George Bosilca
8502661385
Update atomic operations for x86 64 bits architectures.
...
This commit was SVN r1103.
2004-04-30 20:04:11 +00:00
Jeff Squyres
e4bbefd57b
Fix for bug 845: elminiated an #if/#endif block that is only necessary
...
if we're going to suport non-ANSI compilers (in which case we would
need a corresponding #else block with a set of non-ANSI typedefs),
which we're not going to do. :-)
This commit was SVN r1102.
2004-04-28 01:52:56 +00:00
Jeff Squyres
7a81bb118c
Add missing "\" that causes problems with some "make"'s, including
...
Solaris make
This commit was SVN r1101.
2004-04-28 01:28:24 +00:00
George Bosilca
d45d74b36c
Add the type for 2 more MPI predefined datatypes: lam_mpi_2cplex and lam_mpi_2dblcplex.
...
This commit was SVN r1100.
2004-04-27 18:26:05 +00:00
George Bosilca
3084aeeb59
Use the fake element to detect the end of data description (remove some tests about the position in the data description.
...
Correctly define the boundaries in the data description loops.
Remove useless local variables ( like end_loop as it already present on the stack).
Increase the position in the datatype description BEFORE looking for the next data
This commit was SVN r1099.
2004-04-27 18:19:03 +00:00
George Bosilca
e49b59e29c
Remove useless local variables
...
Add the fake last element in the data description.
This commit was SVN r1098.
2004-04-27 18:14:16 +00:00
George Bosilca
2a6e129269
The fake element will be added in lam_ddt_commit function.
...
This commit was SVN r1097.
2004-04-27 18:13:16 +00:00
George Bosilca
a82a47a33e
Remove some memory leaks in the test program. (Thanks valgrind).
...
This commit was SVN r1096.
2004-04-27 18:12:33 +00:00
George Bosilca
258677c453
Add the lam_complex_long_double_t typedef.
...
Add a test to abort the compilation if the DT_MAX_PREFEDINED does not match the expected value.
Add some additional MPI2 predefined datatypes.
This commit was SVN r1095.
2004-04-27 18:11:49 +00:00
George Bosilca
2b34654b62
Change the type of the fields to size-defined types (like u_int32_t or int32t).
...
This commit was SVN r1094.
2004-04-27 18:10:18 +00:00
George Bosilca
4e1709241a
complain if the data to be deleted is a predefined one.
...
This commit was SVN r1093.
2004-04-27 18:09:02 +00:00
George Bosilca
04473f4eb9
OBJ_RELEASE not OBJ_DESTRUCT should be use here.
...
This commit was SVN r1092.
2004-04-27 18:08:11 +00:00
George Bosilca
e8237b95d3
If the duplicated datatype was already commited, copy the last fake entry in the data description.
...
This commit was SVN r1091.
2004-04-27 18:07:18 +00:00
George Bosilca
434ebd2dfc
Smarter computation of the required space for the struct.
...
This commit was SVN r1090.
2004-04-27 18:06:07 +00:00
George Bosilca
1fb66840e1
Nearly a complete rewrite. Not the logic is more clean, and the overall execution time a lot smaller. And I think
...
I remove all (hmmm almost) bugs.
This commit was SVN r1089.
2004-04-27 18:04:17 +00:00
George Bosilca
977ef57289
Move all dumping functions in dt_module.c
...
On destroy set the fields back to 0 or NULL
Always allocate one more element in the datatype description. This element will containt a fake DT_END_LOOP entry, that enable us to remove some tests in all parsing functions. (more about that in the design document soooooon)
This commit was SVN r1088.
2004-04-27 18:00:57 +00:00
George Bosilca
9a9bf88781
- Move all dumping functions in dt_module.c. Now we are able to free the fake convertor used to
...
dump the datatype contents.
- Rewrite the macros DECLARE_MPI2_COMPOSED_STRUCT_DDT and DECLARE_MPI2_COMPOSED_BLOCK_DDT used to instantiate
some predefined MPI datatypes to allow them to overwrite the statically allocated basic datatype. (I know it's ugly
but it work).
- Add a new macro COPY_DATA_DESC who can copy a data description and flags from one datatype to another.
This commit was SVN r1087.
2004-04-27 17:57:52 +00:00
Tim Woodall
b9a0e941b4
changes required for tcp testing
...
This commit was SVN r1086.
2004-04-23 20:53:32 +00:00
George Bosilca
ba91bcf530
Check for long double
...
Avoid the lookup for sizeof(TYPE) if TYPE is not defined.
This commit was SVN r1085.
2004-04-23 20:52:42 +00:00
Tim Woodall
d1e599d3ea
*** no code review - needed to get this in so others could run benchmarks ***
...
- optional pml/ptl statistics for diagnostics
- fixed libevent select case, added support for waking up progress thread
that is asleep in select/poll to pickup changes to fdset
- setup tcp socket options when socket is created
- added mca parameter for tcp snd/rcv buffer size
This commit was SVN r1084.
2004-04-23 01:38:41 +00:00
George Bosilca
d355978053
Prefix complex_float_t and complex_double_t with lam_
...
Add 2 new structures for the element description, dt_loop_desc_t
and dt_endloop_desc_t. It's just a easy way to describe those 2 particular elems.
Add 2 functions to create a fake stack (homogeneous and heterogeneous environments).
They are used to start packing/unpacking
the data from an arbitrary offset.
This commit was SVN r1083.
2004-04-22 20:07:53 +00:00
Tim Woodall
673eb325f1
allow for non-fragmented message
...
