Martin Quinson [Mon, 21 Nov 2016 00:20:06 +0000 (01:20 +0100)]
code simplification in the VM creation + inline a function
Martin Quinson [Sun, 20 Nov 2016 19:12:50 +0000 (20:12 +0100)]
inline another VM function, and kill another one, unused
Martin Quinson [Sun, 20 Nov 2016 18:42:50 +0000 (19:42 +0100)]
how many functions will I inline in the VMs????
Martin Quinson [Sun, 20 Nov 2016 18:35:50 +0000 (19:35 +0100)]
Kill a VM-related function out of HostImpl
Martin Quinson [Sun, 20 Nov 2016 18:18:49 +0000 (19:18 +0100)]
(hopefully) fix the memory issues in the VMs
- Create and populate an host extension, used by the VM plugin to
check the total amount of memory of that host and whether this host
allows overcommiting VM above this value.
- This is a great step toward the pluginization of the VMs
- Unfortunately, there is no way to change these values for an host
yet :-|
Martin Quinson [Sun, 20 Nov 2016 18:13:19 +0000 (19:13 +0100)]
don't use deprecated functions, and don't hide the corresponding warnings
Martin Quinson [Sun, 20 Nov 2016 17:27:46 +0000 (18:27 +0100)]
avoid using C exceptions as valgrind don't like them too much
Martin Quinson [Sun, 20 Nov 2016 17:21:06 +0000 (18:21 +0100)]
more VM functions inlining (plus, getters don't need to be a simcall)
Martin Quinson [Sun, 20 Nov 2016 16:59:16 +0000 (17:59 +0100)]
inline 2 more VM functions, and kill a simcall
Martin Quinson [Sat, 19 Nov 2016 21:54:47 +0000 (22:54 +0100)]
inline a bunch of VM functions
Martin Quinson [Fri, 18 Nov 2016 21:56:29 +0000 (22:56 +0100)]
more verbose checks
Martin Quinson [Fri, 18 Nov 2016 21:56:09 +0000 (22:56 +0100)]
cosmetics
Martin Quinson [Fri, 18 Nov 2016 21:49:44 +0000 (22:49 +0100)]
test to suspend an exec too
Martin Quinson [Thu, 17 Nov 2016 22:18:17 +0000 (23:18 +0100)]
fix copy/paste errors (found by sonar as dupplicated blocks ;)
Martin Quinson [Thu, 17 Nov 2016 10:59:43 +0000 (11:59 +0100)]
finalize the traces in SD_exit also
Thanks to Alexey Nazarenko for the bug report and the fix.
Martin Quinson [Thu, 17 Nov 2016 10:41:46 +0000 (11:41 +0100)]
Merge branch 'master' of scm.gforge.inria.fr:/gitroot/simgrid/simgrid
Martin Quinson [Thu, 17 Nov 2016 10:34:35 +0000 (11:34 +0100)]
kill useless simcall handlers
Martin Quinson [Thu, 17 Nov 2016 10:23:26 +0000 (11:23 +0100)]
no need to know who's resuming an actor
Martin Quinson [Thu, 17 Nov 2016 10:02:37 +0000 (11:02 +0100)]
kill 2 dumb functions in simix_vm
Martin Quinson [Thu, 17 Nov 2016 09:50:55 +0000 (10:50 +0100)]
kill a bunch of brain-dead functions in surf
Martin Quinson [Thu, 17 Nov 2016 09:38:30 +0000 (10:38 +0100)]
inline a stupid function
Martin Quinson [Thu, 17 Nov 2016 09:34:54 +0000 (10:34 +0100)]
don't store the VM implem where the host stores its implem
Martin Quinson [Thu, 17 Nov 2016 09:25:43 +0000 (10:25 +0100)]
inline a stupid function
Martin Quinson [Thu, 17 Nov 2016 09:19:26 +0000 (10:19 +0100)]
move an enum to the right location
degomme [Thu, 17 Nov 2016 05:46:57 +0000 (22:46 -0700)]
note for next time: try to not forget half of the commit when pushing.
degomme [Thu, 17 Nov 2016 05:05:48 +0000 (22:05 -0700)]
Replace deque by boost:circular_buffer_space_optimized for memory reasons.
Deques are pre-allocated to a specific size and use ~1KB of memory, even when empty.
