Re: [boost] Re: multidimensional arrays (was Boost Mathematicians)

From: Matthias Schabel (boost_at_[hidden])
Date: 2004-07-12 12:10:18

>> these points to some extent and provides some other nice properties   
>> such as allowing transposition of arbitrary
>> indices with no need to reorder the underlying storage (that is   
>> mat.transpose(0,1) causes all subsequent calls to mat(i,j) to provide  
>>  the transposed element without moving the elements themselves).  I  
>> have
>
> I'd be interested in seeing how this was done, especially in comparison
> with Timothy Budd's methods in _An Apl Compiler_.  I had an
> implementation of this; however, it was only for dense matrices since
> that's all Budd had in his book.

What I do is maintain a signature of permutations (essentially the  
index order : for example, in a 3D C array the signature would be  
(0,1,2) while a FORTRAN order array would be (2,1,0).  An accessor  
class is responsible for converting between the multidimensional index  
and a one dimensional offset corresponding to a dense matrix; the  
allocator then uses this offset to get the element out of storage - if  
the matrix is dense, the allocator just passes the already computed  
offset along, while if it is sparse, a map is used to look for the  
existence of the specified element.

------------------------------------------------------------------------ 
---------------------------
Matthias Schabel, Ph.D.                           	
Utah Center for Advanced Imaging Research
729 Arapeen Drive
Salt Lake City, UT  84108
801-587-9413 (work)
801-585-3592 (fax)
801-706-5760 (cell)
801-484-0811 (home)
mschabel at ucair med utah edu