Subject: mcuts
From: Joel Malard (joel@epcc.ed.ac.uk)
Date: Tue Apr 15 1997 - 02:50:02 MDT
I am working with a research group that does eco-system modelling using
the ccm3 on a Cray T3D. We ran into problems after changing the value
of kpt. To summarise: has anyone had problems running the ccm3 with a
value of kpt other than 6009?
The first problem we encountered was related to subroutine mcuts and I
don' t see how this could be related to kpt. From what seems to be out
of the blues the ccm3 would no longer run with 64 processes at T42. It
looks like subroutine mcuts allocated spurious work to processes 43 to
64. What I did was to:
a. change the lines in mcuts:
mnstrt = nstart(begm(iam)) + 1
mnend = nstart(endm(iam)) + nlen(endm(iam))
to:
if( iam.le.npessp ) then
mnstrt = nstart(begm(iam)) + 1
mnend = nstart(endm(iam)) + nlen(endm(iam))
else
mnstrt = 1
mnend = 0
end if
b. replace the next two calls in resume:
call getmem('RESUME ',len*(plat/2),palpm)
call getmem('RESUME ',len*(plat/2),pdalpm)
by the next three statements where len2 is an integer variable
len2 = max(len, 1)
call getmem('RESUME ',len2*(plat/2),palpm)
call getmem('RESUME ',len2*(plat/2),pdalpm)
Again I am unclear about the origin of the problem, it could be due to
an unfortunate combination of compiler switches :-(. At any rate
these changes allowed us to start an initial run with a value of kpt
smaller than 6009.
Am I missing something obvious?
Cheers,
Joel Malard
This archive was generated by hypermail 2b27 : Thu Jun 01 2000 - 09:10:54 MDT