change to FindMPI.cmake and -G"MSYS Makefiles" in build_opencoarrays.sh)
• Output of uname -a: MINGW64_NT-6.1 PE-MGR-LAPTOP 2.10.0(0.325/5/3) 2018-02-09 15:25 x86_64 Msys
• MPI library being used: Intel(R) MPI Library for Windows* OS, Version 2018 Update 2 Build 20180125
• Machine architecture and number of physical cores: Intel Haswell i7, 4 cores
• Version of CMake: 3.11.1

Observed Behavior

Cmake running on Win64 under the MSYS2 bash prompt is a native Win64 application, and as such uses the default windows cmake generator (looking for the MS compiler), unless the -G"MSYS Makefiles" cmake command line option is used; which causes cmake to use gcc, gfortran, and ar correctly.

Expected Behavior

Introspection with CMAKE variable MSYS, or output from uname (output of "uname -o" is "Msys") would allow application of the -G"MSYS Makefiles" cmake command line option to select the correct cmake generator when executing cmake from build_opencoarrays.sh

Steps to Reproduce

Allows success with install.sh.

1. Build an import library, libmpi.a, interfacing to the Intel-MPI impi.dll. This is done using the gendef (tools-git package), and dlltool (binutils package) utilities.
2. Set up mpifort, mpicc bash scripts modified from MPICH 3.2 and a simple mpirun script, so that gcc, gfortran, and the Intel-MPI mpiexec.exe work from the msys2 bash prompt the same way as on linux to build and execute MPI programs.
3. Edit FindMPI.cmake to allow execution of the bash scripts under the bash shell.

$ diff -Nu FindMPI.cmake.orig FindMPI.cmake
--- FindMPI.cmake.orig  2018-05-17 16:29:45.352467900 -0500
+++ FindMPI.cmake       2018-05-20 14:58:14.090710500 -0500
@@ -337,7 +337,7 @@
     separate_arguments(_MPI_COMPILER_WRAPPER_OPTIONS NATIVE_COMMAND "${MPI_COMPILER_FLAGS}")
   endif()
   execute_process(
-    COMMAND ${MPI_${LANG}_COMPILER} ${_MPI_COMPILER_WRAPPER_OPTIONS} ${QUERY_FLAG}
+    COMMAND "bash" ${MPI_${LANG}_COMPILER} ${_MPI_COMPILER_WRAPPER_OPTIONS} ${QUERY_FLAG}
     OUTPUT_VARIABLE  WRAPPER_OUTPUT OUTPUT_STRIP_TRAILING_WHITESPACE
     ERROR_VARIABLE   WRAPPER_OUTPUT ERROR_STRIP_TRAILING_WHITESPACE
     RESULT_VARIABLE  WRAPPER_RETURN)

A workaround not requiring any edit to FindMPI.cmake might be to prepend "bash " to the MPI_${LANG}_COMPILER variable if the cmake variable MSYS is used for introspection?

4. Edit OpenCoarrays-2.0.0/prerequisites/install-functions/build_opencoarrays.sh to include the -G"MSYS Makefiles" cmake command line option (done manually rather than via introspection here).

