Charm++ parallel programming system

[Loriano Storchi <redo AT thch.unipg.it>]

Dear Aaron and dear Phil,
	thanks a lot for the answer. This, I guess,  explain why the 
problem comes out only when using big matrices, and I guess there is
no simple solution to the problem.

all the best
loriano

On Wed, 10 Feb 2010, Phil Miller wrote:

> The following response was accidentally sent to only our internal list:
>
> ---------- Forwarded message ----------
> From: Aaron Becker 
> <abecker3 AT illinois.edu>
> Date: Sat, Feb 6, 2010 at 16:01
>
> Some BLACS functions, including dgebr2d and dgebs2d, which are used in
> pzhegvx, create a new MPI datatype each time they're called, and
> destroy it at the end of the function. AMPI does not handle this
> situation correctly at all, and it never reclaims memory from
> destroyed datatypes. I think that's the most likely cause of the
> problem.
>
> Aaron