GFX Q3 goals

[Joe Drew]

Hi everyone,

The graphics team has come up with a preliminary list of our goals for
Q3 of this year, and we'd like your feedback & input. It's a pretty
full list, but we'd really like to see what other people need from GFX
in the next few months, and how we can prioritize other things to make
that happen. (In particular, we'd like to know what the pain points
for Cairo performance are.)

Right now the GFX goals are available on the wiki at

https://wiki.mozilla.org/Platform/2009-Q3-Goals#GFX

At the moment, our goals can be roughly summarized as: multi-process
support, hardware acceleration, Cairo performance, Cairo capabilities,
and font capabilities.

Please feel free to edit and reformat as necessary. I've tried to
format them in a Goal: Supporting Items sort of bullet point-y way,
but I failed for some of them.

If you have items to propose, please add them to the wiki with your
name/nick in parentheses, or just reply to dev.planning.

All the best,

Joe

[Boris Zbarsky]

Joe Drew wrote:
> At the moment, our goals can be roughly summarized as: multi-process
> support, hardware acceleration, Cairo performance, Cairo capabilities,
> and font capabilities.

Would "stop relying on the X server to not suck performance-wise" fit in
with those goals?

-Boris

[Clint Talbert]

On 6/22/09 4:17 PM, Joe Drew wrote:

> Right now the GFX goals are available on the wiki at
>
> https://wiki.mozilla.org/Platform/2009-Q3-Goals#GFX
>
> At the moment, our goals can be roughly summarized as: multi-process
> support, hardware acceleration, Cairo performance, Cairo capabilities,
> and font capabilities.

The list looks pretty full, but it looks like the right set of things to
work on. I'm trying to figure out what the testing impact will be from
Hardware acceleration. Our existing tests and harnesses might catch
some bugs, but I'm not confident that our current harnesses would
provide the best testing for correctness of this change since they don't
have any ability to compare drawing in software mode versus hardware
mode. Do you have ideas about what would work? I'm going to do some
research into what other projects do for this type of testing...games
come to mind most quickly, but I'm sure there are many other projects
that do similar optimizations.

I'll be checking out the latest games if anyone needs me... ;)

Thanks,
Clint

[Joe Drew]

On Jun 24, 2009, at 9:49 PM, Clint Talbert wrote:

> I'm trying to figure out what the testing impact will be from
> Hardware acceleration. Our existing tests and harnesses might catch
> some bugs, but I'm not confident that our current harnesses would
> provide the best testing for correctness of this change since they
> don't have any ability to compare drawing in software mode versus
> hardware mode.

In fact it's kind of worse than that, because many of our automated
testing machines don't even have access to hardware acceleration. (For
example, my understanding is that our Linux test machines use Xvfb,
which doesn't interact with hardware in any way. Similarly, I think we
won't reliably have OpenGL on our Windows VMs.) So we'll be able to
test our software fallbacks (which will need to remain for machines
that don't support hardware acceleration), but the actual hardware
acceleration will be trickier.

I'd love to hear some ideas on this. Getting automated testing on any
sort of hardware acceleration would be fantastic, even leaving aside
the fact that we'll need to support different driver-OS combinations
like (ATI/AMD vs NVIDIA vs Intel) x (Linux vs Windows vs Mac).

Joe

[Robert O'Callahan]

Looks good. A couple of questions:
-- No canvas3D stuff here?
-- Is it really worth working on a Win7/Vista-only font-variant-*
implementation? I think ideally in the future we'll be using Harfbuzz
for most text shaping on all platforms, and implementing font-variant-*
there should be pretty easy, so maybe we should just focus on that?

Rob

[Jeff Muizelaar]

Not specifically, I suppose it's worth looking at though. I wonder if
we should perhaps have a "eliminate existing hardware deceleration"
goal. I think we also suffer on win32 because of our attempt to use
GDI as much as possible...

-Jeff

[moz....@windingwisteria.com]

> Is it really worth working on a Win7/Vista-only font-variant-*
> implementation? I think ideally in the future we'll be using Harfbuzz
> for most text shaping on all platforms, and implementing font-variant-*
> there should be pretty easy, so maybe we should just focus on that?

Yes, I think it's important to have an experimental implementation to
play with to work out the definition of the spec. I wasn't proposing
pushing this into a product. Variations in the Uniscribe API available
on Vista/Windows 7 make it fairly easily to do this. This would also
provide a comparison point for a Harfbuzz implementation.

