Re: Struggling with render performance

Around 11 o'clock on Dec 9, Juliusz Chroboczek wrote:

> But then I'll need to do all core rendering twice, once to the pixmap,
> once to the alpha channel, right?  Is it worth the hassle, or shall I
> fall back to render rendering when there's an alpha channel?

I'm afraid it would be worth the hassle.  There are several reasons why
accelerated drawing hasn't happened yet:

 1)	The spec isn't quite complete, we need to figure out how
	gradients and shading fit in.

 2)	The precise sematics for filling trapezoids haven't been carefully
	written down and verified.

 3)	We have no test suite to check the current software implementation
	let alone hardware accelerated implementations

 4)	There aren't comprehensive performance tests for the whole 
	extension.

Carl Worth and I are working on building a test suite for the parts of the 
extension which are stable; I've got half a dozen students working on the 
basic extension while Carl is building tests for trapezoids.  The test 
suite development should point out many inconsistencies and inaccuracies 
in the specification which should make that a lot more solid.  Carl has 
already found a small issue with trapezoid rendering which needs some 
study.

Figuring out how to specify gradients and shading seems like something we 
need to get a handle on so that the specification can be finalized.  My
current thoughts on this matter are that somehow shading, gradients and 
image transformations are all related; each of these relate a point in the 
destination with some function operating on a collection of colors.  I
think it would be possible to relate the basic projective image 
transformation somehow to operations for shading trapezoids and performing 
gradients across fills.

Keith Packard        XFree86 Core Team        HP Cambridge Research Lab