$ diff -Nu build_opencoarrays.sh.orig build_opencoarrays.sh
--- build_opencoarrays.sh.orig  2018-05-21 12:38:25.164838800 -0500
+++ build_opencoarrays.sh       2018-05-21 12:38:46.151258600 -0500
@@ -38,7 +38,7 @@
   fi
   info "Configuring OpenCoarrays in ${PWD} with the command:"
   info "CC=\"${CC}\" FC=\"${FC}\" $CMAKE \"${opencoarrays_src_dir}\" \"${WDEVFLAG}\" -DCMAKE_INSTALL_PREFIX=\"${install_path}\" -DMPIEXEC=\"${MPIEXEC}\" -DMPI_C_COMPILER=\"${MPICC}\" -DMPI_Fortran_COMPILER=\"${MPIFC}\""
-  CC="${CC}" FC="${FC}" $CMAKE "${opencoarrays_src_dir}" "${WDEVFLAG}" -DCMAKE_INSTALL_PREFIX="${install_path}" -DMPIEXEC="${MPIEXEC}" -DMPI_C_COMPILER="${MPICC}" -DMPI_Fortran_COMPILER="${MPIFC}"
+  CC="${CC}" FC="${FC}" $CMAKE "${opencoarrays_src_dir}" "${WDEVFLAG}" -G"MSYS Makefiles" -DCMAKE_INSTALL_PREFIX="${install_path}" -DMPIEXEC="${MPIEXEC}" -DMPI_C_COMPILER="${MPICC}" -DMPI_Fortran_COMPILER="${MPIFC}"
   info "Building OpenCoarrays in ${PWD} with the command make -j${num_threads}"
   make "-j${num_threads}"
   if [[ ! -z ${SUDO:-} ]]; then

5. install.sh and ctest demonstrate building and execution of CAF programs, with all tests passed except for those with failed images. Note that -DUSE_FAILED_IMAGES was turned on when building the libcaf_mpi.a and libcaf_mpi.dll libraries (verified from output of make clean; make VERBOSE=1).

Excerpt from install.sh.

-- Looking for signal.h
-- Looking for signal.h - found
-- Looking for SIGKILL
-- Looking for SIGKILL - not found
-- Looking for SIGKILL
-- Looking for SIGKILL - found
-- Looking for include files mpi.h, mpi-ext.h
-- Looking for include files mpi.h, mpi-ext.h - not found
-- Looking for MPIX_ERR_PROC_FAILED
-- Looking for MPIX_ERR_PROC_FAILED - found
-- Looking for MPIX_ERR_REVOKED
-- Looking for MPIX_ERR_REVOKED - found
-- Looking for MPIX_Comm_failure_ack
-- Looking for MPIX_Comm_failure_ack - found
-- Looking for MPIX_Comm_failure_get_acked
-- Looking for MPIX_Comm_failure_get_acked - found
-- Looking for MPIX_Comm_shrink
-- Looking for MPIX_Comm_shrink - found
-- Looking for MPIX_Comm_agree
-- Looking for MPIX_Comm_agree - found
-- Looking for include file mpi.h
-- Looking for include file mpi.h - found
-- Looking for I_MPI_VERSION
-- Looking for I_MPI_VERSION - found

Excerpt from ctest output, where all tests passed, except:

      Start 64: image_fail_test_1
64/72 Test #64: image_fail_test_1 ......................***Failed  Required regular expression not found.Regex=[Test passed.
]  0.23 sec
      Start 65: image_fail_and_sync_test_1
65/72 Test #65: image_fail_and_sync_test_1 .............***Failed  Required regular expression not found.Regex=[Test passed.
]  0.23 sec
      Start 66: image_fail_and_sync_test_2
66/72 Test #66: image_fail_and_sync_test_2 .............***Failed  Required regular expression not found.Regex=[Test passed.
]  0.24 sec
      Start 67: image_fail_and_sync_test_3
67/72 Test #67: image_fail_and_sync_test_3 .............***Failed  Required regular expression not found.Regex=[Test passed.
]  0.22 sec
      Start 68: image_fail_and_status_test_1
68/72 Test #68: image_fail_and_status_test_1 ...........***Failed  Required regular expression not found.Regex=[Test passed.
]  0.23 sec
      Start 69: image_fail_and_failed_images_test_1
69/72 Test #69: image_fail_and_failed_images_test_1 ....***Failed  Required regular expression not found.Regex=[Test passed.
]  0.22 sec
      Start 70: image_fail_and_stopped_images_test_1
70/72 Test #70: image_fail_and_stopped_images_test_1 ...***Failed  Required regular expression not found.Regex=[Test passed.
]  0.23 sec
      Start 71: image_fail_and_get_test_1
71/72 Test #71: image_fail_and_get_test_1 ..............***Failed  Required regular expression not found.Regex=[Test passed.
]  0.23 sec


$ caf image_fail_test_1.f90

$ cafrun -n 4 ./a.exe
mpiexec.exe -n 4 --disable-auto-cleanup ./a.exe
IMAGE FAILED!

