> hi-
>
> in there you wrote:
>
> "Two versions of Jet3D are tested here: the original double precision
> scalar code ("scalar"), and a mixed single and double precision vector
> code with approximate spectral integration ("vector")."
>
> since the Navier-Stokes equations are ill-posed in certain regions, how
> did the result (the vector fields) of float v.s. double compare over
> long time values?
>
> IMHO, the G5 results are impressive compared to a G4 and compare to a
> P4. A native G5 compiler should be even better, as you wrote. Finally,
> a mac that can be respected in the SciTech market :-)
>
Well, first of all, Jet3D postprocesses a CFD solution, so any accuracy
issues pertaining to the NS equations happen upstream in flow solvers like
PAB3D, CFL3D, etc. They are run single or double precision depending on the
flow regime we're tackling.
Internal to Jet3D, portions of the code converted to single precision for
vectorization do just fine with single precision. There was a bit of
overhead to implement single precision and vectorization, but it's more than
made up for by the performance gain.
Craig
--
Dr. Craig Hunter
NASA Langley Research Center
AAAC/Configuration Aerodynamics Branch
email@hidden
(757) 864-3020
(Dual G4 - OS X)
_______________________________________________
scitech mailing list | email@hidden
Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/scitech
Do not post admin requests to the list. They will be ignored.
