Re: Raspberry Pi + JavaFX 3D demo
No, it's not possible. These have not yet been implemented. On Mon, Jul 15, 2013 at 4:19 PM, Tobias Bley t...@ultramixer.com wrote: Is it possible to activate the optimizations for me in RoboVM 0.0.2 or 0.0.3 (github)? Am 12.07.2013 um 17:11 schrieb Niklas Therning nik...@therning.org: RoboVM doesn't do release builds yet (virtually none of LLVM's optimizations are enabled) and virtual and interface method dispatch is horribly slow (linear search on every call!!!). So there are A LOT of opportunities for improvements. The focus so far has been to get something up and running which can actually be used to make apps. I hope I will get the time to optimize things later this year. Is this demo open source? It would be interesting to test it on iOS. On Fri, Jul 12, 2013 at 5:03 PM, Richard Bair richard.b...@oracle.comwrote: BTW, we've run a VM performance benchmark against HotSpot on PI vs. RoboVM on iOS and for raw power RoboVM seems faster (lower time to invoke a method, read a field, etc etc). However in the real world RoboVM is slow and I don't know why (GC overhead maybe)? Richard On Jul 12, 2013, at 7:52 AM, Richard Bair richard.b...@oracle.com wrote: That should be encouraging, since the CPU on the PI is *way* worse than the CPU on an iPhone or iPad. Is the difference HotSpot vs. RoboVM? The graphics code being executed should be pretty much exactly the same, and I would expect the PowerVR to be able to handle this without any trouble. Richard On Jul 12, 2013, at 7:14 AM, Tobias Bley t...@ultramixer.com wrote: The performance is much better than JavaFX8 on iOS :( Am 12.07.2013 um 15:37 schrieb August Lammersdorf, InteractiveMesh s...@interactivemesh.com: Found this on YouTube: http://www.youtube.com/watch?v=-scxqJjTJKI August
Re: Raspberry Pi + JavaFX 3D demo
Hi Niklas, thanks for your answer. I looked in the current code form github (RoboVM 0.0.3) and found a „Level“ enum with -o3 optimization flag... so are you working on the optimization support for LLVM? Best, Tobi Am 16.07.2013 um 08:47 schrieb Niklas Therning nik...@therning.org: No, it's not possible. These have not yet been implemented. On Mon, Jul 15, 2013 at 4:19 PM, Tobias Bley t...@ultramixer.com wrote: Is it possible to activate the optimizations for me in RoboVM 0.0.2 or 0.0.3 (github)? Am 12.07.2013 um 17:11 schrieb Niklas Therning nik...@therning.org: RoboVM doesn't do release builds yet (virtually none of LLVM's optimizations are enabled) and virtual and interface method dispatch is horribly slow (linear search on every call!!!). So there are A LOT of opportunities for improvements. The focus so far has been to get something up and running which can actually be used to make apps. I hope I will get the time to optimize things later this year. Is this demo open source? It would be interesting to test it on iOS. On Fri, Jul 12, 2013 at 5:03 PM, Richard Bair richard.b...@oracle.com wrote: BTW, we've run a VM performance benchmark against HotSpot on PI vs. RoboVM on iOS and for raw power RoboVM seems faster (lower time to invoke a method, read a field, etc etc). However in the real world RoboVM is slow and I don't know why (GC overhead maybe)? Richard On Jul 12, 2013, at 7:52 AM, Richard Bair richard.b...@oracle.com wrote: That should be encouraging, since the CPU on the PI is *way* worse than the CPU on an iPhone or iPad. Is the difference HotSpot vs. RoboVM? The graphics code being executed should be pretty much exactly the same, and I would expect the PowerVR to be able to handle this without any trouble. Richard On Jul 12, 2013, at 7:14 AM, Tobias Bley t...@ultramixer.com wrote: The performance is much better than JavaFX8 on iOS :( Am 12.07.2013 um 15:37 schrieb August Lammersdorf, InteractiveMesh s...@interactivemesh.com: Found this on YouTube: http://www.youtube.com/watch?v=-scxqJjTJKI August
Re: Raspberry Pi + JavaFX 3D demo
