fixed point, multithreading, blitjit

[Petr Kobalíček]

Hi devs,

are somebody tried to implement fixed point calculations in antigrain?
It seems that doing some things in fixed point is performance boost on
all machines. But today I found something much more interesting:

  http://mlab.taik.fi/~kkallio/antialiasing/

If you download windows binary and run the tests you can see that
presented rasterizer is twice as fast as GDI+ and AntiGrain (both
included) and quality is also very good. Very interesting, the
algorithm is different to antigrain and very optimized.

-----

Currently I have ideas and plans to introduce multithreaded rendering
in my library and I have question how can this play with antigrain. I
have several ideas how multithreaded rendering can improve the speed
and I'd like to present them:

1. Multiple threads can be used to blit rendered scanlines into output
buffer. This means that I will need some way how to use agg rasterizer
from multiple threads. Is this possible ? I mean only read/only access
is needed (when everything is flushed out). Currently the
rasterizer.sweep_scanline() is in my interest.

2. Threads can be also used to render different parts in image that
not intersects. This way can improve rendering in GUI toolkits.
Imagine for example list box or grid controls, the rows / columns are
separate and if painter is able to serialize graphics operations there
is no reason to not use threads.

3. Threads can be used to precalculate next path operations and
rasterizing while main thread is filling something

Of course the main idea is to mix all of these ideas to create
powerful multithreaded painter, but I have currently not time to
implement all. If there is some related work about this really I'm
interested.

-----

I have also news about blitjit library. Few days ago I tried to
connect it with antigrain and my graphics library and it works. Still
blitjit implements very few operations and I tried only CompositeOver.
If there is anybody interested in making benchmark contact me (I will
help with it), but currently I haven't time to do it myself (In future
I will do).

Everything is under MIT licence and can be found here:
http://code.google.com/p/fog/

-----

Cheers and thanks for possible ideas
- Petr

------------------------------------------------------------------------------
Register Now & Save for Velocity, the Web Performance & Operations
Conference from O'Reilly Media. Velocity features a full day of
expert-led, hands-on workshops and two days of sessions from industry
leaders in dedicated Performance & Operations tracks. Use code vel09scf
and Save an extra 15% before 5/3. http://p.sf.net/sfu/velocityconf
_______________________________________________
Vector-agg-general mailing list
Vector-agg-general@...
https://lists.sourceforge.net/lists/listinfo/vector-agg-general

[Lorne Laliberte]

Very cool news Petr! And very interesting find.

As for fixed point calculations, some parts of AGG are already
operating in fixed point (24.8) internally.

Lorne Laliberte
Senior Software Developer, Indigo Rose Software



On Tue, Apr 28, 2009 at 7:53 AM, Petr Kobalíček
<kobalicek.petr@...> wrote:

> Hi devs,
>
> are somebody tried to implement fixed point calculations in antigrain?
> It seems that doing some things in fixed point is performance boost on
> all machines. But today I found something much more interesting:
>
>  http://mlab.taik.fi/~kkallio/antialiasing/
>
> If you download windows binary and run the tests you can see that
> presented rasterizer is twice as fast as GDI+ and AntiGrain (both
> included) and quality is also very good. Very interesting, the
> algorithm is different to antigrain and very optimized.
>
> -----
>
> Currently I have ideas and plans to introduce multithreaded rendering
> in my library and I have question how can this play with antigrain. I
> have several ideas how multithreaded rendering can improve the speed
> and I'd like to present them:
>
> 1. Multiple threads can be used to blit rendered scanlines into output
> buffer. This means that I will need some way how to use agg rasterizer
> from multiple threads. Is this possible ? I mean only read/only access
> is needed (when everything is flushed out). Currently the
> rasterizer.sweep_scanline() is in my interest.
>
> 2. Threads can be also used to render different parts in image that
> not intersects. This way can improve rendering in GUI toolkits.
> Imagine for example list box or grid controls, the rows / columns are
> separate and if painter is able to serialize graphics operations there
> is no reason to not use threads.
>
> 3. Threads can be used to precalculate next path operations and
> rasterizing while main thread is filling something
>
> Of course the main idea is to mix all of these ideas to create
> powerful multithreaded painter, but I have currently not time to
> implement all. If there is some related work about this really I'm
> interested.
>
> -----
>
> I have also news about blitjit library. Few days ago I tried to
> connect it with antigrain and my graphics library and it works. Still
> blitjit implements very few operations and I tried only CompositeOver.
> If there is anybody interested in making benchmark contact me (I will
> help with it), but currently I haven't time to do it myself (In future
> I will do).
>
> Everything is under MIT licence and can be found here:
> http://code.google.com/p/fog/
>
> -----
>
> Cheers and thanks for possible ideas
> - Petr
>
> ------------------------------------------------------------------------------
> Register Now & Save for Velocity, the Web Performance & Operations
> Conference from O'Reilly Media. Velocity features a full day of
> expert-led, hands-on workshops and two days of sessions from industry
> leaders in dedicated Performance & Operations tracks. Use code vel09scf
> and Save an extra 15% before 5/3. http://p.sf.net/sfu/velocityconf
> _______________________________________________
> Vector-agg-general mailing list
> Vector-agg-general@...
> https://lists.sourceforge.net/lists/listinfo/vector-agg-general
>

------------------------------------------------------------------------------
Register Now & Save for Velocity, the Web Performance & Operations
Conference from O'Reilly Media. Velocity features a full day of
expert-led, hands-on workshops and two days of sessions from industry
leaders in dedicated Performance & Operations tracks. Use code vel09scf
and Save an extra 15% before 5/3. http://p.sf.net/sfu/velocityconf
_______________________________________________
Vector-agg-general mailing list
Vector-agg-general@...
https://lists.sourceforge.net/lists/listinfo/vector-agg-general