Stéphane Castelli [Mon, 14 Apr 2014 07:50:13 +0000 (09:50 +0200)]
Definition of a new class to represent the routing in fat trees
Marion Guthmuller [Thu, 10 Apr 2014 07:27:22 +0000 (09:27 +0200)]
model-checker : add commande line in comment to run the example
Marion Guthmuller [Thu, 10 Apr 2014 07:14:41 +0000 (09:14 +0200)]
model-checker : worst case considered for bugged1_liveness example (bug on the last process)
Arnaud Giersch [Tue, 8 Apr 2014 14:59:11 +0000 (16:59 +0200)]
Remove useless declaration.
Arnaud Giersch [Tue, 8 Apr 2014 12:57:38 +0000 (14:57 +0200)]
Add missing XBT_PUBLIC*.
Arnaud Giersch [Tue, 8 Apr 2014 12:31:47 +0000 (14:31 +0200)]
Fix doc.
Arnaud Giersch [Tue, 8 Apr 2014 12:31:10 +0000 (14:31 +0200)]
No need to have this variable global.
Arnaud Giersch [Tue, 8 Apr 2014 11:43:26 +0000 (13:43 +0200)]
Update tesh.pl.
* avoid to launch a shell for subcommands
* improve substitutions for variables
All tests pass on Linux, let's see what's happening on windows.
Arnaud Giersch [Tue, 8 Apr 2014 09:23:28 +0000 (11:23 +0200)]
Rewrite command line without using find, which is not the same on windows.
Arnaud Giersch [Tue, 8 Apr 2014 08:27:10 +0000 (10:27 +0200)]
Remove warnings about unused parameters.
Paul Bédaride [Fri, 4 Apr 2014 08:46:08 +0000 (10:46 +0200)]
Add example of java cpu model
Gabriel Corona [Tue, 8 Apr 2014 08:08:39 +0000 (10:08 +0200)]
Merge remote-tracking branch 'origin/mc'
Conflicts:
src/mc/mc_checkpoint.c
Augustin Degomme [Mon, 7 Apr 2014 15:33:58 +0000 (17:33 +0200)]
add defines to windows build, to avoid breaking ucontexts examples
Augustin Degomme [Mon, 7 Apr 2014 14:14:51 +0000 (16:14 +0200)]
oops
Augustin Degomme [Mon, 7 Apr 2014 14:05:04 +0000 (16:05 +0200)]
also document torus networks
Augustin Degomme [Mon, 7 Apr 2014 13:41:41 +0000 (15:41 +0200)]
document loopback tags for clusters
Arnaud Giersch [Mon, 7 Apr 2014 12:02:21 +0000 (14:02 +0200)]
Missed by previous commit.
Arnaud Giersch [Mon, 7 Apr 2014 11:51:01 +0000 (13:51 +0200)]
Hope that this time win32_ucontext.h will be found.
Arnaud Giersch [Mon, 7 Apr 2014 11:14:15 +0000 (13:14 +0200)]
Print output when prog_AC_CHECK_MCSC.c fails to compile.
Arnaud Giersch [Mon, 7 Apr 2014 09:34:53 +0000 (11:34 +0200)]
stack-overflow tests are expected to fail on win32 too.
Arnaud Giersch [Mon, 7 Apr 2014 09:28:27 +0000 (11:28 +0200)]
Try to fix MCSC check.
Arnaud Giersch [Mon, 7 Apr 2014 08:33:16 +0000 (10:33 +0200)]
startsWith doesn't expect a regexp.
Arnaud Giersch [Mon, 7 Apr 2014 08:00:19 +0000 (10:00 +0200)]
Since SG-java is now reintegrated into the main tree, this kind of test is not needed anymore.
Arnaud Giersch [Mon, 7 Apr 2014 07:57:17 +0000 (09:57 +0200)]
Update version strings.
Augustin Degomme [Mon, 7 Apr 2014 08:52:32 +0000 (10:52 +0200)]
Force windows jar to bundle into x86, as we only build/support 32 bits version
Augustin Degomme [Mon, 7 Apr 2014 07:54:57 +0000 (09:54 +0200)]
fix surf-java library win compilation
Arnaud Giersch [Sat, 5 Apr 2014 20:30:42 +0000 (22:30 +0200)]
Add one more failing test.
