Navarrop [Mon, 12 Dec 2011 15:09:00 +0000 (16:09 +0100)]
Allow selective update to be independently set for cpu and network model.
Navarrop [Mon, 12 Dec 2011 15:07:27 +0000 (16:07 +0100)]
Cosmetics
Navarrop [Mon, 12 Dec 2011 14:35:51 +0000 (15:35 +0100)]
Apply the optimization mechanism as requested in config flag.
Navarrop [Mon, 12 Dec 2011 14:29:01 +0000 (15:29 +0100)]
Start using optimization mode flag in network.
Navarrop [Mon, 12 Dec 2011 14:26:01 +0000 (15:26 +0100)]
New values from INRIA RR-7821 (callibration when phase effects are removed).
Navarrop [Mon, 12 Dec 2011 14:08:17 +0000 (15:08 +0100)]
Check for failed links on back_route when crosstraffic is activated
Navarrop [Mon, 12 Dec 2011 13:40:32 +0000 (14:40 +0100)]
Rename misleading "fullduplex" keyword to "crosstraffic".
Navarrop [Mon, 12 Dec 2011 13:39:36 +0000 (14:39 +0100)]
Remove unused variable
Navarrop [Mon, 12 Dec 2011 13:38:41 +0000 (14:38 +0100)]
Optimization + cleanup. Test for optimization only once at startup.
Navarrop [Mon, 12 Dec 2011 13:37:44 +0000 (14:37 +0100)]
Do not create a dict for every resource especially when it is empty anyway.
Navarrop [Mon, 12 Dec 2011 11:48:56 +0000 (12:48 +0100)]
Cleanup (no more _im_ but lazy or full when needed).
Navarrop [Mon, 12 Dec 2011 11:38:43 +0000 (12:38 +0100)]
Rename for clarity.
Navarrop [Mon, 12 Dec 2011 11:35:00 +0000 (12:35 +0100)]
Cleanup to allow separation of model specification and of optimization mode for each model.
Clean some cruft in the comments.
Navarrop [Tue, 6 Dec 2011 15:28:44 +0000 (16:28 +0100)]
Revert new time for token-ring
Navarrop [Fri, 2 Dec 2011 12:09:05 +0000 (13:09 +0100)]
Argh I was confused between UM_FULL and UM_LAZY. This commit sets it right.
Navarrop [Tue, 29 Nov 2011 10:40:45 +0000 (11:40 +0100)]
Merge network.c and network_im.c
Navarrop [Mon, 28 Nov 2011 17:20:14 +0000 (18:20 +0100)]
Need to know if have smpi when compiling
Navarrop [Mon, 28 Nov 2011 12:48:54 +0000 (13:48 +0100)]
Add update mechanism for cpu instead of cpu_model name.
Navarrop [Wed, 16 Nov 2011 10:31:38 +0000 (11:31 +0100)]
Merge cpu.c and cpu_im.c
Christophe Thiéry [Sat, 10 Dec 2011 09:50:11 +0000 (10:50 +0100)]
Parmap test did not compile
Christophe Thiéry [Fri, 9 Dec 2011 15:31:11 +0000 (16:31 +0100)]
Alternative implementation or parmap with busy waits instead of futexes
When the user contexts are run in parallel, you can choose the
synchronization mode of their parmap with:
--cfg=contexts/parallel_mode:{posix|futex|busy_wait}.
The default is futex. Posix synchronization is not implemented yet.
Christophe Thiéry [Fri, 9 Dec 2011 13:22:07 +0000 (14:22 +0100)]
Parmap: process the whole dynar, don't force workers to fetch work
Christophe Thiéry [Fri, 9 Dec 2011 13:15:30 +0000 (14:15 +0100)]
This variable has no reason to be global
Arnaud Giersch [Thu, 8 Dec 2011 15:31:19 +0000 (16:31 +0100)]
Kill duplicated macro.
Arnaud Giersch [Thu, 8 Dec 2011 14:01:41 +0000 (15:01 +0100)]
Use xbt_swag_reset wherever possible.
Arnaud Giersch [Thu, 8 Dec 2011 13:45:00 +0000 (14:45 +0100)]
Define xbt_swag_reset().
