Augustin Degomme [Mon, 11 Feb 2013 17:22:17 +0000 (18:22 +0100)]
activate a bunch of tests using mpi_ssend, and change back those where they had been changed to mpi_send
Augustin Degomme [Mon, 11 Feb 2013 16:43:44 +0000 (17:43 +0100)]
add support for Ssend/Issend in SMPI
Augustin Degomme [Mon, 11 Feb 2013 15:33:26 +0000 (16:33 +0100)]
Revert "always have display_timing activated for smpi, as it is quite useful"
This reverts commit
f641b68a3c807956396f8f70965ce9c4644dcfa5.
Augustin Degomme [Mon, 11 Feb 2013 14:50:16 +0000 (15:50 +0100)]
add option to set the threshold where smpi send stops behaving like isend.
This was hardcoded to 65536, but experiments show that sometimes 327680 might be better
Augustin Degomme [Mon, 11 Feb 2013 14:49:08 +0000 (15:49 +0100)]
small change for smpi win
Augustin Degomme [Mon, 11 Feb 2013 14:46:59 +0000 (15:46 +0100)]
bug : smpi tracing options were replacing those the user had set
Augustin Degomme [Tue, 5 Feb 2013 17:27:08 +0000 (18:27 +0100)]
small fixes to repair unix build
Augustin Degomme [Tue, 5 Feb 2013 17:19:47 +0000 (18:19 +0100)]
rename main function for windows
Augustin Degomme [Tue, 5 Feb 2013 17:19:27 +0000 (18:19 +0100)]
several changes in the way cmake handles windows
Augustin Degomme [Mon, 11 Feb 2013 15:19:13 +0000 (16:19 +0100)]
resolve cmakelists conflict
Augustin Degomme [Tue, 5 Feb 2013 17:16:09 +0000 (18:16 +0100)]
remove a few unused things for win, but we have more bugs after ..
Augustin Degomme [Tue, 5 Feb 2013 17:14:38 +0000 (18:14 +0100)]
various fixes to hopefully compile under win32
Augustin Degomme [Mon, 11 Feb 2013 14:50:39 +0000 (15:50 +0100)]
Merge branch 'master' of git+ssh://scm.gforge.inria.fr//gitroot/simgrid/simgrid
suter [Fri, 8 Feb 2013 15:13:17 +0000 (16:13 +0100)]
handle new action trace files properly
suter [Fri, 8 Feb 2013 15:12:35 +0000 (16:12 +0100)]
add three more tests to smpi_replay : alltoall, alltoallv and waitall
suter [Fri, 8 Feb 2013 15:11:53 +0000 (16:11 +0100)]
new traces files were added in examples/smpi/replay. treat them as the
others
George Markomanolis [Fri, 8 Feb 2013 12:02:24 +0000 (14:02 +0200)]
Fixing a possible bug, I should use parse_double() instead of atoi() if I want to handle values like 5e8. Adding examples for MPI_Alltoallv, MPI_Alltoall, MPI_Bcast/MPI_Reduce with various datatypes and MPI_Waitall.
suter [Fri, 8 Feb 2013 09:56:07 +0000 (10:56 +0100)]
rewriting
suter [Fri, 8 Feb 2013 09:52:28 +0000 (10:52 +0100)]
revalidate tesh file as the reduce action has been modified
Arnaud Giersch [Thu, 7 Feb 2013 22:13:55 +0000 (23:13 +0100)]
Make this part automatically generated.
And avoid to give a direct link to the archive, since it's too error prone.
Arnaud Giersch [Thu, 7 Feb 2013 21:54:51 +0000 (22:54 +0100)]
Typo.
Martin Quinson [Thu, 7 Feb 2013 20:15:21 +0000 (21:15 +0100)]
Fix the download link in the install documentation
Martin Quinson [Thu, 7 Feb 2013 20:10:50 +0000 (21:10 +0100)]
ops, didn't knew how gforge downloads work, actually
Martin Quinson [Thu, 7 Feb 2013 01:18:30 +0000 (02:18 +0100)]
a more welcoming README file
Martin Quinson [Thu, 7 Feb 2013 01:14:02 +0000 (02:14 +0100)]
Seed the random generators before the main()
Otherwise, it is possible that the user set the seed s/he wants before
initializing simgrid, that would then override it with another seed.
That would be a *very mean* side effect of initializing simgrid :)
George Markomanolis [Thu, 7 Feb 2013 17:12:24 +0000 (19:12 +0200)]
Updating the Changelog for the SMPI replay tool
George Markomanolis [Thu, 7 Feb 2013 16:38:39 +0000 (18:38 +0200)]
Now the SMPI replay tool supports the modified MPE profiling tool.
What is new:
1) For most of the collective communications we can know exactly which rank is the root one.
