Version française
Home     About     Download     Resources     Contact us    
Browse thread
[Caml-list] Num library
[ Home ] [ Index: by date | by threads ]
[ Search: ]

[ Message by date: previous | next ] [ Message in thread: previous | next ] [ Thread: previous | next ]
Date: -- (:)
From: Sebastien Furic <sebastien.furic@t...>
Subject: Re: [Caml-list] Num library
Pierre Weis wrote:
> 
> >  I made some benchmarks with Dolphin Smalltalk (a pure Smalltalk
> > bytecode interpreter) and O'Caml (using ocamlopt) a few months ago. To
> > my great surprise, Dolphin Samlltalk outperformed O'Caml by a factor of
> > 4 over various tests IIRC. I think the same results may be obtained with
> > other modern big numbers implementations against O'Caml's one.
> [...]
> 
> What do you know about Caml big numbers library ? Are you sure to have
> used the proper layer of the library tuned to performance or have you
> used the casual layer tuned to coding facility ?
> 
> What do you know of the Smalltalk big numbers library ? In which
> language is it written ? What are the algorithms used ?
> 
> What were your benchmark programs ? What are your figures ?
> 
> Be serious on such a topic please! Comparing arithmetic packages is
> not a trivial task and you cannot state ``Dolphin Samlltalk
> outperformed O'Caml by a factor of 4'' without strong arguments and
> evidence that this is indeed true.
> 
> Pierre Weis

 My intention was not to begin a flame war.

 Dolphin's numerical algorithms are written in assembly language, like
all time-critical parts of the interpreter. Unfortunately, Dolphin's
team just announced "new faster algorithms" (and didn't give information
about the algorithms they use).
 Of course, I tried to tune my O'Caml code but I couldn't beat Dolphin's
implementation (despite the use of Nat etc.). Dolphin's bignums are
natively supported by the language and there is no way to hack them.
This is however very convenient for occasional users like me because I
don't have many time to "hack" numerical packages.
 Please note that I'm perfectly happy with O'Caml though.

 Regards,
 Sebastien.
-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners