[Papervision3D] Speed of Flash Player 10 3D APIs

[Joachim Lous]

> Rick Seeler
> Sent: Monday, September 29, 2008 7:09 PM
> To: papervision3d at osflash.org
> Subject: Re: [Papervision3D] Speed of Flash Player 10 3D APIs

> So, this is a lame comparison...

I suppose. The FP9 demos use "precise"-style rendering only
after you stop moving, so I made the assumption (the mother
of all fuckups) that doing so was necessary for good flow,
and doing each frame in full quality would be significantly
more jerky. Otherwise why use the non-precise method at all?
Of course, even should all that be true, it says
nothing about the speed of PV3D's precise rendering.

Didn't stop to think that the VM itself might be faster in 10.
Is it?

> FP10 is a little faster at doing such correction 
> computations, so you will see some improvement -- but the 
> amount you get depends on a lot of factors (I'm getting about 
> 30% improvement).

I'll admit that is less than I thought. I assumed (again) 
that client-code perspective correction would mean doing
actual per-pixel processing in AS, and that full-screen
rendering that way would be prohibitively slow (for smooth
FPS-like movement), just in terms of pixel flow regardless
of geometry. And that the presumably lower-level
implementation inside FP10 would be vastly more efficient.

30% is still quite significant though.

> If you haven't already done so, please check out the FP10 
> optimized rendered in the CS4 branch of Papervision.  
> 'AstroRenderEngine.as' is the file of interest.

Will do, thanks. Still a bit of a novice in flash/flex, so I've
only used the precompiled modules of PV3D. It also seems
I misunderstood what astro was about.

-joachim.