An assertion is failing in lmm_print when running with --log=surf.thres:debug:
** SimGrid: UNCAUGHT EXCEPTION received on ./bug-17132(0): category: unknown error; value: 0
** Incorrect value (
10000000000.000002 is not smaller than
10000000000.000000): 1.90735e-06
** Thrown by () in this process
[0.000013] [...]/src/xbt/ex.c:144: [xbt_ex/CRITICAL] Incorrect value (
10000000000.000002 is not smaller than
10000000000.000000): 1.90735e-06
** In _Z9lmm_printP10lmm_system() at [...]/src/surf/maxmin.cpp:568
** In lmm_solve() at [...]/src/surf/maxmin.cpp:799
** In _ZN5Model18shareResourcesLazyEd() at [...]/src/surf/surf_interface.cpp:531
** In _ZN5Model14shareResourcesEd() at [...]/src/surf/surf_interface.cpp:512
** In _ZN21WorkstationCLM03Model14shareResourcesEd() at [...]/src/surf/workstation_clm03.cpp:82
** In surf_solve() at [...]/src/surf/surf_c_bindings.cpp:80
** In SIMIX_run() at [...]/src/simix/smx_global.c:377
** In smpi_main() at [...]/src/simgrid/sg_config.c:1001
** In main() at [...]/src/smpi/smpi_global.c:472
Aborted
It looks like a rounding error, but I'm not sure if the problem in lmm_print
itself, or before.
Arnaud Giersch [Sat, 5 Apr 2014 20:23:24 +0000 (22:23 +0200)]
Complete tesh documentation.
Martin Quinson [Sat, 5 Apr 2014 20:44:13 +0000 (22:44 +0200)]
add a flatifying test with a multicore cluster -- it's broken for now
Martin Quinson [Sat, 5 Apr 2014 20:28:03 +0000 (22:28 +0200)]
requests the flatifier to display the amount of core if not equal to 1
Bad news: this commit does not break the tests, meaning that we have
no flatifying test that involves multi-cores...
Martin Quinson [Sat, 5 Apr 2014 20:27:23 +0000 (22:27 +0200)]
ignore more generated files
Martin Quinson [Sat, 5 Apr 2014 20:17:45 +0000 (22:17 +0200)]
give SimDag the ability to retrieve the amount of cores per host
Marion Guthmuller [Sat, 5 Apr 2014 17:22:13 +0000 (19:22 +0200)]
model-checker : privatization_regions in snapshot may be NULL
Augustin Degomme [Fri, 4 Apr 2014 14:53:21 +0000 (16:53 +0200)]
add missing include dir
Augustin Degomme [Fri, 4 Apr 2014 14:38:22 +0000 (16:38 +0200)]
add XBT_PUBLIC_CLASS macro, and a bunch of XBT_PUBLIC declarations
This should make surf-java swig bindings compile on Win32
Gabriel Corona [Fri, 4 Apr 2014 13:36:01 +0000 (15:36 +0200)]
[mc] Fix issues with register translation code
Gabriel Corona [Fri, 4 Apr 2014 13:27:27 +0000 (15:27 +0200)]
[mc] Fix bAd text
Gabriel Corona [Fri, 4 Apr 2014 13:19:43 +0000 (15:19 +0200)]
Merge branch 'mc'
Gabriel Corona [Fri, 4 Apr 2014 13:14:08 +0000 (15:14 +0200)]
[mc] Implement translation of register number
Libunwind and DWARF do not associated the same regsiter number (at
least in i386) so me need to translate from DWARf register numbers
into libunwind register numbers.
Fix broken expression evaluation on i386.
Augustin Degomme [Fri, 4 Apr 2014 12:37:58 +0000 (14:37 +0200)]
oops, forgot to perform the actual change of path in
3854ce9a5 .
Augustin Degomme [Fri, 4 Apr 2014 12:21:58 +0000 (14:21 +0200)]
line not needed without the handler, and SIGSTKSZ is not defined on win32
Augustin Degomme [Fri, 4 Apr 2014 12:19:46 +0000 (14:19 +0200)]
file win32_ucontext.h was moved by
6e893f95c
oops, this changes the endlines ... let's say it's wanted.