This commit was SVN r1082.
2004-04-22 19:27:59 +00:00
Jeff Squyres
2510143b4e
Move documentation to the .h file -- where it should be
...
This commit was SVN r1081.
2004-04-22 16:59:56 +00:00
Jeff Squyres
fc6b964284
-Werror-implicit-function-delcaration isn't necessary for C++.
...
This commit was SVN r1080.
2004-04-22 16:50:24 +00:00
Prabhanjan Kambadur
5f25ccf3b1
Should have returned MPI_ERR_ARG if ndims is invalid in MPI_Cart_create
...
This commit was SVN r1079.
2004-04-21 22:11:42 +00:00
Prabhanjan Kambadur
bcc8a53e31
This is the right names
...
This commit was SVN r1078.
2004-04-21 21:08:11 +00:00
Prabhanjan Kambadur
7723dc1dc0
THis is not the right name
...
This commit was SVN r1077.
2004-04-21 21:07:44 +00:00
Prabhanjan Kambadur
8d28895224
Forgot to commit this ... goes with the previous batch of commits
...
This commit was SVN r1076.
2004-04-21 20:57:35 +00:00
Prabhanjan Kambadur
ca48b3962b
Changing some prototypes and also changing the functions. There were some spelling mistakes and other problems. Also commiting the MPI topology functions
...
This commit was SVN r1075.
2004-04-21 20:55:54 +00:00
Jeff Squyres
cc569843f6
Implement barrier
...
This commit was SVN r1074.
2004-04-21 00:40:23 +00:00
Jeff Squyres
5e5655e778
Add an inline hibit calculation function
...
This commit was SVN r1073.
2004-04-21 00:40:08 +00:00
Jeff Squyres
c9f6089bc3
Avoid a potential buffer overflow
...
This commit was SVN r1072.
2004-04-21 00:39:05 +00:00
Jeff Squyres
94e45c07d4
Fix typo -- doh!
...
This commit was SVN r1071.
2004-04-21 00:38:11 +00:00
Prabhanjan Kambadur
9abec1b45f
Saving the information about the topology for easy access later
...
This commit was SVN r1070.
2004-04-21 00:29:08 +00:00
Prabhanjan Kambadur
5d6678256d
Pushing c_cube_dim to communicator since it is used by more than one type
...
This commit was SVN r1069.
2004-04-21 00:17:54 +00:00
Prabhanjan Kambadur
a8d9e4ac7d
Pushing c_cube_dim to communicator since it is used by more than one type
...
This commit was SVN r1068.
2004-04-21 00:16:05 +00:00
Jeff Squyres
2b0cd649b7
Also link in the newly-created libop.la
...
This commit was SVN r1067.
2004-04-21 00:14:57 +00:00
Jeff Squyres
5681513b6a
Add top-level interface functions for the MPI_OP* calls
...
This commit was SVN r1066.
2004-04-20 23:12:34 +00:00
Jeff Squyres
9db9d4fd58
- Fix the #define fortran handle number names
...
- Add lam_op_create()
This commit was SVN r1065.
2004-04-20 23:11:11 +00:00
Jeff Squyres
65b9686184
- Ensure user-created errhandlers are not marked as intrinsic
...
- fix typo in comment
This commit was SVN r1064.
2004-04-20 23:10:44 +00:00
Jeff Squyres
4be63d8cd2
Forgot to fix op global names, and forgot to add global for op_null.
...
Oops.
This commit was SVN r1063.
2004-04-20 23:10:05 +00:00
Jeff Squyres
8991c661c0
Add calls to
...
- op init and finalize
- ddt finalize
- comm finalize
This commit was SVN r1062.
2004-04-20 22:38:22 +00:00
Jeff Squyres
585aab46b7
First cut of op handling -- now link-compatible. Still have work to
...
be done:
- need to define internal op_create function (that MPI_OP_CREATE will
call)
- need to fill in all the handler functions that are the back-ends to the
intrinsic MPI operations (e.g., MPI_SUM, MPI_MAX, etc.)
This commit was SVN r1061.
2004-04-20 22:37:46 +00:00
Jeff Squyres
7177358f23
Doh! Fix typo.
...
This commit was SVN r1060.
2004-04-20 22:35:13 +00:00
Jeff Squyres
ea6b6c4fd5
- Remove some extraneous ()'s
...
- Add type for MPI_OP_NULL
- Rename global op intrinsics
- Change errhandler globals back to instances, not pointers
This commit was SVN r1059.
2004-04-20 22:18:26 +00:00
Jeff Squyres
22cd1bc116
Change the public variables back to instances -- not pointers
...
This commit was SVN r1058.
2004-04-20 22:17:20 +00:00
Prabhanjan Kambadur
f1fa051d5c
Changing the MPI_Info and lam_info functions to incorporate both error handlers and the fortran to C translation table
...
This commit was SVN r1057.
2004-04-20 22:13:39 +00:00
Vishal Sahay
17694418c1
Add comments about rte discovery for developer
...
This commit was SVN r1056.
2004-04-20 22:04:32 +00:00
Prabhanjan Kambadur
a0983ab473
Changing the bindings files to have the #define of MPI_ functions to PMPI_ functions after the #pragma weak statement
...
This commit was SVN r1055.
2004-04-20 18:50:43 +00:00
Jeff Squyres
49ea6ecf4b
Oops. Need to have a *pointer* to the errhandlers in the mpi.h file,
...
not the actual struct. Duh!
This commit was SVN r1054.
2004-04-20 18:30:15 +00:00