Arnaud Giersch [Wed, 7 Dec 2011 14:40:37 +0000 (15:40 +0100)]
Remove redundant assignment in lmm_solve.
Arnaud Giersch [Wed, 7 Dec 2011 10:31:13 +0000 (11:31 +0100)]
Those are constraints, and not elements.
(no functional change)
Arnaud Giersch [Tue, 6 Dec 2011 20:44:20 +0000 (21:44 +0100)]
Update comment (name was changed 7 years ago).
Arnaud Giersch [Thu, 1 Dec 2011 20:43:07 +0000 (21:43 +0100)]
Remove unused log categories.
Christophe Thiéry [Wed, 7 Dec 2011 16:57:19 +0000 (17:57 +0100)]
Parmap: make maestro also do some work.
The gain looks much bigger than I expected (Chord 10k, constant
network: 55s -> 20s and 160,000 syscalls -> 88,000 syscalls).
Not sure to understand why.
Christophe Thiéry [Wed, 7 Dec 2011 16:19:32 +0000 (17:19 +0100)]
The worker id is now useless in the parmap code
Christophe Thiéry [Wed, 7 Dec 2011 16:17:29 +0000 (17:17 +0100)]
Add comments
Christophe Thiéry [Wed, 7 Dec 2011 15:34:28 +0000 (16:34 +0100)]
Remove xbt_parmap_get_worker_id to simplify the parmap interface
Christophe Thiéry [Tue, 6 Dec 2011 14:47:39 +0000 (15:47 +0100)]
Remove a useless private header and rename variables
Laurent Bobelin [Tue, 6 Dec 2011 13:06:57 +0000 (14:06 +0100)]
Added 2 version of a script to generate hierarchical clusters
Navarrop [Tue, 6 Dec 2011 11:51:52 +0000 (12:51 +0100)]
Rewrite the token ring example
Christophe Thiéry [Mon, 5 Dec 2011 17:25:27 +0000 (18:25 +0100)]
Parmaps: remove the event structure to simplify the code, add comments
Christophe Thiéry [Mon, 5 Dec 2011 16:17:46 +0000 (17:17 +0100)]
Make those functions static
Christophe Thiéry [Mon, 5 Dec 2011 16:09:33 +0000 (17:09 +0100)]
Remove a redundant include and reindent a bit
Christophe Thiéry [Mon, 5 Dec 2011 16:05:10 +0000 (17:05 +0100)]
Remove parmap_private.h since parmap.c is the only file using it
Christophe Thiéry [Mon, 5 Dec 2011 15:54:38 +0000 (16:54 +0100)]
In sequential mode, don't use thread-specific data
unknown [Mon, 5 Dec 2011 13:36:27 +0000 (14:36 +0100)]
Make the compilation succed for windows.
THIS is forbidden on win so rename it for THIS_AS
unknown [Mon, 5 Dec 2011 13:35:17 +0000 (14:35 +0100)]
Use the good link for latency.
Navarrop [Fri, 2 Dec 2011 16:04:37 +0000 (17:04 +0100)]
Add a data_center platform.
Navarrop [Fri, 2 Dec 2011 15:38:09 +0000 (16:38 +0100)]
Add a cloud platform.
Arnaud Giersch [Thu, 1 Dec 2011 15:52:42 +0000 (16:52 +0100)]
Kill old CVS $Id$ lines.
Arnaud Giersch [Thu, 1 Dec 2011 13:56:51 +0000 (14:56 +0100)]
Always define SIMGRID_VERSION_*.
Arnaud Giersch [Mon, 28 Nov 2011 14:43:51 +0000 (15:43 +0100)]
Use homogeneous dictionaries whenever possible.
The only remaining users of heterogeneous dicts are sets and multidicts.
Arnaud Giersch [Mon, 28 Nov 2011 13:55:16 +0000 (14:55 +0100)]
Use homogeneous dicts for xbt_lib.
Arnaud Giersch [Fri, 25 Nov 2011 22:05:59 +0000 (23:05 +0100)]
Add some tests for homogeneous dicts (global free functions).
Arnaud Giersch [Mon, 28 Nov 2011 13:39:44 +0000 (14:39 +0100)]
Define xbt_dict_new_homogeneous().
Arnaud Giersch [Mon, 28 Nov 2011 13:22:11 +0000 (14:22 +0100)]
Make dicts ready to accept homogeneous dictionaries.
Arnaud Giersch [Mon, 28 Nov 2011 12:56:53 +0000 (13:56 +0100)]
Define a struct s_xbt_het_dictelm for holding the free function of a dict element.
Create a mallocator for these structs, and use them.
Arnaud Giersch [Fri, 25 Nov 2011 21:06:19 +0000 (22:06 +0100)]
Define xbt_dictelm_set_data, to reduce code duplication.
Arnaud Giersch [Fri, 25 Nov 2011 17:45:13 +0000 (18:45 +0100)]
Add a dict parameter to dictelm functions.
Unused for now.
Arnaud Giersch [Mon, 28 Nov 2011 12:50:46 +0000 (13:50 +0100)]
Add two fields to struct dict, in prevision of homogeneous dicts.
Arnaud Giersch [Fri, 25 Nov 2011 17:30:36 +0000 (18:30 +0100)]
Rewrite integer dicts using regular dicts.
The goal here is to reduce the code complexity for further changes.
It's certainly not optimized anymore, but who cares?
Nobody seems to use them anymore. Integer dicts could be marked as deprecated.
The main difference is that memory is malloc'ed to store a copy of
the key.
Arnaud Giersch [Thu, 1 Dec 2011 10:25:50 +0000 (11:25 +0100)]
A mallocator for dict structs is a bit overkill.
Frederic Suter [Thu, 1 Dec 2011 08:13:01 +0000 (09:13 +0100)]
two more clobbered variables that prevent me to compile. Don't ask
Midjy this time.
Arnaud Giersch [Wed, 30 Nov 2011 22:29:03 +0000 (23:29 +0100)]
tesh: fix the stable sort.
Use a double indirection to be sure that the addresses of the
string pointers do not move.
Arnaud Giersch [Wed, 30 Nov 2011 22:28:52 +0000 (23:28 +0100)]
tesh: fix memory leaks.
Navarrop [Wed, 30 Nov 2011 16:29:06 +0000 (17:29 +0100)]
Correction of a bug. Always return 0.
Navarrop [Wed, 30 Nov 2011 15:58:46 +0000 (16:58 +0100)]
Remove memleaks
Navarrop [Wed, 30 Nov 2011 15:52:09 +0000 (16:52 +0100)]
Add a masterslave example with arguments
number_of_jobs
number_of_slaves
Navarrop [Wed, 30 Nov 2011 15:30:51 +0000 (16:30 +0100)]
Add a platform with one cluster and a host
Arnaud Giersch [Tue, 29 Nov 2011 13:38:54 +0000 (14:38 +0100)]
Free mutex in xbt_dynar_to_array.
Arnaud Giersch [Tue, 29 Nov 2011 13:38:23 +0000 (14:38 +0100)]
Avoid the temporary malloc in xbt_dynar_to_array.
Christophe Thiéry [Tue, 29 Nov 2011 13:39:17 +0000 (14:39 +0100)]
Revert "The latency is constant with the constant network model."
Because actually, it's not.
This reverts commit
123eb94b4e6a8db258c498d38ac1b21656f0f840.
Christophe Thiéry [Tue, 29 Nov 2011 12:57:42 +0000 (13:57 +0100)]
The latency is constant with the constant network model.
Arnaud Giersch [Mon, 28 Nov 2011 22:27:20 +0000 (23:27 +0100)]
Fix uninitialized variable.
Arnaud Giersch [Mon, 28 Nov 2011 17:08:05 +0000 (18:08 +0100)]
Include surf_resource.h, for surf_resource_name().
Arnaud Giersch [Mon, 28 Nov 2011 17:16:14 +0000 (18:16 +0100)]
s/add/free/ in doc.
Arnaud Giersch [Mon, 28 Nov 2011 14:38:21 +0000 (15:38 +0100)]
Don't lie about xbt_xfg_t.
Arnaud Giersch [Tue, 29 Nov 2011 10:36:47 +0000 (11:36 +0100)]
Correclty handle the case p1 < p2 in cmpstringp.
Arnaud Giersch [Fri, 25 Nov 2011 13:58:37 +0000 (14:58 +0100)]
Function does not exist.
