Re: [O-MPI devel] OpenIB results
This is symptomatic of you not installing all 3 tools in the same
$prefix (I mentioned this on the phone yesterday). You *must* install
all 3 tools together (they have interdependencies). Check out the
HACKING file for details...
On Aug 25, 2005, at 7:19 PM, Troy Benjegerdes wrote:
failing autogen.sh output at:
http://scl.ameslab.gov/~troy/ompi-autogen.log
___
devel mailing list
de...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel
--
{+} Jeff Squyres
{+} The Open MPI Project
{+} http://www.open-mpi.org/
[O-MPI devel] maffinity framework
I just committed a new "maffinity" (memory affinity) framework with
only a single, trivial component right now. It's API is still changing
a bit, but it's getting there.
Right now, memory affinity is automatically enabled during
ompi_mpi_init() (*not* automatically during opal_init()) if processor
affinity is both enabled and successful in biinding the process to a
processor. Nothing really uses the memory affinity yet except my
un-committed shared memory collectives, but the API/components will
likely be expanded in the near future to hint to the OS that all
malloc's should return local memory, etc.
Thanks to Troy for pointing out the libnuma project (out of the SUSE
Labs) that is usable on any NUMA machine (e.g., Opterons -- I'm asking
to have it installed on the Odin IU cluster). I'll be making a
maffinity component for that as well.
--
{+} Jeff Squyres
{+} The Open MPI Project
{+} http://www.open-mpi.org/
[O-MPI devel] F90 support
With help from Brian and Craig, we have now F90 MPI bindings support
working properly. Because of the time required to compile the F90
bindings, they are automatically disabled in SVN checkout builds. You
can enable them with --enable-f90.
*** The F90 bindings are now automatically enabled in tarball and VPATH
builds ***
Unfortunately, the current generation of Libtool (1.5.x) does not
support building F90 shared libraries, so the F90 support library can
only be built statically (even if you --enable-shared, the F90 library
will be built statically). Bummer. :-\
Ralf Wildenhues, one of the Libtool developers and a long-time Friend
of LAM/MPI, assures me that F90 support will be included in the
forthcoming Libtool 2.0 series. He's even testing with the Open MPI
code base to ensure that everything will work properly. Woo hoo --
thanks Ralf!
--
{+} Jeff Squyres
{+} The Open MPI Project
{+} http://www.open-mpi.org/
Re: [O-MPI devel] OpenIB results
On Thu, Aug 25, 2005 at 05:53:27PM +0300, Gleb Natapov wrote: > On Thu, Aug 25, 2005 at 09:43:31AM -0500, Troy Benjegerdes wrote: > > On Thu, Aug 25, 2005 at 04:35:41PM +0300, Gleb Natapov wrote: > > > On Wed, Aug 24, 2005 at 09:21:37PM -0500, Troy Benjegerdes wrote: > > > > I also get the following messages on startup: > > > > > > > > libibverbs: Warning: no userspace device-specific driver found for > > > > uverbs0 driver search path: /usr/lib/infiniband > > > > > > > > [0,1,1][btl_openib_component.c:267:mca_btl_openib_component_init] > > > > No hca's found on this host! > > > > > > > Look for RTLD_GLOBAL in opal/libltdl/ltdl.c. If it is not there openib > > > will not work if compiled as dynamic module. What version of libtool are > > > you using? Older libtool haven't used RTLD_GLOBAL for opening libraries. > > > > It's not there... > > > > troy@opteron1:~$ libtool --version > > ltmain.sh (GNU libtool) 1.5.6 (1.1220.2.95 2004/04/11 05:50:42) Debian: 224 > > $ > 1.5.6 from debian do the same for me. But 1.5.8 works OK. If it is a problem > for you > to reinstall libtool edit file opal/libltdl/ltdl.c and change line > #define LT_LAZY_OR_NOW RTLD_LAZY > to > #define LT_LAZY_OR_NOW (RTLD_LAZY|RTLD_GLOBAL) > And recompile. (don't run autogen.sh, this will regenerate the file). Did you manage to get this working on a debian system with just the newer version of libtool? Jeff suggests I build libtool and all the other autotools as well, but I'd like to see if I can avoid that, and just update libtool.
[O-MPI devel] SLURM support
Brian and I finished up the SLURM support (more to the point; I did all
the easy stuff and Brian did all the hard stuff). So now on IU's odin
and thor clusters, for example, you don't need to specify hostfiles and
you won't be using rsh/ssh. For example:
odin% srun -N 4 -A
odin% orterun -np 4 hostname
odin027
odin026
odin025
odin024
Woo hoo!
