1
1
Граф коммитов

3251 Коммитов

Автор SHA1 Сообщение Дата
Rich Graham
52cfe186c2 adding an implementation for mca_ptl_sm_request_alloc and
mca_ptl_sm_request_return.  Fix some bugs found in code review.
Shared Memory implementation reviewed by Tim W. and Ralph C.

This commit was SVN r3222.
2004-10-19 23:37:30 +00:00
George Bosilca
9ba3ae9c9d Reorder the datatype definition and use a storage class coprresponding to the kind of data it's supposed to hold.
Change the stack usage to make things faster:
 - the first of the stack is always used by the count argument
 - the second is initialized to the first position in the elements list.
 - when the ddt engine advance the last stack position hold the informations about the last processed element (the one from where we have to start the next time).

  Modify the pack/unpack relative functions to reflect the changes in the stack usage. In same time I decrease the amount of code (yup yup) and reorder it a little bit. I get some speedup on my laptop ...

This commit was SVN r3221.
2004-10-19 23:36:31 +00:00
George Bosilca
faef97cf25 No more signess warnings ...
This commit was SVN r3220.
2004-10-19 23:31:03 +00:00
George Bosilca
0b2b7cd2c7 Make the compiler happy by comparing similar types.
This commit was SVN r3219.
2004-10-19 23:30:19 +00:00
Tim Woodall
53ddf3e0fc correct typo
This commit was SVN r3217.
2004-10-19 22:12:40 +00:00
Tim Woodall
c65b706c67 more cleanup`
This commit was SVN r3216.
2004-10-19 22:11:59 +00:00
Tim Woodall
b1801b88bf new allocator for bsend
This commit was SVN r3215.
2004-10-19 22:06:20 +00:00
Tim Woodall
7628c9f16a cleanup to correct mpi semantics
This commit was SVN r3214.
2004-10-19 22:00:19 +00:00
Brian Barrett
d107efd77b * add help file to go with last commit
This commit was SVN r3213.
2004-10-19 21:07:35 +00:00
Brian Barrett
d5431d95fd * Clean up of the rms pcm (still some to go on job killing)
- use show_help() instead of printf()
  - shut up the job reaping ocde
  - add the ability to specify partition and extra prun args (fixing bug
    # 1023)

This commit was SVN r3212.
2004-10-19 21:05:23 +00:00
Tim Woodall
d23227214a increased bsend overhead
This commit was SVN r3211.
2004-10-19 20:53:15 +00:00
Tim Woodall
9092ffcc9e init member variables
This commit was SVN r3209.
2004-10-19 16:56:45 +00:00
Brian Barrett
a5623776d1 * remove the pcm component-specific debugging handle - make all components
share the same handle

This commit was SVN r3208.
2004-10-19 16:56:27 +00:00
Brian Barrett
eec0a0c9c4 * fix silly duplicate variable definition
This commit was SVN r3206.
2004-10-19 14:42:27 +00:00
Brian Barrett
911ba67f75 * enable slurm pcm component
* fix stupid leftover from RMS code base

This commit was SVN r3205.
2004-10-19 14:36:20 +00:00
Rich Graham
ba596cc476 First cut at shared memory implementation finished - not debugged,
and not working.

This commit was SVN r3203.
2004-10-18 23:24:53 +00:00
Brian Barrett
09458c7dc2 * First take at a slurm pcm / pcmclient set of components. Will compile if
slurm is available, but will not allow itself to be selected yet.  Still
  have to figure out a couple of things about slurm

This commit was SVN r3202.
2004-10-18 22:29:02 +00:00
Brian Barrett
af59f53ff7 * fix a couple of errors in the RMS configure stub that would appear only if
building outside the default library search path

This commit was SVN r3201.
2004-10-18 20:19:44 +00:00
Brian Barrett
71d7289f6d * add ending condition for extern "C" linkage in op_predefined
This commit was SVN r3200.
2004-10-18 19:35:54 +00:00
Prabhanjan Kambadur
d9e0ad22f4 file/file.c
info/info.c
group/mpi_group_init.c
communicator/comm_init.c
communicator/comm.c

This commit was SVN r3199.
2004-10-18 16:55:20 +00:00
Brian Barrett
d48aa34dec * changes to proc code to make it compile with a C++ compiler
This commit was SVN r3198.
2004-10-18 16:11:14 +00:00
Brian Barrett
dd9726963c * fixes to make runtime code compile with a C++ compiler
This commit was SVN r3197.
2004-10-18 16:08:52 +00:00
Brian Barrett
c360c8c490 * Make sure we get all the MAX/MIN values when using a C++ compiler so that
the nameserver includes are correct when using a C++ compiler

This commit was SVN r3196.
2004-10-18 16:08:35 +00:00
Brian Barrett
61c8d9592f * make thread directory build with C++ compiler (at least, on OS X)
This commit was SVN r3195.
2004-10-18 15:41:40 +00:00
Brian Barrett
1b1702ce00 * changes to make the util code safe to build with a C++ compiler
This commit was SVN r3194.
2004-10-18 15:38:19 +00:00
Jeff Squyres
d600401f77 Add -Wsign-compare in C -- the Windows CL compiler caught a whole
bunch of these.  GCC can, too, but doesn't by default in C (even with
-Wall).

This commit was SVN r3190.
2004-10-16 19:51:04 +00:00
George Bosilca
116654c1c1 The next one who claim that gcc is not a good compiler for the warnings will have to explain me that ...
warning: ISO C forbids conversion of object pointer to function pointer type

The warning was because we try to use the pointer provided by Fortran as is, not the value stored inside this pointer as it should.

This commit was SVN r3189.
2004-10-16 18:31:27 +00:00
Jeff Squyres
2d5c0d1edc These declarations are repeated in src/util/printf.h, and apparently
everywhere in the tree that uses these functions includes that header
file.  So these are redundant.

This commit was SVN r3188.
2004-10-16 12:36:53 +00:00
Jeff Squyres
85c2d9edd8 Update the HACKING file
This commit was SVN r3180.
2004-10-15 23:19:33 +00:00
Jeff Squyres
3b5f8c87a2 Fix for bug 1007 -- save the selected coll component on the
communicator so that it can be passed down through MPI_COMM_DUP as the
"preferred" component during coll selection.

This commit was SVN r3179.
2004-10-15 22:32:48 +00:00
Jeff Squyres
3ddccde21f Remove some old junk that is now outdated because cmd_line_param_t is
now an OBJ (and this stuff is handled in the destructor).  This fixes
bug 1018.

This commit was SVN r3178.
2004-10-15 21:59:11 +00:00
Jeff Squyres
5686df5a37 - The declaration for ompi_mpi_param_check really didn't belong in
ompi_config_bottom.h -- it's already in src/mpi/runtime/params.h.
- Add /src/mpi/runtime/params.h to a small number of places.
- Remove some useless run-time checks in MPI fortran functions

This commit was SVN r3177.
2004-10-15 21:50:59 +00:00
Jeff Squyres
f82aa9802f Similar to the commits earlier today about multiple input files for
MCA param files, ensure that we reset the flex buffer properly when we
start parsing a 2nd (or 3rd or 4th or ...) file.

This commit was SVN r3176.
2004-10-15 21:38:42 +00:00
Jeff Squyres
ea6f7be7ba Fix $HEADERS$ -> $HEADER$ -- will make for easier replacement later
This commit was SVN r3173.
2004-10-15 19:31:47 +00:00
Jeff Squyres
a965e2dd62 Remove those pesky ^M's. :-)
This commit was SVN r3170.
2004-10-15 16:31:54 +00:00
Tim Woodall
418920f50f correct field name
This commit was SVN r3167.
2004-10-15 15:47:50 +00:00
Tim Woodall
abb4003a26 check for truncation
This commit was SVN r3166.
2004-10-15 15:20:31 +00:00
Tim Woodall
a2dd19f3a0 dont allow ptls at lower exclusivity to be selected
This commit was SVN r3164.
2004-10-15 15:12:31 +00:00
Jeff Squyres
d78003659d Minor fix for allreduce intercomm algorithm.
This commit was SVN r3160.
2004-10-15 13:19:20 +00:00
Jeff Squyres
353e8a7a9c Correct a misleading typo in the explanation comment, and clarify
another comment.

This commit was SVN r3159.
2004-10-15 12:14:09 +00:00
Jeff Squyres
51b2da657a Fix a problem where internal flex buffers were not getting properly
reset when scanning a second file.

This commit was SVN r3158.
2004-10-15 11:42:54 +00:00
Jeff Squyres
fceb5c122f - Add new MCA param: component_show_load_errors. Setting this param
to != 0 makes the MCA base display errors when it fails to load or
  open() a component (normally, such errors cause the MCA base to
  silently ignore the component).  Enabling this can be helpful to
  developers instead of turning on the massively-huge-output "verbose"
  param. 
- Add prefixes to the "verbose" output to make it more easily grep-able.

This commit was SVN r3157.
2004-10-15 10:54:39 +00:00
Jeff Squyres
6669e44107 Move the MPI MCA params into the "mpi" type (it's always bothered me
that they were listed under "base")

This commit was SVN r3156.
2004-10-15 10:52:08 +00:00
Jeff Squyres
530ae63e74 Accidentally allocating too much space -- only need to malloc enough
space for pointers to requests, not the actual requests.  Thank to Tim
for spotting this one.

This commit was SVN r3155.
2004-10-15 10:19:01 +00:00
Brian Barrett
7f347651ae * move ioexecvp from pcm base to the rsh pcm. There is a bunch of windows
unfriendly code in there and it is only used in the rsh pcm

This commit was SVN r3150.
2004-10-14 23:37:42 +00:00
Tim Woodall
08ec73a546 changed proc_modex to ompi_object_t
This commit was SVN r3142.
2004-10-14 21:04:45 +00:00
Tim Woodall
0b3c452cde use ns routines to generate jobid string
This commit was SVN r3141.
2004-10-14 21:04:12 +00:00
Tim Woodall
f557b215ee reimplemented module exchange to use registry publish/subscribe
This commit was SVN r3140.
2004-10-14 20:50:06 +00:00
Tim Woodall
3dbd342d43 minor cleanup
This commit was SVN r3139.
2004-10-14 20:49:30 +00:00
Brian Barrett
d7528d1fd3 * remove the pre-ANSI C vararg code, only have the ANSI-C stdarg code
This commit was SVN r3138.
2004-10-14 19:39:21 +00:00