[ODE] Quad precision anyone?

[Bill Sellers]

Like several other people on the list I'm wrestling with getting the  
same answer out of ode on different platforms. I can use gcc on all  
platforms which is a start but I certainly get different results on  
the 3 platforms I'm currently running on (32 bit macos, 32 bit AMD,  
64 bit AMD). One option I am toying with is moving to a quad  
precision implementation which is available on gcc since I imagine  
this must be a largely software implemented floating point option but  
I'm a bit worried that there might be implications of this I hadn't  
thought of (things like the presence or otherwise of quad  
trigonometric functions and various dEpsilon issues). I quite like  
the idea of increased stability (if double is more stable than float,  
is double double more stable than double) but I've no idea what sort  
of performance hit I'd see.

Cheers
Bill

PS. I've uploaded an ODE demo of my bipedal runner at (quite a big  
file I'm afraid):

http://sellers.sbc.man.ac.uk/trex-50fpsAt25fps.mov

--
Dr. Bill Sellers, Editor Primate Eye   Email:  
William.Sellers at manchester.ac.uk
Faculty of Life Sciences               Tel.   0161 275 1719
3.614 Stopford Building                Fax:   0161 275 3938
The University of Manchester           Mob:   0775 107 8997
Oxford Road, Manchester, M13 9PT, UK   http://sellers.sbc.man.ac.uk/ASL/

Primate Society of Great Britain WWW site http://www.psgb.org/