[Mitgcm-support] Random numbers and trig. (PR#18)

mitgcm-support at dev.mitgcm.org mitgcm-support at dev.mitgcm.org
Wed Jul 9 15:27:40 EDT 2003


The following is worth remembering when running hs94* and
internal_wave experiments on different platforms. It would
be nice to fix this but is not vital at this point.

J-M and I implimented portrand() in order to get the same
numbers with the Held and Saurez expts using g77 and pgf77.
Of course, these numbers are completely different on other
machines. Also, a parallel version is needed so that the same
numbers are not generated on each tile. ie. we should make the
random numbers random but a reproducable function of geographic
position!

Ignoring this, comparison of output on different platforms seems
to indicate that hs94.* and internal_wave only "pass" after
throwing out 8 or 9 of the lasts digits. These expts
all use trig. functions for initializing the fields.

A.
PS> Putting this in as a bug so we remember to do something
oneday.

Jean-Michel Campin wrote:
> Will try to compare a run of 1 month (or longer) on DEC
> and on Linux to see if it is OK.




More information about the MITgcm-support mailing list