32-bit Memory Limitations

[John Demme]

Hello all-

I apologize since I'm sure this question has been asked numerous times 
previous, but I could not find it in the last 10k messages.

Is there a rough time line for 64-bit DMD 2 support? (For Linux--I don't 
care about Windows.)  I understand that Walter is working on it and 
certainly don't expect a firm date, but I have no sense for the amount of 
work involved... Is this a one man-week feature or several man-months?

I know GDC has 64-bit support, but it has not been synced up in some time.  
Does anyone know if this will be updated in the near future.

I ask because I have written a scientific application I would like to  
operate on a very large matrix of doubles--in the range of 200^4 elements-- 
requiring about 12GB of memory, far larger than the current ~4GB limit.  
Ideally, I'd like to ramp this up to even 20GB matrices.  I'm currently 
running on machines with 24GB of RAM (and we may upgrade a few next year) so 
this is not a performance issue, merely a software issue.

Additionally, is anyone aware of any extreme cleverness to transparently 
work around this issue?  I would imagine not, but I'm constantly amazed by 
some of the hacks I've seen.

4GB limits me to about 150^4 elements, which is acceptable for the time 
being.  As such, I'm not terribly interested in any extreme hacks to get 
around this.  I could obviously multi-process the computation (which would 
help in distributing it) but I don't need to do this yet.

(Yes, all of those exponents are 4, not 2.  This is actually a 4 dimensional 
matrix, but for the purpose of most parts of the computation, I can treat it 
like a typical 2-dim matrix.  Not relevant, I suppose, but perhaps 
interesting.)

~John