[RFC weston 13/16] compositor: Add a function to test if images transformed by a matrix should be bilinearly filtered

[Bill Spitzak]

On 10/03/2014 11:05 AM, Pekka Paalanen wrote:

>> The really lovely part about pixman is that (if I'm "getting" it
>> correctly) you transform the source image instead of transforming
>> destination vertices.  So pixman needs things exactly backwards.
>
> Yeah, backwards sounds familiar. Just reverse the polarity... err,
> invert the matrix. It's only 3x3 at that point, and pixman can
> probably do it. We can optimize it later if needed. Or if it looks
> like the inverse form is more popular in general, we could just
> construct that to begin with. Later. :-)

Since pixman is trying to figure out what color to make a particular 
output pixel, it needs the inverse matrix because that is how it 
translates the xy of an output pixel to a location in the input image.

OpenGL and other libraries invert the matrix internally as they need 
this same information when doing texture mapping.

I think you will have to keep track of both the matrix and the inverse 
one. The inverse is needed to translate input events to client space.