| Re: Fortran and byte swapping [message #7673] |
Fri, 13 December 1996 00:00 |
Tim Patterson
Messages: 65
Registered: October 1995
|
Member |
|
|
If the DEC Fortran on the OSF side supports the byteswap flag,
perhaps you could use some Fortran code and CALL_EXTERNAL
to read in the data, or convert the datafile before using IDL?
Neither solution is particularly elegant.
Tim
|
|
|
|
| Re: Fortran and byte swapping [message #7674 is a reply to message #7673] |
Fri, 13 December 1996 00:00  |
Simon Williams
Messages: 5
Registered: October 1996
|
Junior Member |
|
|
Liam Gumley wrote:
>
> Simon Williams wrote:
>> I have a large dataset which is in Unformatted variable-length
>> records written in FORTRAN on a SUN (big endian). I want to
>> read it on a DEC alpha running OSF (little endian) using IDL.
>> It is not really practical to convert the whole dataset.
>>
>> 1. I Can't use "openr, ..... ,/f77_unformatted" because the
>> record length information stored in the file is byte-swapped
>> so it falls over on the first read.
>>
>> 2. I could drop the /f77_unformatted keyword and handle the
>> record structure myself, but even then I can't see a simple
>> way of doing it - I would need an IDL function which behaves
>> like readu, so that I can pass an arbitrary number of arguments
>> to it or I will have to clog up my routine with a load of
>> calls to byteorder for each parameter I read.
>>
>> Any ideas? The nicest solution would be a "byteswap keyword" to
>> the open routines - like the one provided by dec fortran, but I
>> guess that's a bit much to ask!
>
> If you don't want to convert the datafile, then:
>
> First, open the file without the /F77_UNFORMATTED keyword.
>
> Create a record length index array by reading (READU) the first length
> word, swapping it using SWAP_ENDIAN, store it in the array, skip to the
> next length word (POINT_LUN), swap it, etc. to the end of the file.
>
> You now have a record length index array that you can use to read the
> file. Point back to the beginning of the file, and skip the length
> word, read the number of data words indicated in the length array, and
> then swap the data array using SWAP_ENDIAN.
>
> Seems simple enough. I'd write some code for you but I haven't had my
> coffee yet.
>
> Cheers,
> Liam.
--
Thanks Liam,
What you suggested is more or less what I've come up with - (although
I'd not come across SWAP_ENDIAN - my documentation's out of
date - thanks).
However it doesn't really solve the problem - I've been given
a stack of read code which runs on a SUN and contains
MANY lines like this:
readu, lun, param1, param2 ......paramN
readu, lun, next_param1, next_param2 ....
....
....
where the parameters may be scalars or arrays of any type. If I
can possibly avoid it I'd rather not have to call swap_endian for
each parameter after each readu call - I'd have to do it for
several hundred named parameters. It would be much nicer to
incorporate the swapping into a "readu-like" call, but I can't
see any way of doing it. I'm likely to have to do similar things
with other datasets in the future, so a more general solution would
be welcome.
Simon Williams
------------------------------------------------------------ -------
Simon Williams British Atmospheric Data Centre
s.r.williams@rl.ac.uk Tel : 01235 446314 FAX : 01235 445848
------------------------------------------------------------ -------
|
|
|
|
| Re: Fortran and byte swapping [message #7676 is a reply to message #7673] |
Fri, 13 December 1996 00:00 |
Liam Gumley
Messages: 473
Registered: November 1994
|
Senior Member |
|
|
Simon Williams wrote:
> I have a large dataset which is in Unformatted variable-length
> records written in FORTRAN on a SUN (big endian). I want to
> read it on a DEC alpha running OSF (little endian) using IDL.
> It is not really practical to convert the whole dataset.
>
> 1. I Can't use "openr, ..... ,/f77_unformatted" because the
> record length information stored in the file is byte-swapped
> so it falls over on the first read.
>
> 2. I could drop the /f77_unformatted keyword and handle the
> record structure myself, but even then I can't see a simple
> way of doing it - I would need an IDL function which behaves
> like readu, so that I can pass an arbitrary number of arguments
> to it or I will have to clog up my routine with a load of
> calls to byteorder for each parameter I read.
>
> Any ideas? The nicest solution would be a "byteswap keyword" to
> the open routines - like the one provided by dec fortran, but I
> guess that's a bit much to ask!
If you don't want to convert the datafile, then:
First, open the file without the /F77_UNFORMATTED keyword.
Create a record length index array by reading (READU) the first length
word, swapping it using SWAP_ENDIAN, store it in the array, skip to the
next length word (POINT_LUN), swap it, etc. to the end of the file.
You now have a record length index array that you can use to read the
file. Point back to the beginning of the file, and skip the length
word, read the number of data words indicated in the length array, and
then swap the data array using SWAP_ENDIAN.
Seems simple enough. I'd write some code for you but I haven't had my
coffee yet.
Cheers,
Liam.
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
