Re: [pooma-dev] allocation assert

[Stephen A. Smith]

I don't remember seeing this error, even though I would
have run this code in the past.  Currently, Node is the
only pooled object, so the error could be one of three
things:

1) A layout is not deleted by the end of the code.  (There
are layouts in the implementations, but the benchmark should
delete the implementations at the end of the code.)

2) A node object is created in some temporary structure
that isn't deleted.

3) The Pool code could be faulty.  This has happened in
the past because of some compiler reorderings of code
using reinterpreted pointers looking at the same piece of
memory.

It would be nice to track this down.  If we have a real
memory leak, that's bad.  If the pool code has problems,
we should replace it with something safer or eliminate it's
use.

(There is a known leak with the new field, so you will
see this error from some of the NewField test codes.)

    Stephen

Allan Stokes wrote:

> I'm in benchmarks/Solvers/Krylov with a linux/kcc --opt build.  All the
> benchmarks run and produce correct results, the timing table prints out
> (with some unflattering results for PoomaII MP) and then this:
>
> POOMA Assertion Failure
> Not all of the pooled memory was freed!
> File Utilities/Pool.cmpl.cpp; Line 82.
> ...
> Aborted
>
> Is this news?