(similar things happen in PBS/Torque environments, in case you forgot
:-) )
--
{+} Jeff Squyres
{+} The Open MPI Project
{+} http://www.open-mpi.org/
[O-MPI devel] configure hangs in libtool..
Hello folk I just checked out 7046 and my configure hangs at: configure: creating ./config.status config.status: creating Makefile config.status: creating config.h config.status: executing depfiles commands configure: /bin/sh '../../../opal/libltdl/configure' succeeded for opal/libltdl checking for libtool-supplied linker flags... ps -ef f gives me: ...configure \_ /bin/sh /home/cap/openmpi/ompi-trunk/vbuild/opal/libltdl/libtool --mode=compile icc -O3 -DNDEBUG foo.c -c -o foo.o \_ sleep 2 is it just currently broken (and if so, how do I check out an older revision) or did I fumble the spell? I did (in top level of a cleanly checked out 7046): ./autogen.sh && mkdir vbuild && cd vbuild && ../configure \ --prefix=/usr/local/openmpi-svn7045/intel-8.1e-029 \ --with-btl-mvapi=/opt/ibgd/driver/infinihost \ --with-platform=optimized tia, Peter -- Peter Kjellström | National Supercomputer Centre | Sweden | http://www.nsc.liu.se pgpoXX7Ewto2h.pgp Description: PGP signature
Re: [O-MPI devel] configure hangs in libtool..
Hi Peter, * Peter Kjellström wrote on Fri, Aug 26, 2005 at 05:21:27PM CEST: > > I just checked out 7046 and my configure hangs at: > > configure: creating ./config.status > config.status: creating Makefile > config.status: creating config.h > config.status: executing depfiles commands > configure: /bin/sh '../../../opal/libltdl/configure' succeeded for > opal/libltdl > checking for libtool-supplied linker flags... > > ps -ef f gives me: > > ...configure > \_ /bin/sh /home/cap/openmpi/ompi-trunk/vbuild/opal/libltdl/libtool > --mode=compile icc -O3 -DNDEBUG foo.c -c -o foo.o > \_ sleep 2 > > is it just currently broken (and if so, how do I check out an older > revision) or did I fumble the spell? Weird. Libtool thought icc does not understand `-c -o'? Can you search and post the respective output from config.log? Thanks, Ralf
Re: [O-MPI devel] configure hangs in libtool..
On Friday 26 August 2005 17.27, Ralf Wildenhues wrote: > Hi Peter, > > * Peter Kjellström wrote on Fri, Aug 26, 2005 at 05:21:27PM CEST: > > I just checked out 7046 and my configure hangs at: > > > > configure: creating ./config.status > > config.status: creating Makefile > > config.status: creating config.h > > config.status: executing depfiles commands > > configure: /bin/sh '../../../opal/libltdl/configure' succeeded for > > opal/libltdl checking for libtool-supplied linker flags... > > > > ps -ef f gives me: > > > > ...configure > > \_ /bin/sh /home/cap/openmpi/ompi-trunk/vbuild/opal/libltdl/libtool > > --mode=compile icc -O3 -DNDEBUG foo.c -c -o foo.o \_ sleep 2 > > > > is it just currently broken (and if so, how do I check out an older > > revision) or did I fumble the spell? > > Weird. Libtool thought icc does not understand `-c -o'? > > Can you search and post the respective output from config.log? configure:130480: checking for libtool-supplied linker flags configure:130509: /home/cap/openmpi/ompi-trunk/opal/libltdl/libtool --mode=compile icc -O3 -DNDEBUG foo.c -c -o foo.o Waiting for -c.o.lock to be removed Waiting for -c.o.lock to be removed Waiting for -c.o.lock to be removed ...and so on for inf /Peter pgpUyxT1CtYeR.pgp Description: PGP signature
Re: [O-MPI devel] configure hangs in libtool..
Here's a bit more info:
* Ralf Wildenhues wrote on Fri, Aug 26, 2005 at 05:27:44PM CEST:
> * Peter Kjellström wrote on Fri, Aug 26, 2005 at 05:21:27PM CEST:
> >
> > I just checked out 7046 and my configure hangs at:
> > checking for libtool-supplied linker flags...
> >
> > ps -ef f gives me:
> >
> > ...configure
> > \_ /bin/sh /home/cap/openmpi/ompi-trunk/vbuild/opal/libltdl/libtool
> > --mode=compile icc -O3 -DNDEBUG foo.c -c -o foo.o
> > \_ sleep 2
> >
> > is it just currently broken (and if so, how do I check out an older
> > revision) or did I fumble the spell?
>
> Weird. Libtool thought icc does not understand `-c -o'?
>
> Can you search and post the respective output from config.log?
Just go to $top_builddir and execute these:
$ find . -name config.log | xargs sed -n '/ supports -c -o
/{N;N;N;N;N;N;N;N;p;}'
$ ./libtool --version
Libtool version previous to 1.5.16 had a bug in the macro to detect
this; it was often wrong before. Now it should be pretty bogus.
Libtool still has a bug that might lead to the hang you describe
when both the macro detects that the compiler (seemingly or not)
does not understand "-c -o" and the source tree and the build tree
live on different mount points, so that a hard link fails.
It's on our TODO list, but as we thought it to be a very rare bug,
it had rather low priority so far.
If you have use Libtool >= 1.5.16 and still experience this issue,
one workaround is to have the build tree below the same mount point.
But I'd like to see the failure of the macro: icc understands "-c -o",
and we should be able to detect that.
Cheers,
Ralf
Re: [O-MPI devel] configure hangs in libtool..
* Ralf Wildenhues wrote on Fri, Aug 26, 2005 at 06:04:19PM CEST: > > Libtool version previous to 1.5.16 had a bug in the macro to detect > this; it was often wrong before. Now it should be pretty bogus. Gah. That should've been: Now it should be pretty good. Sorry.
Re: [O-MPI devel] OpenIB results
* Troy Benjegerdes wrote on Fri, Aug 26, 2005 at 04:18:06PM CEST: > > Did you manage to get this working on a debian system with just the > newer version of libtool? Jeff suggests I build libtool and all the > other autotools as well, but I'd like to see if I can avoid that, and > just update libtool. If you want to update libtool alone (and thus have it in a different $prefix than the other autotools), you need to help aclocal to find the new libtool macro files. - either: everywhere where aclocal is executed in autogen.sh, you replace it with aclocal -I$prefix_of_libtool_installation/share/aclocal - or: in the normal directory where aclocal searches, add a file named "dirlist" which lists other directories to be searched. Note however, that _these_ directories are searched _after_ the normal one, so if you have libtool macros in both, a mixup will occur and you'll only get angry. :) (It would be good to augment the autogen.sh script so that it honored a setting like ACLOCAL="path/to/aclocal -I ..." (in Bourne shell speak) and similarly for AUTOMAKE, AUTOCONF, LIBTOOLIZE). Cheers, Ralf
Re: [O-MPI devel] configure hangs in libtool..
On Friday 26 August 2005 18.04, Ralf Wildenhues wrote:
> > >...
> >
> > Weird. Libtool thought icc does not understand `-c -o'?
> >
> > Can you search and post the respective output from config.log?
>
> Just go to $top_builddir and execute these:
> $ find . -name config.log | xargs sed -n '/ supports -c -o
> /{N;N;N;N;N;N;N;N;p;}' $ ./libtool --version
>
> Libtool version previous to 1.5.16 had a bug in the macro to detect
> this; it was often wrong before.
> ...
my libtool is indeed rather ancient compared to 1.5.16, I knew ompi needed a
recent one but had assumed that an updated centos-4.1 (rhel4u1) would be
enough... (I have 1.5.6)
Either way, I have managed to build ompi before so I looked deeper and found
out that the intel compilers had been updated under my feet (-027 to -029).
Switching back made configure run just fine...
/Peter
--
Peter Kjellström |
National Supercomputer Centre |
Sweden | http://www.nsc.liu.se
pgpHfbp7W5Ql2.pgp
Description: PGP signature
Re: [O-MPI devel] OpenIB results
On Aug 26, 2005, at 12:19 PM, Ralf Wildenhues wrote:
(It would be good to augment the autogen.sh script so that it honored
a setting like ACLOCAL="path/to/aclocal -I ..." (in Bourne shell speak)
and similarly for AUTOMAKE, AUTOCONF, LIBTOOLIZE).
I have no objections to this... :-)
--
{+} Jeff Squyres
{+} The Open MPI Project
{+} http://www.open-mpi.org/
Re: [O-MPI devel] configure hangs in libtool..
