Re: slow LU decomposiotion? for octave 4 for windows

[Tatsuro MATSUOKA]

----- Original Message -----

> From: Tatsuro MATSUOKA 
> To: tmacchant siko1056 "octave-maintainers
> Cc: 
> Date: 2016/6/1, Wed 17:40
> Subject: Re: slow LU decomposiotion? for octave  4 for windows
> 
> ----- Original Message -----
> 
>>  From: Tatsuro MATSUOKA 
>>  To: siko1056  "octave-maintainers
>>  Cc: 
>>  Date: 2016/6/1, Wed 07:50
>>  Subject: Re: slow LU decomposiotion? for octave  4 for windows
>> 
>>  ----- Original Message -----
>> 
>>>   From: siko1056 
>>>   To: octave-maintainers
>>>   Cc: 
>>>   Date: 2016/5/31, Tue 21:07
>>>   Subject: Re: slow LU decomposiotion? for octave  4 for windows
>>> 
>>>   Dear Tatsuro MATSUOKA,
>>> 
>>>   Do you also this performance issue for the vectorized test script in 
> all
>>>   your versions? I see similar good results on my PC without for-loops 
> and
>>>   Octave 4.0.1 and dev.
>>> 
>>>   ------------------
>>>   more off 
>>> 
>>>   Num=1000;
>>>   rand('seed',1); 
>>>   A=rand(Num)-0.5; 
>>>   rand('seed',2); 
>>>   B=rand(Num,ItNum)-0.5; 
>>>   [L U P]=lu(A); 
>>>   % 
>>>   disp('Simple left division'); 
>>>   tic; 
>>>   x=A\B;  % vectorized
>>>   toc; 
>>>   x1=x(:,end); 
>>>   % 
>>>   disp('LU decomposition'); 
>>>   tic; 
>>>   c=P*B; y=L\c; x=U\y;  % vectorized
>>>   toc; 
>>>   x2=x(:,end); 
>>>   id=1:Num; 
>>>   plot(id,x1, 'o1',id,x2, '+2'); 
>>>   ------------------
>>> 
>>>   This would make me struggle. For-loops are slow (Rik held an excellent 
> talk
>>>   about this at the Octconf 2015, http://wiki.octave.org/OctConf_2015) 
> and
>>>   maybe older versions of Octave handled loops in another way?!
>>> 
>>>   Regards,
>>>   Kai
>>> 
>>  Kai
>> 
>>  Thank you for vectorized test script.
>> 
>>  One correction
>> 
>>  Num=1000;
>>  |
>>  V
>>  Num=1000; ItNum=10;
>> 
>> 
>> 
>>  ************************************ 
>>  Octave-3.2.4 mingw 
>>>>   lutest_v 
>> 
>>  Simple left division
>>  Elapsed time is 0.114076 seconds.
>>  LU decomposition
>>  Elapsed time is 0.00600397 seconds.************************************ 
>>  Octave-3.6.4 msvc 
>> 
>> 
>>>>   lutest_v
>>  Simple left division
>>  Elapsed time is 0.11 seconds.
>>  LU decomposition
>>  Elapsed time is 0.00399995 seconds.
>>  ************************************ 
>>  Octave-3.6.4 mingw 
>> 
>> 
>>>>   lutest_v
>>  Simple left division
>>  Elapsed time is 0.0790549 seconds.
>>  LU decomposition
>>  Elapsed time is 0.00300312 seconds.
>>  ************************************ 
>>  Octave-3.8.2 mingw 
>> 
>> 
>>>>   lutest_v
>>  Simple left division
>>  Elapsed time is 0.096066 seconds.
>>  LU decomposition
>>  Elapsed time is 0.00300384 seconds.
>>  ************************************ 
>>  Octave-4.0.0 mingw (32 bit) 
>> 
>> 
>>>>   lutest_v
>>  Simple left division
>>  Elapsed time is 0.07305 seconds.
>>  LU decomposition
>>  Elapsed time is 0.01701 seconds.
>>  ************************************ 
>>  Octave-4.0.2 mingw (32 bit) 
>> 
>>  Simple left division
>>  Elapsed time is 0.060039 seconds.
