[MITgcm-devel] oversized matrix in mitgcm?

[Jean-Michel Campin]

Hi Oliver,

Default factor is only 2 (not 10), to leave some space
for blank tiles; so minimum number for W2_maxNbTiles is
nSx*nSy*nPx*nPy + number_of_blank_tiles

Regarding this new array, will check-in a new CPP option
to disable it, and until I have time to put an alternative
way to compute global cumul-sum, will just add a stop.

Cheers,
Jean-Michel

On Mon, Sep 19, 2011 at 12:25:46PM -0400, An Nguyen wrote:
> There is a new array,
> INTEGER W2_cumSum_tiles( 2, W2_maxNbTiles, W2_maxNbTiles )
> that's W2_maxNbTiles squared. Now in W2_EXCH2_SIZE.h, you set
> W2_maxNbTiles = nSx*nSy*nPx*nPy * 10
> which causes this array to be ~3GB.  Is the 10 really necessary or could it
> be set to something smaller?
> 
> Cheers,
> Oliver, using An's email.