Re: [AD] 4.9.20

[Pavel Sountsov <slabode@xxxxxxxxxx>]

> > On 17 May 2010, at 21:04 , Pavel Sountsov wrote:
> > > I'll look into this again. I had a (perhaps) promising idea, but 
it
> > > might not be viable since it requires shaders in D3D. This is a 
problem
> > > since Intel GPU's might not support enough (or any) shader 
features
> > > that we need, disqualifying this solution (or forcing us to 
convert
> > > colors for Intel cards).
> >
> > Sounds excellent.
> > Is there an alternative option where we can store the required data 
in the
> colour struct in the case we're using the D3D driver (perhaps only if 
the
> shaders can't be used as a work-around)?
>
> That cure sounds worse than the disease.
>
> I don't think we should push too hard on this.  Sometimes one
> representation makes more sense than another, like cartesian vs polar
> coordinates.
>
> Peter

Okay, I removed ALLEGRO_PRIM_COLOR. Basically, what I did is made the 
D3D driver use shaders to make it understand the OpenGL color format. 
Specifically, I pass the 4 float components of the color to the D3D 
driver as a 4D texture coordinate. That color components are then set 
to those texture coordinate components in the shader. Since shaders 
don't work in some older cards (e.g. kenmasters1976 @ allegro.cc has 
one like that) so I re-added the Intel card workabout, which just 
involves transcribing the vertices into a temporary buffer. I wrapped 
that up in a mutex for thread safety. For Windows dislikers, this 
solution is notable for not affecting OpenGL code at all. It doesn't 
affect D3D driver much either (except for the older Intel cards), so 
everyone should be happy.

-SiegeLord