>>  LU decomposition
>>  Elapsed time is 0.0180109 seconds.
>>  ************************************ 
>>  Octave-4.0.2 mingw (64 bit) 
>> 
>>>>   lutest 
>> 
>>  Simple left division
>>  Elapsed time is 0.0500329 seconds.
>>  LU decomposition
>>  Elapsed time is 0.00900698 seconds.
>>  ************************************ 
>>  octave-4.1.0+ mingw (64bit) 
>>  (hg clone on May 28, 2016) 
>> 
>>>>   lutest 
>> 
>>  Simple left division
>>  Elapsed time is 0.213143 seconds.
>>  LU decomposition
>>  Elapsed time is 0.0230169 seconds.************************************
>> 
>> 
>>  As you told, there is issue of slow loop.
>>  However, even vetorized, LU decomposition on version 4 on windows is slower 
> than 
>>  version  3.
>>  Surprisingly, results on 4.1.0+ are the worst. This is a bad situation and 
>>  should be improved.
>> 
>>  Is it better to be file to a bug?
>> 
>>  Tasuro 
> 
> 
> Before filing this to the bug tracker I show test results on Ubuntu 14.04 
> 64bit.
> (Athlon X2 not so fast)
> 
> 
> % lutest_v.m
> more off 
>  
> Num=1000; ItNum=10;
> rand('seed',1); 
> A=rand(Num)-0.5; 
> rand('seed',2); 
> B=rand(Num,ItNum)-0.5; 
> [L U P]=lu(A); 
> % 
> disp('Simple left division'); 
> tic; 
> x=A\B;  % vectorized
> toc; 
> x1=x(:,end); 
> % 
> disp('LU decomposition'); 
> tic; 
> c=P*B; y=L\c; x=U\y;  % vectorized
> toc; 
> x2=x(:,end); 
> id=1:Num; 
> plot(id,x1, 'o1',id,x2, '+2'); 
> 
> % end of lutest_v.m
> 
> 
> ************************************
> 
> 3.8.1 from Ubuntu repository 
>>>  lutest_v
> Simple left division
> Elapsed time is 0.396069 seconds.
> LU decomposition
> Elapsed time is 0.0427001 seconds.
> ************************************
> 4.0.0 bult myself (gcc 4.8.4) 
>>>  lutest_v
> Simple left division
> Elapsed time is 0.378708 seconds.
> LU decomposition
> Elapsed time is 0.071897 seconds.
> ************************************
> 
> 4.0.2 bult myself (gcc 4.8.4) 
>>>  lutest_v
> Simple left division
> Elapsed time is 0.377592 seconds.
> LU decomposition
> Elapsed time is 0.0599658 seconds.
> ************************************
> 
> 4.1.0+ bult myself (gcc 4.8.4) (cloned 2016-06-01 JST)
>>>  lutest_v
> Simple left division
> Elapsed time is 0.378928 seconds.
> LU decomposition
> Elapsed time is 0.0592132 seconds.
> ************************************
> 
> 
> For simple division, the differences within tolerance.
> The LU decomposition is the slowest on 4.0.0 and fastest on 3.8.1.
> But difference are small comparing the cases on windows.
> In addition. slowness of simple division observed on 4.1.0+ does not appear 
> on 
> Ubuntu 14.04 64 bit.
> 
> Slowness of LU decomposition on octave 4 windows is not allowable, I think.
> 
> I would like to see test by other people and opinions.
> 
> Regards
> 
> ******************************************
> Tatsuro MATSUOKA
> 
> Department of Chemical Engineering
> Nagoaya University, Japan
> ******************************************



I have filed to the bug tracker.
Further discussion will be made there.

http://savannah.gnu.org/bugs/?48085


Tatsuro