[ODE] Improved QuickStep stability...?

[Billy Zelsnack]

Improved QuickStep stability...?

Adam D. Moss wrote:
Don't hit me -- I added random constraint reordering to QuickStep and
it 
helped stability a lot, partially in terms of stacking stability 
(test_crash's wall) but mostly in terms of stopping test_crash's buggy 
from jiggling and flying apart (test_crash now does a good impression
of 
being stable at very very few iterations). This trumps 
REORDER_CONSTRAINTS, so there's no point in using them together unless 
you like wasting time.


I played around with testcrash a little bit. I used a little trick that

I use in my own non-matrix-method solver and it seemed to help. I'm not

sure if it breaks other things.

float ramp=(1-((float)(iteration+1)/(float)num_iterations));
delta*=ramp;

Put that just above:

// update fc.
// @@@ potential optimization: SIMD for this and the b2 >= 0 case

Sorry for not using a proper diff.

Of course that divide can be precomputed or other fancier functions can

be used. Also. I have found randomizing ordering in my own solver to 
work quite well.




	
		
__________________________________
Do you Yahoo!?
SBC Yahoo! - Internet access at a great low price.
http://promo.yahoo.com/sbc/