Author: Vincent Diepeveen
Date: 03:11:09 08/01/03
Go up one level in this thread
On August 01, 2003 at 06:06:38, Vincent Diepeveen wrote: >On July 31, 2003 at 17:32:47, Dan Andersson wrote: > >> ML or SML stands for Meta Language. A functional language commonly used for >>teaching functional programming. Best variant IMO is OCaml that includes object >>orientation and blazing performance when running both bytecode and natively. But >>Scheme is my preferred functional language. Second is Haskell and it sports the >>most clean implementation of a lazy functional PL. >> >>MvH Dan Andersson > >With all respect Dan, but functional languages are a pure waste of your time. > >Haskell in combination with the pure functional Gofer for example is a factor >10000 slower (interpreted of course, because you can't compile it unless you use >the outdated turbo c++) litterary. > >I had written a checkers program (8x8 polish rules though) in Gofer/Haskell. > >It searched 1 node each so many seconds thanks to the dead slow lazy evaluation >of the language. > >Then i ported it straight on to C and that ran at 10000 nodes a second. > >That was very unoptimized C code. > >Note that a person who promoted to doctor (not sure what it is called in >english) has put in a lot of effort speeding up my gofer code. > >He managed after months of work to get it 2 times faster. > >We talk about a factor 5000 difference. > >At the same time i managed to get the checkersprogram in C also a factor 4 >faster by writing a smarter move generator, which thanks to the imperative >nature of it (it's using pregenerated tables of course) is not possible to use >in Gofer without slowing it down a factor X first. > >The most important thing however is that any serious application cannot be >developed in such functional languages, even if it was a factor 5000 faster. > >You end up with zillions of short functions. > >In 100 line programs that might perhaps still sound cool, but as you understand >a simple checkers program is already quite some longer and then it already tends >to get a mess, even when written by doctor Gofer. > >Best regards, >Vincent If you wonder about compiled gofer code. It ran a factor 50 slower than the original C code. So that's nearly factor 200 for the optimized C code versus optimized gofer code. Note this was performed under DOS, because it could only compile using turbo c++. As we know that's a dos compiler which is doing very poor optimizations for the pentium 100 processor where i compiled it upon. Basically most of the programming was done upon HP-UX machines so i have no good idea how slow those are when compared to x86 architecture. Best regards, Vincent
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.