As we need 2 deques per mailbox, this was quite a lot in the end.
degomme [Wed, 16 Nov 2016 23:53:03 +0000 (16:53 -0700)]
fix dist
Martin Quinson [Wed, 16 Nov 2016 22:44:21 +0000 (23:44 +0100)]
kill two more useless functions in VM
Martin Quinson [Wed, 16 Nov 2016 22:36:21 +0000 (23:36 +0100)]
kill 2 unused fields in MSG_Global o_O
Martin Quinson [Wed, 16 Nov 2016 22:32:35 +0000 (23:32 +0100)]
remove unused functions in the VMs
Martin Quinson [Wed, 16 Nov 2016 22:19:35 +0000 (23:19 +0100)]
move virtual machines to the directory src/plugins/vm
Martin Quinson [Wed, 16 Nov 2016 21:49:18 +0000 (22:49 +0100)]
move the VM-related data out of MSG's private data for hosts
This is the beginning of turning the VMs into a separate plugin. I'm
sure that reconcentrating all VM code into a few files will greatly
simplify it.
This change is really messy because I did not want to be too intrusive
in msg_vm.c and had to expose much much of the VM internals for that.
It will be improved in the future (it cannot get any worse).
This reveals a little bug where the dirtyPages things of a VM that
runs on a host which turns down are not properly cleaned up. There is
nothing I can do right now as I don't have a onHostOff event or
something, so I simply spit a warning in this case. For now.
Martin Quinson [Wed, 16 Nov 2016 07:46:51 +0000 (08:46 +0100)]
rename AsImpl::getRouteRecursive to AsImpl::getGlobalRoute (+doc improvment)
Martin Quinson [Wed, 16 Nov 2016 07:28:21 +0000 (08:28 +0100)]
rename As::getRouteAndLatency into As::getLocalRoute
+ many small cleanups
Martin Quinson [Wed, 16 Nov 2016 01:40:51 +0000 (02:40 +0100)]
Merge branch 'master' of scm.gforge.inria.fr:/gitroot/simgrid/simgrid
Martin Quinson [Wed, 16 Nov 2016 01:28:45 +0000 (02:28 +0100)]
various small cosmetics in the routing
Martin Quinson [Wed, 16 Nov 2016 01:12:36 +0000 (02:12 +0100)]
reduce the visibility of some functions
Martin Quinson [Wed, 16 Nov 2016 01:01:35 +0000 (02:01 +0100)]
reduce the visibility of one field
degomme [Wed, 16 Nov 2016 00:55:39 +0000 (17:55 -0700)]
oops, add the previously documented option to the list.
Martin Quinson [Wed, 16 Nov 2016 00:54:20 +0000 (01:54 +0100)]
remove another now useless function from RoutingPlat
Martin Quinson [Wed, 16 Nov 2016 00:53:21 +0000 (01:53 +0100)]
I switched to NS 3.26 locally
degomme [Wed, 16 Nov 2016 00:50:00 +0000 (17:50 -0700)]
add missing include (only for osx?)
degomme [Wed, 16 Nov 2016 00:42:16 +0000 (17:42 -0700)]
that was useless.
Martin Quinson [Wed, 16 Nov 2016 00:40:07 +0000 (01:40 +0100)]
code simplification: write the recursivity properly
degomme [Wed, 16 Nov 2016 00:26:23 +0000 (17:26 -0700)]
Don't allocate 5KB per host created when it's not needed.
These were for VM and files only, which are optional. Instead they are now allocated when needed.
degomme [Wed, 16 Nov 2016 00:24:54 +0000 (17:24 -0700)]
document new option
Martin Quinson [Wed, 16 Nov 2016 00:19:55 +0000 (01:19 +0100)]
unXBTize a bit
Martin Quinson [Tue, 15 Nov 2016 23:38:36 +0000 (00:38 +0100)]
cosmetics
Martin Quinson [Tue, 15 Nov 2016 23:16:23 +0000 (00:16 +0100)]
let's do standard C++ here
Martin Quinson [Tue, 15 Nov 2016 23:05:48 +0000 (00:05 +0100)]
cosmetics
Martin Quinson [Tue, 15 Nov 2016 23:02:38 +0000 (00:02 +0100)]
reduce the amount of dynars created by getOneLinkRoutes()
Martin Quinson [Tue, 15 Nov 2016 22:54:10 +0000 (23:54 +0100)]
less void*
Martin Quinson [Tue, 15 Nov 2016 22:41:10 +0000 (23:41 +0100)]
kill a now useless function
Martin Quinson [Tue, 15 Nov 2016 22:35:26 +0000 (23:35 +0100)]
Merge branch 'master' of scm.gforge.inria.fr:/gitroot/simgrid/simgrid
Martin Quinson [Tue, 15 Nov 2016 22:35:16 +0000 (23:35 +0100)]
empty RoutingPlatf: move the loopback link to the network_model(s)
degomme [Mon, 14 Nov 2016 23:04:53 +0000 (16:04 -0700)]
Add a smpi/grow-injected-times option.
This is to allow triggering off the default behavior of injected timings growth in MPI_Iprobe and MPI_Test.
This behavior allows to simulate faster and avoid looping too much over these calls. But it could also hurt precision of the simulation.
degomme [Mon, 14 Nov 2016 03:18:23 +0000 (20:18 -0700)]
some more tiny optims for the smpirun generation
degomme [Sun, 13 Nov 2016 22:35:54 +0000 (15:35 -0700)]
Add a hint about a non trivial error.
mmap reports no memory left, while the error is actually the number of allowed mappings per process.
Arnaud Legrand [Mon, 22 Aug 2016 22:54:48 +0000 (00:54 +0200)]
cosmetics
degomme [Sun, 13 Nov 2016 19:11:19 +0000 (12:11 -0700)]
get rid of most of the awfully slow cut commands in smpirun.
Those for smpi_replay are still here, though, as the set solution only works one list at a time.
Let's hope this works everywhere.
Martin Quinson [Sun, 13 Nov 2016 15:18:20 +0000 (16:18 +0100)]
less strdup/free
Martin Quinson [Sun, 13 Nov 2016 15:13:40 +0000 (16:13 +0100)]
also take future releases of NS3
Martin Quinson [Sun, 13 Nov 2016 14:27:50 +0000 (15:27 +0100)]
reindent, no change
Martin Quinson [Sun, 13 Nov 2016 11:17:46 +0000 (12:17 +0100)]
the property holder does not take a dict in its constructor
Martin Quinson [Sun, 13 Nov 2016 10:54:57 +0000 (11:54 +0100)]
don't give the properties to the Link constructor either
Martin Quinson [Sun, 13 Nov 2016 10:42:28 +0000 (11:42 +0100)]
don't pass property set to the storage constructor, there is no need
Martin Quinson [Sat, 12 Nov 2016 20:28:02 +0000 (21:28 +0100)]
Move some content out of HostImpl (into VirtualMachineImpl)
Martin Quinson [Sat, 12 Nov 2016 18:15:31 +0000 (19:15 +0100)]
the AS can now create hosts.
Martin Quinson [Sat, 12 Nov 2016 18:09:16 +0000 (19:09 +0100)]
useless cosmetics around the host creation
Martin Quinson [Sat, 12 Nov 2016 17:51:45 +0000 (18:51 +0100)]
document my last changes
Martin Quinson [Sat, 12 Nov 2016 00:50:17 +0000 (01:50 +0100)]
code simplification: we don't use non-LMM resources
there was a specific constructor for non-LMM resources that was not
given the lmm constraint, which leaded to many dupplicated code.
But actually, this is not needed: NetworkConstant does not create a
link resource anyway, and NS3 can easily pass nullptr to the
constructor of its ancestor.
Martin Quinson [Sat, 12 Nov 2016 00:28:31 +0000 (01:28 +0100)]
less useless new and explicit mallocs
Martin Quinson [Fri, 11 Nov 2016 23:15:16 +0000 (00:15 +0100)]
New function: MSG_parallel_task_execute_with_timeout
This is the fist time that executions can finish with a timeout, and
I'm not really proud of the resulting code. Sorry, that's just a quick
(fix #115) while I was fixing the platform generation.
At some point the whole activity thingy should be reworked, too.
Martin Quinson [Tue, 8 Nov 2016 14:38:58 +0000 (15:38 +0100)]
Make a proper extension out of the Vivaldi coordinates
Martin Quinson [Tue, 8 Nov 2016 14:35:23 +0000 (15:35 +0100)]
small cleanups in Java advised by eclipse
Martin Quinson [Mon, 7 Nov 2016 21:38:58 +0000 (22:38 +0100)]
normalize a bit the inst callbacks
Martin Quinson [Sun, 6 Nov 2016 22:27:28 +0000 (23:27 +0100)]
Spell check, again O:-)
Martin Quinson [Sat, 5 Nov 2016 12:27:28 +0000 (13:27 +0100)]
this file is not used, actually
degomme [Mon, 7 Nov 2016 13:47:25 +0000 (14:47 +0100)]
Set missing cmake variables to 0
Matthieu Volat [Mon, 7 Nov 2016 13:21:42 +0000 (14:21 +0100)]
Bump mmorecore buffer size to 256 bytes on FreeBSD
It seems to need more memory than Linux.
