[Gcl-devel] Re: tespol benchmark

[Camm Maguire]

Greetings, and thanks for the kind words.

Here is what I'm really striving for:

(defun eval-pol1 (n x)
  (declare (single-float x))
  (let ((su 0.0) (mu 10.0) (pu 0.0)
        (pol (make-array 100 :element-type 'single-float)))
    (dotimes (i n)
             (setf su 0.0)
             (dotimes (j 100)
                      (setf mu (/ (+ mu 2.0) 2.0))
                      (setf (aref pol j) mu))
             (dotimes (j 100)
                      (setf su (+ (aref pol j) (* su x))))
             (setf pu (+ pu su))
             )
    (prin1 pu)))

(defun eval-pol (n x)
  (declare (fixnum n) (single-float x))
  (let ((su 0.0) (mu 10.0) (pu 0.0)
        (pol (make-array 100 :element-type 'single-float)))
    (declare (single-float su) (single-float mu) (single-float pu))
    (dotimes (i n)
             (declare (fixnum i))
             (setf su 0.0)
             (dotimes (j 100)
                      (declare (fixnum j))
                      (setf mu (the single-float (/ (+ mu 2.0) 2.0)))
                      (setf (aref pol j)
                            (the single-float mu)))
             (dotimes (j 100)
                      (declare (fixnum j))
                      (setf su (the single-float
                                    (+ (aref pol j) (the single-float (* su 
x))))))
             (setf pu (the single-float (+ pu su)))
             )
    (prin1 pu)
    ))

(compile 'eval-pol)
(compile 'eval-pol1

(time (eval-pol1 500000 0.2))
1250000.0
real time       :      1.890 secs
run-gbc time    :      1.890 secs
child run time  :      0.000 secs
gbc time        :      0.000 secs
1250000.0

>(time (eval-pol 500000 0.2))
1250000.0
real time       :      1.890 secs
run-gbc time    :      1.890 secs
child run time  :      0.000 secs
gbc time        :      0.000 secs
1250000.0

Take care,


Robert Boyer <address@hidden> writes:

