suter [Thu, 4 Oct 2012 08:04:44 +0000 (10:04 +0200)]
add a little dax to make a real test with dax_test
Martin Quinson [Thu, 4 Oct 2012 07:55:05 +0000 (09:55 +0200)]
Merge branch 'master' of scm.gforge.inria.fr:/gitroot/simgrid/simgrid
Martin Quinson [Thu, 4 Oct 2012 06:03:22 +0000 (08:03 +0200)]
Thank the Jean Noel dream team in the Changelog
Arnaud Giersch [Wed, 3 Oct 2012 21:18:31 +0000 (23:18 +0200)]
Fix errors about uninitialized variables.
Not sure if it is correct but it cannot be worse and, at least,
it compiles..
Arnaud Giersch [Wed, 3 Oct 2012 14:02:26 +0000 (16:02 +0200)]
Use xbt_abort instead of abort.
Maybe this will improve the code coverage analysis.
Augustin Degomme [Wed, 3 Oct 2012 17:16:32 +0000 (19:16 +0200)]
avoid breaking the shiny new and working functionality from jean-noel
Augustin Degomme [Wed, 3 Oct 2012 16:52:01 +0000 (18:52 +0200)]
remove some warnings
Augustin Degomme [Wed, 3 Oct 2012 16:24:16 +0000 (18:24 +0200)]
merge with jean noel's commit
jean-noel quintin [Wed, 3 Oct 2012 15:28:45 +0000 (16:28 +0100)]
Merge remote-tracking branch 'origin/master'
Conflicts:
src/smpi/smpi_base.c
jean-noel quintin [Wed, 3 Oct 2012 15:24:39 +0000 (16:24 +0100)]
[smpi] add a gestion of non-contignous data
Augustin Degomme [Tue, 2 Oct 2012 16:06:29 +0000 (18:06 +0200)]
add support of MPI_PROC_NULL, correct behavior of waitall and waitsome for some special cases (when some requests are NULL), in order to conform a little more to the MPI standard
Augustin Degomme [Tue, 2 Oct 2012 11:06:54 +0000 (13:06 +0200)]
provide Test and Wait functions with a better MPI_STATUS_IGNORE and MPI_REQUEST_NULL handling, as well as returning correct empty statuses when needed
Also add MPI_Testsome support
Arnaud Giersch [Tue, 2 Oct 2012 12:37:30 +0000 (14:37 +0200)]
Call __gcov_flush() on abort so that gcda files are correctly written.
This is obviously only activated with enable_coverage=ON.
Arnaud Giersch [Tue, 2 Oct 2012 09:30:46 +0000 (11:30 +0200)]
Mark as volatile variables that "might be clobbered by ‘longjmp’ or ‘vfork’".
Arnaud Giersch [Tue, 2 Oct 2012 08:45:18 +0000 (10:45 +0200)]
Use atexit to call xbt_postexit.
Martin Quinson [Tue, 2 Oct 2012 09:51:38 +0000 (11:51 +0200)]
Do not kill all processes when MSG shuts down: simix will cleanup anyway
Plus, it broke Java since this function is now called using atexit(),
and it seems that in Java, the thread running these functions is not
maestro. So our mechanism was trying to fire a simcall from this
thread, which is a Java system thread that is unknown from SimGrid.
As a result, the JVM was segfaulting when exiting.
Martin Quinson [Tue, 2 Oct 2012 08:37:15 +0000 (10:37 +0200)]
Deprecate MSG_clean
Martin Quinson [Tue, 2 Oct 2012 07:32:36 +0000 (09:32 +0200)]
Ensures that SIMIX_clean is called automatically
Arnaud Giersch [Thu, 27 Sep 2012 14:17:51 +0000 (16:17 +0200)]
Fix paths for test smpi-replay in out-of-source builds.
Arnaud Giersch [Thu, 27 Sep 2012 14:17:27 +0000 (16:17 +0200)]
Fix paths for test simdag-test-fail in out-of-source builds.
