Mark Stillwell [Sun, 3 Jul 2011 14:23:15 +0000 (16:23 +0200)]
fixed several problems in gather, gatherv, allgather, allgatherv, and scan
Mark Stillwell [Fri, 1 Jul 2011 07:32:15 +0000 (09:32 +0200)]
removed check that sent/received requests are for the same size message.
Pierre-Nicolas Clauss [Thu, 9 Jun 2011 08:37:42 +0000 (10:37 +0200)]
Merge back master branch
Pierre-Nicolas Clauss [Thu, 9 Jun 2011 08:35:58 +0000 (10:35 +0200)]
Revert "Use private mapping (we already have early unlinking)."
Apparently, this blows out memory.
This reverts commit
b3f59725051f14d38fc6d9f63474bda6c207830e.
Pierre-Nicolas Clauss [Wed, 8 Jun 2011 19:57:31 +0000 (21:57 +0200)]
Use private mapping (we already have early unlinking).
Pierre-Nicolas Clauss [Wed, 8 Jun 2011 14:13:19 +0000 (16:13 +0200)]
Use shared segments to implement SHARED_MALLOC.
Mark Stillwell [Thu, 26 May 2011 14:13:43 +0000 (16:13 +0200)]
tree_bcast and antibcast don't actually use the root parameter, so I removed it.
Mark Stillwell [Thu, 26 May 2011 08:39:56 +0000 (10:39 +0200)]
changed match_recv and match_send...They were requiring matching communicators,
which was causing communication to fail after MPI_comm_dup. Real mpi only tests for
matching size, src, tag.
Mark Stillwell [Thu, 26 May 2011 07:36:51 +0000 (09:36 +0200)]
fixed build_tree to enable MPI broadcasts with root nodes other than 0.
Martin Quinson [Wed, 25 May 2011 15:57:17 +0000 (17:57 +0200)]
Let's assume that Hurd systems provide working MMAP services
Martin Quinson [Tue, 24 May 2011 20:04:27 +0000 (22:04 +0200)]
light update to the FAQ: we're not using autotools anymore
Martin Quinson [Tue, 24 May 2011 15:18:12 +0000 (17:18 +0200)]
Get ready for RC2, to take last minute change from Pedro
Lucas Schnorr [Tue, 24 May 2011 15:01:55 +0000 (17:01 +0200)]
updating NEWS about tracing
Lucas Schnorr [Tue, 24 May 2011 14:54:01 +0000 (16:54 +0200)]
[trace] to proceed, check if platform is traced in functions that need it
Pedro Velho [Tue, 24 May 2011 14:21:36 +0000 (16:21 +0200)]
Cope with new way of destroying actions, now if the action is destroyed assume that it finished and the remain is zero.
Pedro Velho [Tue, 24 May 2011 13:17:28 +0000 (15:17 +0200)]
Bug fix. Cope with the new way actions are handled.
Martin Quinson [Tue, 24 May 2011 14:21:58 +0000 (16:21 +0200)]
get ready for 3.6-rc1
Martin Quinson [Tue, 24 May 2011 14:14:11 +0000 (16:14 +0200)]
typo
Martin Quinson [Tue, 24 May 2011 14:05:38 +0000 (16:05 +0200)]
merge
Martin Quinson [Tue, 24 May 2011 14:02:11 +0000 (16:02 +0200)]
Java and Ruby were killed away
Pierre-Nicolas Clauss [Tue, 24 May 2011 13:49:03 +0000 (15:49 +0200)]
Added changes for SMPI.
Martin Quinson [Tue, 24 May 2011 13:42:10 +0000 (15:42 +0200)]
proposition of NEWS content for the 3.6 release
Arnaud Giersch [Fri, 20 May 2011 13:59:20 +0000 (15:59 +0200)]
Regenerate files in maintainer mode.
Arnaud Giersch [Fri, 20 May 2011 13:57:01 +0000 (15:57 +0200)]
Remove variable parts from flexml generated files (maintainer mode).
That is:
- remove the comment about generated date;
- use a relative path when generating dax_dtd.l.
The purpose of this patch is to avoid to generate files with the only
difference being a comment line.