Arnaud Giersch [Fri, 25 Nov 2011 17:17:15 +0000 (18:17 +0100)]
Rename dict structs to follow README.coding.
Arnaud Giersch [Mon, 28 Nov 2011 12:21:41 +0000 (13:21 +0100)]
Remove useless internal functions.
Arnaud Giersch [Mon, 28 Nov 2011 12:19:18 +0000 (13:19 +0100)]
Allow NULL for the reset function of mallocators.
Arnaud Giersch [Mon, 28 Nov 2011 12:07:46 +0000 (13:07 +0100)]
Remove unused empty function.
Arnaud Giersch [Thu, 24 Nov 2011 10:57:22 +0000 (11:57 +0100)]
Reorganize structs to save a few bytes.
The general idea is to move the fields, in order to group pointers
apart from integers. Thus, the padding between fields is reduced,
and the size of the struct is reduced too.
It can save up to 32 bytes per node (host and/or process) on 64bits
architectures.
Arnaud Giersch [Wed, 23 Nov 2011 15:25:13 +0000 (16:25 +0100)]
Remove an useless (but optimized!) memcpy.
Arnaud Giersch [Wed, 23 Nov 2011 14:54:14 +0000 (15:54 +0100)]
Avoid to clear mem for the last element.
And save a few calls to memset.
Arnaud Giersch [Wed, 23 Nov 2011 12:55:19 +0000 (13:55 +0100)]
_xbt_clear_mem is used only once. Use memset directly.
Arnaud Giersch [Wed, 23 Nov 2011 12:53:16 +0000 (13:53 +0100)]
Factorize code for xbt_dynar_shrink/_xbt_dynar_expand.
Arnaud Giersch [Wed, 23 Nov 2011 12:51:27 +0000 (13:51 +0100)]
There's no point in clearing mem in _xbt_dynar_expand.
It's already done in _xbt_dynar_set_at_ptr.
Arnaud Giersch [Wed, 23 Nov 2011 12:39:43 +0000 (13:39 +0100)]
Kill dead (commented) code.
Arnaud Giersch [Wed, 23 Nov 2011 12:37:32 +0000 (13:37 +0100)]
Factorize code for xbt_dynar_set/xbt_dynar_set_at_ptr.
Arnaud Giersch [Wed, 23 Nov 2011 12:31:36 +0000 (13:31 +0100)]
Unlock dynar at proper place.
Arnaud Giersch [Wed, 23 Nov 2011 12:22:46 +0000 (13:22 +0100)]
Don't call _xbt_dynar_expand when it's not needed.
Arnaud Giersch [Tue, 22 Nov 2011 12:39:39 +0000 (13:39 +0100)]
Don't clear mem before free.
Use a tool like valgrind to check for memory errors.
Arnaud Giersch [Tue, 22 Nov 2011 14:48:47 +0000 (15:48 +0100)]
Push last_route in global_routing.
Now we are able to free it at exit.
Arnaud Giersch [Mon, 21 Nov 2011 11:23:26 +0000 (12:23 +0100)]
Need to have my own link list.
Else, it is freed while I'm still using it.
Arnaud Giersch [Mon, 21 Nov 2011 11:20:06 +0000 (12:20 +0100)]
Free cost table on finalize.
Arnaud Giersch [Mon, 21 Nov 2011 11:18:59 +0000 (12:18 +0100)]
Strings are already dup'ed by caller.
Arnaud Giersch [Mon, 21 Nov 2011 11:17:02 +0000 (12:17 +0100)]
memcheck_tests.cmake: regenerate
Arnaud Giersch [Mon, 21 Nov 2011 11:15:05 +0000 (12:15 +0100)]
Set path for binary in token ring tesh file.
Arnaud Giersch [Wed, 23 Nov 2011 08:10:08 +0000 (09:10 +0100)]
xbt_dict_new_ext() doesn't exist anymore.
Arnaud Giersch [Wed, 23 Nov 2011 13:13:30 +0000 (14:13 +0100)]
Fix warning about clobbered variable.
Arnaud Giersch [Wed, 23 Nov 2011 15:40:10 +0000 (16:40 +0100)]
Fixup for commit
08ac8cf2900.
Looks like I forgot this changed line.