RE: communication pattern causes corrupted data

On Tue, 2003-03-25 at 17:12, Joshua J. England wrote:
> In doing some other tests, I have found that under gm-1.4/mpich-1.2..5
> the problem does not exist under RH7.1.

That's interesting. If you can reduce the difference between a good
setup and a bad one to only GM, it would make sense to work on this
variable only and see which version increment of GM produces the error.
You can find all released GM tarballs in /pub/GM/ARCHIVE.
You know that it works with 1.4, but not with 1.5.2.1 (from your
previous email). You can then try 1.5.1.1, 1.5, 1.4.1preX, etc. 

There is way to much changes between 1.4 and 1.5.2.1 to look at the
diff, AFAIK. Now, if you can isolate the incremental version step, it
can reduce a lot the scope of the differences.

Patrick
-- 

Patrick Geoffray
Myricom, Inc.
http://www.myri.com