Augustin Degomme [Mon, 1 Oct 2012 10:51:56 +0000 (12:51 +0200)]
fix bug in smpi tracing for wait
Augustin Degomme [Fri, 28 Sep 2012 09:17:51 +0000 (11:17 +0200)]
update the smpi replay tesh file again, as the problems with bandwidth and latency factors are now fixed
Augustin Degomme [Fri, 28 Sep 2012 08:05:21 +0000 (10:05 +0200)]
changes the way smpi tracing handles process ranks, to be able to use subgroups
Augustin Degomme [Fri, 28 Sep 2012 08:04:38 +0000 (10:04 +0200)]
various fixes for probing/testing
Augustin Degomme [Fri, 28 Sep 2012 08:01:36 +0000 (10:01 +0200)]
add support of a function (necessary to conceptual)
Martin Quinson [Thu, 27 Sep 2012 11:45:46 +0000 (13:45 +0200)]
Merge branch 'master' of scm.gforge.inria.fr:/gitroot/simgrid/simgrid
Martin Quinson [Thu, 27 Sep 2012 11:43:03 +0000 (13:43 +0200)]
update this code so that it matches my thesis document
Martin Quinson [Thu, 27 Sep 2012 11:41:01 +0000 (13:41 +0200)]
do not do useless and potentially harmful debug messages in datadesc_postexit
It may happen that xbt_log already exited at this point, leading to
erratic segfaults. At least we saw such behavior with the coroutine
factory of the Java world, where MSG_process_get_pid was segfaulting
when called from here.
Better safe than sorry.
Martin Quinson [Thu, 27 Sep 2012 11:39:59 +0000 (13:39 +0200)]
improve a tiny bit the doc
jean-noel quintin [Thu, 27 Sep 2012 11:17:05 +0000 (12:17 +0100)]
[beginning smpi non contignous]
Augustin Degomme [Thu, 27 Sep 2012 10:47:07 +0000 (12:47 +0200)]
yet another correction for those factors, hope this is the last one ..
Augustin Degomme [Thu, 27 Sep 2012 10:21:42 +0000 (12:21 +0200)]
yet another correction for those factors, hope this is the last one ..
Augustin Degomme [Thu, 27 Sep 2012 08:53:39 +0000 (10:53 +0200)]
the smpi factor parsing was still returning wrong values
suter [Wed, 26 Sep 2012 15:21:43 +0000 (17:21 +0200)]
mention last commits
suter [Wed, 26 Sep 2012 13:13:15 +0000 (15:13 +0200)]
revalidate tesh file after Augustin's commit
suter [Wed, 26 Sep 2012 12:59:46 +0000 (14:59 +0200)]
moved to examples/smpi/replay/replay.c
suter [Wed, 26 Sep 2012 12:59:05 +0000 (14:59 +0200)]
add smpi-replay to the SMPI+Tracing part of the test suite
suter [Wed, 26 Sep 2012 12:46:40 +0000 (14:46 +0200)]
add a tesh file for trace replay with SMPI
Three tests are performed
* single trace
* single trace with tracing
* multiple traces (one per process)
suter [Wed, 26 Sep 2012 12:38:58 +0000 (14:38 +0200)]
instrument broadcast and reduce actions too
suter [Wed, 26 Sep 2012 12:33:22 +0000 (14:33 +0200)]
slight modifications to make things work
suter [Wed, 26 Sep 2012 09:25:44 +0000 (11:25 +0200)]
What? give the same name to the directory and the executable file isn't
a good idea ? Come on ...
suter [Wed, 26 Sep 2012 09:20:41 +0000 (11:20 +0200)]
update the CMake file
suter [Wed, 26 Sep 2012 09:20:19 +0000 (11:20 +0200)]
add files to test trace replay with smpi
suter [Wed, 26 Sep 2012 09:17:46 +0000 (11:17 +0200)]
put the replay stuff in its own directory
Augustin Degomme [Wed, 26 Sep 2012 12:54:42 +0000 (14:54 +0200)]
sort latency and bandwidth value submitted in config files for smpi, before using them (If the user did provide them sorted from bigger to smaller sizes, only the first and biggest value was being used)
Augustin Degomme [Tue, 25 Sep 2012 16:23:39 +0000 (18:23 +0200)]
add MPI_BOTTOM definition (needed by AMG benchmark)
Augustin Degomme [Tue, 25 Sep 2012 12:56:36 +0000 (14:56 +0200)]
MPI_Address didn't return MPI_SUCCESS on success
suter [Tue, 25 Sep 2012 14:28:36 +0000 (16:28 +0200)]
add the sd_fail test to the others
suter [Tue, 25 Sep 2012 14:24:04 +0000 (16:24 +0200)]
add a tesh file for sd_fail
suter [Tue, 25 Sep 2012 14:14:40 +0000 (16:14 +0200)]
circumvent the fact that if you let the simulation runs, trace events
are chomped once for all. However, if watch points are added, everything
is fine.
suter [Tue, 25 Sep 2012 13:25:33 +0000 (15:25 +0200)]
extend the availability trace for the second part of the sd_fail test.
suter [Tue, 25 Sep 2012 13:24:54 +0000 (15:24 +0200)]
test works fine with typed tasks. Add extra testing for parallel task
(scheduled on one host), but there is something strange happening ...
suter [Tue, 25 Sep 2012 13:22:42 +0000 (15:22 +0200)]
solve bug on FAILED tasks that are rescheduled
seems to have raised another bug ...
suter [Tue, 25 Sep 2012 09:47:49 +0000 (11:47 +0200)]
add a stupid set/get test (wasn't covered)
navarro [Tue, 25 Sep 2012 12:26:55 +0000 (14:26 +0200)]
We need to have this function public for ruby binding
Augustin Degomme [Fri, 21 Sep 2012 16:07:08 +0000 (18:07 +0200)]
Add support for various MPI_Type functions, to handle creation of new MPI types
Augustin Degomme [Fri, 21 Sep 2012 10:34:55 +0000 (12:34 +0200)]
change the behavior of mpi_testall to loop over mpi_test and no longer mpi_testany.
Augustin Degomme [Fri, 21 Sep 2012 10:26:35 +0000 (12:26 +0200)]
really remove the previously commented code
schnorr [Fri, 21 Sep 2012 10:48:41 +0000 (12:48 +0200)]
[trace] save/restore previous state when pause/resume tracing
Augustin Degomme [Thu, 20 Sep 2012 17:54:24 +0000 (19:54 +0200)]
add support of MPI_Testall
Augustin Degomme [Thu, 20 Sep 2012 17:19:03 +0000 (19:19 +0200)]
forgot a declaration, and two others were removed by mistake
Augustin Degomme [Thu, 20 Sep 2012 17:02:15 +0000 (19:02 +0200)]
I thought I had done this one a long time ago. Deactivate sender_gap aggregation.
Augustin Degomme [Thu, 20 Sep 2012 17:01:20 +0000 (19:01 +0200)]
get the right sender for tracing, because it might not be known if MPI_ANY_SOURCE is used as sender
Augustin Degomme [Thu, 20 Sep 2012 17:00:19 +0000 (19:00 +0200)]
add MPI_Probe and MPI_Iprobe support, and better handling of the MPI_Status structure, which will need more testing
schnorr [Wed, 19 Sep 2012 15:24:21 +0000 (17:24 +0200)]
[trace] give users the possibility to pause/resume all tracing activities
schnorr [Wed, 19 Sep 2012 14:48:20 +0000 (16:48 +0200)]
typos
schnorr [Wed, 19 Sep 2012 13:21:35 +0000 (15:21 +0200)]
[trace] factorize + cosmetics + debug messages
Use this to understand how the graph is extracted from a platform file:
--log=instr_routing.thres:debug
Augustin Degomme [Wed, 19 Sep 2012 13:24:06 +0000 (15:24 +0200)]
add the MPI_Wtick function to smpi to allow compilation with IRS benchmark (returns the value of the maxmin/precision parameter)
suter [Tue, 18 Sep 2012 19:41:07 +0000 (21:41 +0200)]
Doesn't see why MSG_get_host_number should be deprecated.
Marion Guthmuller [Tue, 18 Sep 2012 17:05:59 +0000 (19:05 +0200)]
model-checker : first part to retrieve local variables of the application with dwarfdump
Marion Guthmuller [Thu, 13 Sep 2012 11:55:25 +0000 (13:55 +0200)]
model-checker : remove unused variable
navarro [Mon, 17 Sep 2012 08:56:28 +0000 (10:56 +0200)]
Oups use the right value
Arnaud Giersch [Mon, 17 Sep 2012 08:08:45 +0000 (10:08 +0200)]
Move instruction at proper place.
This instruction was wrongly displaced by commit
98897032e770b9a1437d6aa5820fb448e71bd224
(add an immediate asynchronous send possibility for messages [...]), leading some
simulations to failure.
One such example is "chord" with large platforms (cluster with 10000+ hosts).
Arnaud Giersch [Fri, 14 Sep 2012 19:18:22 +0000 (21:18 +0200)]
Q&D reindent.
