MathGroup Archive 2010

[Date Index] [Thread Index] [Author Index]

Search the Archive

Re: Compiling in Mathematica 8

  • To: mathgroup at smc.vnet.net
  • Subject: [mg115043] Re: Compiling in Mathematica 8
  • From: Richard Fateman <fateman at cs.berkeley.edu>
  • Date: Wed, 29 Dec 2010 05:57:56 -0500 (EST)
  • References: <if1o60$p7u$1@smc.vnet.net> <ifd0vg$e11$1@smc.vnet.net>

On 12/28/2010 7:50 AM, DmitryG wrote:

>>
>> 2. What about compiling of Mathematica's own routines such as NDSolve?

If Mathematica's compiler could compile NDSolve to make it
run significantly faster, then I expect that WRI would compile it. So it
is probably pointless for you to do so.

Why does compiling of NDSolve not do anything?
Because, unlike your "user code" it is not spending a lot of its time 
doing floating-point arithmetic in the slow, Mathematica-like software way.

It is doing stuff like matching patterns, computing derivatives,
or possibly calling already-compiled numerical routines written in
C or some similar language. While in principal the "call" to match
a pattern might be made a little faster, the time spent in the matching
routine will still vastly dominate the time.

so..

The compiler probably will not make any of these any faster.

RJF




  • Prev by Date: Re: about the use of button
  • Next by Date: Font problem under Linux
  • Previous by thread: Re: Compiling in Mathematica 8
  • Next by thread: Re: Compiling in Mathematica 8