Re: [Scilab-users] ?==?utf-8?q? Scilab 6.1 too slow to list large vectors
Hello Pierre,
FYI your reported issue is probably not related at all to the matrix printing
in the console. The Xcos simulation engine is kind of a gateway from the Scilab
interpreter point of view (like optim() for instance), let's keep them separate
issues.
Thanks,
--
Clément
> -Original Message-
> From: users On Behalf Of Perrichon
> Sent: Tuesday, March 31, 2020 12:05 PM
> To: 'Users mailing list for Scilab'
> Subject: Re: [Scilab-users] ?==?utf-8?q? Scilab 6.1 too slow to list large
> vectors
>
> Hello,
>
> On another way, see bugzilla #16359 in Xcos
>
> I've provided the program to ESI
>
> Results :
> Xcos Scilab 5.5.2 :
> Durée de la simulation : 258.3 s
>
> Xcos Sciab 6.1.0 :
> Durée de la simulation : 4353.4 s
>
> Stack? Mallloc ?
>
> Regards
>
>
>
> Before printing, think about ENVIRONMENTAL responsabity
>
>
> -Message d'origine-
> De : users De la part de Antoine Monmayrant
> Envoyé : mardi 31 mars 2020 11:55 À : Users mailing list for Scilab
> Objet : Re: [Scilab-users] ?==?utf-8?q? Scilab 6.1
> too
> slow to list large vectors
>
> Hello,
>
> I also ran some tests and it seems that the regression is clearly line-based:
> ie the
> regression is worse for vector than for matrix because there more lines to
> print
> to display a vector than a matrix of similar length.
> See my test code below.
>
>
> n1=300;
> n2=100;
> mat=rand(n1,n2);
> tn=[];
>
> /* Ctrl+E the code below several times*/
>
> tic
> mat // matrix
> //mat(:) // vector
> t=toc()
> tn=[tn,t];
>
> disp('-')
> disp('N_run = '+string(length(tn)))
> disp('T='+string(mean(tn))+'+/-'+string(stdev(tn)));
>
>
> /*
> 6.0.2 matrix
> N_run = 20
> T=0.6392742+/-0.056698
> 1.6x slower but plotting 4 values per line ie 4x less lines to plot than for
> the
> matrix version
>
> 6.1.0 matrix
> N_run = 10
> T=1.0335109+/-0.0317414
>
> 6.0.2 vector mat(:)
> N_run = 20
> T=0.4943449+/-0.0662727
>
> 6.1.0 vector mat(:)
> N_run = 10
> T=3.800698+/-0.1121251
> 7.7x slower ie same slow down per line than for the vector version as we plot
> 4
> values per line so 4x more lines
>
> */
>
>
>
> Le Mardi, Mars 31, 2020 11:36 CEST, Federico Miyara
> a écrit:
>
> >
> > Stéphane,
> >
> > I simplified and diversified the test:
> >
> > tic
> > u = rand(100,1000)
> > toc
> >
> > takes
> > 16 s in 6.1
> > 10 s in 6.0.2
> >
> > tic
> > u = u(:)'
> > toc
> >
> > takes
> > 66 s in 6.1
> > 1.29 s in 6.0.2
> >
> > tic
> > u = u(:)
> > toc
> >
> > takes
> > 107 s in 6.1
> > 1.52 s in 6.0.2
> >
> > tic
> > u = matrix(u,1000,100)
> > toc
> >
> > takes
> > 16 s in 6.1
> > 10 s in 6.0.2
> >
> > tic
> > u = matrix(u,100, 100, 10)
> > toc
> >
> > takes
> > 0.5 s in 6.1
> > 0.5 s in 6.0.2
> >
> > Seems as if vertical (many line feeds) print took more time than
> > horizontal print.
> >
> > Regards,
> >
> > Federico Miyara
> >
> >
> >
> > On 31/03/2020 05:23, Stéphane Mottelet wrote:
> > >
> > > Hi Frederico,
> > >
> > > Thanks for reporting. Can you test if the regression also holds for
> > > matrices ?
> > >
> > > S.
> > >
> > > Le 30/03/2020 à 11:56, Federico Miyara a écrit :
> > >>
> > >> Fs = 44100
> > >> T = 2.5
> > >> t = [0:T*Fs]/Fs;
> > >> ximp = exp(-t/0.3).*rand(t,"normal") + 0.004*rand(t,"normal");
> > >>
> > >> tic
> > >> ximp = ximp(:)
> > >> toc
> > > --
> > > Stéphane Mottelet
> > > Ingénieur de recherche
> > > EA 4297 Transformations Intégrées de la Matière Renouvelable
> > > Département Génie des Procédés Industriels Sorbonne Universités -
> > > Université de Technologie de Compiègne CS 60319, 60203 Compiègne
> > > cedex Tel : +33(0)344234688 http://www.utc.fr/~mottelet
> > >
> > > ___
> > > users mailing list
> > > users@lists.scilab.org
> > > http://lists.scilab.org/mailman/listinfo/users
> >
>
> ___
> users mailing list
> users@lists.scilab.org
> http://lists.scilab.org/mailman/listinfo/users
>
> ___
> users mailing list
> users@lists.scilab.org
> http://lists.scilab.org/mailman/listinfo/users
___
users mailing list
users@lists.scilab.org
http://lists.scilab.org/mailman/listinfo/users
Re: [Scilab-users] ?==?utf-8?q? Scilab 6.1 too slow to list large vectors
Hello,
On another way, see bugzilla #16359 in Xcos
I've provided the program to ESI
Results :
Xcos Scilab 5.5.2 :
Durée de la simulation : 258.3 s
Xcos Sciab 6.1.0 :
Durée de la simulation : 4353.4 s
Stack? Mallloc ?
Regards
Before printing, think about ENVIRONMENTAL responsabity
-Message d'origine-
De : users De la part de Antoine Monmayrant
Envoyé : mardi 31 mars 2020 11:55
À : Users mailing list for Scilab
Objet : Re: [Scilab-users] ?==?utf-8?q? Scilab 6.1 too slow to list large
vectors
Hello,
I also ran some tests and it seems that the regression is clearly line-based:
ie the regression is worse for vector than for matrix because there more lines
to print to display a vector than a matrix of similar length.
See my test code below.
n1=300;
n2=100;
mat=rand(n1,n2);
tn=[];
/* Ctrl+E the code below several times*/
tic
mat // matrix
//mat(:) // vector
t=toc()
tn=[tn,t];
disp('-')
disp('N_run = '+string(length(tn)))
disp('T='+string(mean(tn))+'+/-'+string(stdev(tn)));
/*
6.0.2 matrix
N_run = 20
T=0.6392742+/-0.056698
1.6x slower but plotting 4 values per line ie 4x less lines to plot than for
the matrix version
6.1.0 matrix
N_run = 10
T=1.0335109+/-0.0317414
6.0.2 vector mat(:)
N_run = 20
T=0.4943449+/-0.0662727
6.1.0 vector mat(:)
N_run = 10
T=3.800698+/-0.1121251
7.7x slower ie same slow down per line than for the vector version as we plot 4
values per line so 4x more lines
*/
Le Mardi, Mars 31, 2020 11:36 CEST, Federico Miyara
a écrit:
>
> Stéphane,
>
> I simplified and diversified the test:
>
> tic
> u = rand(100,1000)
> toc
>
> takes
> 16 s in 6.1
> 10 s in 6.0.2
>
> tic
> u = u(:)'
> toc
>
> takes
> 66 s in 6.1
> 1.29 s in 6.0.2
>
> tic
> u = u(:)
> toc
>
> takes
> 107 s in 6.1
> 1.52 s in 6.0.2
>
> tic
> u = matrix(u,1000,100)
> toc
>
> takes
> 16 s in 6.1
> 10 s in 6.0.2
>
> tic
> u = matrix(u,100, 100, 10)
> toc
>
> takes
> 0.5 s in 6.1
> 0.5 s in 6.0.2
>
> Seems as if vertical (many line feeds) print took more time than
> horizontal print.
>
> Regards,
>
> Federico Miyara
>
>
>
> On 31/03/2020 05:23, Stéphane Mottelet wrote:
> >
> > Hi Frederico,
> >
> > Thanks for reporting. Can you test if the regression also holds for
> > matrices ?
> >
> > S.
> >
> > Le 30/03/2020 à 11:56, Federico Miyara a écrit :
> >>
> >> Fs = 44100
> >> T = 2.5
> >> t = [0:T*Fs]/Fs;
> >> ximp = exp(-t/0.3).*rand(t,"normal") + 0.004*rand(t,"normal");
> >>
> >> tic
> >> ximp = ximp(:)
> >> toc
> > --
> > Stéphane Mottelet
> > Ingénieur de recherche
> > EA 4297 Transformations Intégrées de la Matière Renouvelable
> > Département Génie des Procédés Industriels Sorbonne Universités -
> > Université de Technologie de Compiègne CS 60319, 60203 Compiègne
> > cedex Tel : +33(0)344234688 http://www.utc.fr/~mottelet
> >
> > ___
> > users mailing list
> > users@lists.scilab.org
> > http://lists.scilab.org/mailman/listinfo/users
>
___
users mailing list
users@lists.scilab.org
http://lists.scilab.org/mailman/listinfo/users
___
users mailing list
users@lists.scilab.org
http://lists.scilab.org/mailman/listinfo/users
Re: [Scilab-users] ?==?utf-8?q? Scilab 6.1 too slow to list large vectors
Hello,
I also ran some tests and it seems that the regression is clearly line-based:
ie the regression is worse for vector than for matrix because there more lines
to print to display a vector than a matrix of similar length.
See my test code below.
n1=300;
n2=100;
mat=rand(n1,n2);
tn=[];
/* Ctrl+E the code below several times*/
tic
mat // matrix
//mat(:) // vector
t=toc()
tn=[tn,t];
disp('-')
disp('N_run = '+string(length(tn)))
disp('T='+string(mean(tn))+'+/-'+string(stdev(tn)));
/*
6.0.2 matrix
N_run = 20
T=0.6392742+/-0.056698
1.6x slower but plotting 4 values per line ie 4x less lines to plot than for
the matrix version
6.1.0 matrix
N_run = 10
T=1.0335109+/-0.0317414
6.0.2 vector mat(:)
N_run = 20
T=0.4943449+/-0.0662727
6.1.0 vector mat(:)
N_run = 10
T=3.800698+/-0.1121251
7.7x slower ie same slow down per line than for the vector version as we plot 4
values per line so 4x more lines
*/
Le Mardi, Mars 31, 2020 11:36 CEST, Federico Miyara
a écrit:
>
> Stéphane,
>
> I simplified and diversified the test:
>
> tic
> u = rand(100,1000)
> toc
>
> takes
> 16 s in 6.1
> 10 s in 6.0.2
>
> tic
> u = u(:)'
> toc
>
> takes
> 66 s in 6.1
> 1.29 s in 6.0.2
>
> tic
> u = u(:)
> toc
>
> takes
> 107 s in 6.1
> 1.52 s in 6.0.2
>
> tic
> u = matrix(u,1000,100)
> toc
>
> takes
> 16 s in 6.1
> 10 s in 6.0.2
>
> tic
> u = matrix(u,100, 100, 10)
> toc
>
> takes
> 0.5 s in 6.1
> 0.5 s in 6.0.2
>
> Seems as if vertical (many line feeds) print took more time than
> horizontal print.
>
> Regards,
>
> Federico Miyara
>
>
>
> On 31/03/2020 05:23, Stéphane Mottelet wrote:
> >
> > Hi Frederico,
> >
> > Thanks for reporting. Can you test if the regression also holds for
> > matrices ?
> >
> > S.
> >
> > Le 30/03/2020 à 11:56, Federico Miyara a écrit :
> >>
> >> Fs = 44100
> >> T = 2.5
> >> t = [0:T*Fs]/Fs;
> >> ximp = exp(-t/0.3).*rand(t,"normal") + 0.004*rand(t,"normal");
> >>
> >> tic
> >> ximp = ximp(:)
> >> toc
> > --
> > Stéphane Mottelet
> > Ingénieur de recherche
> > EA 4297 Transformations Intégrées de la Matière Renouvelable
> > Département Génie des Procédés Industriels
> > Sorbonne Universités - Université de Technologie de Compiègne
> > CS 60319, 60203 Compiègne cedex
> > Tel : +33(0)344234688
> > http://www.utc.fr/~mottelet
> >
> > ___
> > users mailing list
> > users@lists.scilab.org
> > http://lists.scilab.org/mailman/listinfo/users
>
___
users mailing list
users@lists.scilab.org
http://lists.scilab.org/mailman/listinfo/users
Re: [Scilab-users] ?==?utf-8?q? Scilab 6.1 too slow to list large vectors
Antoine, Thank you for testing. I have filed bug 16397: https://bugzilla.scilab.org/show_bug.cgi?id=16397 Regards, Federico Miyara On 30/03/2020 09:05, Antoine Monmayrant wrote: Hello Frederico, I can confirm this (6.0.2 vs 6.1) on linux Ubuntu 18.04 64bits: 6.0.2 -> ~1s 6.1.0 -> ~50s Could you fill a bug report? Antoine Le Lundi, Mars 30, 2020 11:56 CEST, Federico Miyara a écrit: Dear All, I have observed that Scilab 6.1 seems to have a regression respect to 6.0.2. Sometimes one forgets to put semicolon after the coputation of a vector with tens of thousands components. Scilab 6.0.2 listed all the components very fast. That was nice because one hadn't to cancel the computation, it took about 1 s. With 6.1 it takes much longer. For instance Fs = 44100 T = 2.5 t = [0:T*Fs]/Fs; ximp = exp(-t/0.3).*rand(t,"normal") + 0.004*rand(t,"normal"); tic ximp = ximp(:) toc This takes 1.36 s in 6.0.2 The same code in 6.1 takes 182 s Any idea why? I know I can cancel at any moment the script execution, but if there were a process that took a lot of time one wouldn't be willing to stop it... Regards, Federico Miyara ___ users mailing list users@lists.scilab.org http://lists.scilab.org/mailman/listinfo/users ___ users mailing list users@lists.scilab.org http://lists.scilab.org/mailman/listinfo/users
Re: [Scilab-users] ?==?utf-8?q? Scilab 6.1 too slow to list large vectors
Hi again, Just tested using the cli (no window, no java): it's even more : 0.34s vs 46s. Antoine Le Lundi, Mars 30, 2020 11:56 CEST, Federico Miyara a écrit: > > Dear All, > > I have observed that Scilab 6.1 seems to have a regression respect to > 6.0.2. Sometimes one forgets to put semicolon after the coputation of a > vector with tens of thousands components. Scilab 6.0.2 listed all the > components very fast. That was nice because one hadn't to cancel the > computation, it took about 1 s. With 6.1 it takes much longer. For instance > > Fs = 44100 > T = 2.5 > t = [0:T*Fs]/Fs; > ximp = exp(-t/0.3).*rand(t,"normal") + 0.004*rand(t,"normal"); > > tic > ximp = ximp(:) > toc > > This takes 1.36 s in 6.0.2 > > The same code in 6.1 takes 182 s > > Any idea why? > > I know I can cancel at any moment the script execution, but if there > were a process that took a lot of time one wouldn't be willing to stop it... > > Regards, > > Federico Miyara ___ users mailing list users@lists.scilab.org http://lists.scilab.org/mailman/listinfo/users
Re: [Scilab-users] ?==?utf-8?q? Scilab 6.1 too slow to list large vectors
Hello Frederico, I can confirm this (6.0.2 vs 6.1) on linux Ubuntu 18.04 64bits: 6.0.2 -> ~1s 6.1.0 -> ~50s Could you fill a bug report? Antoine Le Lundi, Mars 30, 2020 11:56 CEST, Federico Miyara a écrit: > > Dear All, > > I have observed that Scilab 6.1 seems to have a regression respect to > 6.0.2. Sometimes one forgets to put semicolon after the coputation of a > vector with tens of thousands components. Scilab 6.0.2 listed all the > components very fast. That was nice because one hadn't to cancel the > computation, it took about 1 s. With 6.1 it takes much longer. For instance > > Fs = 44100 > T = 2.5 > t = [0:T*Fs]/Fs; > ximp = exp(-t/0.3).*rand(t,"normal") + 0.004*rand(t,"normal"); > > tic > ximp = ximp(:) > toc > > This takes 1.36 s in 6.0.2 > > The same code in 6.1 takes 182 s > > Any idea why? > > I know I can cancel at any moment the script execution, but if there > were a process that took a lot of time one wouldn't be willing to stop it... > > Regards, > > Federico Miyara ___ users mailing list users@lists.scilab.org http://lists.scilab.org/mailman/listinfo/users
