Ada Annex G (was Re: question on GtkAda)

["Nasser M. Abbasi" <nma@12000.org>]

On 12/18/2010 9:57 AM, Simon Wright wrote:

> "Nasser M. Abbasi"<nma@12000.org>  writes:
>
>> Now, given all the above, for scientific applications, I still
>> need to be able to access lapack functions from Ada. Ada Annex G
>> just does not cut it. Very limited. As well as what other small
>> Ada math libraries I could find.
>>
>>
>> So far, I could not find full binding to these. There are 100's of
>> functions there, that has been written over span of more than 30
>> years or more:
>>
>> http://www.netlib.org/lapack/release_notes.html
>>
>> that one would need to call to do any type of scientific
>> work.

>
> As a matter of interest, what main functionality are you missing?
> (bearing in mind that many functions come in 4 variants (s, d, c, z) and
> that there are many that are optimized for a particular kind of matrix,
> eg tridiagonal).

Ok, here is what I found. But I preface this by saying that I am
no expert in Ada, nor in linear algebra for that matter, but what
I did is look at the lapack functions, and looked at the AnnexG,
and determined if that lapack function can be implemented
using only functions in the Ada 2005 annexG.

I only looked at the real (single) functions listed here

http://www.netlib.org/lapack/single/

and went only to the first 58 functions (up to where it says
"Available COMPUTATIONAL routines" )

I made a list of the result here (including Matlab and
Mathematica functions).

http://12000.org/my_notes/lapack_analysis/lapack/index.htm

Please correct me if you see any errors, and I will
fix the note above.

Conclusion on Ada:

1) no support complex types Ax=b, only real, but
this does not seem too important. Not sure now.

2) No condition number function

3) NO solver for Constrained Linear Least Squares

4) No Generalized Linear Regression Model

5) No Schur Decomposition

6) No Singular Value Decomposition  (SVD)

7) No generalized Singular Value Decomposition

8) No support for finding Eigenvalues/Eigenvectors for the generalized case.

It is very well possible that some of the missing
functionality above is present is some other Ada
libraries over the net, but I was only looking at the
annex G.  I also, did not try to code any Ada examples
to verify any of the above, I was just looking at the
manual.

It seems to me it would have been more efficient to provide an
"offical" bindings to the most stable known math libraries
out there, than to try to provide this in an Ada standard
library such as the annex? May be a thich binding? But
this is not an easy problem. Whcih libraries? which
versions? who will make such an "official" binding, etc...

--Nasser