Re: [OMPI devel] Announcing Open MPI v5.0.0rc2
On 01.01.2022 20:07, Barrett, Brian wrote:
Marco -
There are some patches that haven't made it to the 5.0 branch to make this
behavior better. I didn't get a chance to back port them before the holiday
break, but they will be in the next RC. That said, the issue below is a
warning, not an error, so you should still end up with a build that works (with
an included PMIx). The issue is that png-config can't be found, so we have
trouble guessing what libraries are dependencies of PMIx, which is a potential
problem in complicated builds with static libraries.
Brian
Thanks Brian,
the build error was in reality in setting for threads.
I was using up to v4.1
--with-threads=posix
that currently is not accepted anymore but no error is reported,
causing a different setting that does not work in CYGWIN.
Removing the configuration seems to work
I have however found a logic error in prrte that
probably need a verification of all the HAVE_*_H
between configuration and code
/pub/devel/openmpi/v5.0/openmpi-5.0.0-0.1.x86_64/src/openmpi-5.0.0rc2/3rd-party/prrte/src/mca/odls/default/odls_default_module.c:114:14:
fatal error: sys/ptrace.h: No such file or directory
114 | #include
| ^~
caused by
$ grep -rH HAVE_SYS_PTRACE_H .
./3rd-party/prrte/config.log:| #define HAVE_SYS_PTRACE_H 0
./3rd-party/prrte/config.log:| #define HAVE_SYS_PTRACE_H 0
./3rd-party/prrte/config.log:#define HAVE_SYS_PTRACE_H 0
./3rd-party/prrte/config.status:D["HAVE_SYS_PTRACE_H"]=" 0"
./3rd-party/prrte/src/include/prte_config.h:#define HAVE_SYS_PTRACE_H 0
while the code in
3rd-party/prrte/src/mca/odls/default/odls_default_module.c
has
#ifdef HAVE_SYS_PTRACE_H
#include
#endif
currently I am stacked at
0rc2/3rd-party/prrte/src/mca/oob/tcp/oob_tcp_connection.c:61:
/pub/devel/openmpi/v5.0/openmpi-5.0.0-0.1.x86_64/src/openmpi-5.0.0rc2/3rd-party/prrte/s
rc/mca/oob/tcp/oob_tcp_connection.c: In function
‘prte_oob_tcp_peer_try_connect’:
/pub/devel/openmpi/v5.0/openmpi-5.0.0-0.1.x86_64/src/openmpi-5.0.0rc2/3rd-party/prrte/src/mca/oob/tcp/oob_tcp_connection.c:163:16:
error: expected identifier or ‘(’ before ‘struct’
163 | prte_if_t *interface;
|^
/pub/devel/openmpi/v5.0/openmpi-5.0.0-0.1.x86_64/src/openmpi-5.0.0rc2/3rd-party/prrte/src/mca/oob/tcp/oob_tcp_connection.c:180:19:
error: expected ‘{’ before ‘=’ token
180 | interface = PRTE_NEW(prte_if_t);
| ^
not sure if it is caused by new GCC 11 requirement or from wrong headers
being pulled in.
Has anyone built with GCC 11 ?
Regards
Marco
Re: [OMPI devel] Announcing Open MPI v5.0.0rc2
Hi Marco
Try the patch here (for the prrte 3rd-party subdirectory):
https://github.com/openpmix/prrte/pull/1173
Ralph
> On Jan 9, 2022, at 12:29 AM, Marco Atzeri via devel
> wrote:
>
> On 01.01.2022 20:07, Barrett, Brian wrote:
>> Marco -
>> There are some patches that haven't made it to the 5.0 branch to make this
>> behavior better. I didn't get a chance to back port them before the holiday
>> break, but they will be in the next RC. That said, the issue below is a
>> warning, not an error, so you should still end up with a build that works
>> (with an included PMIx). The issue is that png-config can't be found, so we
>> have trouble guessing what libraries are dependencies of PMIx, which is a
>> potential problem in complicated builds with static libraries.
>> Brian
>
> Thanks Brian,
>
> the build error was in reality in setting for threads.
>
> I was using up to v4.1
>
> --with-threads=posix
>
> that currently is not accepted anymore but no error is reported,
> causing a different setting that does not work in CYGWIN.
> Removing the configuration seems to work
>
>
> I have however found a logic error in prrte that
> probably need a verification of all the HAVE_*_H
> between configuration and code
>
>
> /pub/devel/openmpi/v5.0/openmpi-5.0.0-0.1.x86_64/src/openmpi-5.0.0rc2/3rd-party/prrte/src/mca/odls/default/odls_default_module.c:114:14:
> fatal error: sys/ptrace.h: No such file or directory
> 114 | #include
> | ^~
>
> caused by
>
> $ grep -rH HAVE_SYS_PTRACE_H .
> ./3rd-party/prrte/config.log:| #define HAVE_SYS_PTRACE_H 0
> ./3rd-party/prrte/config.log:| #define HAVE_SYS_PTRACE_H 0
> ./3rd-party/prrte/config.log:#define HAVE_SYS_PTRACE_H 0
> ./3rd-party/prrte/config.status:D["HAVE_SYS_PTRACE_H"]=" 0"
> ./3rd-party/prrte/src/include/prte_config.h:#define HAVE_SYS_PTRACE_H 0
>
> while the code in
>3rd-party/prrte/src/mca/odls/default/odls_default_module.c
> has
>
> #ifdef HAVE_SYS_PTRACE_H
> #include
> #endif
>
>
> currently I am stacked at
>
> 0rc2/3rd-party/prrte/src/mca/oob/tcp/oob_tcp_connection.c:61:
> /pub/devel/openmpi/v5.0/openmpi-5.0.0-0.1.x86_64/src/openmpi-5.0.0rc2/3rd-party/prrte/s
> rc/mca/oob/tcp/oob_tcp_connection.c: In function
> ‘prte_oob_tcp_peer_try_connect’:
> /pub/devel/openmpi/v5.0/openmpi-5.0.0-0.1.x86_64/src/openmpi-5.0.0rc2/3rd-party/prrte/src/mca/oob/tcp/oob_tcp_connection.c:163:16:
> error: expected identifier or ‘(’ before ‘struct’
> 163 | prte_if_t *interface;
> |^
> /pub/devel/openmpi/v5.0/openmpi-5.0.0-0.1.x86_64/src/openmpi-5.0.0rc2/3rd-party/prrte/src/mca/oob/tcp/oob_tcp_connection.c:180:19:
> error: expected ‘{’ before ‘=’ token
> 180 | interface = PRTE_NEW(prte_if_t);
> | ^
>
>
> not sure if it is caused by new GCC 11 requirement or from wrong headers
> being pulled in.
>
> Has anyone built with GCC 11 ?
>
> Regards
> Marco
>
Re: [OMPI devel] Announcing Open MPI v5.0.0rc2
On 10.01.2022 06:50, Marco Atzeri wrote: On 09.01.2022 15:54, Ralph Castain via devel wrote: Hi Marco Try the patch here (for the prrte 3rd-party subdirectory): https://github.com/openpmix/prrte/pull/1173 Ralph Thanks Ralph, I will do on the next build as I need still to test the current build. The test are not satisfactory I have only one test fail FAIL: dlopen_test.exe that I supect is due to a wrong name on test but a simple run fails $ mpirun -n 4 ./hello_c.exe [116] /pub/devel/openmpi/v5.0/openmpi-5.0.0-0.1.x86_64/src/openmpi-5.0.0rc2/3rd-party/openpmix/src/mca/ptl/base/ptl_base_listener.c:498 bind() failed for socket 13 storage size 16: Cannot assign requested address Hello, world, I am 0 of 1, (Open MPI v5.0.0rc2, package: Open MPI Marco@LAPTOP-82F08ILC Distribution, ident: 5.0.0rc2, repo rev: v5.0.0rc2, Oct 18, 2021, 125) -- 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): PML add procs failed --> Returned "Not found" (-13) instead of "Success" (0) -- [LAPTOP-82F08ILC:0] *** An error occurred in MPI_Init [LAPTOP-82F08ILC:0] *** reported by process [36547002369,1] [LAPTOP-82F08ILC:0] *** on a NULL communicator [LAPTOP-82F08ILC:0] *** Unknown error [LAPTOP-82F08ILC:0] *** MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort, [LAPTOP-82F08ILC:0] ***and MPI will try to terminate your MPI job as well) -- Suggestion for what to look for ? Regards Marco
Re: [OMPI devel] Announcing Open MPI v5.0.0rc2
On 09.01.2022 15:54, Ralph Castain via devel wrote:
Hi Marco
Try the patch here (for the prrte 3rd-party subdirectory):
https://github.com/openpmix/prrte/pull/1173
Ralph
Thanks Ralph,
I will do on the next build
as I need still to test the current build.
To complete the build I also needed the attached:
5.0.0rc2-amend-DEF.patch
to correct a missing def in OPAL
CYGWIN-undefined.patch
to pass "-no-undefined" to pmix and romio to allow shared lib
CYGWIN-interface-workaround.patch
temporary workaround to avoid the "interface" collision with
Windows headers
I will look to provide a better solution for this point.
Regards
Marco
--- origsrc/openmpi-5.0.0rc2/opal/util/minmax.h 2021-10-18 17:27:42.0
+0200
+++ src/openmpi-5.0.0rc2/opal/util/minmax.h 2022-01-09 12:46:07.148969800
+0100
@@ -40,7 +40,7 @@ OPAL_DEFINE_MINMAX(float, float)
OPAL_DEFINE_MINMAX(double, double)
OPAL_DEFINE_MINMAX(void *, ptr)
-#if OPAL_C_HAVE__GENERIC
+#ifdef OPAL_C_HAVE__GENERIC
#define opal_min(a, b) \
(_Generic((a) + (b),\
int8_t: opal_min_8,\
---
origsrc/openmpi-5.0.0rc2/3rd-party/prrte/src/mca/oob/tcp/oob_tcp_connection.c
2021-10-18 17:28:05.0 +0200
+++ src/openmpi-5.0.0rc2/3rd-party/prrte/src/mca/oob/tcp/oob_tcp_connection.c
2022-01-09 11:13:29.687212300 +0100
@@ -160,6 +160,9 @@ void prte_oob_tcp_peer_try_connect(int f
prte_oob_tcp_peer_t *peer;
prte_oob_tcp_addr_t *addr;
bool connected = false;
+#if defined interface
+# undef interface
+#endif
prte_if_t *interface;
char *host;
--- origsrc/openmpi-5.0.0rc2/opal/mca/btl/tcp/btl_tcp_proc.c2021-10-18
17:27:42.0 +0200
+++ src/openmpi-5.0.0rc2/opal/mca/btl/tcp/btl_tcp_proc.c2022-01-09
12:48:27.075225600 +0100
@@ -160,6 +160,9 @@ static int mca_btl_tcp_proc_create_inter
fields needed in the proc version */
for (i = 0; i < btl_proc->proc_addr_count; i++) {
/* Construct opal_if_t objects for the remote interfaces */
+#ifdef interface
+# undef interface
+#endif
opal_if_t *interface = OBJ_NEW(opal_if_t);
if (NULL == interface) {
rc = OPAL_ERR_OUT_OF_RESOURCE;
--- origsrc/openmpi-5.0.0rc2/3rd-party/openpmix/configure.ac2021-10-18
17:28:09.0 +0200
+++ src/openmpi-5.0.0rc2/3rd-party/openpmix/configure.ac2022-01-09
04:51:14.271907200 +0100
@@ -337,6 +337,23 @@ AC_ARG_ENABLE(werror,
])
+# no-undefined needed on some platform for shared lib
+
+
+AC_MSG_CHECKING([if libtool needs -no-undefined flag to build shared
libraries])
+case "`uname`" in
+ CYGWIN*|MINGW*|AIX*)
+## Add in the -no-undefined flag to LDFLAGS for libtool.
+AC_MSG_RESULT([yes])
+LDFLAGS="$LDFLAGS -no-undefined"
+;;
+ *)
+## Don't add in anything.
+AC_MSG_RESULT([no])
+;;
+esac
+
+
# Version information
--- origsrc/openmpi-5.0.0rc2/3rd-party/romio341/configure.ac2021-10-18
17:27:42.0 +0200
+++ src/openmpi-5.0.0rc2/3rd-party/romio341/configure.ac2022-01-09
04:58:32.150499100 +0100
@@ -1784,6 +1784,19 @@ AM_PROG_LIBTOOL
# support gcov test coverage information
PAC_ENABLE_COVERAGE
+AC_MSG_CHECKING([if libtool needs -no-undefined flag to build shared
libraries])
+case "`uname`" in
+ CYGWIN*|MINGW*|AIX*)
+## Add in the -no-undefined flag to LDFLAGS for libtool.
+AC_MSG_RESULT([yes])
+LDFLAGS="$LDFLAGS -no-undefined"
+;;
+ *)
+## Don't add in anything.
+AC_MSG_RESULT([no])
+;;
+esac
+
AC_MSG_NOTICE([setting CC to $CC])
AC_MSG_NOTICE([setting F77 to $F77])
AC_MSG_NOTICE([setting TEST_CC to $TEST_CC])
--- origsrc/openmpi-5.0.0rc2/3rd-party/romio341/mpl/configure.ac
2021-10-18 17:27:42.0 +0200
+++ src/openmpi-5.0.0rc2/3rd-party/romio341/mpl/configure.ac2022-01-09
05:01:27.462723200 +0100
@@ -1077,6 +1077,19 @@ CFLAGS=""
AX_GCC_FUNC_ATTRIBUTE(fallthrough)
PAC_POP_ALL_FLAGS
+AC_MSG_CHECKING([if libtool needs -no-undefined flag to build shared
libraries])
+case "`uname`" in
+ CYGWIN*|MINGW*|AIX*)
+## Add in the -no-undefined flag to LDFLAGS for libtool.
+AC_MSG_RESULT([yes])
+LDFLAGS="$LDFLAGS -no-undefined"
+;;
+ *)
+## Don't add in anything.
+AC_MSG_RESULT([no])
+;;
+esac
+
dnl Final output
AC_CONFIG_FILES([Makefile localdefs include/mpl_timer.h])
AC_OUTPUT
