I've begun some testing of the current implementation of the RENDER's
Trapezoids request. I'm comparing server results against the output of
the nickle code that was developed along with the current trapezoid
algorithm.
If anyone is interested in following along, check out "trap" from
Keith's CVS tree:
http://keithp.com/cvs.html
Along with scripts to generate random trapezoids, I've include an
example file with over 10000 trapezoids resulting from overzealous
tessellation of a stroked bezier spline, (courtesy of Xr).
After a couple of fixes to the nickle code, (missing rounding and
precision loss in the intersection), I'm pleased with the test results
so far. For the spline trapezoids, over 95% pass already, and many of
the failures are single-bit alpha errors for single pixels. It's
encouraging to see that the implementation is mostly correct, but more
importantly, its encouraging to have confirmed failures I can now use
to fix the implementation.
The random trapezoids are a little more abusive -- only 78% or so of
those are currently passing.
This test setup outputs data files suitable for dropping directly into
the xtest suite. I've also got code for testing trapezoids in that
framework, which I'll contribute along with patches for the trapezoid
implementation once I've made them.
-Carl
--
Carl Worth
USC Information Sciences Institute [EMAIL PROTECTED]
_______________________________________________
Render mailing list
[EMAIL PROTECTED]
http://XFree86.Org/mailman/listinfo/render