Lucas Schnorr [Tue, 24 May 2011 12:37:14 +0000 (14:37 +0200)]
[trace] remove old source file with functions that are no longer used
Lucas Schnorr [Tue, 24 May 2011 12:29:45 +0000 (14:29 +0200)]
[trace] cosmetics, remove TRACE_is_active, use TRACE_is_enabled instead
Lucas Schnorr [Tue, 24 May 2011 12:12:18 +0000 (14:12 +0200)]
[trace] graphicator needs the platform to creata a graph of it
details:
- users should specify --cfg=tracing/platform:1 from now on
Lucas Schnorr [Tue, 24 May 2011 12:01:32 +0000 (14:01 +0200)]
[trace] adding a new boolean parameter (tracing/platform) to register platform in the trace file
Lucas Schnorr [Tue, 24 May 2011 11:54:38 +0000 (13:54 +0200)]
[trace] updating tracing-related section of the FAQ
Lucas Schnorr [Tue, 24 May 2011 10:51:27 +0000 (12:51 +0200)]
[trace] interface cosmetics: removing pre-processor defines, substitute by function calls
Lucas Schnorr [Tue, 24 May 2011 09:19:11 +0000 (11:19 +0200)]
[trace] adding new way of set user variables for links, by providing src and dst
details:
- instrumentation gets the routes between src and dst
and set the variable for all the links of the route
Lucas Schnorr [Tue, 24 May 2011 08:50:25 +0000 (10:50 +0200)]
[trace] cosmetics, cleaner public include tracing header
Lucas Schnorr [Mon, 23 May 2011 15:09:29 +0000 (17:09 +0200)]
fix compilation error if gtnets is to be compiled without tracing
Lucas Schnorr [Mon, 23 May 2011 14:59:45 +0000 (16:59 +0200)]
[trace] fix tests: tracing-tasks and smpi-tracing-ptp
details:
- command line parameters for tracing smpi have changed
Lucas Schnorr [Wed, 18 May 2011 09:09:24 +0000 (11:09 +0200)]
[trace] bugfix, parameters were wrong
Lucas Schnorr [Mon, 2 May 2011 10:34:53 +0000 (12:34 +0200)]
[trace] new smpi tracing options
details:
- to activate the traditional gantt-chart visualization, pass -trace
as parameter. Then, the tracefile can be visualized with Paje using
$ Paje smpi_simgrid.trace
- to activate the graph-view and resource utilization, pass
-trace -trace-triva -trace-resource as parameters. Then, the
tracefile can be visualization with Triva using
$ Triva --graph --gc_conf smpi_uncat.plist smpi_simgrid.trace
- other options include:
- group mpi processes by where they are executed: -trace-grouped
- specify a different trace filename: -trace-file <tracefile>
Lucas Schnorr [Fri, 29 Apr 2011 13:44:12 +0000 (15:44 +0200)]
[trace] default colors for mpi states
Lucas Schnorr [Thu, 28 Apr 2011 15:19:58 +0000 (17:19 +0200)]
[trace] updating tesh's to follow recent changes in tracing system
Lucas Schnorr [Thu, 28 Apr 2011 15:19:38 +0000 (17:19 +0200)]
[trace] msg process and tasks tracing also needs platform
Lucas Schnorr [Thu, 28 Apr 2011 14:37:35 +0000 (16:37 +0200)]
[trace] removing volume case study, deprecated at 3.6
Lucas Schnorr [Thu, 28 Apr 2011 14:15:19 +0000 (16:15 +0200)]
[trace] smpi grouped trace implies user wants to trace smpi as a whole
Lucas Schnorr [Thu, 28 Apr 2011 13:59:53 +0000 (15:59 +0200)]
[trace] let the tracing system (and not the instrumentation interface) decide what to trace
Lucas Schnorr [Thu, 28 Apr 2011 13:59:10 +0000 (15:59 +0200)]
[trace] only need to declare resource utilization types if categorized/uncategorized are selected by user
Lucas Schnorr [Thu, 28 Apr 2011 13:44:26 +0000 (15:44 +0200)]
[trace] isolating buffer dumping from the simulation, only dump if trace is enabled
Lucas Schnorr [Thu, 28 Apr 2011 13:41:59 +0000 (15:41 +0200)]
[trace] controlling the callbacks and push/pop of containers based on TRACE_needs_platform
Lucas Schnorr [Thu, 28 Apr 2011 13:24:03 +0000 (15:24 +0200)]
[trace] new function returns true if user options need the platform structure
Lucas Schnorr [Thu, 28 Apr 2011 12:02:24 +0000 (14:02 +0200)]
correcting mode of a perl script to be executable
Christophe Thiéry [Thu, 19 May 2011 13:49:14 +0000 (15:49 +0200)]
Let the user set the latency with the constant network model
Pedro Velho [Thu, 19 May 2011 12:32:10 +0000 (14:32 +0200)]
Added an optinal timer for those who want to sample the remaining throughout execution.