Matthieu Volat [Mon, 7 Nov 2016 13:17:15 +0000 (14:17 +0100)]
When mmap-ing, ensute that offset is set to 0 if anon mmap
Linux tolerate non-zero offset parameter (and ignore it),
but others do not.
Matthieu Volat [Mon, 7 Nov 2016 11:25:21 +0000 (12:25 +0100)]
Check for a dlfunc() function to get handle if present
dlfunc() is a (for now) FreeBSD-only function that is to be used when
trying to retrieve function handles from code segments, rather
than dlsym(), but let's make the code generic in case somebody else
likes the idea...
Matthieu Volat [Sun, 6 Nov 2016 15:36:46 +0000 (16:36 +0100)]
Allow to use BSD ptrace().
Matthieu Volat [Sat, 5 Nov 2016 14:22:59 +0000 (15:22 +0100)]
Build ID are somewhat gnu/linux specific.
Martin Quinson [Fri, 4 Nov 2016 15:05:59 +0000 (16:05 +0100)]
fix the handling of bypassASroutes
Fred used to think he needs it for his work, so here is the fix. It
was wrong anyway.
Now that it works, we should make the code better (too much redundency
for now) at some point.
Martin Quinson [Fri, 4 Nov 2016 11:35:34 +0000 (12:35 +0100)]
tesh: report the command line in messages, not the currently parsed line
This consistency is much more convenient for the user.
Martin Quinson [Fri, 4 Nov 2016 11:34:52 +0000 (12:34 +0100)]
don't sort the output of the flatifier, it makes debugging much harder
degomme [Fri, 4 Nov 2016 13:55:59 +0000 (14:55 +0100)]
clang complains about these seemingly unecessary std::move.
Matthieu Volat [Fri, 4 Nov 2016 10:18:55 +0000 (11:18 +0100)]
Make simgrid::mc::HeapLocation instanciation C++11 compliant
simgrid::mc::HeapLocation is a typedef of std:array which require use
of double brace to declare subobjects.
degomme [Fri, 4 Nov 2016 11:28:56 +0000 (12:28 +0100)]
Merge branch 'master' of git+ssh://scm.gforge.inria.fr/gitroot/simgrid/simgrid
Martin Quinson [Fri, 4 Nov 2016 09:39:06 +0000 (10:39 +0100)]
fix NS3
Martin Quinson [Thu, 3 Nov 2016 21:49:17 +0000 (22:49 +0100)]
unPERLifies a bit the routing: use the object, not its name
Martin Quinson [Thu, 3 Nov 2016 21:34:56 +0000 (22:34 +0100)]
less char*, more std::string
degomme [Thu, 3 Nov 2016 14:09:33 +0000 (15:09 +0100)]
replace multiples allocs by one.
degomme [Thu, 3 Nov 2016 11:01:15 +0000 (12:01 +0100)]
SMPI tracing : match links with sender/receiver + tag, as in MPI, as messages could be mixed up
degomme [Wed, 2 Nov 2016 15:07:56 +0000 (16:07 +0100)]
oops
degomme [Wed, 2 Nov 2016 14:38:53 +0000 (15:38 +0100)]
The sed version used on macos doesn't like the I flag..
So let's turn normal ugly regexpes into plainly awful ones.
degomme [Wed, 2 Nov 2016 14:31:09 +0000 (15:31 +0100)]
Use the libunwind from apple directly
degomme [Wed, 2 Nov 2016 13:40:00 +0000 (14:40 +0100)]
Revert "Fix MC builds after my fix for Mac OSX around libunwind :-("
This reverts commit
182a4405d85c87f52db5fcd8eaf8b04d3b27a3c8.
degomme [Wed, 2 Nov 2016 13:39:37 +0000 (14:39 +0100)]
Revert "cmake: better log message to get more useful information"
This reverts commit
438bc4767b28f0113c8f64d9d46ed4417988b6f7.