ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность
3 540 коммитов 3 Ветки 0 Теги 130 MiB
6323a0ac4f
Граф коммитов

3540 Коммитов

This Branch
This Branch
Все ветки
Автор SHA1 Сообщение Дата
George Bosilca
884c54decc Small improvements. 
This commit was SVN r3929.
 2005-01-10 08:22:41 +00:00
George Bosilca
70839e8a7a Some major changes: 
- rendez-vous protocol for long messages (internally used by the GM driver)
 - first step toward overlapping registration/transfert
 - use the correct type (size_t and uint64_t) for some internal types to conform to the PML layer
Some minor changes:
 - remove some useless macros
 - clean-up the GM defines
 - renames some GM MCA parameters
 - correctly use the limit between eager and rendez-vous protocol
 - speedup the code a little (dont allocate useless fragments).
 - when allocation fragment set to ZERO all usefull fields in the struct

This commit was SVN r3927.
 2005-01-09 19:42:14 +00:00
Craig E Rasmussen
019b769568 1. Added tests for Fortran 90 types and sizes. 
2. Added creation of src/mpi/f90/fortran_sizes.h.

This commit was SVN r3925.
 2005-01-07 21:21:46 +00:00
Craig E Rasmussen
51092cd7f5 Added lkinds list 
This commit was SVN r3924.
 2005-01-07 21:18:48 +00:00
Craig E Rasmussen
a1bb424e92 Initial version. 
This commit was SVN r3923.
 2005-01-07 21:17:48 +00:00
Craig E Rasmussen
e3140f5026 Initial version. 
This commit was SVN r3922.
 2005-01-07 19:22:49 +00:00
Craig E Rasmussen
c91eb82532 Replaced call to sizeof (not supported by all compilers) and 
replaced it by parameter obtained from fortran_sizes.h, created
by configure.

This commit was SVN r3921.
 2005-01-07 19:22:00 +00:00
Tim Woodall
ea698d2165 correct includes 
This commit was SVN r3919.
 2005-01-07 17:59:14 +00:00
Ralph Castain
bc6973cb02 Add a new interface function to the name server that returns the highest vpid allocated to a given job. This lets the caller determine how many vpids have been allocated to the job since vpids are always allocated contiguously. 
Also, blow away a static-components.h file that accidentally got committed - should only be locally generated.

This commit was SVN r3918.
 2005-01-07 16:03:55 +00:00
Craig E Rasmussen
b349d0ebcf Added printing of kind values (in addition to selected_int_kind) 
This commit was SVN r3917.
 2005-01-07 15:28:22 +00:00
Craig E Rasmussen
a6a7277d8e Initial version. 
This commit was SVN r3916.
 2005-01-07 14:40:46 +00:00
George Bosilca
e31cf84924 Remove useless functions and make the size of the GM basic segment a MCA parameter. 
This commit was SVN r3910.
 2005-01-06 08:20:45 +00:00
Prabhanjan Kambadur
b7ddd7a730 Removing debug from the configuration settings 
This commit was SVN r3909.
 2005-01-06 05:48:58 +00:00
Prabhanjan Kambadur
fab3f680d7 Removing debug from the configuration settings 
This commit was SVN r3908.
 2005-01-06 05:48:39 +00:00
Prabhanjan Kambadur
e4a7af525f Removing debug from the configuration settings 
This commit was SVN r3907.
 2005-01-06 05:48:09 +00:00
George Bosilca
f7d7b2be15 Dont forget to correctly return the amount of data converted in the max_data. 
This commit was SVN r3906.
 2005-01-06 00:03:44 +00:00
George Bosilca
20793328d6 Decrease the lacenty for small messages. Now we are closer to the MPICH-GM performances. 
Use the GM FAST event to avoid a call to gm_unknown.
Dont allocate a fragment for the match message (either on the send or the receive side).

This commit was SVN r3905.
 2005-01-06 00:02:47 +00:00
Craig E Rasmussen
3d922d8947 Initial version. 
This commit was SVN r3897.
 2005-01-04 17:46:10 +00:00
Jeff Squyres
9802822b7b Make the opening of the io framework be lazy -- only occurs at the 
first invocation of MPI_File_open or MPI_File_delete (whichever is
first).  The io framework is then only closed down if it was
successfully opened.  

This is the first [atomic] step to having a progress thread in the
ROMIO component; it wasn't strictly *necessary*, but it's logically
the same direction and provided a good test case.

This commit was SVN r3895.
 2005-01-04 15:43:26 +00:00
George Bosilca
8b9b8cec9d Move the generic part of the convertor initialization in a inlined function. 
This commit was SVN r3893.
 2005-01-03 17:27:48 +00:00
Prabhanjan Kambadur
f87da25e34 Adding the new files and deleting some old files, mostly from ralph's new gpr re-org 
This commit was SVN r3887.
 2004-12-31 20:23:59 +00:00
Edgar Gabriel
f5abd20a57 adding the basic framework for the hierarchical collective module. 
This commit was SVN r3880.
 2004-12-27 08:56:33 +00:00
Craig E Rasmussen
6342e5f139 Added real and complex array types. 
This commit was SVN r3879.
 2004-12-26 17:07:26 +00:00
Craig E Rasmussen
dffa250111 Added real kinds. 
This commit was SVN r3878.
 2004-12-25 21:19:32 +00:00
George Bosilca
3a0fdc49d5 Cleanups and correctly deregister the sender memory. I did some experiments with eager messages. Basically I send small packets (16K) 
independing on the size of the data. The strange fact is that I get nearly the same performances as the NetPipe GM (that use registered memory) who are really close to the maximum performances with the Myrinet cards available on the cluster. However the load on the CPU is higher.
I still have to investigate how exactly this fact fit with the send/recv of non-contiguous datatypes.

This commit was SVN r3877.
 2004-12-23 17:30:39 +00:00
Edgar Gabriel
9b2056c345 adding support for the optional attribute MPI_APPNUM. The information is now correctly available at the child processes, however the attribute is not yet set. 
This commit was SVN r3876.
 2004-12-23 15:33:16 +00:00
Edgar Gabriel
cfd5ba940d fix a typo, which reversed the logic 
This commit was SVN r3875.
 2004-12-23 13:25:16 +00:00
Edgar Gabriel
674bafc6ed adding support for the info-object 'wdir' in Comm_spawn/Comm_spawn_multiple 
This commit was SVN r3874.
 2004-12-23 13:04:58 +00:00
George Bosilca
6e60619434 Decrease the latency. A new protocol for long messages. It's just the begining as it still can be improved but 
it increase the bandwidth by nearly 10%. Now we are very close to the NetPipe (GM) bandwidth.

This commit was SVN r3873.
 2004-12-23 00:32:32 +00:00
George Bosilca
bfa4f158b2 Add some comments. 
This commit was SVN r3872.
 2004-12-22 17:45:16 +00:00
George Bosilca
e6fa352437 Remove one layer of functions. 
Lookup for the peer information only when we need it for later usage.
Small optimizations (moving some function in .h and transform them in static inline).
Cleanups, cleanups and finally cleanups ...

This commit was SVN r3870.
 2004-12-22 07:20:05 +00:00
George Bosilca
f0d86178bc Remove useless file (all functions moved in others .c and .h files). 
This commit was SVN r3869.
 2004-12-22 06:51:40 +00:00
George Bosilca
4acb2b853c Start the cleanup process. I try to remove all useless operations from the critical path. 
Start of the pool of memory that will be used for unexpected messages (just the framework at this moment).

This commit was SVN r3867.
 2004-12-22 00:14:31 +00:00
George Bosilca
92e94f5921 Decrease the creation time for a convertor. I target here specifically the contiguous data or/and the data with a short description. Each convertor 
have now a limited stack attached. If we handle contiguous data then we will use this stack, avoiding the free/malloc for the stack management. In 
all others cases the convertor work as before a stack containing the required number of elements will be allocated. This small modification
decrease the latency for GM by nearly 0.7 micro-sec as reported by NetPipe.

