[MITgcm-support] Forcing dataset check
Martin Losch
Martin.Losch at awi.de
Wed Jul 1 03:53:31 EDT 2015
Hi Jody,
I am not really the expert on this, but input files all need to be big endian, and so are the (mds) output files, independent of the endian-ness of your platform. Nowadays all (nearly all) compilers have a flag that converts to big endian somewhere internally. If you compiler does not have this flat, then you can use a _BYTESWAPIO flag in your build option file (see eg. linux_ia32_g77: DEFINES='-D_BYTESWAPIO -DWORDLENGTH=4 -DNML_EXTENDED_F77’). This flag turn one code (S/R MDS_BYTESWAPR4 or MDS_BYTESWAPR8) that does the conversion, and yes, I assume that there is a penalty, because this conversion is done every time a file is written or read.
Martin
> On 01 Jul 2015, at 09:43, Jody Klymak <jklymak at uvic.ca> wrote:
>
> Hi Martin et al,
>
>
>> On Jul 1, 2015, at 8:40 AM, Martin Losch <Martin.Losch at awi.de> wrote:
>>
>> (ieee-big endian unblocked, so that they can be opened in fortran with direct access)
>
> Off topic, but I keep hearing this. Is this true if your underlying architecture is little-endian? Is there some kind of performance penalty to little-endian i/o on an intel cluster?
>
> Thanks, Jody
>
>
>
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support
More information about the MITgcm-support
mailing list