JavaFX graphics performance and suitability for advanced animations (BrickBreaker)
pavel.safrata at oracle.com
Mon Jun 3 08:03:18 PDT 2013
As I see it, our animation API can be viewed as four levels:
- simple transitions
- custom transitions
If you know how an object will move, you should be fine with the upper
two. If you want to react on user actions you can still adjust the
transitions in input event handlers (to change directions), you can also
have property listeners that will be called every time the transition
updates the property which is every frame. The benefit of the
transitions is that they compute the time-based interpolation for you
and you don't need to do that manually (avoiding the mentioned issue
with the patch). If there is something complicated that needs to be done
in each frame and the above is not enough, then yes, AnimationTimer may
be the right way. I agree this may be the case for many video games, but
I don't think it justifies for using the low-level API where it is not
necessary. If this demo is about "how to write brick breaker", then
diving right to the deepest API level is not really educative. If it is
about "how to write a typical video game", then I think we should rather
produce a typical ShootEmUp game where this API usage would fit more
naturally. I realize it is easy to say but hard to find time for it..
With regards to the pulses - pulse is a sync from scenegraph to render
tree - it tells to the render tree how the current scenegraph looks like
and render tree then keeps displaying that state until the next pulse.
Scenegraph also does some updates during the pulse (yes, layouts fall
into this category). By default on desktop the pulses have an upper
limit of 60fps (means 60 pulses per second), without any connection to
actual display refresh rate. I think (and I'm not certain here) that the
pulse requests are there only to make sure the next pulse is scheduled
(we don't schedule pulses if nothing changes).
On 3.6.2013 15:30, Scott Palmer wrote:
> In this particular example you may be right. The trajectory of the
> ball can be pre-computed for at least the entire length of the
> straight line that it will travel in. You could go a couple bounces
> ahead even, to the point where the ball would get back to the bottom.
> However, that is definitely not the typical case for a video game.
> Per-frame or per-update computations are the norm for video games.
> Change the game style to a shoot'em up where every frame you need to
> determine collisions, the direction of the player or enemy may change
> etc, and it actually isn't any simpler trying to use a pre-computed
> timeline. The calculation of such a timeline would be more work. I
> think in the general case for a video game, AnimationTimer is more
> likely to apply than other techniques. It fits the basic pattern
> commonly used and therefore makes better sample code. There is a
> small problem with Richard's patch though...
> I would like to know a little more about the relation of the "pulse"
> to the refresh rate. I know that there are places in JavaFX where a
> "pulse" is requested or maybe forced. Presumably to trigger
> calculations related to general scene graph layout issues. (I find
> often that the layout of the scene is wrong actually - that is when I
> do actions that force a refresh I will see the layout adjust to a
> state where it clearly should have been in the first place.)
> One would think that in general one pulse per display refresh would be
> the minimum though. Anything less means the animation will not be
> smooth, anything more shouldn't cause problems. The key is that the
> "now" time is used to update any time-based values. I think Richard's
> patch to BrickBreaker is actually incorrect in this regard, as I don't
> see the actual time passed being used in the calculations.
> On Mon, Jun 3, 2013 at 7:56 AM, Pavel Safrata
> <pavel.safrata at oracle.com <mailto:pavel.safrata at oracle.com>> wrote:
> I'm a bit behind with this thread but I want to make a few
> comments on AnimationTimer as there is a hidden message in the
> discussion that AnimationTimer is the way to go.
> First, AnimationTimer is called in each pulse, which doesn't have
> much in common with display refresh rate (if I understand the term
> correctly). More importantly, AnimationTimer is kind of extreme
> low-level animation API that should not be needed in vast majority
> of cases. I think a better way to code BrickBraker would be to use
> a single TranslateTransition for the entire straight part of the
> ball's trajectory; it would automatically compute the
> interpolation and sync the position on every pulse, which should
> have the same result as doing everything manually with the
> AnimationTimer (but expressed in simpler code).
> On 31.5.2013 22:45, Richard Bair wrote:
> I pushed the fix to graphics. Thanks Scott for tracking that
> down! It looks 10x better.
> On May 31, 2013, at 9:25 AM, Richard Bair
> <richard.bair at oracle.com <mailto:richard.bair at oracle.com>> wrote:
> Patch attached to
> https://javafx-jira.kenai.com/browse/RT-29801. I'm not
> seeing any stutter on my Mac, interested to hear the
> experience on Windows.
> On May 31, 2013, at 8:44 AM, Richard Bair
> <richard.bair at oracle.com <mailto:richard.bair at oracle.com>>
> Ya I did the same, am now adjusting it so the factor
> by which things move is better.
> On May 31, 2013, at 8:32 AM, Scott Palmer
> <swpalmer at gmail.com <mailto:swpalmer at gmail.com>> wrote:
> Richard, I suspect you made a typo. I think you
> mean "*40*ms is a really odd number..." (it was 25
> FPS, not 25ms)
> I quickly hacked it to use AnimationTimer and the
> animation is very smooth now. Though I didn't
> make the required changes to adjust the speeds
> based on the refresh rate. The quick conversion
> to AnimationTimer is trivial.. but going through
> and adjusting all the translations and increments
> to be relative to the time between consecutive
> frames is something I don't have time for.
> On Fri, May 31, 2013 at 11:21 AM, Kevin Rushforth
> <kevin.rushforth at oracle.com
> <mailto:kevin.rushforth at oracle.com>> wrote:
> Btw, there is a JIRA issue filed against
> BrickBreaker specifically:
> Richard Bair wrote:
> Have you tried to determine what the FPS is?
> My guess is that FPS is not anywhere near the
> limit and it is the occasional stutter that is
> the problem, but I'm not certain. Knowing that
> helps to point in which direction to go. The
> fact that it runs pretty well on a PI is
> indication that it isn't the framerate.
> On May 31, 2013, at 4:26 AM, Scott Palmer
> <swpalmer at gmail.com
> <mailto:swpalmer at gmail.com>> wrote:
> Speaking of poor animation in Ensemble...
> Is anyone able to run Brick Breaker
> without choppy animation or poor framerate
> performance on the ball?
> Now, I suspect the issue there is in the
> balls animation implementation in the
> application rather than the JavaFX
> framework, as the bat moves smoothly when
> I move the mouse, but the overall
> perception of JavaFX performance for this
> demo app is not good. I would go so far as
> to say that Brick Breaker has had the
> opposite effect it was intended too -
> simply because the animation of the ball
> is not smooth. That's something that
> would run smoothly on a Commodore 64,yet
> the last time I tried it (5 minutes ago)
> with JavaFX 8.0-b91 on a quad-core 3GHz
> Windows 7 box with a decent NVIDIA card,
> it didn't run as smoothly as I would
> expect. Just a single ball with a shadow
> bouncing around the screen seemed to have
> a low framerate and the occasional skipped
> frame. It just didn't look that great.
> The fact that Brick Breaker ships as a
> sample app from Oracle and it's animation
> looks bad is harming JavaFX's reputation
> in my opinion. I think it could run much
> better on the existing JavaFX runtime.
> The simple animations in the Ensemble app
> run much smoother for example.
> On Thu, May 30, 2013 at 11:11 AM, Richard
> Bair <richard.bair at oracle.com
> <mailto:richard.bair at oracle.com>> wrote:
> Then you mention Halo 5. I have to
> say the subtext here troubles me
> greatly. If I read you correctly then
> you are saying that JavaFX is not
> really suitable for games (at least
> anything beyond the demands of something
> like Solitaire). As someone else
> pointed out, what is point of developing
> 3D support in JavaFX if it is not
> really suitable for games? To say it is
> not suitable for games implies that it
> is not really suitable for *any*
> application that requires performant
> animations and visualisations. What
> use then is the 3D API?
> That's not fair at all. There are a *lot*
> of enterprise use cases for 3D, and we get
> these requests all the time. Whether we're
> taking about 3D visualizations for medical
> or engineering applications or consumer
> applications (product display, etc), there
> is a requirement for 3D that are broader
> than real time first person shooters.
> Game engines often have very specialized
> scene graphs (sometimes several of them)
> as well as very specialized tricks for
> getting the most out of their graphics
> cards. When we expose API that allows
> people to hammer the card directly, then
> it would be possible for somebody to build
> some of the UI in FX and let their game
> engine be hand written (in Unity or JOGL
> or whatever).
> However, I am not sure that having me
> preparing "reproducible" test cases
> will actually help. In my experience,
> the Ensemble app already serves this
> purpose. The choppiness I describe is
> *always* prevalent when I run the
> animations and transitions in Ensemble
> (including Ensemble 8). The only
> variation is in the degree of that
> Then start with that, something absolutely
> dead simple like a path animation or
> rotate transition and lets figure out how
> to measure the jitter and get it into our
> benchmark suite.
More information about the openjfx-dev