Do the "found" messages (from searching mpi.h) that turn on -DUSE_FAILED_IMAGES not mean that failed images are supported in the MPI implementation? I seem to recall that MPICH 3.2 had the functions defined in mpi.h, but also did not work with failed images.

[rouson]

@jbmaggard thanks for the thorough report! My guess is that you are correct that the Intel MPI compiler doesn't yet support what is required for failed-image detection to work. My understanding is that the features are not yet part of the MPI standard but are proposed to become part of an upcoming MPI standard (MPI 4.0?). @zbeekman will likely have more detailed feedback on the suggested solution. We'll discuss this issue in an OpenCoarrays teleconference this Friday. FYI, we are also very close to releasing OpenCoarrays 2.1.0 so that will be the first order of business and the amount of time that takes will determine what else we do.

[zbeekman]

@jbmaggard I'm still a bit flustered/confused by the need to generate an interface lib for Intel MPI. Can you please confirm that you've followed the steps on the intel website for setting up MPI on windows and then paste the output of mpicc -show and mpiexec -show (or whatever the equivalent Windows flags are).

[jbmaggard]

Discussion of native Win64 CAF implementation using GCC 7.3.0 / MSYS2 / Intel-MPI / and OpenCoarrays-2.1.0.

1. Install MSYS2. I'd read the wiki pages on introduction, installation, and using packages.
   1.a. There are three MSYS2 subsystems (msys2, mingw32, and mingw64). You'll need to know about the msys2 subsystem, and the mingw64 subsystem, where each has it's own bash prompt, but the mingw64 bash prompt has /mingw64/bin in the path before all the bin folders for msys2 packages. The msys2 packages (beginning with msys/ like output from "pacman -Ss openssh") require the /usr/bin/msys-2.0.dll. The mingw64 packages (beginning with mingw64/ like output from "pacman -Ss gcc-fortran") are Win64 native.
   1.a.i After initial MSYS2 installation and update, I added several msys2 packages (see pacman -Qe): diffutils, openssh, tar, and vim.
   1.a.ii. After initial MSYS2 installation and update, I added several mingw64 packages (see pacman -Qe): mingw-w64-x86_64-cmake, mingw-w64-x86_64-gcc-fortran, mingw-w64-x86_64-make, mingw-w64-x86_64-tools-git.
   1.b. I did not install any packages for the mingw32 subsystem.

2. Install Intel-MPI. Note the environment variable I_MPI_ROOT. In bash, the "$I_MPI_ROOT"/intel64/bin folder has impi.dll, and mpiexec.exe.
   2.a. mpi.h (located in ~/mpi/include) is from the MPI SDK, however, you only need the "free" runtime version of MPI (for impi.dll and mpiexec.exe).

3. Build an import library using gendef.exe (from tools-git), and dlltool.exe (from binutils dependency of gcc-fortran). Here are scripts that I put in my ~/mpi/lib folder:

Generate the impi.def file.

#!/bin/bash
gendef "$I_MPI_ROOT"/intel64/bin/impi.dll

Generate the import library (libmpi.a) to impi.dll

#!/bin/bash
dlltool -l libmpi.a -d impi.def -D "$I_MPI_ROOT"/intel64/bin/impi.dll

4. Set up mpifort (and mpigfortran), mpicc (and mpigcc), and mpirun (and mpiexec) bash scripts that can work with FindMPI.cmake and install.sh. I took the mpifort and mpicc bash scripts from MPICH-3.2, modified to assume I have ~/mpi/bin, ~/mpi/lib, ~/mpi/include folders tree, and fix rpath_flag.

setcaf.sh (I_MPI_ROOT is unset after setting PATH to help with FindMPI.cmake)

$ cat ~/setcaf.sh
#!/bin/bash
export MPI_HOME="$HOME"/mpi
export PATH="$HOME"/caf/bin:"$MPI_HOME"/bin:"$I_MPI_ROOT"/intel64/bin:"$PATH"
unset I_MPI_ROOT
echo $PATH
echo $MPI_HOME

For mpifort, since I don't know how to get the expanding ellipsis that hides huge quotations, I posted here, and show diffs below. The diffs (+) show the final version, and where I had to make changes.

