[Pixman] Is circle rendering possible?

Bill Spitzak spitzak at gmail.com
Thu Feb 7 06:52:39 UTC 2019


If you are rendering a solid color (so the result is either that color or
transparent black) then you can just fill the entire RGB with that color to
turn the premulitplied output into unpremultiplied.
However I am unsure what your precision problem is. You would have to
composite several dozen mostly-transparent pixels for it to make any
visible difference.


On Wed, Feb 6, 2019 at 2:28 AM Indi Kernick <kerndog73 at gmail.com> wrote:

> I thought pixman allowed you to choose between premultiplied and not
> premultiplied but I checked and apparently not. I think I'm getting my
> libraries mixed up. The problem I have with premultiplied alpha is that you
> lose precision when you have to un-premultiply. That's something I need in
> my application. I want to get the color of a pixel without losing precision
> with semi-transparent pixels.
>
> So if I don't want to lose precision, I'll have to store images in regular
> alpha and then multiply the alpha every time I want to composite. That's
> kind of annoying.
>
> To render a polygon, I could divide it into triangles and render those
> with pixman.
>
> Thanks for your help. I think I'll explore other libraries.
>
>
> On Wed, 6 Feb 2019 at 20:24 Pekka Paalanen <ppaalanen at gmail.com> wrote:
>
>> On Wed, 6 Feb 2019 19:25:48 +1030
>> Indi Kernick <kerndog73 at gmail.com> wrote:
>>
>> > The only problem I can find with Cairo is that it uses premultiplied
>> alpha.
>>
>> Pixman uses premultiplied alpha. Also Wayland uses premultiplied alpha
>> FWIW. I think you will find it quite common in general.
>>
>> > That's so annoying! Also, I'm not sure if the Cairo renderer is GPU or
>> CPU
>> > based.
>>
>> You choose through the API which one you use. Use the image API, and
>> you get CPU rendering.
>>
>> > I'm using Qt to render the images. Qt can render surfaces so I can use
>> > pixman to manipulate a surface and then Qt to upload it to the GPU when
>> > it's time to render. Cairo can do the same thing but then I have to
>> worry
>> > about premultiplied alpha. I could use pixman to do the compositing and
>> > Cairo to do other stuff but that's just a mess.
>> >
>> > I'm guessing that using a GPU backend for Cairo and then making it play
>> > nice with the Qt layout engine will be a nightmare. If I have to go back
>> > and forth between CPU and GPU, it might actually be faster to use CPU
>> > rendering. The only thing I'm worried about is compositing. Compositing
>> is
>> > much faster on the GPU (I presume) but if I use Cairo, I'll have to
>> fiddle
>> > around with the alpha channel. That means a round trip between CPU and
>> GPU
>> > memory.
>>
>> You don't need to fiddle with alpha. Just write your fragment shaders
>> and blending state to expect it.
>>
>> The point of using premultiplied alpha is to pre-compute values you
>> would be computing in any case during blending.
>>
>> > I only really need non-antialiased circle rendering (antialiasing is a
>> > nice-to-have). I can figure out to do that myself (Bresenham comes to
>> > mind). I think I'll use pixman and do everything on the CPU.
>> >
>> > Another question: can pixman render a filled polygon without
>> antialiasing?
>> > If not, I might have to do that myself too. 😢
>>
>> That I don't know. I have never used Pixman beyond
>> pixman_image_composite32(). I am guessing the trapezoids or triangles
>> API could be used, but there does not seem to be anything about polygons
>> per se. You can see the header yourself, that is all the documentation
>> there is, which is practically none. Cairo at least has documentation.
>>
>> I believe there are also other CPU rendering libraries, but I wouldn't
>> know to point to them.
>>
>>
>> Thanks,
>> pq
>>
> _______________________________________________
> Pixman mailing list
> Pixman at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/pixman
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/pixman/attachments/20190206/dc5d4761/attachment.html>


More information about the Pixman mailing list