Memory issue

[Marc Glisse]

On Mon, 20 Dec 2010, Craig Helfgott wrote:

> I ran with valgrind, using the flags suggested.  It claims there is no
> memory error or memory leak, and that all the blocks are reachable.
> However, it was also claiming several 100MB of memory in "reachable" blocks
> when I killed the run, and I know this isn't correct.  I've tracked the
> problem down to probably being somewhere in my complex Gamma subroutine as
> opposed to my real (or complex) exp subroutine.
>
> I've basically defined
>
> struct mpc_t {
>       mpf_class re;
>       mpf_class im;
> }
>
> and my mpc_gamma returns one of those, but most of the gamma computation is
> done with arrays of mpz_class variables (again a binary splitting
> algorithm):
>
> mpz_class NumerTermsRe[LgPrecisionForGam+3],
>             NumerTermsIm[LgPrecisionForGam+3];
> mpz_class DenomTermsRe[LgPrecisionForGam+3],
>             DenomTermsIm[LgPrecisionForGam+3];


VLA in C++ are a non-standard gcc extension.

> My guess is that somehow my subroutine is not properly freeing these up when
> it finishes.  But I've tried delete and free and can't get my program to
> compile properly with them.  Can anyone give me a clue as to how I can
> ensure that the memory bound up in those arrays is freed up when my
> subroutine exits?

In gmpxx.h, make the mp*_class constructors and destructors print 
something.

-- 
Marc Glisse