Martin Quinson [Tue, 11 Apr 2017 21:38:19 +0000 (23:38 +0200)]
regenerate the smpi_location thingy
Martin Quinson [Tue, 11 Apr 2017 21:37:47 +0000 (23:37 +0200)]
ignore the newly created test files
degomme [Tue, 11 Apr 2017 21:53:08 +0000 (23:53 +0200)]
Avoid trailing whitespaces when dl not found
... which is an issue by itself, but too late, martin will fix it tomorrow 8 a.m.
degomme [Tue, 11 Apr 2017 21:32:20 +0000 (23:32 +0200)]
Ooooops ²
degomme [Tue, 11 Apr 2017 21:25:51 +0000 (23:25 +0200)]
Fix .. something ?
degomme [Tue, 11 Apr 2017 17:09:10 +0000 (19:09 +0200)]
fix mc build
degomme [Tue, 11 Apr 2017 15:21:08 +0000 (17:21 +0200)]
ooooops
degomme [Tue, 11 Apr 2017 13:43:29 +0000 (15:43 +0200)]
fix smpi tests
degomme [Tue, 11 Apr 2017 13:22:41 +0000 (15:22 +0200)]
Yet another dlopen merge
Martin Quinson [Tue, 11 Apr 2017 20:53:01 +0000 (22:53 +0200)]
kill useless code
Martin Quinson [Tue, 11 Apr 2017 20:31:36 +0000 (22:31 +0200)]
simplify SMPI process creation a tiny bit
Martin Quinson [Tue, 11 Apr 2017 17:17:45 +0000 (19:17 +0200)]
try to cleanup how MPI process are created. First act.
Martin Quinson [Tue, 11 Apr 2017 16:00:33 +0000 (18:00 +0200)]
travis: activate optimizations, since the tests are taking for ever
Martin Quinson [Tue, 11 Apr 2017 13:30:02 +0000 (15:30 +0200)]
please sonar by killing useless parameters and cleanups
Martin Quinson [Tue, 11 Apr 2017 13:09:25 +0000 (15:09 +0200)]
typos
degomme [Tue, 11 Apr 2017 12:28:55 +0000 (14:28 +0200)]
Iprobe was actually ignored if MPI_STATUS_IGNORE was set, which is a bug.
Martin Quinson [Mon, 10 Apr 2017 23:26:26 +0000 (01:26 +0200)]
MC: this bug was disabling state equality detection
The variable is an integer being the maximal amount of state that we
save at the same time (typically 1000). I'm not sure which int value
is equal to "true", but I guess it's not 1000.
Martin Quinson [Mon, 10 Apr 2017 23:21:38 +0000 (01:21 +0200)]
typos
Martin Quinson [Mon, 10 Apr 2017 22:21:34 +0000 (00:21 +0200)]
better handling of the MC debug on unknown simcalls
degomme [Mon, 10 Apr 2017 16:28:54 +0000 (18:28 +0200)]
Due to recent "improvements" in the smpi mpich rma tests, netbsd timeouts. Avoid this.
degomme [Mon, 10 Apr 2017 16:06:17 +0000 (18:06 +0200)]
Add two new mpich tests.
Taken from https://github.com/pmodels/mpich/commit/
c4d2d5b88d530f1f8d3a74fe725b8fa30f57137e and https://github.com/pmodels/mpich/commit/
cc7f28a1442f8bf731e174df334ab5e770fc88b9 .
They are rather simple and useless for us, but I'm tracking all changes to mpich test tree now to ease updates.
Christian Heinrich [Fri, 7 Apr 2017 12:20:18 +0000 (14:20 +0200)]
[SMX] Fixed wrong termination timing when daemons are present
The check whether daemons need to be killed because no real process is present
anymore needs to be performed every time a process could potentially have exited.
Martin Quinson [Fri, 7 Apr 2017 08:56:43 +0000 (10:56 +0200)]
fix a dead link
Martin Quinson [Thu, 6 Apr 2017 14:57:19 +0000 (16:57 +0200)]
be more robust to plugins creating processes a bit early
degomme [Thu, 6 Apr 2017 09:03:23 +0000 (11:03 +0200)]
try to appease ubsan a bit
Frederic Suter [Thu, 6 Apr 2017 06:47:09 +0000 (08:47 +0200)]
Merge branch 'master' of git+ssh://scm.gforge.inria.fr//gitroot/simgrid/simgrid
Frederic Suter [Thu, 6 Apr 2017 06:46:36 +0000 (08:46 +0200)]
plug some leak by delete last message before leaving
degomme [Wed, 5 Apr 2017 22:39:03 +0000 (00:39 +0200)]
activate some more rma tests
degomme [Wed, 5 Apr 2017 22:28:10 +0000 (00:28 +0200)]
cornercases handling
Gabriel Corona [Wed, 5 Apr 2017 22:06:49 +0000 (00:06 +0200)]
[mc] Install simgrid-mc
Frederic Suter [Wed, 5 Apr 2017 18:25:53 +0000 (20:25 +0200)]
try to please MC
Frederic Suter [Wed, 5 Apr 2017 16:56:53 +0000 (18:56 +0200)]
new s4u example: chord
leaking a lot, not perfect, but hopefully better than the C version
degomme [Wed, 5 Apr 2017 16:24:20 +0000 (18:24 +0200)]
Super lazily avoid globals in this test.
degomme [Wed, 5 Apr 2017 15:06:25 +0000 (17:06 +0200)]
Activate some more tests.
We almost have everything in MPI3 RMAs, now.
I will try to activate more tests soon.
degomme [Wed, 5 Apr 2017 15:05:22 +0000 (17:05 +0200)]
Potential crash --
degomme [Wed, 5 Apr 2017 15:05:08 +0000 (17:05 +0200)]
Have lock and unlock actually behave like they should (terminating both remote and local requests)
degomme [Wed, 5 Apr 2017 15:04:19 +0000 (17:04 +0200)]
Potential crash --
degomme [Wed, 5 Apr 2017 15:04:05 +0000 (17:04 +0200)]
Actually we can have locks and fences (erk) in the same program.
degomme [Wed, 5 Apr 2017 15:03:26 +0000 (17:03 +0200)]
Add MPI_Rput, Rget, Raccumulate and Rget_accumulate calls.
They return the request for separate termination instead of putting it in the queue.
degomme [Wed, 5 Apr 2017 15:01:59 +0000 (17:01 +0200)]
Add MPI_Compare_and_swap (normally atomic) call
degomme [Wed, 5 Apr 2017 13:11:14 +0000 (15:11 +0200)]
We need this order of parameters, as the user may want to override the default privatization command.
Christian Heinrich [Wed, 5 Apr 2017 12:50:44 +0000 (14:50 +0200)]
Christian Heinrich [Wed, 5 Apr 2017 12:32:55 +0000 (14:32 +0200)]
[PLUGINS] Added HostLoad::getAverageLoad() to the HostLoad plugin
This functions returns the average load since the last reset
degomme [Tue, 4 Apr 2017 08:06:04 +0000 (10:06 +0200)]
activate tests
degomme [Tue, 4 Apr 2017 08:05:58 +0000 (10:05 +0200)]
Add MPI_Win_lock_all MPI_Win_unlock_all MPI_Win_flush MPI_Win_flush_local MPI_Win_flush_all MPI_Win_flush_local_all calls
degomme [Tue, 4 Apr 2017 00:27:50 +0000 (02:27 +0200)]
Oops, this has to be secured by the mutex
degomme [Mon, 3 Apr 2017 12:42:22 +0000 (14:42 +0200)]
Add MPI_Fetch_and_op call, which is just a get_accumulate with specific params
degomme [Mon, 3 Apr 2017 12:41:23 +0000 (14:41 +0200)]
Oops, we need to store a pointer to the simix process in the smpi process now.
degomme [Mon, 3 Apr 2017 12:38:20 +0000 (14:38 +0200)]
No_op has a specific effect, actually, and turns the get_accumulate in a get.
The standard says it may differ a bit internally.. Or not.
degomme [Mon, 3 Apr 2017 12:37:19 +0000 (14:37 +0200)]
We can start the recv requests directly now, and don't have to wait for a synchro.
Doesn't change much as they were async anyway, but it's simpler.
degomme [Mon, 3 Apr 2017 12:36:13 +0000 (14:36 +0200)]
cleanup, and allow an external process to start a MPI request.
degomme [Mon, 3 Apr 2017 12:35:19 +0000 (14:35 +0200)]
close the window if we explicitely tell it's closed
Frederic Suter [Tue, 4 Apr 2017 07:45:28 +0000 (09:45 +0200)]
no fun if you don't break evertyhing 'cos you're dumb...
Frederic Suter [Tue, 4 Apr 2017 07:30:18 +0000 (09:30 +0200)]
add send with timeout and irecv to this_actor API
Frederic Suter [Tue, 4 Apr 2017 07:03:53 +0000 (09:03 +0200)]
fix bug in asynchronous receive
(https://github.com/simgrid/simgrid/issues/144)
degomme [Fri, 31 Mar 2017 13:03:39 +0000 (15:03 +0200)]
Really don't do anything in the copy callback if we are in a shared zone.
Attempt to fix https://github.com/simgrid/simgrid/issues/147
degomme [Fri, 31 Mar 2017 12:01:59 +0000 (14:01 +0200)]
Increase our max limit for tag.. because 1M was low, indeed.
degomme [Fri, 31 Mar 2017 09:45:08 +0000 (11:45 +0200)]
Hide xbt_cfg cruft from smpi tesh files.
We need this as privatization may not be activated on some systems, thus outputting a different thing.
degomme [Fri, 31 Mar 2017 09:44:05 +0000 (11:44 +0200)]
Stop lying. Really activate privatization by default (when supported)
Frederic Suter [Fri, 31 Mar 2017 07:19:00 +0000 (09:19 +0200)]
kill dead code
heterogenous dicts disappeared a while ago
Frederic Suter [Fri, 31 Mar 2017 07:08:08 +0000 (09:08 +0200)]
fix a bunch of recently introduced smells
Frederic Suter [Thu, 30 Mar 2017 14:48:48 +0000 (16:48 +0200)]
another leak plugged
Frederic Suter [Thu, 30 Mar 2017 14:46:32 +0000 (16:46 +0200)]
leak plug
Frederic Suter [Thu, 30 Mar 2017 14:06:53 +0000 (16:06 +0200)]
do not initialize surf for nothing
Frederic Suter [Thu, 30 Mar 2017 12:17:10 +0000 (14:17 +0200)]
attempt to plug all leaks for simdag.
only fill a dynar if you need it, and in this case allocate and free
it yourself
Frederic Suter [Thu, 30 Mar 2017 11:19:56 +0000 (13:19 +0200)]
remove dead code
Frederic Suter [Thu, 30 Mar 2017 10:49:54 +0000 (12:49 +0200)]
modernize two simcalls
and use a third in msg_io
Frederic Suter [Thu, 30 Mar 2017 10:37:10 +0000 (12:37 +0200)]
Revert "Revert "cast once for all at surf level and not in the APIs""
This reverts commit
fa900b346af7c55195fd90c3345cc4107c2afc6d.
Frederic Suter [Thu, 30 Mar 2017 10:35:31 +0000 (12:35 +0200)]
Revert "cast once for all at surf level and not in the APIs"
This reverts commit
14696475e1a6145501ae0f9d6104a82ab2ffa64d.
Frederic Suter [Thu, 30 Mar 2017 10:15:59 +0000 (12:15 +0200)]
cast once for all at surf level and not in the APIs
Frederic Suter [Thu, 30 Mar 2017 08:56:31 +0000 (10:56 +0200)]
Merge branch 'master' of git+ssh://scm.gforge.inria.fr//gitroot/simgrid/simgrid
Frederic Suter [Thu, 30 Mar 2017 08:53:03 +0000 (10:53 +0200)]
complete s4u::Storage interface, what a mess...
not all storages are put in the static storages_ map
degomme [Wed, 29 Mar 2017 20:17:17 +0000 (22:17 +0200)]
Replace linear search by a faster one.
degomme [Wed, 29 Mar 2017 11:08:35 +0000 (13:08 +0200)]
oops
degomme [Wed, 29 Mar 2017 10:51:38 +0000 (12:51 +0200)]
Add a smpi/shared-malloc-blocksize option. This is relevant only when global shared mallocs mode is used.
This allows to change the size of the fake file used (default 1MB), to potentially limit the number of mappings for large runs.
degomme [Wed, 29 Mar 2017 10:49:37 +0000 (12:49 +0200)]
tiny optim
degomme [Wed, 29 Mar 2017 10:30:51 +0000 (12:30 +0200)]
ignore some more memcpys/serialization when a buffer is detected as shared
degomme [Wed, 29 Mar 2017 10:30:10 +0000 (12:30 +0200)]
connect logs
degomme [Tue, 28 Mar 2017 16:59:52 +0000 (18:59 +0200)]
nothing to see here.
degomme [Tue, 28 Mar 2017 16:51:01 +0000 (18:51 +0200)]
ok, let's mess up the includes a bit more.
Frederic Suter [Tue, 28 Mar 2017 15:37:13 +0000 (17:37 +0200)]
Merge branch 'master' of git+ssh://scm.gforge.inria.fr//gitroot/simgrid/simgrid
Frederic Suter [Tue, 28 Mar 2017 15:36:44 +0000 (17:36 +0200)]
some compilers don't like unused variables
degomme [Tue, 28 Mar 2017 11:55:37 +0000 (13:55 +0200)]
Split up smpi_bench.cpp .
Privatization is sent to smpi_memory.cpp (will be renamed later).
smpi_shared* functions are sent to smpi_shared.cpp.
I don't know what to do with trace_call_location yet. bench should only be timing related functions, smpi_sample* and sleep stuff.
Frederic Suter [Tue, 28 Mar 2017 15:28:40 +0000 (17:28 +0200)]
I think I just killed a simcall
Frederic Suter [Tue, 28 Mar 2017 15:26:33 +0000 (17:26 +0200)]
leak-- smell++ (C cast to avoid const_cast)
Frederic Suter [Tue, 28 Mar 2017 15:12:28 +0000 (17:12 +0200)]
pleasing sonar was a bad idea there
Frederic Suter [Tue, 28 Mar 2017 11:56:06 +0000 (13:56 +0200)]
test conversion
not completely satisfying, but storage is a bit messy
Frederic Suter [Tue, 28 Mar 2017 08:48:22 +0000 (10:48 +0200)]
leak plug and test AS properties
Frederic Suter [Tue, 28 Mar 2017 08:01:56 +0000 (10:01 +0200)]
typos and leak
Lucas M. Schnorr [Tue, 28 Mar 2017 07:39:28 +0000 (09:39 +0200)]
tracing TODO marks
Martin Quinson [Tue, 28 Mar 2017 07:04:45 +0000 (09:04 +0200)]
please sonar: this datatype is not used anymore
Martin Quinson [Sun, 26 Mar 2017 02:03:08 +0000 (04:03 +0200)]
s4u: fix documentation
degomme [Tue, 28 Mar 2017 06:25:46 +0000 (08:25 +0200)]
2 a.m. commits are always a good way to fail.
degomme [Tue, 28 Mar 2017 00:05:53 +0000 (02:05 +0200)]
Mainly on 32 bits systems, target_disp can be seen as negative in one case.
This is due to the fact that in this case, the disp is actually an address, which is stored in a signed type.
We can't do much about that, and all MPI libraries I checked have this particular check disabled for this case.
degomme [Mon, 27 Mar 2017 21:19:51 +0000 (23:19 +0200)]
Add an smpi_is_shared call, and ignore memcpys in SMPI copy callback when buffers seem to be in shared memory.
For the global case this means we need to store the metadata as it was done in the local case
degomme [Sun, 26 Mar 2017 20:48:14 +0000 (22:48 +0200)]
support MPI_Op_commutative call, as it was already implemented internally
degomme [Sun, 26 Mar 2017 20:47:41 +0000 (22:47 +0200)]
activate some more tests
degomme [Fri, 24 Mar 2017 14:52:34 +0000 (15:52 +0100)]
Add MPI_Win_allocate, MPI_Win_create_dynamic, MPI_Win_attach, MPI_Win_detach, MPI_Win_set_info, MPI_Win_get_info.
Frederic Suter [Mon, 27 Mar 2017 18:27:59 +0000 (20:27 +0200)]
plug leaks
Frederic Suter [Mon, 27 Mar 2017 14:36:08 +0000 (16:36 +0200)]
finish to plug ptaskL07 leak
network and cpu models share the same lmm as the host model. the host
model frees the lmm, but both sub models have to set their local
reference to nullptr before the parent destructors are called to
prevent double free/delete.