[MITgcm-devel] exf_interp_read

[Martin Losch]

OK, I apologize. I was only having a quick look last night. The  
EXF_INTERP_USE_DYNALLOC restores the "old" behavior, which is fine  
with me, still I think that the default for the bufferSize should be  
20000, or the comment should be changed accordingly.

Martin

On 12 May 2007, at 00:02, Jean-Michel Campin wrote:

> Hi Martin,
>
> This is documented in the tag-index:
>> - Add new option: EXF_INTERP_USE_DYNALLOC to use Dynamical  
>> Allocation when
>>   reading the file ; Important: This options needs to be defined  
>> to recover
>>   previous code.
> And the reason for choosing to do this way, is that the most
> "safer" case is with #undef EXF_INTERP_USE_DYNALLOC, which is
> with no dynamical allocation.
>
> Jean-Michel
>
> On Fri, May 11, 2007 at 09:51:23PM +0200, Martin Losch wrote:
>> Hi there,
>>
>> something has happened to exf_interp_read
>>
>> After updating the code I get this:
>>
>> mpiexec /opt/pce/bin/xd1launcher ../build/mitgcmuv
>> EXF_INTERP_READ: exf_interp_bufferSize too small
>> EXF_INTERP_READ: exf_interp_bufferSize too small
>> EXF_INTERP_READ: exf_interp_bufferSize too small
>> EXF_INTERP_READ: exf_interp_bufferSize too small
>> STOP
>>
>> in EXF_PARAMS.h I see
>> C  To read input data without dynamical allocation
>> (EXF_INTERP_USE_DYNALLOC undef),
>> C  buffer size currently set to 20000 (allows to read-in a 2x2 global
>> data set)
>>       INTEGER    exf_interp_bufferSize
>>       PARAMETER( exf_interp_bufferSize = 10000 )
>>
>> wouldn't it make sense to set the default really to 20000 (as
>> promised in the comment) which is larger than 192x94=18048 (the
>> "standard" size of ncep/core fields that we use for the cs510 for
>> example)?
>>
>> Plus with the fields that I downloaded from Dimitiris, I have a
>> runoff file which is 360x180 = 64800.
>> I am not happy about having a local copy of EXF_PARAMS.h in every
>> single experiment that uses EXF_USE_INTERPOLATION. How do you feel
>> about this? Couldn't we have a better default (more like 64800)?
>>
>>
>> Martin
>>
>> _______________________________________________
>> MITgcm-devel mailing list
>> MITgcm-devel at mitgcm.org
>> http://mitgcm.org/mailman/listinfo/mitgcm-devel
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel