Why run xserver on the OpenGL API?

[Marcelo E. Magallon]

On Tue, Nov 25, 2003 at 03:44:35PM -0800, Keith Packard wrote:

 > operations not supported by GL that might make good candidates (plane
 > masks, per-component alpha, conjoint/ disjoint compositing
 > operators).  

 Hmm... I can't think of a simple way of emulating plane masks.  For a
 mask m, and source and destination fragments s and d, you want, as far
 as I understand the operation of X plane masks:

        d' = (s | ~m) & d

 which is doable, but in three passes

  1 draw d
    copy to tex
  2 draw m
    GL_COLOR_LOGIC_OP -> GL_OR_REVERSE
    draw s
    copy to tex
  3 draw d from tex
    GL_COLOR_LOGIC_OP -> GL_AND
    draw (s| ~m) from tex

 but not many implementations hardware accelerate this.

 As for per-component alpha, I guess you mean per component weighting
 factors.  That's certainly doable, it's called GL_blend_func_separate
 and it's part of OpenGL 1.5 (AFAIR).  It's not commonly supported in
 hardware (or better said, only recent commodity hardware supports it).

 And all the operators in the Duff & Porter paper can be implemented
 with OpenGL as long as you can change the order of rendering (i.e., if
 "A" and "B" are fixed -- that is, always the source and destination
 images -- you need to reverse the order for some operators, or use
 destination alpha, which might or might not be so nice).  It's just a
 matter of playing with the blending factors.

 Cheers,

 Marcelo