This commit was SVN r3866.
 2004-12-22 00:12:26 +00:00
Tim Woodall
fb203852d3 I/O forwarding framework 
This commit was SVN r3865.
 2004-12-21 22:16:09 +00:00
George Bosilca
d4434bac91 The eager protocol is defined directly in number of bytes not in number of "minimum fragments". 
In the case of a zero length message avoid any convertor creation on both sides.

This commit was SVN r3857.
 2004-12-21 06:39:06 +00:00
George Bosilca
d70a22b057 Start to add a MCA parameter to fix the limit between eager and rendez-vous protocol (gm_max_eager_fragments). 
This commit was SVN r3856.
 2004-12-21 06:23:42 +00:00
Jeff Squyres
7f5c731cdf Fix what looks like bit-rot. 
This commit was SVN r3850.
 2004-12-20 17:12:34 +00:00
Ralph Castain
b860872343 Fix the module declaration - thanks to Jeff for catching the warning. 
This commit was SVN r3849.
 2004-12-20 14:38:55 +00:00
Ralph Castain
832f4b566c Add a new function to the name server that takes a base and offset and returns the corresponding vpid. 
Fix the soh code.

This commit was SVN r3848.
 2004-12-20 01:48:37 +00:00
Ralph Castain
45f77e2fc1 Take a quick break and build type checking into the buffer packing system. Add a new constant OMPI_PACK_MISMATCH - return it whenever the caller asks for a different type than that of the next item in the buffer. 
This commit was SVN r3847.
 2004-12-18 20:18:41 +00:00
Jeff Squyres
18a2960f59 Remove dereferences to (void*) 
This commit was SVN r3846.
 2004-12-18 01:38:22 +00:00
Brian Barrett
43d22fce30 * Fix for ticket #1040. Make sure that we have enough space in the 
buffer for the SIOCGIFCONF ioctl to complete successfully.  Also, use
  the sa_len member of if ifreq's ifr_addr member, if available, for
  computing offsets of the ifreq structures.

  Since this has the potential to break people, setting the env
  variable OMPI_orig_if will result in the old code being used.  This
  will be removed once the new code survives a couple days in the
  wild.

This commit was SVN r3845.
 2004-12-17 16:27:37 +00:00
George Bosilca
5d50b025b8 Decrease the latency. Just keep everything minimal. 
This commit was SVN r3844.
 2004-12-17 09:27:10 +00:00
George Bosilca
17f37130d2 First fully working version of GM. Next step ... improuvements ... 
This commit was SVN r3843.
 2004-12-17 08:39:07 +00:00
Craig E Rasmussen
c134c911e3 Changed declared type of function pointers to void* from MPI_Fint* 
This commit was SVN r3842.
 2004-12-17 05:18:01 +00:00
Jeff Squyres
ae5180aa3b Ok, it's *not* enough to only have one of the files compilable. Need 
to make them all compilable.

#if 0'ed out one section in soh_base_open.c that needs to be fixed.

This commit was SVN r3841.
 2004-12-17 03:20:39 +00:00
Jeff Squyres
63474e0232 Have to have *one* compilable file or some platforms complain. 
Grrr...

This commit was SVN r3840.
 2004-12-17 00:26:16 +00:00
Jeff Squyres
c58a99b6ac Add a missing Makefile.am for soh, and add some svn:ignore's. 
Shame on me for not doing a sanity build :-\ -- thanks Brian for
pointing it out...

This commit was SVN r3839.
 2004-12-16 23:28:02 +00:00
Brian Barrett
bb7de89545 * Add configure tests for SA_RESTART signal constant and sa_len 
* protect use of SA_RESTART in event library

This commit was SVN r3838.
 2004-12-16 21:37:57 +00:00