[cairo] span datastructure

[M Joonas Pihlaja]

On Fri, 14 Nov 2008, Jeff Muizelaar wrote:

> blit_with_span_renderer():
>
> I was wondering why you create the span array instead of just blitting
> directly from the cells.

The spans array is a compromise between having a fast 
implementation and a retargetable API.  Blitting directly from 
the cell list would be letting the particular implementation leak 
out of the scan converter.  While it's not apparent anymore given 
that there's only one scan converter and span renderer 
implementations left, it's been pretty useful to separate the two 
using a simple interface.

An even simpler interface would be a single "render this span" 
call back, but since there are usually lots of short spans from 
antialiasing, the span renderer call overhead starts rearing its 
head.  A span array per pixel row seemed like the next simplest 
thing to do.

> It seems like doing the extra step of converting to a span 
> array including counting the number of entries in the list 
> isn't going to be particularily helpful for performance.

That's true, but I haven't really seen it in the profiles so much 
compared to advancing edges or the compositing done later, 
especially if we're supersampling on that pixel row.

Regarding counting the entries on the list it's only needed to 
make sure we don't overflow the span array when forming it.  It 
looks a bit shoddy and I'll admit I didn't think very hard when 
writing that bit.  I guess we could use the width of the clip 
region to allocate a span array we'll know to always be large 
enough up front.

Joonas