Augustin Degomme [Fri, 4 Apr 2014 12:18:04 +0000 (14:18 +0200)]
fix compil of smpi_bench.c on win32
Augustin Degomme [Fri, 4 Apr 2014 12:17:46 +0000 (14:17 +0200)]
get page size info on win32
Augustin Degomme [Fri, 4 Apr 2014 12:17:36 +0000 (14:17 +0200)]
disable segv handler and guards on windows platforms
Arnaud Giersch [Fri, 4 Apr 2014 11:03:29 +0000 (13:03 +0200)]
Remove unused default constructors.
Arnaud Giersch [Fri, 4 Apr 2014 09:27:38 +0000 (11:27 +0200)]
Factorize initializations for Action constructors.
Also initialize m_hat to NOTSET.
Gabriel Corona [Fri, 4 Apr 2014 10:19:36 +0000 (12:19 +0200)]
[mc] Fix MC_free_snapshot (NP reading NULL snapshot->privatization_regions)
Arnaud Giersch [Thu, 3 Apr 2014 09:46:56 +0000 (11:46 +0200)]
Fix memory leak on file_unlink.
Arnaud Giersch [Thu, 3 Apr 2014 08:18:22 +0000 (10:18 +0200)]
Older gcc warns that variable may be uninitialized. Initialize it.
Gabriel Corona [Thu, 3 Apr 2014 16:18:46 +0000 (18:18 +0200)]
[mc] Make snapshots compatible with SMPI privatization of global variables
Not handled yet:
* State comparaison is not handled in this commit;
* Address process/snapshot translation support.
Still to do: make a clean API for the privatization.
Arnaud Legrand [Thu, 3 Apr 2014 23:48:41 +0000 (01:48 +0200)]
Have an informative final message.
Arnaud Legrand [Thu, 3 Apr 2014 23:48:11 +0000 (01:48 +0200)]
Distinguish between time precision and sharing precision.
- Set up two distinct configuration variables and use them in the right places
- Set them to legacy values especially in smpirun
- Revalidate tesh files accordingly (only the "Configuration change" message)
Arnaud Legrand [Thu, 3 Apr 2014 15:47:11 +0000 (17:47 +0200)]
Use more explicit variable names and add a bit of debug.
There is still something "wrong" with this code, which is that it aborts whenever an action gets a share of 0 (in which case I am not able to assign it a completion date and insert it into the calendar). This could happen but is more likely to be a bug in lmm or in surf, which is why I leave it as such for the moment.
Arnaud Legrand [Thu, 3 Apr 2014 15:44:46 +0000 (17:44 +0200)]
Fix bug #17132 (surf.c:366: The Impossible Did Happen (yet again)).
This occured only when a fatpipe had two (or more) flows with different bandwidth bounds. Only one would get bandwidth and because of early termination in this loop, the constraint was never considered ever again, hence the other flow would stay with a share of 0.
Arnaud Legrand [Thu, 3 Apr 2014 15:42:17 +0000 (17:42 +0200)]
Add more debug
Arnaud Legrand [Thu, 3 Apr 2014 12:18:19 +0000 (14:18 +0200)]
Start documenting how lmm works
Arnaud Giersch [Tue, 1 Apr 2014 14:51:38 +0000 (16:51 +0200)]
[mc] Initialize expression->ops before calling mc_dwarf_expression_init.
Pierre Veyre [Tue, 1 Apr 2014 12:52:35 +0000 (14:52 +0200)]
Fix MSG_file_rcopy() and MSG_file_rmove()
Arnaud Giersch [Tue, 1 Apr 2014 09:41:10 +0000 (11:41 +0200)]
Check if MAP_ANONYMOUS is defined after including sys/mman.h.
Some systems define MAP_ANON instead of MAP_ANOYMOUS.
Arnaud Giersch [Tue, 1 Apr 2014 09:37:56 +0000 (11:37 +0200)]
Update copyright headers.
Arnaud Giersch [Tue, 1 Apr 2014 09:15:57 +0000 (11:15 +0200)]
Cosmetics: move SimGrid copyright on top.
