[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Scheme & number crunching
>>>>> "MF" == Matthias Felleisen <matthias@cs.rice.edu> writes:
MF> If you're willing to do whole-program analysis, "Scheme" programs can beat
MF> equivalent C number crunching programs. Both Gambit and Stalin (argh) have
MF> proven this point over the years. The problem of infering decent types for
MF> loop-ish programs is no harder than in ML. These Scheme implementations
MF> unbox, allocate in stack regions, call closures directly, etc.
MF> Having said that, I must admit that I am not sure Schemers should play this
MF> game and I am sure that PLT won't play this game. I have watched this for
MF> many years, and I just don't think this is going to give Scheme the break
MF> it needs.
MF> We need to compete with the scripters and show that a good semantics
MF> matters and that scripts with a good semantics can evolve into real
MF> programs.
Matthias is absolutely right. Scheme is not a primary contestant for number
crunching applications even if it could be made to perform as well as C.
Please, do not underestimate the conservatism of number crunching community.
Over the years Fortran lost most of its appeal but it is still alive and
kicking in the numeric applications. Even if somebody comes up with a very good
Scheme compiler optimized for numeric calc and sims I would not bet my money on
its acceptance.
Scripting on the other hand is the area where Scheme (MzScheme) can leverage its
muscles faster and with more chances for success.
-- Leo
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+ Leonid Razoumov, + E-mail: lrazoumov@qualcomm.com +
+ Qualcomm Inc., + http://www.qualcomm.com +
+ 5775 Morehouse Drive, + +
+ San Diego, CA 92121-1714, + VOICE: +1-858/651-5163 +
+ USA + FAX: +1-858/658-2113 +
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++