[gmx-developers] Suggestion: move bonded forces in mdrun

[Berk Hess]

David van der Spoel wrote:

>On Wed, 29 Jun 2005, Carsten Kutzner wrote:
>
>  
>
>>On Wed, 29 Jun 2005, Berk Hess wrote:
>>
>>    
>>
>>>Do you mean that one could compensate for the larger bonded interaction load
>>>by doing less non-bonded interactions on such a cpu?
>>>      
>>>
>>Yes, but that is not possible if PME is inbetween.
>>
>>    
>>
>>>But after looking in the code I was that also PME uses whole molecules,
>>>so it is very easy to change indeed.
>>>      
>>>
>>Could anybody who has CVS write access do that change or is there
>>any reason against it?
>>    
>>
>
>Actually I have a tree in which this is implemented, but it is not 
>completely trivial to get it working with shifts. One does need an extra 
>coordinate array to do this. In fact the code doesn't work for me yet. 
>This patch also removes the need for multi-cpu tpr files and 
>splits all bonded forces over all processors, at the cost of doing 
>communication of all coordinates and forces over all the processors.
> 
>  
>
But you don't need to shift, I think.

PME uses whole molecules, just like the bondeds.

Berk.