← Back to team overview

yade-dev team mailing list archive

Re: [Yade-users] Optimized collider needs testing

 

Ok thanks. It explains.

lp:~bruno-chareyre/yade/collide2 contains further improvements (and code
cleaning) as compared to lp:~bruno-chareyre/yade/collide, which was
already faster than trunk. But before yesterday, collide2 was simply trunk.

collide without the "2" should not exist actually, it's not listed here
for instance: https://code.launchpad.net/yade
I'm surprised that you could download it.

Bruno

On 28/09/11 04:49, Klaus Thoeni wrote:
> Hi Bruno,
>
> I checked out your branch lp:~bruno-chareyre/yade/collide (without 2 at the 
> end as indicated in the bzr co part of your first posting) when I tested it. I 
> just saw now that there is as well a branch lp:~bruno-chareyre/yade/collide2. 
> Probably that's the one to test. This explains why it was working. I will do 
> the tests with this branch now.
>
> So guys if you want to test Bruno's new implementation make sure you check out 
> the right branch:
> bzr checkout lp:~bruno-chareyre/yade/collide2
>
> Cheers,
> Klaus
>
> On Tue, 27 Sep 2011 10:43:28 pm Bruno Chareyre wrote:
>> There was indeed a bug with clumps (fixed), but the thing I don't
>> understand is the branch did not contain the changes. Bazaar branches
>> are not quite user-friendly for commits/updates...
>> Klaus, you probably didn't try the good code (but then how did you spot
>> the bug? I'm lost...).
>> Could you send me a copy of the core/bound.hpp you downloaded, just to
>> be sure?
>>
>> Bruno
>>
>> On 26/09/11 10:37, Bruno Chareyre wrote:
>>> (moving to yade-dev)
>>>
>>> No guess yet. Thanks for notifying. Which scripts gives the segfault?
>>>
>>> B
>>>
>>> On 26/09/11 07:43, Klaus Thoeni wrote:
>>>> Hi Bruno,
>>>>
>>>> I tested your implementation with my examples (dynamic problem with
>>>> about 10000 particles) and the computational time in
>>>> InsertionSortCollider slightly improved as expected.
>>>>
>>>> However, when running your branch with examples with clumps gives a
>>>> Segmentation Fault. Any guess?
>>>>
>>>> Klaus
>>>>
>>>> On Tue, 20 Sep 2011 07:29:02 AM Bruno Chareyre wrote:
>>>>> Hi all,
>>>>>
>>>>> I worked on some optimizations of contact detection recently and I
>>>>> would like to get feedback from you to detect possible bugs and
>>>>> confirm the speedup for different applications. If everything is ok,
>>>>> the changes can be part of the next release after some code cleanup
>>>>> and documentation.
>>>>>
>>>>> If you have a chance, could you please download and compile the branch
>>>>> lp:~bruno-chareyre/yade/collide2 (bzr checkout
>>>>> lp:~bruno-chareyre/yade/collide, cd yade, scons, as usual) and try this
>>>>> with your own problems?
>>>>> I would be gratefull if you could report differences in terms of
>>>>> computation speed.
>>>>> I don't expect big improvements for small quasistatic problems. The
>>>>> speedup is more likely to happen on large and/or dynamic problems,
>>>>> where it should range between x2 and x3, or even more in multithread.
>>>>> Some results are reported here:
>>>>> https://yade-dem.org/wiki/Colliders_performace#Improved_InsertionSort
>>>>>
>>>>> Thanks.
>>>>>
>>>>> Bruno


-- 
_______________
Bruno Chareyre
Associate Professor
ENSE³ - Grenoble INP
11, rue des Mathématiques
BP 46
38402 St Martin d'Hères, France
Tél : +33 4 56 52 86 21
Fax : +33 4 76 82 70 43
________________



Follow ups

References