Gnash - GNU Flash Player / John Gilmore

[Alex Hudson]

On Tue, 2006-01-03 at 19:13 +0000, MJ Ray wrote:
> Alex Hudson <home at alexhudson.com>
> > As an example of the kind of development that needs 3D hardware
> > acceleration:
> > 	http://www.mozillazine.org/talkback.html?article=6498
> 
> Why are they interested in Cairo rather than any 2d-friendly vector
> library?  Can someone translate that page for non-graphics-fans?

I wouldn't rate myself a graphics fan, but I'll try - I did UI & OpenGL
in a former life...

So - why Cairo. Well, Cairo *is* 2D-friendly - it's an API for drawing
2D stuff, like lines, polygons, that kind of thing. But there's "2D" and
there's "2D".

"Old" 2D hardware did memory-interesting operations: moving blocks of
memory around at high speed (often called 'sprites') for example is very
common and very useful (a nice smooth mouse pointer). It may also do
things like filling areas quickly (usually rectangles), and other
assorted stuff. But, to call this "2D hardware" is really missing the
point - it's acceleration of memory operations that happen to be useful
for bitmaps. 

Now, drawing complex lines and things is as much rasterisation as
drawing 3D objects - in fact, it's basically the same problem, just with
the Z-dimension removed.

Cairo is this 'high level 2D' API, which does things in a nicer, more
abstract 'lines and curves' way, but this is a lot more graphics work
than old cards are used to. If you think of 2D being a simplification of
the 3D mathematics (which is basically is), you can think of 2D being
less work than 3D, but still quite complex (the basic mixing of pixel
colours where things overlap still happens, taking care of transparency,
that kind of thing). That kind of view is completely different to the
"memory buffers" view of older cards.

As an example, look at drawings in OpenOffice.org with Cairo turned on:

	http://primates.ximian.com/~michael/ooo-cairo.png

You can see the polygons on the page have smooth edges. That's what
Cairo does; it's a high-quality 2D library. It's also output-neutral:
the Mozilla people have had trouble generating pages for printing since,
well, years ago, and Cairo nicely solves part of the issue. And this is
all in software, none of this requires 3D anything.

So, where does GL come in? High-quality 2D is a substantial amount of
work, and it just so happens that OpenGL will happily do 2D operations
for you. GL is a convenient way of speeding up Cairo drawing operations,
by offloading the drawing to the graphics card. This is what this
"Glitz" thing is - it's Cairo running on GL, and running pretty fast
compared to the all-software version.

There are alternatives to GL, such as:

	http://www.khronos.org/openvg/

(which is an interesting demonstration of the stuff you need to do for
modern 2D displays in its own right). But, they're even less suitable.

> Is everyone going GL because they are ignorant or just willing to
> ignore what looks like a new Java Trap problem to get the benefits?

Almost certainly the latter, although it's hard to see how you could get
the benefits (and there are benefits) in any other way, so it's not
quite the same.

The benefits aren't just eye-candy. I think we've all seen PDF files
load in front of our eyes because there is so much stuff on the page; a
GL-accelerated PDF reader would render it instantaneously. Being able to
have a resolution-independent display (so that it can be smoothly
zoomed, useful if you have poor eyesight) needs that acceleration. Being
able to drive a high-resolution display would almost certainly need it,
and being able to support real-time vector graphics (e.g., SVG icons,
themes, etc.) across a whole desktop probably needs it.

What I'm trying to get across is that this isn't developers wanting to
spin and rotate windows on three axes, there are lots of 2D reasons for
wanting a GL-based desktop. It makes the display much higher quality,
faster, and takes load off the main CPU which ends up doing much less
drawing. People want to leave the world of bitmaps and memory shunting
behind, and move to a wholly vector-art based system.

> What is the scale of this problem? A few cards have been mentioned,
> but the messages make it sound like they're a small minority of 3d
> hardware on sale today. Are there similar sites to r300.sf.net for
> other manufacturers, or any overview site?

No, not really, it's a mess.

R300 and R200 (the two best drivers, I think) are for a limited range of
older Radeon cards (basically, anything without an X at the start of the
model number). I believe there are some drivers for an Intel chipset,
and a few for some older cards. Nothing nVidia is available, and nothing
modern ATI. 

Basically, anything which supports "modern" 3D is not available in a
free software driver. In terms of "modern" features, I'm talking about
things like vertex shaders - small programs that software downloads onto
the 3D card in order to make them do incredible things. This is how
complex the interfaces with these cards has become.

Cheers,

Alex.