$ diff -Nu mpifort.orig mpifort
--- mpifort.orig        2018-06-05 13:21:58.174714100 -0500
+++ mpifort     2018-05-21 12:20:59.923588500 -0500
@@ -28,12 +28,12 @@
 #
 # Directory locations: Fixed for any MPI implementation.
 # Set from the directory arguments to configure (e.g., --prefix=/usr/local)
-prefix=/c/Intel-MPI
-exec_prefix=/c/Intel-MPI
-sysconfdir=/c/Intel-MPI/etc
-includedir=/c/Intel-MPI/include
-libdir=/c/Intel-MPI/lib
-modincdir=${prefix}/include
+prefix="$MPI_HOME"
+exec_prefix="$MPI_HOME"
+sysconfdir="$MPI_HOME"/etc
+includedir="$MPI_HOME"/include
+libdir="$MPI_HOME"/lib
+modincdir="$MPI_HOME"/include

 # Default settings for compiler, flags, and libraries
 # Determined by a combination of environment variables and tests within
@@ -74,8 +74,7 @@

 # Flag to hardcode $libdir into a binary during linking.
 # This must work even if $libdir does not exist.
-#hardcode_libdir_flag_spec="\${wl}-rpath \${wl}\$libdir -Wl,--enable-new-dtags"
-hardcode_libdir_flag_spec="\${wl}-rpath \${wl}\$libdir"
+hardcode_libdir_flag_spec="\${wl}-rpath \${wl}\$libdir -Wl,--enable-new-dtags"

 # Whether we need a single -rpath flag with a separated argument.
 hardcode_libdir_separator=""
@@ -362,17 +361,17 @@
     else
        rpath_flags=""
     fi
+eval rpath_flags='-Wl,"-rpath,\"$libdir\""'

     if [ "$nativelinking" = yes ] ; then
         $Show $FC $PROFILE_INCPATHS ${final_fcflags} ${final_ldflags} "${allargs[@]}"
         rc=$?
     else
-#        $Show $FC $PROFILE_INCPATHS ${final_fcflags} ${final_ldflags} "${allargs[@]}" $FCINCDIRS $FCMODDIRS -L$libdir -lmpifort $PROFILE_PRELIB $PROFILE_FOO $rpath_flags -lmpi  $PROFILE_POSTLIB ${final_libs}
-        $Show $FC $PROFILE_INCPATHS ${final_fcflags} ${final_ldflags} "${allargs[@]}" $FCINCDIRS $FCMODDIRS -L$libdir $PROFILE_PRELIB $PROFILE_FOO $rpath_flags -lmpi  $PROFILE_POSTLIB ${final_libs}
+        $Show $FC $PROFILE_INCPATHS ${final_fcflags} ${final_ldflags} "${allargs[@]}" "$FCINCDIRS" -L"$libdir" $PROFILE_PRELIB $PROFILE_FOO "$rpath_flags" -lmpi  $PROFILE_POSTLIB ${final_libs}
         rc=$?
     fi
 else
-    $Show $FC $PROFILE_INCPATHS ${final_fcflags} "${allargs[@]}" $FCINCDIRS $FCMODDIRS
+    $Show $FC $PROFILE_INCPATHS ${final_fcflags} "${allargs[@]}" "$FCINCDIRS"
     rc=$?
 fi
 if [ -n "$rmfiles" ] ; then

For mpicc, since I don't know how to get the expanding ellipsis that hides huge quotations, I posted here, and show diffs below. The diffs (+) show the final version, and where I had to make changes.

$ diff -Nu mpicc.orig mpicc
--- mpicc.orig  2018-06-05 13:23:55.457514100 -0500
+++ mpicc       2018-05-21 12:24:37.322139700 -0500
@@ -27,11 +27,11 @@
 #
 # Directory locations: Fixed for any MPI implementation.
 # Set from the directory arguments to configure (e.g., --prefix=/usr/local)
-prefix=/c/Intel-MPI
-exec_prefix=/c/Intel-MPI
-sysconfdir=/c/Intel-MPI/etc
-includedir=/c/Intel-MPI/include
-libdir=/c/Intel-MPI/lib
+prefix="$MPI_HOME"
+exec_prefix="$MPI_HOME"
+sysconfdir="$MPI_HOME"/etc
+includedir="$MPI_HOME"/include
+libdir="$MPI_HOME"/lib

 # Default settings for compiler, flags, and libraries.
 # Determined by a combination of environment variables and tests within
@@ -58,8 +58,7 @@

 # Flag to hardcode $libdir into a binary during linking.
 # This must work even if $libdir does not exist.
-#hardcode_libdir_flag_spec="\${wl}-rpath \${wl}\$libdir -Wl,--enable-new-dtags"
-hardcode_libdir_flag_spec="\${wl}-rpath \${wl}\$libdir"
+hardcode_libdir_flag_spec="\${wl}-rpath \${wl}\$libdir -Wl,--enable-new-dtags"

 # Whether we need a single -rpath flag with a separated argument.
 hardcode_libdir_separator=""
@@ -271,16 +270,16 @@
     else
        rpath_flags=""
     fi
-
+eval rpath_flags='-Wl,"-rpath,\"$libdir\""'
     if [ "$nativelinking" = yes ] ; then
-        $Show $CC ${final_cppflags} $PROFILE_INCPATHS ${final_cflags} ${final_ldflags} "${allargs[@]}" -I$includedir
+        $Show $CC ${final_cppflags} $PROFILE_INCPATHS ${final_cflags} ${final_ldflags} "${allargs[@]}" -I"$includedir"
         rc=$?
     else
-        $Show $CC ${final_cppflags} $PROFILE_INCPATHS ${final_cflags} ${final_ldflags} "${allargs[@]}" -I$includedir -L$libdir $PROFILE_PRELIB $PROFILE_FOO $rpath_flags -lmpi  $PROFILE_POSTLIB ${final_libs}
+        $Show $CC ${final_cppflags} $PROFILE_INCPATHS ${final_cflags} ${final_ldflags} "${allargs[@]}" -I"$includedir" -L"$libdir" $PROFILE_PRELIB $PROFILE_FOO "$rpath_flags" -lmpi  $PROFILE_POSTLIB ${final_libs}
         rc=$?
     fi
 else
-    $Show $CC ${final_cppflags} $PROFILE_INCPATHS ${final_cflags} "${allargs[@]}" -I$includedir
+    $Show $CC ${final_cppflags} $PROFILE_INCPATHS ${final_cflags} "${allargs[@]}" -I"$includedir"
     rc=$?
 fi

mpirun is simply:

#! /bin/bash
echo mpiexec.exe "$@"
mpiexec.exe "$@"

At this point, I have bash scripts for mpifort, mpicc, and mpirun that work like on Linux. I took the simple MPI test programs from OpenCoarrays-1.9.2

$ gfortran --version
GNU Fortran (Rev2, Built by MSYS2 project) 7.3.0
Copyright (C) 2017 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.


$ mpifort -show
gfortran -I/home/maggard.admin/mpi/include -L/home/maggard.admin/mpi/lib -Wl,-rpath,"/home/maggard.admin/mpi/lib" -lmpi

$ mpifort mpi_hello.f90

$ mpirun --version
mpiexec.exe --version
Intel(R) MPI Library for Windows* OS, Version 2018 Update 2 Build 20180125
Copyright 2003-2018 Intel Corporation.

$ mpirun -n 8 ./a.exe
mpiexec.exe -n 8 ./a.exe
 Hello from processor LAPTOP rank            6  out of            8 .
 Hello from processor LAPTOP rank            0  out of            8 .
 Hello from processor LAPTOP rank            1  out of            8 .
 Hello from processor LAPTOP rank            2  out of            8 .
 Hello from processor LAPTOP rank            3  out of            8 .
 Hello from processor LAPTOP rank            4  out of            8 .
 Hello from processor LAPTOP rank            5  out of            8 .
 Hello from processor LAPTOP rank            7  out of            8 .

$ cat mpi_hello.f90
!
! Taken from OpenCoarrays-1.9.2/CMakeLists.txt, lines 378-389.
!
program mpi_hello
implicit none
include 'mpif.h'
integer :: ierr, mpi_world_size, mpi_world_rank, res_len
character*(MPI_MAX_PROCESSOR_NAME) :: proc
call mpi_init(ierr)
call mpi_comm_size(MPI_COMM_WORLD,mpi_world_size,ierr)
call mpi_comm_rank(MPI_COMM_WORLD,mpi_world_rank,ierr)
call mpi_get_processor_name(proc,res_len,ierr)
write(*,*) 'Hello from processor ', trim(proc), ' rank ', mpi_world_rank, ' out of ', mpi_world_size, '.'
call mpi_finalize(ierr)
end program

$ gcc --version
gcc.exe (Rev2, Built by MSYS2 project) 7.3.0
Copyright (C) 2017 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.


$ mpicc -show
gcc -I/home/maggard.admin/mpi/include -L/home/maggard.admin/mpi/lib -Wl,-rpath,"/home/maggard.admin/mpi/lib" -lmpi

$ mpicc mpi_hello.c

$ mpirun -n 8 ./a.exe
mpiexec.exe -n 8 ./a.exe
Hello world from processor LAPTOP, rank 2 out of 8 processors
Hello world from processor LAPTOP, rank 4 out of 8 processors
Hello world from processor LAPTOP, rank 1 out of 8 processors
Hello world from processor LAPTOP, rank 5 out of 8 processors
Hello world from processor LAPTOP, rank 6 out of 8 processors
Hello world from processor LAPTOP, rank 0 out of 8 processors
Hello world from processor LAPTOP, rank 7 out of 8 processors
Hello world from processor LAPTOP, rank 3 out of 8 processors

$ cat mpi_hello.c
/*
 Taken from OpenCoarrays-1.9.2/CMakeLists.txt, lines 303-317.
*/
#include <mpi.h>
#include <stdio.h>
int main(int argc, char** argv) {
  MPI_Init(NULL, NULL);
  int world_size;
  MPI_Comm_size(MPI_COMM_WORLD, &world_size);
  int world_rank;
  MPI_Comm_rank(MPI_COMM_WORLD, &world_rank);
  char processor_name[MPI_MAX_PROCESSOR_NAME];
  int name_len;
  MPI_Get_processor_name(processor_name, &name_len);
  printf("Hello world from processor %s, rank %d out of %d processors\n",
         processor_name, world_rank, world_size);
  MPI_Finalize();
}

After testing the MPI installation, I was ready to go on to CAF. I had to edit FindMPI.cmake as discussed in the first comment of this issue. I also had to edit build_opencoarrays.sh as discussed in the first comment of this issue.

In a mingw64 bash prompt from the ~/OpenCoarrays-2.1.0 folder I executed
./install.sh -i ~/caf -M ~/mpi

I posted a huge text file, here, covering install.sh output, make clean; make VERBOSE=1 output, and ctest output. I have not yet had time to investigate why ctest 75 failed, as I previously had success, as discussed in #550.

Ignore the warnings about maggard.a; FindMPI.cmake apparently interprets the $HOME folder as a library, because it has ".a" in it. These warnings would probably not appear if the folder path doesn't have ".a" in it (please confirm).

I am willing to bring in the three big files (mpifort, mpicc, and output.txt) if you want me to, but I should probably learn how to use the expanding ... first.

One hidden problem I have had is if you are trying to execute mpiexec.exe from ctest and you have not authenticated yet, you don't get any useful output. Be sure to do a manual mpirun (or cafrun) first.

[jbmaggard]

So far, my limited bash scripting skills have not resulted in being able to obtain the -G"MSYS Makefiles" option to the cmake command line, as shown with a manual addition in the initial comment above (diff of build_opencoarrays.sh). I did notice a similar handling of double quotes in the mpifort script of MPICH-3.2, using eval. I'll try to take a look at that as a possibility for a proposed fix.

[zbeekman]

@jbmaggard I think I have a pretty decent idea of how to use introspection in install.sh to add -G"generator-name" for msys2. I hope to take a quick look at it soon, so don't stress too much over this. Still my biggest concern is finding an automated way to generate the Intel interface library (or find where intel has stashed it).

[jbmaggard]

The .lib files provided by Intel-MPI are for Intel Fortran and the Microsoft linker from Visual Studio. The libmpi.a import library discussed here was created with gendef and dlltool to allow gfortran to link to impi.dll.

Similarly the .mod files provided by Intel-MPI (SDK) are for Intel Fortran, which is why decided to work with Intel's provided mpi.h.

So, if somone has Intel-MPI and MSYS2 (with needed packages) installed, you'd like cmake to generate the libmpi.a file? I'm sure cmake could do this, based on the WIN32 and MSYS cmake variables.

[zbeekman]

Yes I can probably look for the executables to do it. I had forgotten the details you just re-shared.

…

On Thu, Jun 7, 2018 at 4:44 PM jbmaggard ***@***.***> wrote: The .lib files provided by Intel-MPI are for Intel-Fortran and the Microsoft linker from Visual Studio. The libmpi.a import library discussed here was created with gendef and dlltool to allow gfortran to link to impi.dll. So, if somone has Intel-MPI and MSYS2 (with needed packages) installed, you'd like cmake to generate the libmpi.a file? I'm sure cmake could do this, based on the WIN32 and MSYS cmake variables. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#541 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAREPLhKxZUjLURGdFPYtXhiPHhRZOBOks5t6ZC1gaJpZM4USBmy> .

[jbmaggard]

Look for the executables in MSYS2? gendef.exe is in the above installation notes, in the mingw-w64-x86_64-tools-git package, and dlltool.exe is in the previously mentioned mingw-w64-x86_64-binutils package. binutils would install as a dependency of the mingw-w64-x86_64-gcc-fortran package, and you would do "pacman -S mingw-w64-x86_64-tools-git" to install it.

If you are looking for an executable, doing "pacman -Fy" updates the packages database, then "pacman -Fs gendef.exe" tells you which package has it. Similarly for "pacman -Fs dlltool.exe". You can use "pacman -Ss mingw-w64-x86_64-tools-git | grep "[installed]" to confirm installation of the package.

[jbmaggard]

Here is an example bash script that appends the cmake generator to the cmake options variable.

#!/bin/bash -xv
CMAKE_OPTIONS="-Bbuild -Hsrc -DCMAKE_MAKE_PROGRAM=/mingw64/bin/mingw32-make.exe -DMPIEXEC_EXECUTABLE="$MPI_HOME"/bin/mpirun -DMPI_Fortran_COMPILER=mpifort -DMPI_Fortran_LIBRARIES="$MPI_HOME"/lib/libmpi.a -DMPI_Fortran_ADDITIONAL_INCLUDE_DIRS="$MPI_HOME"/include"
eval CMAKE_OPTIONS='${CMAKE_OPTIONS}\ -G\"MSYS\ Makefiles\"'
eval cmake "${CMAKE_OPTIONS}"

The mingw-w64-x86_64-make package installs the make utility executable as mingw32-make.exe. You have to use the -DCMAKE_MAKE_PROGRAM, or "ln -s /mingw64/bin/mingw32-make.exe /mingw64/bin/make.exe" (note that ln -s is actually just a file copy for MSYS2).

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs.