Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
simgrid.git
6 years agoFix MPI_Comm_Split, to increment reference counter of the group, as it is shared...
Augustin Degomme [Mon, 8 Jul 2013 12:28:42 +0000 (14:28 +0200)]
Fix MPI_Comm_Split, to increment reference counter of the group, as it is shared between processes

6 years agoadd a reference counter in order to avoide deleting MPI_Comm, and MPI_Group, too...
Augustin Degomme [Mon, 8 Jul 2013 12:28:07 +0000 (14:28 +0200)]
add a reference counter in order to avoide deleting MPI_Comm, and MPI_Group, too soon

6 years agohave persistent communications use a PREPARED flag to implement the fact that they...
Augustin Degomme [Mon, 8 Jul 2013 12:22:31 +0000 (14:22 +0200)]
have persistent communications use a PREPARED flag to implement the fact that they have not already been sent
This is useful when testing their status before launch, to have an empty MPI_Status returned, as specified

6 years agoMake internal collective flags negative (incorrect in MPI), to avoid confusion with...
Augustin Degomme [Mon, 8 Jul 2013 12:18:56 +0000 (14:18 +0200)]
Make internal collective flags negative (incorrect in MPI), to avoid confusion with user's messages
This was seen when user's code was using MPI_ANY_TAG to receive any message (from the userland code), and lead to confusions.

6 years agofix tracing issue with some communicators
Augustin Degomme [Fri, 5 Jul 2013 15:42:00 +0000 (17:42 +0200)]
fix tracing issue with some communicators

6 years agoMake distcheck succeed.
Arnaud Giersch [Fri, 5 Jul 2013 08:06:31 +0000 (10:06 +0200)]
Make distcheck succeed.

6 years agoUse the same parameter name in the code and in the documentation.
Arnaud Giersch [Fri, 5 Jul 2013 08:04:16 +0000 (10:04 +0200)]
Use the same parameter name in the code and in the documentation.

6 years agoFix paths for msg/energy tests.
Arnaud Giersch [Fri, 5 Jul 2013 07:55:52 +0000 (09:55 +0200)]
Fix paths for msg/energy tests.

6 years agoMake it build again after commit 152117af2fbcf4c99d0da5d10a6813aac7eee007
Arnaud Giersch [Fri, 5 Jul 2013 07:46:12 +0000 (09:46 +0200)]
Make it build again after commit 152117af2fbcf4c99d0da5d10a6813aac7eee007
(added dvfs support).

6 years agoadd one missing test
Augustin Degomme [Wed, 3 Jul 2013 13:30:55 +0000 (15:30 +0200)]
add one missing test

6 years agofix dist build
Augustin Degomme [Fri, 5 Jul 2013 07:56:20 +0000 (09:56 +0200)]
fix dist build

6 years agoMerge branch 'dvfs'
acarpena [Fri, 5 Jul 2013 01:53:41 +0000 (01:53 +0000)]
Merge branch 'dvfs'

Added DVFS and energy consumption support.

6 years agoadded tesh tests for DVFS
acarpena [Thu, 4 Jul 2013 23:42:07 +0000 (16:42 -0700)]
added tesh tests for DVFS

6 years agoadded tesh tests for dvfs
acarpena [Thu, 4 Jul 2013 20:19:31 +0000 (13:19 -0700)]
added tesh tests for dvfs

6 years agoadd MSG_host_get_process_list() function.
suter [Wed, 3 Jul 2013 13:45:23 +0000 (15:45 +0200)]
add MSG_host_get_process_list() function.

6 years agocosmetics
suter [Wed, 3 Jul 2013 13:17:13 +0000 (15:17 +0200)]
cosmetics

6 years agopushing but never popping means useless code
suter [Mon, 1 Jul 2013 14:19:49 +0000 (16:19 +0200)]
pushing but never popping means useless code
remove blanks in empty lines

6 years agoThis used to work by accident
Augustin Degomme [Tue, 2 Jul 2013 16:15:56 +0000 (18:15 +0200)]
This used to work by accident

6 years agobetter handling of loop endings
Augustin Degomme [Tue, 2 Jul 2013 16:15:06 +0000 (18:15 +0200)]
better handling of loop endings
The upper bound may not be inside the set,
or the stride can be negative, with upper bound < lower bound

6 years agoget the rank and not the index for process creation
Augustin Degomme [Tue, 2 Jul 2013 16:10:36 +0000 (18:10 +0200)]
get the rank and not the index for process creation

6 years agouse the right index for sendrecv optimization
Augustin Degomme [Tue, 2 Jul 2013 15:34:56 +0000 (17:34 +0200)]
use the right index for sendrecv optimization

6 years agounify collective tags
Augustin Degomme [Tue, 2 Jul 2013 07:51:05 +0000 (09:51 +0200)]
unify collective tags
allows to quickly identify the type of the collective when debugging
avoid some collisions with wrong messages being received

6 years agohandle MPI_IN_PLACE for basic reduce
Augustin Degomme [Mon, 1 Jul 2013 15:54:34 +0000 (17:54 +0200)]
handle MPI_IN_PLACE for basic reduce

6 years agodo not create communicator if process is not inside its group
Augustin Degomme [Mon, 1 Jul 2013 10:12:17 +0000 (12:12 +0200)]
do not create communicator if process is not inside its group

6 years agoMerge branch 'dvfs'
acarpena [Mon, 1 Jul 2013 00:36:14 +0000 (17:36 -0700)]
Merge branch 'dvfs'

Conflicts:
buildtools/Cmake/MakeExe.cmake
src/include/surf/surf.h
src/surf/simgrid_dtd.c

6 years agoadded dvfs support
acarpena [Mon, 1 Jul 2013 00:04:23 +0000 (17:04 -0700)]
added dvfs support

6 years agoadded dvfs support
acarpena [Sun, 30 Jun 2013 23:58:27 +0000 (16:58 -0700)]
added dvfs support

6 years agoChangelog update, more to come
Augustin Degomme [Fri, 28 Jun 2013 13:53:36 +0000 (15:53 +0200)]
Changelog update, more to come

6 years agomake not_yet_implemented() a macro outputting the name of the function
Augustin Degomme [Fri, 28 Jun 2013 12:40:11 +0000 (14:40 +0200)]
make not_yet_implemented() a macro outputting the name of the function
Also tell the user to contact the team if he wants the support for the function.

6 years agoAdd a huge variety of not yet impelmented functions, to compile MPICH3 test suite...
Augustin Degomme [Fri, 28 Jun 2013 11:32:17 +0000 (13:32 +0200)]
Add a huge variety of not yet impelmented functions, to compile MPICH3 test suite (without RMA or fortran)

6 years agoTest -eq is for numerical values.
Arnaud Giersch [Thu, 27 Jun 2013 12:10:52 +0000 (14:10 +0200)]
Test -eq is for numerical values.

Use construct "Use Alternate Value" to achieve the desired effect.

6 years agooptimize MPI_Sendrecv if a process has to exchange with himself only
Augustin Degomme [Wed, 26 Jun 2013 15:01:38 +0000 (17:01 +0200)]
optimize MPI_Sendrecv if a process has to exchange with himself only
This helps avoiding some lockings of lazy network model when one process waits for itself

6 years agoavoid breaking replay, handle MPI_OP_NULL case
Augustin Degomme [Wed, 26 Jun 2013 14:35:53 +0000 (16:35 +0200)]
avoid breaking replay, handle MPI_OP_NULL case

6 years agoFix smpi_sleep to execute the sleep on the simulated host (ignoring running host...
Augustin Degomme [Wed, 26 Jun 2013 11:58:41 +0000 (13:58 +0200)]
Fix smpi_sleep to execute the sleep on the simulated host (ignoring running host power)

6 years agohandle non commutative reduce operations
Augustin Degomme [Wed, 26 Jun 2013 09:14:04 +0000 (11:14 +0200)]
handle non commutative reduce operations

6 years agotry again and again to be more portable with this test
Augustin Degomme [Tue, 25 Jun 2013 15:57:46 +0000 (17:57 +0200)]
try again and again to be more portable with this test

6 years agosame as before, to make ci happy
Augustin Degomme [Tue, 25 Jun 2013 15:01:48 +0000 (17:01 +0200)]
same as before, to make ci happy

6 years agotry again to be 32/64 bit agnostic
Augustin Degomme [Tue, 25 Jun 2013 14:51:50 +0000 (16:51 +0200)]
try again to be 32/64 bit agnostic

6 years agomodify reduce test to provide better coverage
Augustin Degomme [Tue, 25 Jun 2013 13:04:34 +0000 (15:04 +0200)]
modify reduce test to provide better coverage

6 years agotry to make the shared test compatible with 32bit architectures
Augustin Degomme [Tue, 25 Jun 2013 12:31:53 +0000 (14:31 +0200)]
try to make the shared test compatible with 32bit architectures

6 years agompich allgatherv was using wrong size value
Augustin Degomme [Tue, 25 Jun 2013 11:17:49 +0000 (13:17 +0200)]
mpich allgatherv was using wrong size value

6 years agomem leaks
Augustin Degomme [Tue, 25 Jun 2013 09:34:29 +0000 (11:34 +0200)]
mem leaks

6 years agoAdd a test for SMPI shared malloc and free
Augustin Degomme [Tue, 25 Jun 2013 09:34:18 +0000 (11:34 +0200)]
Add a test for SMPI shared malloc and free

6 years agounits for bandwidth properties too
suter [Tue, 25 Jun 2013 10:47:26 +0000 (12:47 +0200)]
units for bandwidth properties too

6 years agoThese "properties" are actually bandwidths: allow users to declare units
suter [Tue, 25 Jun 2013 10:44:40 +0000 (12:44 +0200)]
These "properties" are actually bandwidths: allow users to declare units
in the XML file

6 years agostart to clarify this example
suter [Tue, 25 Jun 2013 10:37:45 +0000 (12:37 +0200)]
start to clarify this example

6 years agoremove an ugly pimple thanks to units in XML
suter [Tue, 25 Jun 2013 10:32:17 +0000 (12:32 +0200)]
remove an ugly pimple thanks to units in XML

6 years agoallow to use units for storage too (bits, or bytes, or information bits
suter [Tue, 25 Jun 2013 10:25:09 +0000 (12:25 +0200)]
allow to use units for storage too (bits, or bytes, or information bits
or bytes, multiplied up to the Tera)

6 years agoseparate allreduce tests, with one large and the other small, to avoid to lose too...
Augustin Degomme [Mon, 24 Jun 2013 13:43:39 +0000 (15:43 +0200)]
separate allreduce tests, with one large and the other small, to avoid to lose too much time in testing, and triggering timeouts.

6 years agofix two nasty bugs in complex datatypes handling, caught with test suite.
Augustin Degomme [Mon, 24 Jun 2013 12:49:51 +0000 (14:49 +0200)]
fix two nasty bugs in complex datatypes handling, caught with test suite.

6 years agomem leak
Augustin Degomme [Mon, 24 Jun 2013 12:49:04 +0000 (14:49 +0200)]
mem leak

6 years agopass collective selectors through mpich-coll test suite
Augustin Degomme [Mon, 24 Jun 2013 12:48:56 +0000 (14:48 +0200)]
pass collective selectors through mpich-coll test suite

6 years agostar-mpi's pairwise alltoall is only valid for power of 2 cases, add openmpi's one...
Augustin Degomme [Thu, 20 Jun 2013 08:11:46 +0000 (10:11 +0200)]
star-mpi's pairwise alltoall is only valid for power of 2 cases, add openmpi's one instead

6 years agocorrect a few bcast algorithms and change the test to actually use them all
Augustin Degomme [Wed, 19 Jun 2013 15:08:45 +0000 (17:08 +0200)]
correct a few bcast algorithms and change the test to actually use them all

6 years agohave allreduce test work on much larger arrays to trigger segmentation in one algorithm
Augustin Degomme [Wed, 19 Jun 2013 12:31:15 +0000 (14:31 +0200)]
have allreduce test work on much larger arrays to trigger segmentation in one algorithm

6 years agoompi selector was using the wrong algo for alltoallv, add the right one and use it
Augustin Degomme [Wed, 19 Jun 2013 11:08:42 +0000 (13:08 +0200)]
ompi selector was using the wrong algo for alltoallv, add the right one and use it

6 years agoincrease tesh timeout for replay
Augustin Degomme [Wed, 19 Jun 2013 09:03:54 +0000 (11:03 +0200)]
increase tesh timeout for replay

6 years agoadd forgotten test for an mpich algorithm
Augustin Degomme [Wed, 19 Jun 2013 09:03:26 +0000 (11:03 +0200)]
add forgotten test for an mpich algorithm

6 years agoapply some coding standard in function naming (agier has to be in
suter [Tue, 18 Jun 2013 13:28:34 +0000 (15:28 +0200)]
apply some coding standard in function naming (agier has to be in
vacations ... ;))

6 years agodon't call things storage when it's mount
suter [Tue, 18 Jun 2013 13:18:26 +0000 (15:18 +0200)]
don't call things storage when it's mount

6 years agodon't call things storage when it's mount
suter [Tue, 18 Jun 2013 13:15:57 +0000 (15:15 +0200)]
don't call things storage when it's mount

6 years agoThis is a bug in the implementation in OpenMPI, with OpenMPI accidentally working.
Augustin Degomme [Tue, 18 Jun 2013 12:02:26 +0000 (14:02 +0200)]
This is a bug in the implementation in OpenMPI, with OpenMPI accidentally working.
That's because MPI_ANY_SOURCE is also -1 in OpenMPI.

6 years agotest for barrier algorithms was testing scatter ...
Augustin Degomme [Tue, 18 Jun 2013 11:57:16 +0000 (13:57 +0200)]
test for barrier algorithms was testing scatter ...

6 years agochange some defined values to less current ones
Augustin Degomme [Tue, 18 Jun 2013 11:38:04 +0000 (13:38 +0200)]
change some defined values to less current ones

6 years agoMPI_ERR_COMM was sometimes returned instead of MPI_ERR_RANK in SMPI
Augustin Degomme [Tue, 18 Jun 2013 11:16:55 +0000 (13:16 +0200)]
MPI_ERR_COMM was sometimes returned instead of MPI_ERR_RANK in SMPI

6 years agoreduce_scatter default implementation was a mess
Augustin Degomme [Mon, 17 Jun 2013 16:13:53 +0000 (18:13 +0200)]
reduce_scatter default implementation was a mess

6 years agoif we send 0 data, don't bother with subtypes
Augustin Degomme [Mon, 17 Jun 2013 16:13:26 +0000 (18:13 +0200)]
if we send 0 data, don't bother with subtypes

6 years agoadd option "smpi/use_shared_malloc", with true as default, to allow disabling the...
Augustin Degomme [Mon, 17 Jun 2013 16:12:28 +0000 (18:12 +0200)]
add option "smpi/use_shared_malloc", with true as default, to allow disabling the shared memory SMPI stuff dynamically
Useful for debug purposes, to see if the shared memory is the cause of a bug

6 years agoyet another simplification of the storage API: we don't manage
suter [Tue, 18 Jun 2013 11:17:12 +0000 (13:17 +0200)]
yet another simplification of the storage API: we don't manage
file contents, no need for a buffer as parameter ...

6 years agosize shouldn't be accessible from the user API.
suter [Tue, 18 Jun 2013 10:53:14 +0000 (12:53 +0200)]
size shouldn't be accessible from the user API.

6 years agoFix of pastry
Paul Bédaride [Mon, 17 Jun 2013 09:07:18 +0000 (11:07 +0200)]
Fix of pastry

6 years agoAdd pastry msg example (First attempt)
Paul Bédaride [Fri, 14 Jun 2013 16:19:46 +0000 (18:19 +0200)]
Add pastry msg example (First attempt)

6 years agoChange xbt_os_timer interface to handle cpu time, thread time and
Guillaume Serrière [Fri, 14 Jun 2013 13:02:19 +0000 (15:02 +0200)]
Change xbt_os_timer interface to handle cpu time, thread time and
wall time distinctly.
Windows section isn't test.

Signed-off-by: Guillaume Serrière <guillaume.serriere@esial.net>
6 years agoUse HEAD as default reference for git archive.
Arnaud Giersch [Fri, 14 Jun 2013 10:58:39 +0000 (12:58 +0200)]
Use HEAD as default reference for git archive.

The default branch may be undefined.

6 years agoAdd a small value to avoid rounding differences between x86 and x86_64.
Arnaud Giersch [Fri, 14 Jun 2013 09:25:47 +0000 (11:25 +0200)]
Add a small value to avoid rounding differences between x86 and x86_64.

6 years agoChange type double for size_t.
Arnaud Giersch [Fri, 14 Jun 2013 09:25:16 +0000 (11:25 +0200)]
Change type double for size_t.

6 years agoRemove unused variables.
Arnaud Giersch [Fri, 14 Jun 2013 07:42:45 +0000 (09:42 +0200)]
Remove unused variables.

6 years agoComment out useless operation.
Arnaud Giersch [Fri, 14 Jun 2013 07:31:02 +0000 (09:31 +0200)]
Comment out useless operation.

6 years agoAdd missing dependency.
Arnaud Giersch [Thu, 13 Jun 2013 14:11:17 +0000 (16:11 +0200)]
Add missing dependency.

