fortran2008: fix mpiext example #3617

ggouaillardet · 2017-05-31T07:53:32Z

in order to solve an egg and the chicken problem, in which mpiext need mpi-f08-types.mod
and/but use-mpi-f08[-desc] needs mpiext, add an extra step

build fortran 2008 modules only
build fortran 2008 mpi extensions
and then build fortran 2008 bindings

Fixes #3605

Signed-off-by: Gilles Gouaillardet gilles@rist.or.jp

ggouaillardet · 2017-05-31T07:55:09Z

@jsquyres could you please review this before it lands into master ?

btw, can you point me to a compiler that will use use-mpi-f08-desc ?
the comments suggest intel compilers can do that, but nor ifort 2017 nor 2018 beta are able to do it.

ggouaillardet · 2017-05-31T07:55:51Z

just to be clear, i blindingly updated use-mpi-f08-desc since i am unable to test it.

ibm-ompi · 2017-05-31T08:05:12Z

The IBM CI (PGI Compiler) build failed! Please review the log, linked below.

Gist: https://gist.github.com/b263311298475efed1441e9f6fde6e04

ibm-ompi · 2017-05-31T08:06:03Z

The IBM CI (XL Compiler) build failed! Please review the log, linked below.

Gist: https://gist.github.com/511b9c2765392e4c916c9b06a914a328

ibm-ompi · 2017-05-31T08:06:29Z

The IBM CI (GNU Compiler) build failed! Please review the log, linked below.

Gist: https://gist.github.com/fbc67eac58ea60a3c8bea7c80a21cf6d

t-kurita · 2017-08-08T12:20:57Z

I confirmed use-mpi-f08-desc can be compiled successfully with gfortran 5.3.0 and the --enable-mpi-f08-subarray-prototype option.
When compiling, I applied the following patch and changed the extension of a file.
(ompi/mpi/fortran/use-mpi-f08-desc/mod/mpi-f08-types.f90 -> ompi/mpi/fortran/use-mpi-f08-desc/mod/mpi-f08-types.F90)

diff --git a/config/ompi_configure_options.m4 b/config/ompi_configure_options.m4
index 3301df0..5bde2b2 100644
--- a/config/ompi_configure_options.m4
+++ b/config/ompi_configure_options.m4
@@ -247,8 +247,7 @@ AS_IF([test $OMPI_TRY_FORTRAN_BINDINGS -lt $OMPI_FORTRAN_USEMPIF08_BINDINGS],
              [OMPI_BUILD_FORTRAN_F08_SUBARRAYS=1
               AC_MSG_RESULT([extra crispy (subarray prototype)])
               AC_MSG_WARN([Sorry, the subarray prototype is no longer available])
-              AC_MSG_WARN([Contact your favorite OMPI developer and ask for it to be re-enabled])
-              AC_MSG_ERROR([Cannot continue])],
+              AC_MSG_WARN([Contact your favorite OMPI developer and ask for it to be re-enabled])]
              [AC_MSG_RESULT([regular (no subarray support)])])
       ])
 AC_DEFINE_UNQUOTED([OMPI_BUILD_FORTRAN_F08_SUBARRAYS],
diff --git a/config/ompi_ext.m4 b/config/ompi_ext.m4
index ae0e91b..1054a74 100644
--- a/config/ompi_ext.m4
+++ b/config/ompi_ext.m4
@@ -216,8 +216,7 @@ EOF
 
     # Only build this mpi_f08_ext module if we're building the "use
     # mpi_f08" module *and* it's the non-descriptor one.
-    AS_IF([test $OMPI_BUILD_FORTRAN_BINDINGS -ge $OMPI_FORTRAN_USEMPIF08_BINDINGS && \
-           test $OMPI_BUILD_FORTRAN_F08_SUBARRAYS -eq 0],
+    AS_IF([test $OMPI_BUILD_FORTRAN_BINDINGS -ge $OMPI_FORTRAN_USEMPIF08_BINDINGS],
           [OMPI_BUILD_FORTRAN_USEMPIF08_EXT=1],
           [OMPI_BUILD_FORTRAN_USEMPIF08_EXT=0])
     AM_CONDITIONAL(OMPI_BUILD_FORTRAN_USEMPIF08_EXT,
diff --git a/ompi/mpi/fortran/use-mpi-f08-desc/Makefile.am b/ompi/mpi/fortran/use-mpi-f08-desc/Makefile.am
index f7d1275..559012f 100644
--- a/ompi/mpi/fortran/use-mpi-f08-desc/Makefile.am
+++ b/ompi/mpi/fortran/use-mpi-f08-desc/Makefile.am
@@ -59,7 +59,7 @@ MOSTLYCLEANFILES = *.mod
 # manually here.  Bummer!
 #
 
-OMPI_Fortran_binding.lo: OMPI_Fortran_binding.f90 mpi-f08-types.lo
+OMPI_Fortran_binding.lo: OMPI_Fortran_binding.f90 mod/mpi-f08-types.lo
 
 
 #
diff --git a/ompi/mpi/fortran/use-mpi-f08-desc/mod/Makefile.am b/ompi/mpi/fortran/use-mpi-f08-desc/mod/Makefile.am
index ec5a76f..2e47016 100644
--- a/ompi/mpi/fortran/use-mpi-f08-desc/mod/Makefile.am
+++ b/ompi/mpi/fortran/use-mpi-f08-desc/mod/Makefile.am
@@ -27,7 +27,7 @@ noinst_LTLIBRARIES = $(module_sentinel_file)
 # f08 support modules
 
 libforce_usempif08_internal_modules_to_be_built_la_SOURCES = \
-        mpi-f08-types.f90 \
+        mpi-f08-types.F90 \
         mpi-f08-interfaces.F90
 
 #
@@ -41,7 +41,7 @@ MOSTLYCLEANFILES = *.mod
 # manually here.  Bummer!
 #
 
-mpi-f08-types.lo: mpi-f08-types.f90
+mpi-f08-types.lo: mpi-f08-types.F90
 mpi-f08-interfaces.lo: mpi-f08-interfaces.F90 mpi-f08-types.lo

Will it be merged to the master?

ggouaillardet · 2017-08-08T12:33:29Z

first thing first, @jsquyres can you please review this PR ?

