I've compiled CLISP from fources, and hacked the Makefile, with this
optimizations:
gcc -O3 -march=pentium3 -fomit-frame-pointer -ffast-math
Next, I ran this little program. With the standar CLISP included in
Debian, I got:
[1]> (defun kk (cuantos) (dotimes (jaja cuantos jaja) t))
KK
[2]> (compile 'kk)
KK ;
NIL ;
NIL
[3]> (time (kk 20000000))
Real time: 11.865342 sec.
Run time: 9.760517 sec.
Space: 51564560 Bytes
GC: 79, GC time: 0.766881 sec.
20000000
But with the new optimized version I got:
[3]> (defun kk (cuantos) (dotimes (jaja cuantos jaja) t))
KK
[4]> (compile 'kk)
KK ;
NIL ;
NIL
[5]> (time (kk 20000000))
Real time: 8.988631 sec.
Run time: 7.465865 sec.
Space: 51564560 Bytes
GC: 72, GC time: 0.636899 sec.
20000000
which is 25% faster. I haven't made any more tests, but this just shows
up the interesting factor. Perhaps in other areas it might be even
faster.
(Note: both are 2.35 version of CLISP)
Bye.
More tests:
Integer arithmetic:
----------------
http://www.maths.bath.ac.uk/~jap/ak1/youtoo/Bench/arith0.lisp
Normal:
Real time: 0.520267 sec.
Run time: 0.474928 sec.
Space: 0 Bytes
Optimized:
Real time: 0.490464 sec.
Run time: 0.360946 sec.
Space: 0 Bytes
float arithmetic
----------------
http://www.cs.bath.ac.uk/~jap/ak1/youtoo/Bench/arith1.lisp
Normal:
Real time: 0.907093 sec.
Run time: 0.644903 sec.
Space: 17600000 Bytes
GC: 27, GC time: 0.245963 sec.
Optimized:
Real time: 0.740176 sec.
Run time: 0.514922 sec.
Space: 17600000 Bytes
GC: 24, GC time: 0.188969 sec.
Bye.
javuchi wrote:
> I've compiled CLISP from fources, and hacked the Makefile, with this
> optimizations:
>
>
You should run the entire test suite that comes with clisp to make
sure you have not broken something. I have had even pretty simple
C++ programs break under O3.
Wade
Wade Humeniuk ha escrito:
> You should run the entire test suite that comes with clisp to make
> sure you have not broken something. I have had even pretty simple
> C++ programs break under O3.
I did "make test" and it passed ok.
Bye.
From: Sam Steingold
Subject: Re: Interesting: CLISP can be 25% faster
Date:
Message-ID: <uk6etracd.fsf@gnu.org>
> * javuchi <·······@tznvy.pbz> [2005-11-27 13:39:03 -0800]:
>
> Wade Humeniuk ha escrito:
>
>> You should run the entire test suite that comes with clisp to make
>> sure you have not broken something. I have had even pretty simple
>> C++ programs break under O3.
>
> I did "make test" and it passed ok.
and the timing was? (vs the stock version)
--
Sam Steingold (http://www.podval.org/~sds) running w2k
http://truepeace.org http://www.iris.org.il http://www.mideasttruth.com/
http://www.honestreporting.com http://www.dhimmi.com/ http://www.memri.org/
Bus error -- please leave by the rear door.