[plt-scheme] Experience Using Mz(Dr)Scheme for Numerical Work

From: Will M. Farr (farr at MIT.EDU)
Date: Mon Apr 18 10:46:50 EDT 2005

Hello all,

Thanks for the great discussion.  In answer to the question about mzc, 
I have used it, and my small benchmark times (factor of 10 slower than 
SBCL, 15 slower than C) are based on a combination of compiled scheme 
code and FFI-ed C code.  mzc did speed up my loops considerably, as 
advertised, but not quite enough....

In fact, I took SICM last term---it's a fantastic class.  What Jerzy 
said is true: Gerry and Jack could have included more geometrical 
mechanics.  However, I think then they would have lost their target 
audience (it's actually the only graduate level mechanics class at MIT, 
so the students run the gamut from theoretical 
astrophysicists---me---to senior-level geology students who need a 
mechanics class).  They did have acceptable performance out of the 
numerical parts of the code (which actually use a nifty scheme for 
automatic differentiation---have a look at the source if you're 
curious) using some special functions (*fp +fp, etc, I think) of MIT 
scheme, which has a good native code compiler.  For me, on PPC, this is 
not available.  If MzScheme had such a native code compiler available I 
would use it without hesitation (though I understand why such a 
compiler has not traditionally been a focus of the PLT group).  For 
those who have access to MIT scheme, I highly recommend the book and 
code; it's too bad that it is so complicated that the porting effort 
lost steam.

Back to my original question: I think I'm settling on... not settling 
on anything yet (though I'm leaning toward CL via SBCL---though I have 
looked at FPC-PPC with openmcl---because I would like to have unboxed 
floating point arrays).  Thanks very much for the suggestions, 
everyone.

Will 



Posted on the users mailing list.