[Mitgcm-support] Re: c34 GM Redi changes

[mitgcm-support at dev.mitgcm.org]

Patrick,

Do you remember why you set MAX_NO_THREADS=32 in EEPARAMS.h v1.7
(c32-33)? I had previously reduced it from 128 to 1 in v1.6 (c29-31)
in order to save memory. The latest changes in GMREDI.h will
increase memory contrary to plans below.

Plans for memory:
One of the main differences between c34 and pre35 is that
the time-stepping can easily be changed to use fewer storage
arrays and reduce memory. We currently have 31 3D arrays
of which 4 will be dropped from time-stepping, 4 from using
local arrays for CD, 4 arrays in KPP that don't need to be exchanged
and should be local and possibly 5 grid arrays that would be
better replaced by recomputation.

I had expected to get down to 14 (or 19) 3D arrays by the end of
the day and it will be a major set back if we can't use local arrays
for things like KPP and GMREDI.

Is there any reason for MAX_NO_THREADS to be anything but 1
for all the current applications? If Chris ever get's
motivated about threads I think setting it as required would be
easy enough.

Alistair.

chris wrote:
> Alistair,
> 
>  Whats your plan for the GMRedi changes in c34. The switch
> to bi,bj is not the architecture I had in mind. I would
> prefer to switch MAX_NO_THREADS default to 1. What do you
> think?
> 
> Chris