On Thu, 4 Dec 2003, Mark Taylor wrote:

> On Thu, 4 Dec 2003, Peter W. Draper wrote:
>
> > Mark produced a document about his work, but I've lost the reference to
> > that.
>
> I did write what I believe to be a reasonably comprehensive
> document about the issues involved for application source code
> and a number of tools for automatically modifying source files
> to be 64-bit friendly. The document is SSN/73(draft) - I seem to
> remember asking Martin to promote it to a proper SSN and put it
> with the other docs, but I don't think he ever did. You can see
> it in HTML at:
>
> http://andromeda.star.bris.ac.uk/~mbt/docs/ssn73.htx/ssn73.html
>
> and/or pick up the package that contains it from ~mbt/ussc/extreme.tar.Z
> at RAL.

Hmm. One thing you did not mention in there is that you are allowed to use
the C pre processor with Fortran code so you could use macros for BIG_INT
just like in C.

Perl can be built in 32 bit or 64 bit mode so there are lots of macros in
there that could be examined.

>
> Quick summary - it's not going to be a breeze.
> Actually applying the tools I wrote would take a reasonable while(?)
> but inevitably they're not going to get everything right and
> picking up the pieces afterwards is not going to be much fun.
> It is quite likely that a lot of things will break;
> discovering all of them will probably be hard.
>

Hmm. Our spec is that we can handle up to 24k x 24k images which will
include a variance array and possible Quality. Most images will be smaller
than that for a given night but I have to demonstrate for the critical
design review that we can handle the big image. The raw data files are
going to be smaller (of order 120 MB x 4 dumped every 2 minutes).

The other problem is that I have to begin coding on this by spring and
need to have the demo by the end of 2004.

> One point that may stave off the inevitable for a while is worth
> noting - if your images have less than 2^31 pixels (rather than bytes)
> it may be possible to get away without doing most of the changes
> in SSN/73 - you'll still need to do the CNF_PVAL business and get
> low level stuff like HDX fixed up, but actually redeclaring all
> integers as (potentially) 8-byte may not be necessary. I think.

but what happens if I ask kappa to manipulate that 2GB file?

It sounds like this needs a lot of thought. Did I mention I want to
videcon into the programmer's meeting? :-)

--
Tim Jenness
JAC software
http://www.jach.hawaii.edu/~timj