First, please have a look at this site:
http://www.timestretch.com/FractalBenchmark.html
I ran the python code on my machine with Python 2.5 and got the result
of 4.36 secs(compared to 15.33 secs shown on the page). Adding import
psyco;psyco.full() at the top of the code, which does run-time code
optimization, increased speed: 0.98 secs. Cool. Now I thought J could
do better, or at least do as well, since this kind of problem is what
J's good at.
Now I translated the code into J. First version, which was more or
less sentence to sentence translated, ran for more than a minute. I
optimised incrementally, from vectorizing x value and then to
vectorizing both x and y. Now I get 9.87 secs with the following code:
BAILOUT=. 16
MAX_ITERATIONS=. 1000
run=: 3 : 0
smoutput 'Rendering...'
i=. mb~ 40%~ _39+i.78
(i<:BAILOUT){' *'
)
mb=: 4 : 0
cr=. y-0.5
ci=. x
zi=. (y,&#x)$0
zr=. (y,&#x)$0
for. i. MAX_ITERATIONS do.
temp=. zr*zi
zr2=. *:zr
zi2=. *:zi
zr=. (cr+zr2)-zi2
zi=. (+:temp)+"1 ci
end.
zi2+zr2
)
Now compare this to the running time of Vectorized Io, which is 0.41
secs according to the benchmark on the web page.
I could use complex numbers, which is more natural for this problem.
What else can I do to speed it up significantly?
Thanks in advance.
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