Pierre Veyre [Tue, 1 Apr 2014 09:27:52 +0000 (11:27 +0200)]
Rename msg_storage_priv->host by msg_storage_priv->hostname
Improve MSG_file_read() and MSG_file_write()
Pierre Veyre [Tue, 1 Apr 2014 09:08:39 +0000 (11:08 +0200)]
Change MSG_file_unlink() return type
Pierre Veyre [Tue, 1 Apr 2014 08:52:52 +0000 (10:52 +0200)]
Merge branch 'master' of git+ssh://scm.gforge.inria.fr//gitroot/simgrid/simgrid
Pierre Veyre [Tue, 1 Apr 2014 08:52:04 +0000 (10:52 +0200)]
remove MSG_file_ls()
Arnaud Giersch [Tue, 1 Apr 2014 08:31:12 +0000 (10:31 +0200)]
Remove binary file.
Pierre Veyre [Tue, 1 Apr 2014 06:52:39 +0000 (08:52 +0200)]
Remove XBT_INFO call
Pierre Veyre [Tue, 1 Apr 2014 06:26:21 +0000 (08:26 +0200)]
Merge branch 'master' of git+ssh://scm.gforge.inria.fr//gitroot/simgrid/simgrid
Pierre Veyre [Tue, 1 Apr 2014 06:25:59 +0000 (08:25 +0200)]
Add bug-17132 tesh suite
Martin Quinson [Mon, 31 Mar 2014 19:43:02 +0000 (21:43 +0200)]
Revert "Remove XBT_INFO call"
This commit changes much much more than wanted. In particular, it
reverts several changes done in the tree recently. I assume that the
merge that occured just before this commit went awfully wrong,
actually.
That's ok, shit happens, don't worry.
Nevertheless, I revert this commit when it's still possible, before
the first unrelated commit on top of it. Pierre, please, pick your
changes back from the diff, and resubmit them.
This reverts commit
f9e6853d2c5b7b867211700bc12f0ee57f640b30.
Pierre Veyre [Mon, 31 Mar 2014 15:19:31 +0000 (17:19 +0200)]
Remove XBT_INFO call
Pierre Veyre [Mon, 31 Mar 2014 14:59:29 +0000 (16:59 +0200)]
Merge branch 'master' of git+ssh://scm.gforge.inria.fr//gitroot/simgrid/simgrid
Pierre Veyre [Mon, 31 Mar 2014 14:56:45 +0000 (16:56 +0200)]
Improve MSG_file_read() and MSG_file_write()
Arnaud Giersch [Mon, 31 Mar 2014 14:34:21 +0000 (16:34 +0200)]
Improve error message.
Arnaud Giersch [Mon, 31 Mar 2014 14:22:45 +0000 (16:22 +0200)]
Fix bug with Java bindings on 32-bit systems.
On 32-bit systems, sizeof(NetworkLink*) < sizeof(jlong), and the elements must
be copied one by one.
Arnaud Giersch [Mon, 31 Mar 2014 11:49:36 +0000 (13:49 +0200)]
[mc] Reduce stack_size for tests mc-bugged*.
Default stack size was recently increased from 128KiB to 8MiB,
which is a bit too much on small machines (e.g. on CI).
Reduce it to 256KiB (2 times the old size) for tests mc-bugged*.
Arnaud Giersch [Fri, 28 Mar 2014 17:15:55 +0000 (18:15 +0100)]
[mc] Better fix for compatibility with old/broken version of dwarf.h.
Avoids to use a hardcoded value.
Augustin Degomme [Mon, 31 Mar 2014 12:05:50 +0000 (14:05 +0200)]
Remove benchmarking calls in MPI functions that don't actually perform comms and are not traced,
We don't need to halt processes and trigger context switches for them (and privatization switches).
This should improve SMPI performance a little.
