Benjamin Kirk wrote:
>> Unless MPI_Probe involves some performance hit (and I don't see why it
>> should) this is the way to go.
> 
> Along these lines, since send_recv is so integral in a lot of algorithms,

That may not be the case forever; I've tried to factor those algorithms 
out into the three parallel_ghost_sync.h templates, so when we have time 
to try and squeeze some actual performance out of them we can work at 
that level too.  I think the efficient implementation won't use a 
blocking send_receive at all.

> should we implement it internally with
> 
> (1) nonblocking send
> (2) blocking probe to get recv size
> (3) resize recv buffer
> (4) blocking recv
> (5) wait on send

That sounds like it could be a real improvement, and much quicker to 
implement than rewriting the sync_ algorithms would be.
---
Roy

-------------------------------------------------------------------------
This SF.Net email is sponsored by the Moblin Your Move Developer's challenge
Build the coolest Linux based applications with Moblin SDK & win great prizes
Grand prize is a trip for two to an Open Source event anywhere in the world
http://moblin-contest.org/redirect.php?banner_id=100&url=/
_______________________________________________
Libmesh-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-devel

Reply via email to