Is it possible to activate the optimizations for me in RoboVM 0.0.2 or 0.0.3 (github)? Am 12.07.2013 um 17:11 schrieb Niklas Therning nik...@therning.org: RoboVM doesn't do release builds yet (virtually none of LLVM's optimizations are enabled) and virtual and interface method dispatch is horribly slow (linear search on every call!!!). So there are A LOT of opportunities for improvements. The focus so far has been to get something up and running which can actually be used to make apps. I hope I will get the time to optimize things later this year. Is this demo open source? It would be interesting to test it on iOS. On Fri, Jul 12, 2013 at 5:03 PM, Richard Bair richard.b...@oracle.com wrote: BTW, we've run a VM performance benchmark against HotSpot on PI vs. RoboVM on iOS and for raw power RoboVM seems faster (lower time to invoke a method, read a field, etc etc). However in the real world RoboVM is slow and I don't know why (GC overhead maybe)? Richard On Jul 12, 2013, at 7:52 AM, Richard Bair richard.b...@oracle.com wrote: That should be encouraging, since the CPU on the PI is *way* worse than the CPU on an iPhone or iPad. Is the difference HotSpot vs. RoboVM? The graphics code being executed should be pretty much exactly the same, and I would expect the PowerVR to be able to handle this without any trouble. Richard On Jul 12, 2013, at 7:14 AM, Tobias Bley t...@ultramixer.com wrote: The performance is much better than JavaFX8 on iOS :( Am 12.07.2013 um 15:37 schrieb August Lammersdorf, InteractiveMesh s...@interactivemesh.com: Found this on YouTube: http://www.youtube.com/watch?v=-scxqJjTJKI August
Re: Raspberry Pi + JavaFX 3D demo
The performance is much better than JavaFX8 on iOS :( Am 12.07.2013 um 15:37 schrieb August Lammersdorf, InteractiveMesh s...@interactivemesh.com: Found this on YouTube: http://www.youtube.com/watch?v=-scxqJjTJKI August
Re: Raspberry Pi + JavaFX 3D demo
That should be encouraging, since the CPU on the PI is *way* worse than the CPU on an iPhone or iPad. Is the difference HotSpot vs. RoboVM? The graphics code being executed should be pretty much exactly the same, and I would expect the PowerVR to be able to handle this without any trouble. Richard On Jul 12, 2013, at 7:14 AM, Tobias Bley t...@ultramixer.com wrote: The performance is much better than JavaFX8 on iOS :( Am 12.07.2013 um 15:37 schrieb August Lammersdorf, InteractiveMesh s...@interactivemesh.com: Found this on YouTube: http://www.youtube.com/watch?v=-scxqJjTJKI August
Re: Raspberry Pi + JavaFX 3D demo
BTW, we've run a VM performance benchmark against HotSpot on PI vs. RoboVM on iOS and for raw power RoboVM seems faster (lower time to invoke a method, read a field, etc etc). However in the real world RoboVM is slow and I don't know why (GC overhead maybe)? Richard On Jul 12, 2013, at 7:52 AM, Richard Bair richard.b...@oracle.com wrote: That should be encouraging, since the CPU on the PI is *way* worse than the CPU on an iPhone or iPad. Is the difference HotSpot vs. RoboVM? The graphics code being executed should be pretty much exactly the same, and I would expect the PowerVR to be able to handle this without any trouble. Richard On Jul 12, 2013, at 7:14 AM, Tobias Bley t...@ultramixer.com wrote: The performance is much better than JavaFX8 on iOS :( Am 12.07.2013 um 15:37 schrieb August Lammersdorf, InteractiveMesh s...@interactivemesh.com: Found this on YouTube: http://www.youtube.com/watch?v=-scxqJjTJKI August
Re: Raspberry Pi + JavaFX 3D demo
I think these are the sample apps: http://interactivemesh.org/testspace/j3dmeetsjfx.html I take that back, the PI HotSpot numbers are generally faster than the RoboVM iOS numbers (but RoboVM is faster than HotSpot when run as an interpreter, obviously). We need to get our benchmark open source (we call it VMBenchmark) so that we can host accurate numbers on the WIKI and everybody can run it on their own systems. Richard On Jul 12, 2013, at 8:11 AM, Niklas Therning nik...@therning.org wrote: RoboVM doesn't do release builds yet (virtually none of LLVM's optimizations are enabled) and virtual and interface method dispatch is horribly slow (linear search on every call!!!). So there are A LOT of opportunities for improvements. The focus so far has been to get something up and running which can actually be used to make apps. I hope I will get the time to optimize things later this year. Is this demo open source? It would be interesting to test it on iOS. On Fri, Jul 12, 2013 at 5:03 PM, Richard Bair richard.b...@oracle.com wrote: BTW, we've run a VM performance benchmark against HotSpot on PI vs. RoboVM on iOS and for raw power RoboVM seems faster (lower time to invoke a method, read a field, etc etc). However in the real world RoboVM is slow and I don't know why (GC overhead maybe)? Richard On Jul 12, 2013, at 7:52 AM, Richard Bair richard.b...@oracle.com wrote: That should be encouraging, since the CPU on the PI is *way* worse than the CPU on an iPhone or iPad. Is the difference HotSpot vs. RoboVM? The graphics code being executed should be pretty much exactly the same, and I would expect the PowerVR to be able to handle this without any trouble. Richard On Jul 12, 2013, at 7:14 AM, Tobias Bley t...@ultramixer.com wrote: The performance is much better than JavaFX8 on iOS :( Am 12.07.2013 um 15:37 schrieb August Lammersdorf, InteractiveMesh s...@interactivemesh.com: Found this on YouTube: http://www.youtube.com/watch?v=-scxqJjTJKI August