6 years agowrong file name in cmake file
Augustin Degomme [Fri, 14 Jun 2013 07:06:23 +0000 (09:06 +0200)]
wrong file name in cmake file

6 years agoadd missing prototype
Augustin Degomme [Fri, 14 Jun 2013 07:02:11 +0000 (09:02 +0200)]
add missing prototype

6 years agoopenmpi recently changed its decision for alltoallv algo
Augustin Degomme [Fri, 14 Jun 2013 06:57:24 +0000 (08:57 +0200)]
openmpi recently changed its decision for alltoallv algo

6 years agoadd missing replay files to dist
Augustin Degomme [Wed, 12 Jun 2013 12:34:41 +0000 (14:34 +0200)]
add missing replay files to dist

6 years agoforgot one file
degomme [Thu, 13 Jun 2013 22:56:09 +0000 (00:56 +0200)]
forgot one file

6 years agowe now have an mpich selector, and a bunch of new algorithms (--cfg=smpi/coll_selecto...
degomme [Thu, 13 Jun 2013 22:51:18 +0000 (00:51 +0200)]
we now have an mpich selector, and a bunch of new algorithms (--cfg=smpi/coll_selector:mpich)

6 years agoMerge branch 'master' of git+ssh://scm.gforge.inria.fr//gitroot/simgrid/simgrid
Jonathan Rouzaud-Cornabas [Thu, 13 Jun 2013 13:52:57 +0000 (15:52 +0200)]
Merge branch 'master' of git+ssh://scm.gforge.inria.fr//gitroot/simgrid/simgrid

6 years agoSIMIX call + MSG function + Java binding for get_core
Jonathan Rouzaud-Cornabas [Thu, 13 Jun 2013 13:52:48 +0000 (15:52 +0200)]
SIMIX call + MSG function + Java binding for get_core

6 years agomodel-checker : restore while statement for bugged1_liveness example
Marion Guthmuller [Thu, 13 Jun 2013 13:00:53 +0000 (15:00 +0200)]
model-checker : restore while statement for bugged1_liveness example

6 years agoMerge branch 'master' of git+ssh://scm.gforge.inria.fr//gitroot/simgrid/simgrid
Jonathan Rouzaud-Cornabas [Thu, 13 Jun 2013 12:50:23 +0000 (14:50 +0200)]
Merge branch 'master' of git+ssh://scm.gforge.inria.fr//gitroot/simgrid/simgrid

6 years agoProperties available for cluster tag
Jonathan Rouzaud-Cornabas [Thu, 13 Jun 2013 12:49:59 +0000 (14:49 +0200)]
Properties available for cluster tag

6 years agomodel-checker : update bugged1 and bugged2 tesh (different compilation flags)
Marion Guthmuller [Thu, 13 Jun 2013 10:45:47 +0000 (12:45 +0200)]
model-checker : update bugged1 and bugged2 tesh (different compilation flags)

6 years agomodel-checker : MC doesn't work with thread context
Marion Guthmuller [Thu, 13 Jun 2013 10:44:45 +0000 (12:44 +0200)]
model-checker : MC doesn't work with thread context

6 years agodisable optimizations for build MC in jenkins
Marion Guthmuller [Thu, 13 Jun 2013 10:30:47 +0000 (12:30 +0200)]
disable optimizations for build MC in jenkins

6 years agounused variable if tracing disabled
Marion Guthmuller [Thu, 13 Jun 2013 10:29:40 +0000 (12:29 +0200)]
unused variable if tracing disabled

6 years agomodel-checker : increase timeout in tesh
Marion Guthmuller [Thu, 13 Jun 2013 10:15:31 +0000 (12:15 +0200)]
model-checker : increase timeout in tesh

6 years agomodel-checker : missing #ifdef
Marion Guthmuller [Thu, 13 Jun 2013 10:08:21 +0000 (12:08 +0200)]
model-checker : missing #ifdef

6 years agomodel-checker : update bugged2 tesh
Marion Guthmuller [Thu, 13 Jun 2013 09:38:02 +0000 (11:38 +0200)]
model-checker : update bugged2 tesh

6 years agomodel-checker : fix (again ..) DPOR for iSend/iRecv and Wait communications
Marion Guthmuller [Thu, 13 Jun 2013 07:21:52 +0000 (09:21 +0200)]
model-checker : fix (again ..) DPOR for iSend/iRecv and Wait communications

- add rdv_cpy in s_smx_action_t structure. comm.rdv is used for
  garbage collection and set to NULL when the communication matches
  with another communication already pushed and waiting in the
  mailbox.
- iSend/iRecv and Wait communications are independant if the
  communications are not in the same mailbox. Otherwise, the
  iSend/iRecv communication process issuer must be different of the
  src_proc or dst_proc of the Wait communication, or, src_buff and
  dst_buff of the communications must be different.

6 years agomodel-checker : improve debug messages
Marion Guthmuller [Wed, 12 Jun 2013 16:30:23 +0000 (18:30 +0200)]
model-checker : improve debug messages

6 years agobarrier tests
degomme [Thu, 13 Jun 2013 09:36:43 +0000 (11:36 +0200)]
barrier tests