2) Now we can include in the traces the real used MPI datatype (till know we were using the MPI_BYTE). Also we can have multiple datatypes in the traces.
3) The tool supports both time-independent traces from the TAU and the modified MPE tools.
4) Fix Reduce action, as there was no information about computation load.
Arnaud Giersch [Thu, 7 Feb 2013 13:39:44 +0000 (14:39 +0100)]
Try to make simdag-test-dotload reproducible.
Comment extracted from this patch:
/* Hack: circumvent a bug in libcgraph, where the edges are not always given
* back in creation order. We sort them again, according to their sequence
* id. The problem appears to be solved (i.e.: I did not test it) in
* graphviz' mercurial repository by the following changeset:
* changeset: 8431:
d5f1fb7e8103
* user: Emden Gansner <erg@research.att.com>
* date: Tue Oct 11 12:38:58 2011 -0400
* summary: Make sure edges are stored in node creation order
* It should be fixed in graphviz 2.30 and above.
*/
Arnaud Giersch [Wed, 6 Feb 2013 21:40:14 +0000 (22:40 +0100)]
Return MPI_UNDEFINED for negative values too.
Arnaud Giersch [Wed, 6 Feb 2013 21:39:51 +0000 (22:39 +0100)]
Use MPI_ANY_SOURCE instead of -1.
Arnaud Giersch [Wed, 6 Feb 2013 20:38:55 +0000 (21:38 +0100)]
Avoid to modify lib while it's been iterated.
Arnaud Giersch [Wed, 6 Feb 2013 20:26:17 +0000 (21:26 +0100)]
MSG_comm_get_status must be called before MSG_comm_destroy.
Arnaud Giersch [Wed, 6 Feb 2013 18:11:01 +0000 (19:11 +0100)]
Exclude Java tests from memcheck.
We don't want to test the JVM.
Arnaud Giersch [Wed, 6 Feb 2013 17:49:43 +0000 (18:49 +0100)]
my_valgrind.pl: don't stop loop if $arg has value 0.
Arnaud Giersch [Wed, 6 Feb 2013 17:07:37 +0000 (18:07 +0100)]
generate_memcheck_tests.pl: add support for any environment variable.
"classpath" is an example of previously unsupported name.
Arnaud Giersch [Wed, 6 Feb 2013 10:10:39 +0000 (11:10 +0100)]
Give state a sensible value, and don't let it uninitialized.
Arnaud Giersch [Wed, 6 Feb 2013 09:56:04 +0000 (10:56 +0100)]
Test is not broken anymore.
Arnaud Giersch [Wed, 6 Feb 2013 09:55:33 +0000 (10:55 +0100)]
Regenerate tesh file for master_slave_virtual_machines.
Arnaud Giersch [Wed, 6 Feb 2013 09:52:04 +0000 (10:52 +0100)]
Wait a while before effective shutdown.
Arnaud Giersch [Wed, 6 Feb 2013 09:48:57 +0000 (10:48 +0100)]
Destroy comm action in SIMIX_process_kill.
Arnaud Giersch [Tue, 5 Feb 2013 22:07:22 +0000 (23:07 +0100)]
Correctly restart all processes on VM reboot.
vm->processes is changed by MSG_process_restart.
Arnaud Giersch [Tue, 5 Feb 2013 20:49:36 +0000 (21:49 +0100)]
Fix memory leak.
Marion Guthmuller [Wed, 6 Feb 2013 08:00:38 +0000 (09:00 +0100)]
model-checker : fix insertion in mc_heap_comparison_ignore
Arnaud Giersch [Tue, 5 Feb 2013 15:31:47 +0000 (16:31 +0100)]
[pipol] Enable Java on MC builds.
Arnaud Giersch [Tue, 5 Feb 2013 15:26:07 +0000 (16:26 +0100)]
[pipol] Synchronize MemCheck.sh and Nightly_memCheck.sh.
Arnaud Giersch [Tue, 5 Feb 2013 15:22:51 +0000 (16:22 +0100)]
[pipol] Remove tests for simgrid-java.
Arnaud Giersch [Tue, 5 Feb 2013 15:18:34 +0000 (16:18 +0100)]
[pipol] Install default version for compilers.
Particularly, default-jdk should fix the search paths for cmake.
Arnaud Giersch [Tue, 5 Feb 2013 15:16:37 +0000 (16:16 +0100)]
[pipol] Sort packages by name.
Arnaud Giersch [Tue, 5 Feb 2013 15:14:07 +0000 (16:14 +0100)]
Merge branch 'v3_9_x'
Arnaud Giersch [Tue, 5 Feb 2013 10:32:10 +0000 (11:32 +0100)]
Fix release date for v3.9.
