Jelena Pjesivac-Grbovic 36156f39c2 Modification to allreduce ring algorithm: ...

- the block sizes are computed in more uniformn way.
  The first k blocks may be 1 element larger than the remaining blocks.
The algorithm passed Intel Allreduce_c and Allreduce_loc_c tests, and 
IMB-3.2 Allreduce, over TCP and both btl and mtl MX (up to 128 processes).
The algorithm still only supports commutative operations.

This commit was SVN r13738.

2007-02-21 19:30:08 +00:00

..

coll_tuned_allgather.c

Fix a bunch of minor typos submitted by Bernhard Fischer.

2007-02-06 12:00:30 +00:00

coll_tuned_allreduce.c

Modification to allreduce ring algorithm:

2007-02-21 19:30:08 +00:00

coll_tuned_alltoall.c

Adding variant of linear alltoall algorithm where the number of

2007-02-20 04:25:00 +00:00

coll_tuned_barrier.c

Cleanup of Barrier implementation:

2007-01-10 22:49:43 +00:00

coll_tuned_bcast.c

Removing compilation warnings about the output format.

2007-02-16 23:32:47 +00:00

coll_tuned_component.c

Adding variant of linear alltoall algorithm where the number of

2007-02-20 04:25:00 +00:00

coll_tuned_decision_dynamic.c

Adding variant of linear alltoall algorithm where the number of

2007-02-20 04:25:00 +00:00

coll_tuned_decision_fixed.c

Removing compilation warnings about the output format.

2007-02-16 23:32:47 +00:00

coll_tuned_dynamic_file.c

Remove a non reacheable statement.

2006-10-18 16:43:55 +00:00

coll_tuned_dynamic_file.h

Update to have public symbols be consistent; use new prefix rule

2005-12-22 13:49:33 +00:00

coll_tuned_dynamic_rules.c

Adding variant of linear alltoall algorithm where the number of

2007-02-20 04:25:00 +00:00

coll_tuned_dynamic_rules.h

Adding variant of linear alltoall algorithm where the number of

2007-02-20 04:25:00 +00:00

coll_tuned_forced.c

Adding variant of linear alltoall algorithm where the number of

2007-02-20 04:25:00 +00:00

coll_tuned_forced.h

Adding variant of linear alltoall algorithm where the number of

2007-02-20 04:25:00 +00:00

coll_tuned_module.c

Adding segmented binary reduce algorithm which works with non-commutative operations.

2007-02-14 22:38:38 +00:00

coll_tuned_reduce.c

Adding segmented binary reduce algorithm which works with non-commutative operations.

2007-02-14 22:38:38 +00:00

coll_tuned_topo.c

Big collective commit. I lightly test it, but I think it should be quite stable. Anyway,

2006-10-26 22:53:05 +00:00

coll_tuned_topo.h

Big collective commit. I lightly test it, but I think it should be quite stable. Anyway,

2006-10-26 22:53:05 +00:00

coll_tuned_util.c

Adding new Allreduce algorithms, updating allreduce decision function, and cleaning up util.

2007-01-23 01:19:11 +00:00

coll_tuned_util.h

Adding new Allreduce algorithms, updating allreduce decision function, and cleaning up util.

2007-01-23 01:19:11 +00:00

coll_tuned.h

Modification to allreduce ring algorithm:

2007-02-21 19:30:08 +00:00

configure.params

Remove unneeded PARAM_INIT_FILE variable in configure.params files used by

2007-01-08 03:44:22 +00:00

Makefile.am

Adding tuned allgather implementation.

2006-12-21 18:40:02 +00:00