Pedro Velho [Thu, 19 May 2011 11:50:09 +0000 (13:50 +0200)]
Shut down the gtnets verbose output when advancing simulation clock to a given amount.
Arnaud Giersch [Wed, 18 May 2011 08:45:00 +0000 (10:45 +0200)]
Apply "coordinates" config option in surf_config, and not in xbt/config.
Arnaud Giersch [Wed, 18 May 2011 12:30:38 +0000 (14:30 +0200)]
[gras] Don't define variables twice.
Declare them "extern" in include files.
Found by compiling with "-fno-common".
Arnaud Giersch [Wed, 18 May 2011 12:28:29 +0000 (14:28 +0200)]
[surf] Don't define variable "platform_filename" twice.
Declare it "extern" in include file, and define it elsewhere.
Found by compiling with "-fno-common".
Arnaud Giersch [Wed, 18 May 2011 12:27:06 +0000 (14:27 +0200)]
[mc] Don't define variable "raw_heap_fd" twice.
Declare it "extern" in include file.
Found by compiling with "-fno-common".
Actually comment the declaration/definition, since it is not used.
Arnaud Giersch [Wed, 18 May 2011 12:25:27 +0000 (14:25 +0200)]
[tesh] Don't define variables twice.
Declare them "extern" in include file.
Found by compiling with "-fno-common".
Arnaud Giersch [Wed, 18 May 2011 11:52:24 +0000 (13:52 +0200)]
Delete unused variable.
Arnaud Giersch [Wed, 18 May 2011 11:43:50 +0000 (13:43 +0200)]
Delete old/unused file.
It seems unused since commit
6f94b475ee53a96bbe5739ad0764b82649451f36.
Arnaud Giersch [Wed, 18 May 2011 14:23:07 +0000 (16:23 +0200)]
Tasks must be initialized.
Arnaud Giersch [Tue, 17 May 2011 20:44:44 +0000 (22:44 +0200)]
Move tests at proper place.
Arnaud Giersch [Tue, 17 May 2011 19:28:12 +0000 (21:28 +0200)]
Fix stupid off-by-one error, and make check survive to NDEBUG.
Pierre-Nicolas Clauss [Tue, 17 May 2011 19:24:33 +0000 (21:24 +0200)]
Merge back master branch.
Conflicts:
src/smpi/smpi_bench.c
Pierre-Nicolas Clauss [Tue, 17 May 2011 19:10:51 +0000 (21:10 +0200)]
Very basic memoizer.
Pierre-Nicolas Clauss [Tue, 17 May 2011 15:28:59 +0000 (17:28 +0200)]
Merge back master branch.
Arnaud Giersch [Tue, 17 May 2011 12:34:04 +0000 (14:34 +0200)]
[trace] Fix tests: tracing-ms, and tracing-categories.
This basically reverts commit
533fc9ef1e41dcc1902b990d489e8b23c8740e15
("update tracing tesh's to new order of events caused by the change in the automatic event ordering").
Arnaud Giersch [Tue, 17 May 2011 12:27:40 +0000 (14:27 +0200)]
[trace] Fix insert_into_buffer.
Arnaud Giersch [Mon, 16 May 2011 15:24:25 +0000 (17:24 +0200)]
Register get_route_latency in global_routing, and use it.
Avoid an unnecessary double call to get_route...
Arnaud Giersch [Mon, 16 May 2011 15:02:46 +0000 (17:02 +0200)]
Define get_route_latency.
Arnaud Giersch [Mon, 16 May 2011 14:47:23 +0000 (16:47 +0200)]
Do not compute a full extended route when it is not needed.
Arnaud Giersch [Thu, 12 May 2011 08:29:17 +0000 (10:29 +0200)]
Merge _get_route and _get_latency.
Arnaud Giersch [Fri, 13 May 2011 09:07:35 +0000 (11:07 +0200)]
Rewrite remplace in surf_routing_rulebase and avoid calls to strlen..
Arnaud Giersch [Tue, 17 May 2011 06:52:44 +0000 (08:52 +0200)]
Save another few percents in elements_father.
Stop using a dynar for path_src/dst, and use an array allocated on stack.
Since elements_father is called very often, it is a non-negligible economy.
If it reveals not sufficient, there are at least three solutions:
- revert this patch; or
- increase ELEMENTS_FATHER_MAXDEPTH; or
- dynamically switch to dynar for large structures.
Arnaud Giersch [Fri, 13 May 2011 06:04:24 +0000 (08:04 +0200)]
Sanitize types, slightly improve, and reindent elements_father.
Arnaud Giersch [Fri, 13 May 2011 06:03:09 +0000 (08:03 +0200)]
Change interface for elements_father, and avoid to allocate a dynar.