Arnaud Giersch [Tue, 5 Feb 2013 10:12:03 +0000 (11:12 +0100)]
Cast value, to please C++ compilers.
Arnaud Giersch [Tue, 5 Feb 2013 10:10:00 +0000 (11:10 +0100)]
Mark parameter as unused.
Paul Bédaride [Thu, 31 Jan 2013 10:14:00 +0000 (11:14 +0100)]
Revert "Add ModelChecker build type"
This reverts commit
07eb7317b31474d0b6702c255741b194dbfd2d6d.
Conflicts:
CMakeLists.txt
Arnaud Giersch [Mon, 4 Feb 2013 16:37:49 +0000 (17:37 +0100)]
Die if fopen() failed.
Arnaud Giersch [Mon, 4 Feb 2013 16:37:19 +0000 (17:37 +0100)]
Kill spaces at eol.
suter [Mon, 4 Feb 2013 17:13:42 +0000 (18:13 +0100)]
don't try to understand why the order changes when jedule is activated
anymore. just consider that it should be off.
This test is really annoying. As valgrind indicates a lot of leaks
coming the graphviz stuff, and that the issue comes from one edge read
after or before another, I'm kind to blame this tool!
Gonna get rid of this dependency one day!
suter [Mon, 4 Feb 2013 17:02:34 +0000 (18:02 +0100)]
better to close file
Arnaud Legrand [Mon, 4 Feb 2013 09:53:17 +0000 (10:53 +0100)]
Add a new test on availability for the SimDAG API.
Issue was reported by Ricardo Rodriguez on 29/12/12 and fixed in
previous commits.
Arnaud Legrand [Mon, 4 Feb 2013 10:49:53 +0000 (11:49 +0100)]
This way of passing arguments has changed since a long time.
Fixing all tesh files accordingly. Yet, now that this mechanism is
effective, absolute paths are output everywhere, hence the need to
hide just like we do for contexts. I wish I had a cleaner solution but
I don't.
Arnaud Legrand [Mon, 4 Feb 2013 13:04:14 +0000 (14:04 +0100)]
Bug fix in trace management.
When using periodicity, the first event indicates when events start and
has undefined value (-1) as the value of the resource is used (e.g., the
bandwidth value precised in the xml file). When looping with periodicity,
the -1 value was used, which resulted in links with bandwidth -1.0 and
could be noticed only in very specific situations.
Arnaud Legrand [Mon, 4 Feb 2013 13:01:18 +0000 (14:01 +0100)]
Bug fix. The date given when updating the state of a resource was incorrect. It could result in incoherent traces.
Arnaud Legrand [Mon, 4 Feb 2013 13:00:12 +0000 (14:00 +0100)]
Additionnal information never hurt.
Arnaud Legrand [Mon, 4 Feb 2013 09:53:33 +0000 (10:53 +0100)]
Docfix
suter [Mon, 4 Feb 2013 16:52:28 +0000 (17:52 +0100)]
don't break anything this way
suter [Mon, 4 Feb 2013 16:42:45 +0000 (17:42 +0100)]
don't break if debug is activated ...
suter [Mon, 4 Feb 2013 16:32:16 +0000 (17:32 +0100)]
mem leak --
suter [Mon, 4 Feb 2013 16:29:53 +0000 (17:29 +0100)]
mem leak--
suter [Mon, 4 Feb 2013 15:52:17 +0000 (16:52 +0100)]
'end' is parsed as other node now, don't use a non-existing workstation
rank, just say nothing instead
suter [Mon, 4 Feb 2013 15:50:54 +0000 (16:50 +0100)]
try again to revalidate this thing... hope it is robust now
suter [Mon, 4 Feb 2013 15:50:20 +0000 (16:50 +0100)]
oups, forgot to check is graph is acyclic or not
suter [Mon, 4 Feb 2013 15:41:31 +0000 (16:41 +0100)]
deal with the predefined scheduling stuff
suter [Mon, 4 Feb 2013 14:51:38 +0000 (15:51 +0100)]
yet another attempt to solve issues with the dotload tester. This code
was very complex, more than it should have ...
Reduce it to something simple:
- parse the compute node
- create transfer and control dependencies by browsing the input edges
of the compute nodes
- add extra dependencies from or to the 'root' and 'end' nodes for the
entry and exit tasks respectively.
The hope is that the order in some output will stay consistent across
cdash distributions.
suter [Mon, 4 Feb 2013 14:42:30 +0000 (15:42 +0100)]
security commit before using the ax!
suter [Mon, 4 Feb 2013 14:23:36 +0000 (15:23 +0100)]
change generated trace format
cosmetics
suter [Fri, 1 Feb 2013 10:56:53 +0000 (11:56 +0100)]
doesn't hurt to use types expected by some xbt_dynar functions
Arnaud Giersch [Mon, 4 Feb 2013 14:54:29 +0000 (15:54 +0100)]
Rewrite xbt_getline.