Augustin Degomme [Mon, 31 Mar 2014 09:14:44 +0000 (11:14 +0200)]
forgotten useless line (should not do anything with our sets of flags for mmapped pages)
Gabriel Corona [Fri, 28 Mar 2014 14:50:58 +0000 (15:50 +0100)]
[mc] Fix type conversion in unit test
Gabriel Corona [Fri, 28 Mar 2014 14:28:25 +0000 (15:28 +0100)]
[mc] Fix type conversion (once again)
Gabriel Corona [Fri, 28 Mar 2014 14:05:55 +0000 (15:05 +0100)]
[mc] Fix type conversion
Gabriel Corona [Fri, 28 Mar 2014 13:25:15 +0000 (14:25 +0100)]
[mc] Fix type conversions
Augustin Degomme [Fri, 28 Mar 2014 12:22:15 +0000 (13:22 +0100)]
privatization : perform the initialization at process launch, not at MPI_Init time
Also switch at destroy time, to avoid deallocation problems in user code
Augustin Degomme [Wed, 26 Mar 2014 16:00:42 +0000 (17:00 +0100)]
replace warnings by debug messages
Gabriel Corona [Fri, 28 Mar 2014 10:29:13 +0000 (11:29 +0100)]
[mc] Change type for type->id
Marion Guthmuller [Fri, 28 Mar 2014 10:27:34 +0000 (11:27 +0100)]
model-checker : smpi mutual exclusion example
Gabriel Corona [Fri, 28 Mar 2014 09:54:52 +0000 (10:54 +0100)]
[mc] Fix type of expression resolution methods (uintpt_t and not Dwarf_Off)
Gabriel Corona [Fri, 28 Mar 2014 09:50:29 +0000 (10:50 +0100)]
[mc] Fix compatibility with old/broken version of dwarf.h
Gabriel Corona [Thu, 27 Mar 2014 15:53:27 +0000 (16:53 +0100)]
[mc] Kernel Samepage Merging support (Linux)
Add MADV_MERGEABLE advice (madvise) for KSM in the heaps and
read/write segments: mark these regions of memory as candidate for
KSM. As the stacks are stored on the heap, they are candidate for KSM
as well.
In order to work, KSM must be enabled on the system (by root):
# Enable:
echo 1 > /sys/kernel/mm/ksm/run
# Do some real work:
echo 10000 > /sys/kernel/mm/ksm/pages_to_scan
See Documentation/vm/ksm.txt:
https://www.kernel.org/doc/Documentation/vm/ksm.txt
We would like to only mark the .data+.bss and not the other sections
of the segment.
Gabriel Corona [Thu, 27 Mar 2014 11:25:04 +0000 (12:25 +0100)]
[mc] Comment unhandled rangelistptr class in DW_AT_start_scope
Gabriel Corona [Thu, 27 Mar 2014 10:00:11 +0000 (11:00 +0100)]
[mc] Correctly handle the end of the stack unwinding
unw_step returns 0 when reaching the bottom of the stack: this return
code was not handled correctly leading to infinite loops in some
cases.
Gabriel Corona [Thu, 27 Mar 2014 09:58:54 +0000 (10:58 +0100)]
[mc] Really fix DW_AT_high_pc for DWARF4
In DWARF4, DW_AT_high_pc can be an offset from DW_AT_low_pc instead of
a relocatable address.
MC works with DWARF4!
Arnaud Giersch [Wed, 26 Mar 2014 19:05:48 +0000 (20:05 +0100)]
Fix parameter names.
Arnaud Giersch [Wed, 26 Mar 2014 19:01:30 +0000 (20:01 +0100)]
Fix directories for bug-17312.
Also add --cfg=smpi/cpu_threshold:-1 and --log=smpi_kernel.thres:warning
to the command line.
Martin Quinson [Wed, 26 Mar 2014 17:03:47 +0000 (18:03 +0100)]
fix 'make dist', stupid me
Martin Quinson [Wed, 26 Mar 2014 16:57:28 +0000 (17:57 +0100)]
integrate the testcase of #17132 into our testsuite
Paul Bédaride [Wed, 26 Mar 2014 16:03:08 +0000 (17:03 +0100)]
Finalize reservation surf plugin
Paul Bédaride [Wed, 26 Mar 2014 16:02:09 +0000 (17:02 +0100)]
Add networkCommunicateCallback, NetworkLink::updateLatency and NetworkLink::updateBandwidth methods