Arnaud Giersch [Fri, 13 May 2011 08:06:10 +0000 (10:06 +0200)]
Use realloc and make _xbt_dynar_expand a bit faster.
Arnaud Giersch [Mon, 16 May 2011 09:24:00 +0000 (11:24 +0200)]
[trace] Do not split comm if trace is not active.
Arnaud Giersch [Mon, 16 May 2011 08:34:56 +0000 (10:34 +0200)]
[trace] Cache config options.
Arnaud Giersch [Mon, 16 May 2011 07:51:45 +0000 (09:51 +0200)]
[trace] Guard expensive computations with TRACE_is_active.
Pedro Velho [Tue, 17 May 2011 08:16:29 +0000 (10:16 +0200)]
Fixed the reamining value, now distinguish between flows that have completed and are still to be started.
Navarrop [Mon, 16 May 2011 14:51:18 +0000 (16:51 +0200)]
Add an hidden option to enable ucontext on Mac.
cmake -Ducontext=on .
For the moment ucontext don't work on MAC OSX
Navarrop [Mon, 16 May 2011 14:12:31 +0000 (16:12 +0200)]
In mac ucontext.h is not detected by cmake but we want to use it.
Set flag "-D_XOPEN_SOURCE" in Flags.cmake
Pierre-Nicolas Clauss [Fri, 13 May 2011 13:21:05 +0000 (15:21 +0200)]
Merge back master branch.
Navarrop [Thu, 12 May 2011 12:01:09 +0000 (14:01 +0200)]
If ucontext.h is not present don't try to run prog_AC_CHECK_MCSC.c.
Christophe Thiéry [Thu, 12 May 2011 11:22:10 +0000 (13:22 +0200)]
Action replayer: don't hardcode the usage of raw contextes
Christophe Thiéry [Thu, 12 May 2011 11:10:16 +0000 (13:10 +0200)]
Use variadic macros
Christophe Thiéry [Thu, 12 May 2011 11:07:56 +0000 (13:07 +0200)]
Action replayer: remove dirty hacks, they didn't work anyway
Navarrop [Thu, 12 May 2011 09:19:48 +0000 (11:19 +0200)]
Add missing test before set properties.
Lucas Schnorr [Thu, 12 May 2011 08:59:44 +0000 (10:59 +0200)]
[trace] cosmetics on user variables tracing, simpler interface
Christophe Thiéry [Wed, 11 May 2011 13:02:45 +0000 (15:02 +0200)]
Update ChangeLog
Lucas Schnorr [Tue, 10 May 2011 15:49:57 +0000 (17:49 +0200)]
[trace] fix on user link variable declaration
Navarrop [Tue, 10 May 2011 14:19:04 +0000 (16:19 +0200)]
Add a script for install ns3
Arnaud Giersch [Tue, 10 May 2011 14:03:49 +0000 (16:03 +0200)]
Fix error message.
Subtraction (optionlist_cpy - name) was upside-down. Change to only keep
the name of the variable.
Navarrop [Tue, 10 May 2011 08:51:27 +0000 (10:51 +0200)]
Optimisation of get_route.
Remove the copie of e_route.
Directely give result of get_route into e_route.
Arnaud Giersch [Tue, 10 May 2011 06:14:56 +0000 (08:14 +0200)]
Define and use a function to run the processes to run.
Navarrop [Mon, 9 May 2011 14:54:15 +0000 (16:54 +0200)]
Don't include generated headers into distrib.
Navarrop [Mon, 9 May 2011 12:40:39 +0000 (14:40 +0200)]
Add some print for debug (trace level)
Navarrop [Mon, 9 May 2011 12:40:17 +0000 (14:40 +0200)]
Use flag -Drelease=on on pipol for the moment.
Arnaud Giersch [Sun, 8 May 2011 08:10:46 +0000 (10:10 +0200)]
Change the handling of SIMIX requests in model checker too.
Arnaud Giersch [Thu, 5 May 2011 08:23:41 +0000 (10:23 +0200)]
Try to improve xbt_parmap.
Do not pre-compute the job share for each thread, and let the load
balance automatically.
Arnaud Giersch [Fri, 6 May 2011 08:19:29 +0000 (10:19 +0200)]
Some more cleanup in smx_context_raw.
Remove smx_ctx_raw_self_parallel which is the same as smx_ctx_base_self.
Then the raw_factory static variable is not needed anymore.
Arnaud Giersch [Thu, 5 May 2011 17:05:39 +0000 (19:05 +0200)]
Remove the get_thread_id field from context_factory.
It is now unused.