> Interesting benchmark below, after
> 
>   http://dan.corlan.net/bench.html
> 
> Conclusion: Here, GCL 2.7.0 is 6 times faster than Allegro
> and is 1.5 times faster than cmu|sbcl but still only half as
> fast as Fortran.
> 
>   For new projects in which one is free to choose the
>   language, one should choose Lisp.  --  Corlan
> 
> What's most amazing, even incredible to me, is that I did
> not put in a single Lisp declaration myself, except for the
> optimize, so that all Lisps were shooting at the same
> target.  Your work on the GCL compiler is awesome.
> 
> Bob
> 
> P. S.  Benchmarking is such a black art that this message
> should not have been written.  But I just can't stop.  Lord
> only knows what all I did wrong this time.
> 
> -------------------------------------------------------------------------------
> 
> Transcript just now on antones.cs.utexas.edu, Ubuntu,
> (Pentium 4 3.00GHz).
> 
> % cat tespol.f
>       program tespol
>       dimension pol(100)
>       real pol
>       integer i,j,n
>       real su,pu,mu
>       real x
> 
>       n = 500000
>       x = 0.2
>       mu = 10.0
>       pu = 0.0
>       do i = 1,n
>          do j=1,100
>             mu = (mu + 2.0) / 2.0
>             pol(j) = mu
>          enddo
>          su = 0.0
>          do j=1,100
>             su = x * su + pol(j)
>          enddo
>          pu = pu + su
>       enddo
>       write (*,*) pu
>       end
> % g77 tespol.f -O6 -o tespol
> time ./tespol
> %   1250000.
> 0.476u 0.000s 0:00.47 100.0%  0+0k 0+0io 0pf+0w
> 
> % cat tespol.l
> (declaim (optimize (safety 0) (speed 3)))
> 
> (defun eval-pol (n x)
>   (declare (fixnum n) (single-float x))
>   (let ((su 0.0) (mu 10.0) (pu 0.0)
>       (pol (make-array 100 :element-type 'single-float)))
>     (declare (single-float su) (single-float mu) (single-float pu))
>     (dotimes (i n)
>            (declare (fixnum i))
>            (setf su 0.0)
>            (dotimes (j 100)
>                     (declare (fixnum j))
>                     (setf mu (the single-float (/ (+ mu 2.0) 2.0)))
>                     (setf (aref pol j)
>                           (the single-float mu)))
>            (dotimes (j 100)
>                     (declare (fixnum j))
>                     (setf su (the single-float
>                                   (+ (aref pol j) (the single-float (* su 
> x))))))
>            (setf pu (the single-float (+ pu su)))
>            )
>     (prin1 pu)
>     ))
> 
> (compile 'eval-pol)
> 
> (time (eval-pol 500000 0.2))
> % xg
> GCL (GNU Common Lisp)  2.7.0 ANSI    Jun 23 2006 08:10:39
> Source License: LGPL(gcl,gmp,pargcl), GPL(unexec,bfd)
> Binary License:  GPL due to GPL'ed components: (BFD UNEXEC)
> Modifications of this banner must retain notice of a compatible license
> Dedicated to the memory of W. Schelter
> 
> Use (help) to get some basic information on how to use GCL.
> 
> Temporary directory for compiler files set to /tmp/
> 
> >(load "tespol.l")
> 
> Loading tespol.l
> ;; Compiling /tmp/gazonk_30441_0.lsp.
> ;; End of Pass 1.  
> ;; End of Pass 2.  
> ;; OPTIMIZE levels: Safety=0 (No runtime error checking), Space=0, Speed=3, 
> (Debug quality ignored)
> ;; Finished compiling /tmp/gazonk_30441_0.o.
> Loading /tmp/gazonk_30441_0.o
> start address -T 0xac23a00 Finished loading /tmp/gazonk_30441_0.o
> 1250000.0
> real time       :      1.080 secs
> run-gbc time    :      1.080 secs
> child run time  :      0.000 secs
> gbc time        :      0.000 secs
> Finished loading tespol.l
> T
> 
> >% acl
> International Allegro CL Enterprise Edition
> 7.0 [Linux (x86)] (Jun 29, 2005 9:23)
> Copyright (C) 1985-2004, Franz Inc., Oakland, CA, USA.  All Rights Reserved.
> 
> This development copy of Allegro CL is licensed to:
>    [TC8015] University of Texas
> 
> ;; Optimization settings: safety 1, space 1, speed 1, debug 2.
> ;; For a complete description of all compiler switches given the
> ;; current optimization settings evaluate (EXPLAIN-COMPILER-SETTINGS).
> CL-USER(1): (load "tespol.l")
> ; Loading /v/filer2/boyer/tespol.l
> 1250000.0
> ; cpu time (non-gc) 6,810 msec user, 10 msec system
> ; cpu time (gc)     0 msec user, 0 msec system
> ; cpu time (total)  6,810 msec user, 10 msec system
> ; real time  6,814 msec
> ; space allocation:
> ;  12 cons cells, 512 other bytes, 0 static bytes
> T
> CL-USER(2): EOF
> Really exit lisp [n]? y
> ; Exiting
> 6.832u 0.012s 0:15.88 43.0%   0+0k 0+0io 0pf+0w
> % sbcl
> This is SBCL 0.9.8, an implementation of ANSI Common Lisp.
> More information about SBCL is available at <http://www.sbcl.org/>.
> 
> SBCL is free software, provided as is, with absolutely no warranty.
> It is mostly in the public domain; some portions are provided under
> BSD-style licenses.  See the CREDITS and COPYING files in the
> distribution for more information.
> * (load "tespol.l")
> ; in: LAMBDA NIL
> ;     (PRIN1 PU)
> ; 
> ; note: doing float to pointer coercion (cost 13) from PU
> ; 
> ; compilation unit finished
> ;   printed 1 note
> 1250000.0
> Evaluation took:
>   1.658 seconds of real time
>   1.656104 seconds of user run time
>   0.0 seconds of system run time
>   0 page faults and
>   0 bytes consed.
> T
> * % cmucl
> CMU Common Lisp 19b (19B), running on antones.cs.utexas.edu
> With core: /v/filer3/v0q027/cmucl/lib/cmucl/lib/lisp.core
> Dumped on: Mon, 2005-06-27 19:09:58-05:00 on lorien
> See <http://www.cons.org/cmucl/> for support information.
> Loaded subsystems:
>     Python 1.1, target Intel x86
>     CLOS based on Gerd's PCL 2004/04/14 03:32:47
> * (load "tespol.l")
> 
> ; Loading #P"/v/filer2/boyer/tespol.l".
> ; Compiling LAMBDA (N X): 
> 
> ; In: LAMBDA (N X)
> 
> ;   (PRIN1 PU)
> ; Note: Doing float to pointer coercion (cost 13) from PU.
> ; 
> ; Compiling Top-Level Form: 
> 
> ; Compilation unit finished.
> ;   1 note
> 
> ; Compiling LAMBDA NIL: 
> ; Compiling Top-Level Form: 
> 1250000.0
> ; Evaluation took:
> ;   1.58 seconds of real time
> ;   1.556097 seconds of user run time
> ;   0.0 seconds of system run time
> ;   4,740,004,193 CPU cycles
> ;   0 page faults and
> ;   1,184 bytes consed.
> ; 
> T
> * (quit)
> % 
> 
> 
> 

-- 
Camm Maguire                                            address@hidden
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah