yade-dev team mailing list archive

["Eulitz, Alexander" <Alexander.Eulitz@xxxxxxxxxxxxxxxx>]

Thanks a lot Burak for your work. If you plan to improve the performance with a lot of particles I could run a test script on our server. Maybe we can compare multicore performance (ranging from 1 to 32 CPU threads) of the new integrator then. Just a suggestion.

greetings,

Alex

 

Von: Yade-dev [mailto:yade-dev-bounces+alexander.eulitz=iwf.tu-berlin.de@xxxxxxxxxxxxxxxxxxx] Im Auftrag von Research Ass. Burak Er
Gesendet: Dienstag, 18. Februar 2014 10:28
An: bruno chareyre
Cc: yade-dev@xxxxxxxxxxxxxxxxxxx
Betreff: Re: [Yade-dev] adaptative integration scheme (Burak ER)

 

Dear Bruno,

 

Thank you. About your questions,

 

->inheritance from TimeStepper is probably not what you want. TimeStepper is only a class for computing the timestep dt, not to effectively execute the timestepping. Inheriting from it is harmless but it does not make much sense.

First of all you are right, deriving from TimeStepper makes not much sense, except its name, which, would be a nice name for all kind of time-steppers. Integrator class may be derived from the GlobalEngine directly which will reduce one level of inheritance.

Also,  TimeStepper's name may be changed to "StepSizeEstimator" or something else to prevent confusion. 

-> I am a bit confused by this sentence: "an engine which calls the given list of engines in a sequential order by using the parallel engine functions". Parallel engines are running slaves in parallel but you mention that they are called in a sequential order. Did you really use anything from ParallelEngine?

Yes, I have used ParallelEngine helper functions in order to create an engine which uses the given list of engines as its slaves, and calls them sequentially. While using the action part of the ParalellEngine, I have removed the nested parallelism , thus, engines in the Integrator are called sequentially, you can see the difference in the system function of the Integrator[1] and the action function of ParallelEngine[2] . 

 

->For the consistency of sources formatting in yade, please don't keep blank lines between each lines of codes.

I will also consider this.





burak

 

[1]https://github.com/yade/trunk/blob/master/pkg/dem/Integrator.cpp#L34-L56

[2]https://github.com/yade/trunk/blob/master/pkg/common/ParallelEngine.cpp#L14-L29





________________________________

From: "Bruno Chareyre" <bruno.chareyre@xxxxxxxxxxx>
To: yade-dev@xxxxxxxxxxxxxxxxxxx
Sent: Monday, February 17, 2014 7:05:27 PM
Subject: Re: [Yade-dev] adaptative integration scheme (Burak ER)

 

Dear Burak,

Nice commit, congrats.
There are a few things not very clear to me:
- inheritance from TimeStepper is probably not what you want. TimeStepper is only a class for computing the timestep dt, not to effectively execute the timestepping. Inheriting from it is harmless but it does not make much sense.
- I am a bit confused by this sentence: "an engine which calls the given list of engines in a sequential order by using the parallel engine functions". Parallel engines are running slaves in parallel but you mention that they are called in a sequential order. Did you really use anything from ParallelEngine?
 
For the consistency of sources formatting in yade, please don't keep blank lines between each lines of codes.

Thanks for the two example scripts and the clear comments. It makes the pull request self consistent.

Bruno




On 17/02/14 17:48, Bruno Chareyre wrote:

	forwarding https://github.com/yade/trunk/pull/40

	Bruno

	______ 

	Hi Mr. Chareyre/Gladky,

	As we all already discussed at the topic here[1] I have created a general integrator interface for both multistep and single step methods. I think the interface is flexible for the integration purposes. It inherits the TimeStepper class, thus, it is an engine which calls the given list of engines in a sequential order by using the parallel engine functions,. I have also created a sample adaptive integrator using odeint library where the integration is based on Runge Kutta. The class is called RungeKuttaCashKarp45 and can be found at pkg/dem. I have added an example to compare the NewtonIntegrator and RungeKuttaCashKarp45 that is based on the simple-scene example. The result is very good and the computation is fast for this scene with a low number of object. But as I tried on other simulations with many objects in it, the integrator is slow and needs to be improved.

	Here I want you to merge my master branch with the yade master branch. If you find any bugs or want to improve the code you can contact me.

	Thank you in advance.

	Burak ER

	Research Assistant
	Mechanical Engineering Department
	Bursa Technical University
	mail:burak.er@xxxxxxxxxx
	phone:00902243141658

	[1] https://answers.launchpad.net/yade/+question/242646

	
	
	

	_______________________________________________
	Mailing list: https://launchpad.net/~yade-dev
	Post to     : yade-dev@xxxxxxxxxxxxxxxxxxx
	Unsubscribe : https://launchpad.net/~yade-dev
	More help   : https://help.launchpad.net/ListHelp

 

 

_______________________________________________
Mailing list: https://launchpad.net/~yade-dev
Post to     : yade-dev@xxxxxxxxxxxxxxxxxxx
Unsubscribe : https://launchpad.net/~yade-dev
More help   : https://help.launchpad.net/ListHelp

 


Bu elektronik posta ve ekleri sadece adreste belirtilen kisi veya kurulusun kullanimi icin gonderilmektedir. Bu mesaj tarafiniza yanlislikla ulasirsa, lutfen gonderen kisiyi bilgilendiriniz ve mesaji sisteminizden siliniz. Mesajda  ve eklerinde yer alan bilgilerin her ne sekilde olursa olsun ucuncu kisiler ile paylasilmasi hukuki ve cezai  sorumluluk dogurabilir. Bursa Teknik Universitesi'nin, bu mesaj ve eklerinin icerigi ve yayimi ile ilgili hicbir  sorumlulugu bulunmamaktadir.

This email and the attachments are sent to the individual or entity defined in the address field only. If you are  not the intended recipient or have received the message in error, please notify the sender and remove the message from your system immediately. Sharing the information in the message or the attachments with the 3rd parties may cause legal rules and penalties to apply. Bursa Technical University has no responsibility on the submission of this message and the attachments.