[Scilab-users] unusable cat() rewritten. Call for reviewers

[Claus Futtrup]

Hi Samuel, et al.

 >Doesn't it fit more to what could be expected?

Yes. Absolutely. Sorry that I don't know how to help verify the code.

/Claus

On 10-08-2017 20:03, Samuel Gougeon wrote:
> Le 10/08/2017 à 19:55, Samuel Gougeon a écrit :
>>
>> Dear co-scilabers,
>>
>> The bug 8297 <http://bugzilla.scilab.org/8297> reporting the 
>> outstanding slowness of the cat() function was reported in 2010
>> but lasts since cat() was introduced. As measured and commented in 
>> the report, this
>> slowness actually makes cat() almost unusable, except for small sized 
>> arrays to be concatenated.
>>
>> For instance, after M = rand(500,500); cat(3,M,M) takes more than 11 
>> days (yes)
>> with Scilab 5.5.2, and roughly 4 hours with Scilab 6.0.0.
>>
>> So, everyone could imagine how much Scilab currently takes just to 
>> build a standard
>> 1000x1000 RGB image by stacking its 3  1000x1000 R,G,B layers using 
>> cat(3,R,G,B).
>> The current algorithm is exponential with respect to the array size...
>>
>> Before using GPU routines to process images, a much simpler thing can 
>> be done
>> to enable cat() for true life usages : rewritting it.
>> This is what is done here: https://codereview.scilab.org/#/c/19278/
>>
>> Some rewievers are needed to make cat() truly available in Scilab 6.0.1.
>> So, do not hesitate to contribute online.
>>
>
> To complete: This new implementation spends 0.25 s (instead of 4 
> hours) to execute
> M = rand(500,500);
> cat(3,M,M);
>
> and 0.42 s to process
>
> M = rand(1000,1000);
> cat(3,M,M,M);
>
> Doesn't it fit more to what could be expected?
> SG
>
>
>
> _______________________________________________
> users mailing list
> users at lists.scilab.org
> http://lists.scilab.org/mailman/listinfo/users


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.scilab.org/pipermail/users/attachments/20170810/7ffb5b70/attachment.htm>