* fix a buffer overflow
* correctly return -1 when nothing can be read
Arnaud Giersch [Mon, 4 Feb 2013 14:05:27 +0000 (15:05 +0100)]
Avoid to return from inside a TRY block.
Arnaud Giersch [Mon, 4 Feb 2013 13:24:27 +0000 (14:24 +0100)]
Initialize variable.
Useless, but gcc whines if not initialized.
Arnaud Legrand [Mon, 4 Feb 2013 09:53:17 +0000 (10:53 +0100)]
Add a new test on availability for the SimDAG API.
Issue was reported by Ricardo Rodriguez on 29/12/12 and fixed in
previous commits.
Arnaud Legrand [Mon, 4 Feb 2013 10:49:53 +0000 (11:49 +0100)]
This way of passing arguments has changed since a long time.
Fixing all tesh files accordingly. Yet, now that this mechanism is
effective, absolute paths are output everywhere, hence the need to
hide just like we do for contexts. I wish I had a cleaner solution but
I don't.
Arnaud Legrand [Mon, 4 Feb 2013 13:04:14 +0000 (14:04 +0100)]
Bug fix in trace management.
When using periodicity, the first event indicates when events start and
has undefined value (-1) as the value of the resource is used (e.g., the
bandwidth value precised in the xml file). When looping with periodicity,
the -1 value was used, which resulted in links with bandwidth -1.0 and
could be noticed only in very specific situations.
Arnaud Legrand [Mon, 4 Feb 2013 13:01:18 +0000 (14:01 +0100)]
Bug fix. The date given when updating the state of a resource was incorrect. It could result in incoherent traces.
Arnaud Legrand [Mon, 4 Feb 2013 13:00:12 +0000 (14:00 +0100)]
Additionnal information never hurt.
Arnaud Legrand [Mon, 4 Feb 2013 09:53:33 +0000 (10:53 +0100)]
Docfix
Martin Quinson [Mon, 4 Feb 2013 12:50:24 +0000 (13:50 +0100)]
Further cleanup in the ignore file
- ignore the files touched to track dependencies around java examples
- don't ignore the files for the gras tutorial, as there is no such
tutorial anymore
Martin Quinson [Mon, 4 Feb 2013 12:12:34 +0000 (13:12 +0100)]
use our xbt_getline on every platform to ease our cmake mess
Also, as stdio.h is not loaded by root headers of the simgrid tree, we
need to explicitely load it in some locations.
Augustin Degomme [Mon, 28 Jan 2013 10:51:26 +0000 (11:51 +0100)]
always have display_timing activated for smpi, as it is quite useful
Arnaud Giersch [Sun, 3 Feb 2013 23:01:00 +0000 (00:01 +0100)]
Rename getline as xbt_getline.
Arnaud Giersch [Sun, 3 Feb 2013 21:28:23 +0000 (22:28 +0100)]
Update documentation about java bindings.
Arnaud Giersch [Sun, 3 Feb 2013 21:21:49 +0000 (22:21 +0100)]
Fix javadoc warnings.
Arnaud Giersch [Sun, 3 Feb 2013 21:09:51 +0000 (22:09 +0100)]
Make javadoc less verbose.
George Markomanolis [Sun, 3 Feb 2013 21:06:59 +0000 (23:06 +0200)]
Add MPI_Alltoallv support to the SMPI replay tool. Make it easy to change MPI_Datatype for the SMPI replay tool (IS benchmark uses MPI_INT and some others use MPI_BYTE)
Arnaud Giersch [Sun, 3 Feb 2013 01:17:02 +0000 (02:17 +0100)]
Connect log categories used in java bindings.
Arnaud Giersch [Sun, 3 Feb 2013 00:35:52 +0000 (01:35 +0100)]
Change message to debug loglevel.
Arnaud Giersch [Sun, 3 Feb 2013 00:14:39 +0000 (01:14 +0100)]
Fix java examples: catch exceptions on failures.
Arnaud Giersch [Sat, 2 Feb 2013 23:57:32 +0000 (00:57 +0100)]
Don't let exceptions pass through the MSG layer.
Arnaud Giersch [Sat, 2 Feb 2013 23:45:16 +0000 (00:45 +0100)]
Typo.
Arnaud Giersch [Sat, 2 Feb 2013 23:38:52 +0000 (00:38 +0100)]
Run Java tests from build directory, instead of source directory.
Arnaud Giersch [Sat, 2 Feb 2013 23:27:39 +0000 (00:27 +0100)]
Remove output file after run.
Arnaud Giersch [Sat, 2 Feb 2013 23:25:17 +0000 (00:25 +0100)]
There's no test for java_io.