Dear Haskellers:
Could Haskell ever be used for serious scientific computing?
What would have to be done to promote it from a modelling
tool to a more serious number crunching engine? Maybe not
necessarily competing with Cray, but not terribly lagging
far behind the other languages?
I hope some of you have some answers, ideas, experiences
that could shed some light here. Below are somewhat unorganized
thoughts and more detailed questions that relate to this topic.
1. We tend to kid ourselves, hiding behind "Pseudoknot benchmarks"
or quoting how fast Sisal is. In fact, Haskell is not so
highly marked in the benchmarks, and Sisal is not a language
of our choice for many reasons.
In the announcement of FISh [FISH] it is said that FISh is
in the order of two magnitudes faster than Haskell, two-four
times faster than Ocaml, and even faster than C because of
the code optimization. Two magnitudes faster than Haskell?
2. It appears to me that over the years several people
have tried using Haskell for scientific purposes, get
disillusioned somehow and either tried some improvements
to Haskell, build something different or abandoned the idea
alltogether. I saw quite ambitious projects, such as
"Functional Programming for Finite Element Analysis"
in Miranda [LIU], but the authors conclusions were quite
discouraging.
What happened to FSC (Functional Scientific Computing)
of Chris Angus [ANGUS], which supposed to address some
ineficiences of Haskell? Is it still alive? I do not know
how related FSC supposes to be to Haskell. Is it a sort of
a Haskell extension, or completely new language that uses
Haskell as a development tool?
Do you agree with Chris'es conclusions in his presentation
in SciTools'96? Mainly that strictness is needed for speed
[ANGUS1]?
Two years have passed since SciTools'96. This is quite a
long time -- considering how fast new ideas evolve nowadays.
Any new development, any new conclusions, new directions
that could be worthwhile to adapt for Haskell?
Why a need for SAC (Single Assignment C)? Aside from
a promotional view that it is still a kind of C? Is there
something definitely inefficient in Haskell that forces some
people to choose an inelegant practicality instead of the
Haskell elegancy?
3. Some researchers conclude that Haskell is not gaining
popularity for scientific applications simply because of
the legacy of Fortran algorithms, which mostly rely on
arrays, indexing and destructive updating. Surely, some
algorithms, such as LU decomposition, do not do destructive
updating -- thus circumventing one of the most time consuming
obstacles.
Some authors [ELLMEN] suggest using indexed lists for
certain algorithms, instead of arrays. These suppose
speed up some algorithms somehow. Any such modules and/or
benchmarks available?
Some people talk about using unboxed arrays in order
to speed up computations. Any pointers to usage of such
in Haskell?
Does anyone working on theory of parallel computations
have any experiences with speed improvements for classical
Haskell algorithms? Polytoping?
Would Psi-calculus be of any help in cranking up some
Haskell computations on arrays? Or is it just something
that helps organize indexing and provides a programmer
with a set of primitives to avoid using error-prone
indexing schemes?
Is the idea of shapes worthy to investigate in this
context, or is it just a concept introduced for other
reasons?
Have quadtrees of David Wise's ([WEISE] and [WEISE1])
proved to be of any importance to scientific computing
in Haskell? Among other things, the quadtree algorithms
supposed to improve array updating schemes. Judging
from the publishing dates (1992, 1995 with a paper version
of the latter scheduled for 1998) this idea is still
very much alive. Has anyone benefited from it yet?
And please, do not argue that Haskell is only a general
purpose language. After all, most of the arguments I found
on Internet for using Haskell at all were of the kind:
"Look Ma, how nicely it resembles the scientific notation!"
Jan
References:
[LIU] - Junxian Liu, Paul Kelly, Stuart Cox. Functional
Programming for Finite Element Analysis, Department of
Computing, Imperial College.
[FISH] - http://www-staff.socs.uts.EDU.AU:8080/~cbj/FISh/
[ANGUS] - http://www.ncl.ac.uk/~n4521971/work.html
[ANGUS1] -
http://www.oslo.sintef.no/SciTools96/noframes/Contrib/angus/p2.html
[ELLMEN] - Nils Ellmenreich and Christian Lengauer. On Indexed Data
Structures and Functional Matrix Algorithms, Faculty of
Mathematics and Informatics, University of Passau
[WEISE] - David S. Wise. Matrix Algorithms using Quadtrees, Invited Talk
ATABLE-92, Technical Report 357, Computer Science Department,
Indiana University, June 1992.
[WEISE1] - David S. Wise. Undulant-Block Elimination and
Integer-Preserving Matrix Inversion, Technical Report 418,
Computer Science Department, Indiana University, August 1995.