@t-kurita do you really need the use-mpi-f08-desc interface ?
(e.g. is the use-mpi-f08 not good enough for you ?)

Jeff, do you have any recollection on why we

Only build this mpi_f08_ext module if we're building the "use
mpi_f08" module and it's the non-descriptor one.

regarding the inline patch, at first glance, i'd rather add yet an other configure option in order to enable the subarray prototype.
out of curiosity, why did you change the extension of mpi-f08-types.f90 to .F90 ?

kawashima-fj · 2017-08-08T12:43:34Z

@t-kurita is my colleague. I'll explain.

In Open MPI master (and release branches), use-mpi-f08-desc is disabled currently because the implementation is incomplete. In @t-kurita's patch, the purpose of changes in ompi_configure_options.m4 and ompi_ext.m4 is temporarily enabling use-mpi-f08-desc to confirm @ggouaillardet's patch. For this PR, only changes in Makefile.am and the filename change are needed.

mpi-f08-types.f90 has a #include preprocessor directive (instead of include Fortran statement). So it need to be handled by the preprocessor. Otherwise, compilation will fail.

jsquyres · 2017-08-08T14:17:58Z

Let me ask a different question: should we delete the descriptor-based mpi_f08 module?

It has sat dormant for years, and I certainly have no cycles to develop it further. Does anyone else? If not, we should probably delete it (it'll always be in git history if someone wants to dig it out / continue the work).

in order to solve an egg and the chicken problem, in which mpiext need mpi-f08-types.mod and/but use-mpi-f08[-desc] needs mpiext, add an extra step - build fortran 2008 modules only - build fortran 2008 mpi extensions - and then build fortran 2008 bindings Fixes open-mpi#3605 Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>

Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>

so Open MPI can be configure'd with --enable-mpi-f08-subarray-prototype Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>

ggouaillardet · 2017-08-09T05:43:53Z

@jsquyres i pushed two commits into master, so this PR is now really only for the last one
(i do not know why github still thinks there are 3 commits in this PR)

@t-kurita @kawashima-fj just to be clear, is the inline patch something you need ?
or is this only here as a proof of concept to help me fix this PR ?
i pushed the fixes, but did not re-enable the --enable-mpi-f08-subarray-prototype option.

i have no strong opinion whether we should (or not) remove use-mpi-f08-desc support

kawashima-fj · 2017-08-09T05:52:33Z

@ggouaillardet Thanks for your work. We Fujitsu don't need the inline patch and we also have no strong option about use-mpi-f08-desc support. The inline patch is created for your comment:

just to be clear, i blindingly updated use-mpi-f08-desc since i am unable to test it.

jsquyres · 2017-08-09T15:37:00Z

If no one has any strong feelings, then we should remove the use-mpi-f08-desc directory/support.

jsquyres · 2017-08-10T13:15:39Z

Per #4068, let's merge this PR and then #4068 will entirely remove the proof of concept descriptor-based mpi_f08 module.

ggouaillardet requested a review from jsquyres May 31, 2017 07:53

ggouaillardet added the bug label May 31, 2017

ggouaillardet force-pushed the topic/f08_mpiext branch from 04d2ee7 to 4393ab6 Compare May 31, 2017 08:13

ggouaillardet mentioned this pull request Jun 2, 2017

MPI extension using use-mpi-f08 build fails with "Can't open module" error #3605

Closed

jsquyres approved these changes Aug 8, 2017

View reviewed changes

ggouaillardet added 3 commits August 9, 2017 13:19

fortran/use-mpi-f08-desc: add a missing include file

dfe7b2b

Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>

configury: enable Fortran 2008 subarray prototype

344f000

so Open MPI can be configure'd with --enable-mpi-f08-subarray-prototype Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>

ggouaillardet force-pushed the topic/f08_mpiext branch 2 times, most recently from 344f000 to 2c71c27 Compare August 9, 2017 04:44

jsquyres mentioned this pull request Aug 10, 2017

ompi/fortran: remove proof-of-concept mpi_f08 module #4068

Merged

jsquyres approved these changes Aug 10, 2017

View reviewed changes

jsquyres merged commit 1a70e5b into open-mpi:master Aug 10, 2017

fortran2008: fix mpiext example #3617

fortran2008: fix mpiext example #3617

Uh oh!

Conversation

ggouaillardet commented May 31, 2017

Uh oh!

ggouaillardet commented May 31, 2017

Uh oh!

ggouaillardet commented May 31, 2017

Uh oh!

ibm-ompi commented May 31, 2017

Uh oh!

ibm-ompi commented May 31, 2017

Uh oh!

ibm-ompi commented May 31, 2017

Uh oh!

t-kurita commented Aug 8, 2017

Uh oh!

ggouaillardet commented Aug 8, 2017

Uh oh!

kawashima-fj commented Aug 8, 2017

Uh oh!

jsquyres commented Aug 8, 2017

Uh oh!

ggouaillardet commented Aug 9, 2017

Uh oh!

kawashima-fj commented Aug 9, 2017

Uh oh!

jsquyres commented Aug 9, 2017

Uh oh!

jsquyres commented Aug 10, 2017

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants