regression between 2.1.71 and 2.9.5 ?

[Laurent Mazet]

Hi,

Based on the email from Shai Ayal, I've done my own tests on octave 2.1.71 and
a fresh build of 2.9.5 and I got this :

GNU Octave, version 2.9.5 (i686-pc-linux-gnu).

octave:7> s0=cputime ;a=[];for ii=1:10000;a=[a,ii];endfor;cputime-s0
ans = 1.9157
octave:8> s0=cputime ;a=[];for ii=1:10000;a=[a,ii];endfor;cputime-s0
ans = 1.9227
octave:9> s0=cputime ;a=[];for ii=1:10000;a(ii)=ii;endfor;cputime-s0
ans = 0.96885
octave:10> s0=cputime ;a=[];for ii=1:10000;a(ii)=ii;endfor;cputime-s0
ans = 0.95586

GNU Octave, version 2.1.71 (i586-mandriva-linux-gnu).

octave:5> s0=cputime ;a=[];for ii=1:10000;a=[a,ii];endfor;cputime-s0
ans = 2.2187
octave:6> s0=cputime ;a=[];for ii=1:10000;a=[a,ii];endfor;cputime-s0
ans = 2.2317
octave:7> s0=cputime ;a=[];for ii=1:10000;a(ii)=ii;endfor;cputime-s0
ans = 0.63690
octave:8> s0=cputime ;a=[];for ii=1:10000;a(ii)=ii;endfor;cputime-s0
ans = 0.64590

I've discussed with David and he thought that the speed incease on
concatenation is due to the special casing same type concatenation but he
didn't understand the speed regression for indexed assignments. Any thoughts
why the regression occurs?

Regards,

         Laurent
-- 
Dr. Laurent Mazet: Research Engineer /V\ Centre de Recherche de MOTOROLA
Tel: +33 1 69 35 48 30 =-=-=-=-=-=-=-= Email: address@hidden