* Peter Kjellström wrote on Fri, Aug 26, 2005 at 06:24:43PM CEST: > On Friday 26 August 2005 18.04, Ralf Wildenhues wrote: > > > > Libtool version previous to 1.5.16 had a bug in the macro to detect > > this; it was often wrong before. > > my libtool is indeed rather ancient compared to 1.5.16, I knew ompi needed a > recent one but had assumed that an updated centos-4.1 (rhel4u1) would be > enough... (I have 1.5.6) > > Either way, I have managed to build ompi before so I looked deeper and found > out that the intel compilers had been updated under my feet (-027 to -029). > Switching back made configure run just fine... OK. But see, I'd still like to know whether libtool-1.5.18 copes with your newer icc. I can't fix it (given there is something to fix) unless I know about the failure. Cheers, Ralf
[O-MPI devel] pml vs bml vs btl
Greetings, I was just curious about the distinction between the pml, bml, and btl layers within OpenMPI and was wondering if there was any place (other than the code ;-) where I could read about the design decisions. Any pointers? Particularly, I'm just curious about the difference between pml-bml-btl and the old pml-ptl (as described in your TEG paper) and why changes were made. I'm also trying to understand the differences between this approach and LAM's RPI as well as the device-channel distinction made in MPICH2; could anyone recommend something I could read or give me a brief description here to satisfy my curiosities? Thanks, brad
Re: [O-MPI devel] OpenIB results
On Fri, Aug 26, 2005 at 12:35:07PM -0400, Jeff Squyres wrote:
> On Aug 26, 2005, at 12:19 PM, Ralf Wildenhues wrote:
>
> > (It would be good to augment the autogen.sh script so that it honored
> > a setting like ACLOCAL="path/to/aclocal -I ..." (in Bourne shell speak)
> > and similarly for AUTOMAKE, AUTOCONF, LIBTOOLIZE).
>
> I have no objections to this... :-)
how's this? (I spent waaay to much time learning obscure shell stuff for
this)
I allows me to do
ACLOCAL='aclocal -I /usr/local/share/aclocal' ./autogen.sh
and it works a little better.. waiting for a build now.
Index: autogen.sh
===
--- autogen.sh (revision 7061)
+++ autogen.sh (working copy)
@@ -22,18 +22,29 @@
##
#
# User-definable parameters (search path and minimum supported versions)
-#
+#
+# Note: use ';' to separate parameters
##
-ompi_aclocal_search="aclocal"
-ompi_autoheader_search="autoheader"
-ompi_autoconf_search="autoconf"
-ompi_libtoolize_search="libtoolize glibtoolize"
-ompi_automake_search="automake"
+if [ -z "$ACLOCAL" ]; then ompi_aclocal_search="aclocal"
+else ompi_aclocal_search=$ACLOCAL
+fi
+if [ -z "$AUTOHEADER" ]; then ompi_autoheader_search="autoheader"
+else ompi_autoheader_search=$AUTOHEADER
+fi
+if [ -z "$AUTOCONF" ]; thenompi_autoconf_search="autoconf"
+else ompi_autoconf_search=$AUTOCONF
+fi
+if [ -z "$LIBTOOLIZE" ]; then ompi_libtoolize_search="libtoolize;glibtoolize"
+else ompi_libtoolize_search=$LIBTOOLIZE
+fi
+if [ -z "$AUTOMAKE" ]; thenompi_automake_search="automake"
+else ompi_automake_search=$AUTOMAKE
+fi
ompi_automake_version="1.7"
ompi_autoconf_version="2.58"
-ompi_libtool_version="1.5"
+ompi_libtool_version="1.5.18"
##
@@ -46,13 +57,6 @@
ompi_autoheader_version="$ompi_autoconf_version"
ompi_libtoolize_version="$ompi_libtool_version"
-# program names to execute
-ompi_aclocal=""
-ompi_autoheader=""
-ompi_autoconf=""
-ompi_libtoolize=""
-ompi_automake=""
-
mca_no_configure_components_file="config/mca_no_configure_components.m4"
mca_no_config_list_file="mca_no_config_list"
mca_no_config_env_file="mca_no_config_env"
@@ -137,11 +141,14 @@
local version="0.0.0"
local min_version="99.99.99"
local found=0
+local tmpIFS=$IFS
+IFS=";"
eval "min_version=\"\$ompi_${app_name}_version\""
eval "search_path=\"\$ompi_${app_name}_search\""
for i in $search_path ; do
+IFS=$tmpIFS
version="`${i} --version 2>&1`"
if test "$?" != 0 ; then
continue
@@ -157,6 +164,8 @@
fi
done
+IFS=$tmpIFS
+
if test "$found" = "0" ; then
cat <
