progress.txt

pairlist optimization 

Calculate multiple atoms per thread when load one shared block. 

Without pairlist, it works.  On talent with gtx880 GPU, it is 12% faster. 
On JYC with Kelper, it is slower.

With pairlist optimization, simulation is unstable.


- GPU and CPU computes balance 

Create Computes -- balance computes between CPU and GPU 

Computes ( Pair, Self) call initilize(), commenting out GPU code

noWork(),  Modified code to calculate cpu computes

Base(), modified code so as to not skip calculation.