Arnaud Giersch [Thu, 13 Sep 2012 12:57:18 +0000 (14:57 +0200)]
Avoid to dereference a null pointer with --log=surf_route.thres:debug.
Arnaud Giersch [Thu, 13 Sep 2012 12:56:50 +0000 (14:56 +0200)]
Use latest flexml for maintainer files.
Arnaud Giersch [Thu, 13 Sep 2012 12:54:27 +0000 (14:54 +0200)]
Regenerated memcheck_tests.cmake.
schnorr [Fri, 14 Sep 2012 08:26:19 +0000 (10:26 +0200)]
[trace] stress the importance of using comments on main options file
schnorr [Fri, 14 Sep 2012 08:26:00 +0000 (10:26 +0200)]
[trace] documentation for new tracing options (comment, comment_file)
schnorr [Fri, 14 Sep 2012 08:25:41 +0000 (10:25 +0200)]
[trace] missing documentation for tracing/basic option
schnorr [Fri, 14 Sep 2012 08:09:39 +0000 (10:09 +0200)]
[trace] update tesh'es to contain simgrid version that generated them
schnorr [Fri, 14 Sep 2012 08:07:17 +0000 (10:07 +0200)]
Revert "[trace] remove version, otherwise we'll have to updates tesh'es at every release"
This reverts commit
b4a1edc6568563a179fc1177a36421094d0bf6ac.
schnorr [Thu, 13 Sep 2012 22:24:19 +0000 (00:24 +0200)]
[trace] update tesh'es
schnorr [Thu, 13 Sep 2012 22:23:00 +0000 (00:23 +0200)]
[trace] remove version, otherwise we'll have to updates tesh'es at every release
and use active voice
schnorr [Thu, 13 Sep 2012 22:18:17 +0000 (00:18 +0200)]
[trace] reflect new tracing options in smpirun script
schnorr [Thu, 13 Sep 2012 22:17:19 +0000 (00:17 +0200)]
cosmetics
schnorr [Thu, 13 Sep 2012 22:10:02 +0000 (00:10 +0200)]
[trace] dumping comment and comment_file to trace files when requested
schnorr [Thu, 13 Sep 2012 22:09:41 +0000 (00:09 +0200)]
fix, removing tabs
schnorr [Thu, 13 Sep 2012 21:53:06 +0000 (23:53 +0200)]
[trace] new tracing options to add comments to trace files
schnorr [Thu, 13 Sep 2012 21:47:51 +0000 (23:47 +0200)]
fix: missing undef
Augustin Degomme [Thu, 13 Sep 2012 07:50:11 +0000 (09:50 +0200)]
fix a compilation warning/error in MSG_task_isend_with_matching instrumentation
schnorr [Wed, 12 Sep 2012 15:24:18 +0000 (17:24 +0200)]
[trace] instrumentation of MSG_task_isend_with_matching
details:
- tracing only works for asynchronous messaging on sender side for now
- let us know if you need to trace a simulation with asynchronous messaging on the receiver side
schnorr [Wed, 12 Sep 2012 15:03:13 +0000 (17:03 +0200)]
[trace] instrumenting MSG_task_dsend
suter [Tue, 11 Sep 2012 16:21:10 +0000 (18:21 +0200)]
add tracing for the trace replay tool. produces a trace, have to
visualize it now.
navarro [Tue, 11 Sep 2012 12:14:12 +0000 (14:14 +0200)]
Update file example for ls command
navarro [Tue, 11 Sep 2012 12:03:31 +0000 (14:03 +0200)]
Add a function ls to storage
suter [Mon, 10 Sep 2012 12:27:06 +0000 (14:27 +0200)]
don't need to synchronize processes anymore
suter [Mon, 10 Sep 2012 09:02:16 +0000 (11:02 +0200)]
slight change in the example that lauchs everything
suter [Mon, 10 Sep 2012 09:01:04 +0000 (11:01 +0200)]
functional implementation of trace replay on top of SMPI.
Major modifications required in the input traces:
* get rid off the "p" before the rank of the process (it was useless
BTW)
* add the message size to the recv actions (simplify a lot of thing, and
we can get it easily from the TAU traces)
suter [Mon, 10 Sep 2012 08:58:13 +0000 (10:58 +0200)]
add some test to not perform some memcpy. Mandatory for trace replay on
top of SMPI. We don't care about data, so the send and recv buffers are
always NULL. memcpy doesn't like that at all.