Re: Raspberry Pi + JavaFX 3D demo
Here's my acid test, put a square grid of circles in a scroll pane. Scale up until pain is felt. GridPane gp = new GridPane(); //int size = 3; int size = 10; //int size = 32; // int size = 100; //int size = 317; for (int x = 0; x size; x++) { for (int y = 0; y size; y++) { //Pane p = new Pane(); //p.setStyle(-fx-background-fill: blue;); //p.setMinSize(20, 20); //Text p = new Text(x + x + y); Circle p = new Circle(50, Color.rgb((x0xf)*16, (y0xf)*16, 128)); GridPane.setColumnIndex(p, x); GridPane.setRowIndex(p, y); GridPane.setMargin(p, new Insets(5,5,5,5)); gp.getChildren().add(p); } } gp.setCacheHint(CacheHint.SPEED); ScrollPane scrollPane = new ScrollPane(gp); scrollPane.setMinHeight(400); scrollPane.setPannable(true); My actual use case is a bit more involved (and it includes expensive non-graphical calculations that make an intel core duo cry) and I've found the windows box can handle two to three more orders of magnitude before lag moves from unpleasant to unusable. p.s. I really dig the scroll pane on skin on the touch screens. Complete with over-drag and hidden unless dragging scroll bars. Well done. On Fri, Jul 12, 2013 at 9:17 AM, Tobi t...@ultramixer.com wrote: Painting a plain rectangle is ok. But using a stackpane to draw a javafx styled button over a plain painted rectangle node is very slow. Am 12.07.2013 um 17:03 schrieb Richard Bair richard.b...@oracle.com: BTW, we've run a VM performance benchmark against HotSpot on PI vs. RoboVM on iOS and for raw power RoboVM seems faster (lower time to invoke a method, read a field, etc etc). However in the real world RoboVM is slow and I don't know why (GC overhead maybe)? Richard On Jul 12, 2013, at 7:52 AM, Richard Bair richard.b...@oracle.com wrote: That should be encouraging, since the CPU on the PI is *way* worse than the CPU on an iPhone or iPad. Is the difference HotSpot vs. RoboVM? The graphics code being executed should be pretty much exactly the same, and I would expect the PowerVR to be able to handle this without any trouble. Richard On Jul 12, 2013, at 7:14 AM, Tobias Bley t...@ultramixer.com wrote: The performance is much better than JavaFX8 on iOS :( Am 12.07.2013 um 15:37 schrieb August Lammersdorf, InteractiveMesh s...@interactivemesh.com: Found this on YouTube: http://www.youtube.com/watch?v=-scxqJjTJKI August
Re: Raspberry Pi + JavaFX 3D demo
Have you tried running with the perf logger turned on? -Djavafx.pulseLogger=true This will tell you a little something about where the time is spent. It might not be detailed enough, but you can then put additional log statements in the code to get more direct answers to questions about where the time is being spent. Richard On Jul 12, 2013, at 8:17 AM, Tobi t...@ultramixer.com wrote: Painting a plain rectangle is ok. But using a stackpane to draw a javafx styled button over a plain painted rectangle node is very slow. Am 12.07.2013 um 17:03 schrieb Richard Bair richard.b...@oracle.com: BTW, we've run a VM performance benchmark against HotSpot on PI vs. RoboVM on iOS and for raw power RoboVM seems faster (lower time to invoke a method, read a field, etc etc). However in the real world RoboVM is slow and I don't know why (GC overhead maybe)? Richard On Jul 12, 2013, at 7:52 AM, Richard Bair richard.b...@oracle.com wrote: That should be encouraging, since the CPU on the PI is *way* worse than the CPU on an iPhone or iPad. Is the difference HotSpot vs. RoboVM? The graphics code being executed should be pretty much exactly the same, and I would expect the PowerVR to be able to handle this without any trouble. Richard On Jul 12, 2013, at 7:14 AM, Tobias Bley t...@ultramixer.com wrote: The performance is much better than JavaFX8 on iOS :( Am 12.07.2013 um 15:37 schrieb August Lammersdorf, InteractiveMesh s...@interactivemesh.com: Found this on YouTube: http://www.youtube.com/watch?v=-scxqJjTJKI August