Re: [glam-devel] OpenGL Display Lists

[Manu S. <ms...@xr...>]

David Longley wrote:
> I'm not quite sure how GL_COMPILE_AND_EXECUTE works such 
> that it slows down the render process--perhaps it compiles as it is 
> executing.

Whoever made the statement that GL_COMPILE_AND_EXECUTE slows down the 
render process probably meant to say "It slows down the first execution 
of the call, thus if you are constantly compiling/executing in the same 
thread... application performance will suffer."

This is because (in bad driver implementations) it doubles the work the 
graphics card has to do:

1. All of the geometry calculations
2. Writes across the AGP bus and display list management (costly 
graphics card state changes)

Whats worse is that the two steps above have to be in sync with each 
other for large lists... which limits throughput to the speed at which 
the AGP bus can keep up... which is much slower than what the graphics 
card is capable of.

The NVidia drivers have performance problems with 
GL_COMPILE_AND_EXECUTE, as do most drivers. The only guaranteed way to 
control performance is to compile and execute seperately (and as Dave 
said, that is what we do in glam, glamrock and CSD).

So if you see GL_COMPILE_AND_EXECUTE anywhere in glam code, it is wrong 
and needs to be changed.

I couldn't find any instances of GL_COMPILE_AND_EXECUTE in glam or glamrock?

-- manu

-- 
Manu Sporny
Director of Product Development
xRhino, Inc.
1700 Kraft Drive, Suite 2408
Blacksburg, Virginia, USA 24060
Telephone: (540) 961-4469
Email    : ms...@xr...
Web      : http://www.xrhino.com/