John

[Gen Kanai]

On Jun 23, 2009, at 8:17 AM, Joe Drew wrote:

> Hi everyone,
>
> The graphics team has come up with a preliminary list of our goals
> for Q3 of this year, and we'd like your feedback & input. It's a
> pretty full list, but we'd really like to see what other people need
> from GFX in the next few months, and how we can prioritize other
> things to make that happen. (In particular, we'd like to know what
> the pain points for Cairo performance are.)
>

> Right now the GFX goals are available on the wiki at
>
> https://wiki.mozilla.org/Platform/2009-Q3-Goals#GFX
>

> At the moment, our goals can be roughly summarized as: multi-process
> support, hardware acceleration, Cairo performance, Cairo
> capabilities, and font capabilities.

There was a separate discussion (with roc, dbaron, jkew, blizzard and
jdaggett) about adding CSS hyphenation support. I don't know if that
would be Q3 or later but I thought I'd add that to the discussion.

Gen

[Robert O'Callahan]

On 26/6/09 6:05 PM, Gen Kanai wrote:
> There was a separate discussion (with roc, dbaron, jkew, blizzard and
> jdaggett) about adding CSS hyphenation support. I don't know if that
> would be Q3 or later but I thought I'd add that to the discussion.

That is a layout issue. It's doable but I don't think it's high enough
priority for Q3.

Rob

[jimm]

Just added some windows integration bullet points to the GFX Q3 Goals.
(Yes it's a little strange to have win integration in GFX but it's the
simplest way to track this work at this point.)

Win integration work
* Improvements to parental controls support (blocked content
management, extension manager support).
* Win7 specific
o Taskbar drop lists support for Firefox and Thunderbird.
o Support for tablet pc pen events.
o Touch input improvements (new gesture events for touch input
rather than mouse event simulations, smarter text selection logic,
"big pointer" UI improvements?, smarter toggle zoom).
o General: make sure new win7 features are covered (including
tests!)

[Damon Sicore]

If it makes more sense to put them in their own section (and it seems
to), we should add a section.

>
> Win integration work
> * Improvements to parental controls support (blocked content
> management, extension manager support).
> * Win7 specific
> o Taskbar drop lists support for Firefox and Thunderbird.
> o Support for tablet pc pen events.
> o Touch input improvements (new gesture events for touch input
> rather than mouse event simulations, smarter text selection logic,
> "big pointer" UI improvements?, smarter toggle zoom).
> o General: make sure new win7 features are covered (including
> tests!)

> _______________________________________________
> dev-planning mailing list
> dev-pl...@lists.mozilla.org
> https://lists.mozilla.org/listinfo/dev-planning

[Joe Drew]

On Jun 25, 2009, at 11:05 PM, Robert O'Callahan wrote:

> -- No canvas3D stuff here?

No; not enough cycles remaining from members of the graphics team.

[Joe Drew]

On Jun 22, 2009, at 7:17 PM, Joe Drew wrote:

> The graphics team has come up with a preliminary list of our goals
> for Q3 of this year, and we'd like your feedback & input.

Thanks for all your feedback and input, everyone. While we won't be
able to do everything that people suggested, we hope that GFX's final
Q3 goals, available now at https://wiki.mozilla.org/Platform/2009-Q3-Goals
#GFX , will help everyone who depends on us at least a little.

Some of the changes from when I initially posted the list:
- Multiprocess image decoding was dropped as not immediately
necessary, as we don't believe that images will be shared in a
meaningful way
- Decode-on-draw was added as an explicit goal; we have been working
on it but had never had a quarterly goal for it.
- A whole slew of Windows (7 and Vista) integration issues were added.
- Some specific requests for Cairo capabilities were added, but aren't
guaranteed to be finished.

Please check out the updated list and make sure that we haven't missed
anything obvious or forgotten your feedback. If there are no
objections, the goals currently on the wiki (or something very much
like them) will become our goals for Q3.

https://wiki.mozilla.org/Platform/2009-Q3-Goals#GFX

Thanks lots!

Joe

[Question]

Hi,
As the experimental HW ACCL been kicked on, would it break the
mozilla-headless?
If it does break, can we get it working much easier then before?

--
>: ~

[Joe Drew]

Implicit in any HW acceleration is a software fallback path, most
likely involving Cairo. So I wouldn't be too concerned about the
headless branch work.

Joe