benchmarks vs racket

[Andy Wingo]

Hi all,

I was recently reminded of the following benchmarks:

  http://www.cs.utah.edu/~mflatt/benchmarks-20100126/log3/Benchmarks.html

All the caveats I just sent to guile-user still apply, but benchmarks
are fun.  I re-ran this one with current Racket (v5.1.3 from Debian) and
Guile (git).  The results are here:

                      fastest  guile  mzscheme 
    conform         1433 ms        1         - 
    cpstack          216 ms     7.62         1 
    ctak            2143 ms        1      1.01 
    dderiv           219 ms     5.00         1 
    deriv            191 ms     4.86         1 
    destruct        1570 ms        1         - 
    div              252 ms     5.75         1 
    dynamic         1423 ms        1         - 
    dynamic2        1213 ms     1.18         1 
    earley           323 ms     1.87         1 
    fft              728 ms     5.53         1 
    graphs           249 ms     8.90         1 
    lattice        12290 ms        1         - 
    lattice2        1584 ms     7.75         1 
    maze            1440 ms        1         - 
    maze2            180 ms     8.51         1 
    mazefun          415 ms     5.64         1 
    nboyer          4428 ms     1.53         1 
    nestedloop       333 ms     8.14         1 
    nfa              491 ms    14.12         1 
    nothing            0 ms        1         1 
    nqueens          225 ms     5.73         1 
    nucleic2         948 ms     5.72         1 
    paraffins        172 ms     2.67         1 
    peval           1840 ms        1         - 
    puzzle           119 ms     5.46         1 
    sboyer          3911 ms     5.81         1 
    scheme           647 ms        1         - 
    scheme2          169 ms     3.86         1 
    sort1           2443 ms        1         - 
    tak              169 ms     8.47         1 
    takl             269 ms     8.81         1 
    takr            1284 ms     1.16         1 
    takr2           1352 ms     1.09         1 
    triangle         439 ms     7.71         1 

The dashes indicate tests that use mutable pairs.  The benchmarks are
from
http://svn.plt-scheme.org/plt/trunk/collects/tests/mzscheme/benchmarks/common/,
checked out today.  I don't know if running racket as mzscheme makes it
slower.  I did fix the ms-per-tick when reading Guile's timings.

Anyway, it's clear that Racket is generally about 5 times as fast as
Guile right now, which is to be expected for the difference between
bytecode and native code.  So I think that we do have a chance, once we
get native code, whenever that happens.

Andy
-- 
http://wingolog.org/