NewtonSetSolverModel and NewtonSetFrictionModel

[Witek900]

Hi

I wolud like to know what what exactly NewtonSetSolverModel and NewtonSetFrictionModel do, and what values are better for performance.
And the same question about NewtonSetMultiThreadSolverOnSingleIsland and NewtonSetPlatformArchitecture. I would like to have high performance.

[Julio Jerez]

there are two solve algorithm in Newton, one is colacual ethe reaction form ces in at most a fix number of steps.
teh time complxity is propostional to teh square of eh number of consteration forces.
NewtonSetSolverModel when set to zero selct thsi solver
when set to anthong oteh than zero it uses teh iteratiove solve that calculater teh foces making a smal contribuition in each pass.
teh numeb defien teh number of passes. teh mor epasses the more teh smaller teh errope betwenn teh solution and the slected forces.

NewtonSetFrictionModel each solver use a frition model
mode exact means the friction force will be calculated corrently in each call to newton update.
the adaptive usen teh normal force of eth pass ferame as the values to calculate teh frition forces for the corrent frame.
the metho is faste but have a mush bigger error in eh force calculation specially when bodies ar moving and bouncing.


NewtonSetPlatformArchitecture selct if the COU is x87, Mac, Simd, altvect.
in teh futrue it will be used to selcet newer platform liek Larrabee, or GPU.

NewtonSetMultiThreadSolverOnSingleIsland when using mltithreaded mode the engine grups the bodies th aare in contact an dsolve each gropu in a seprated thread.
howevet if you haev an big unbalance, for exampel you may have a stack with 500 bodies and anoteh stack with 2 bodis.
using two thread on CPU will do a lot of work solvion 500 body stack while teh othe remaind idle.

Multithread in single island uses all the thread solve work in the 500 stack, and if teh secund stack in bigger tha some value it will use or mor ethread to work on it.

Muttheread on single island I develeper for GPUs, when you need to have lot of work load to have teh core4 busses bu teh I realize it can work for PC too.
It hav eteh probl;em that teh convergence rate of teh solve is lower, so it need more itertaions,
threfore Mutthreaded on singel isnald is only useful for system with lots of cores like Nahalam, GPU, or Larrabee.
for system with one ot two cores is better not using that option.

[Witek900]

OK. Thanks