The testsuite run times are somewhat inconsistent. Maybe something like
f(n) := block([], modedeclare([i, n], integer), for i : 1 thru n do (sin(i), cos(i), log(i), tan(i), atan2(1,i)));
(compiled) would be a better test? For f(10^6), I got 19.3300 seconds without rearrangement,
and 17.7900 after.
--Barton
________________________________________
From: Richard Fateman [fateman at eecs.berkeley.edu]
Sent: Sunday, August 05, 2012 19:38
To: Barton Willis
Cc: Maxima - list
Subject: Re: [Maxima] property lisp hack
On 8/5/2012 5:31 PM, Barton Willis wrote:
> Using the re-arrangement code:
>
> took 214,214,000 microseconds (214.214000 seconds) to run.
> 18,728,831 microseconds ( 18.728832 seconds, 8.74%) of which was spent in GC.
>
> Without:
>
> took 217,601,000 microseconds (217.601000 seconds) to run.
> 17,750,023 microseconds ( 17.750025 seconds, 8.16%) of which was spent in GC.
>
> This is about a 1.6% speed up---I only tried the test once.
>
>
The re-arrangement should not have affected the garbage collection at
all. Your test suggests it
slowed it down! If we factor out the GC time, it seems the
rearrangement was even more effective.
Thanks for trying it out.
RJF