JavaFX 8 on BeagleBone Black (SGX driver problem?)

[littlef...@gmail.com]

Hello,

I am trying to run the newest JavaFX 8 snapshot release on the beaglebone black. So far I haven't got it running, the debug output shows the following:

Multi-Threading Enabled
Prism pipeline init order: es2
Using platform text rasterizer
Using java-based Pisces rasterizer
Using dirty region optimizations
Not using texture mask for primitives
Not forcing power of 2 sizes for textures
Using hardware CLAMP_TO_ZERO mode
Opting in for HiDPI pixel scaling
Prism pipeline name = com.sun.prism.es2.ES2Pipeline
Loading ES2 native library ... prism_es2_eglfb
Loaded /opt/jdk1.8.0/jre/lib/ext/../arm/libprism_es2_eglfb.so from relative path
    succeeded.
GLFactory using com.sun.prism.es2.EGLFBGLFactory
eglGetConfigAttrib failed!(X) Got class = class com.sun.prism.es2.ES2Pipeline
GraphicsPipeline.createPipeline: error initializing pipeline com.sun.prism.es2.ES2Pipeline
Graphics Device initialization failed for :  es2
Error initializing QuantumRenderer: no suitable pipeline found


What I've done:

- Installed the Ubuntu 13.10 image from Robert C Nelson (https://rcn-ee.net/deb/rootfs/saucy/ubuntu-13.10-console-armhf-2013-12-17.tar.xz)
- Compiled the kernel from https://github.com/RobertCNelson/linux-dev/tree/am33x-v3.12 and installed it
- Used the sgx_build_modules.sh script from the git repo above to download and compile the SGX drivers
- Copied the drivers to the SD card and installed it on the BBB
- I had to manually load the omaplfb module, as it doesn't seem to be loaded automatically (bufferclass_ti and pvrsrvkm are loaded automatically)
- After that, the SDK demos (OGLES2ChameleonMan and OGLES2MagicLantern) seem to work fine
- When starting a JavaFX application, it fails during startup with the log mentioned above (tried to enable all possible debug output)


I've found the error string in the OpenJFX sources, but to get more information, I'd need to create a debug build of the prism-es2 module (or maybe from the whole OpenJDK), which at this point I don't know how to do. Maybe somebody knows what the problem might be.

From <OpenJFX>/rt/modules/graphics/src/main/native-prism-es2/eglfb/EGLFBGLFactory.c:

...
if (!eglChooseConfig(egldisplay, eglAttrs, &config, 1, &numconfigs)) {
        fprintf(stderr, "Failed to get a FBconfig with requested attrs\n");
        //cleanup
        return 0;
    }

#ifdef DEBUG
    printf("eglChooseConfig return %d configs\n", numconfigs);
#endif

    if (!eglGetConfigAttrib(egldisplay, config, EGL_CONFIG_ID, &configId)) {
        fprintf(stderr, "eglGetConfigAttrib failed!");
        return 0;
    }

#ifdef DEBUG
    printf("EGL: Using config #%d\n", configId);
    printConfig(egldisplay, config);
#endif
...


Regards,
Michael

[Giuseppe Iellamo]

Sorry I can't help you but maybe you can help me.

I'm using your exact configuration...

But I still read

root@arm:/opt/Graphics_SDK_setuplinux_5_00_00_01/gfxsdkdemos/ogles2# ./OGLES2ChameleonMan
PVRShell: EGL 1.4 initialized
Exit message has been set to: "PVRShell: Unable to create surface
".
InitAPI failed!
PVRShell: Unable to create surface

I've loaded the modules, I've installe the libraries in
gfx_rel_es8.x

Please tell me what did you load from the Graphics SDK?

[John Syne]

From: Giuseppe Iellamo <pep...@gmail.com>
Reply-To: <beagl...@googlegroups.com>
Date: Thursday, January 9, 2014 at 8:11 AM
To: <beagl...@googlegroups.com>
Subject: [beagleboard] Re: JavaFX 8 on BeagleBone Black (SGX driver problem?)

Sorry I can't help you but maybe you can help me.

I'm using your exact configuration...

But I still read

root@arm:/opt/Graphics_SDK_setuplinux_5_00_00_01/gfxsdkdemos/ogles2# ./OGLES2ChameleonMan
PVRShell: EGL 1.4 initialized
Exit message has been set to: "PVRShell: Unable to create surface
".
InitAPI failed!
PVRShell: Unable to create surface

I've loaded the modules, I've installe the libraries in
gfx_rel_es8.x

Please tell me what did you load from the Graphics SDK?

Hi Giuseppe,

Here is what I did to make this work:

Download the Graphics SDK from here:

http://software-dl.ti.com/dsps/dsps_public_sw/gfxsdk/5_01_00_01/index_FDS.html

Use the hardfp version if you are using Debian.

Make the file executable and install. Update the Rules.make file to reflect your environment. Here is an example:

=================

# <userid> is your login userid

HOME=/home/<userid>

# Set Toolchain path. I installed Linaro via apt-get install arm-linux-gnueabihf-gcc

CSTOOL_DIR=/usr

# Set Tool chain prefix (ex arm-arago-linux-gnueabi- )

CSTOOL_PREFIX=arm-linux-gnueabihf-

# Set kernel installation path ( ex /home/user/linux-04.00.01.13 )

KERNEL_INSTALL_DIR=${HOME}/GIT/linux-dev-3.12/KERNEL

# Set Target filesystem path ( ex /home/user/targetfs )

TARGETFS_INSTALL_DIR=/mnt/<userid>/rootfs

# Set installation folder

GRAPHICS_INSTALL_DIR=${HOME}/Graphics/Graphics_SDK_5_01_00_01_hardfp

==================

Host > make BUILD=release OMAPES=8.x PM_RUNTIME=1 all

# Insert SDCARD and I’m assuming the rootfs is mounted at /mnt/<userid>/rootfs from Rules.make file above

Host > sudo make BUILD=release OMAPES=8.x PM_RUNTIME=1 install

After you boot your BBB, run

BBB > sudo /etc/init.d/335x-demo

# Reboot

BBB > sudo /etc/init.d/rc.pvr start

Now run the demos and everything should work fine.

Regards,

John

--
For more options, visit http://beagleboard.org/discuss
---
You received this message because you are subscribed to the Google Groups "BeagleBoard" group.
To unsubscribe from this group and stop receiving emails from it, send an email to beagleboard...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

[Rusty Wright]

Do you need the JavaFX 8 graphics capabilities?  If not, can you run it with graphics disabled?

On the chance that there might be useful information in them, here are two links for java I have from my days of noodling around with the raspi:

http://www.oracle.com/technetwork/articles/java/raspberrypi-1704896.html

http://www.savagehomeautomation.com/raspi-jdk8

[Giuseppe Iellamo]

And it worked!!!!!

Thank you very much John and Robert!

Now I'm heading to QT5 realm!

[littlef...@gmail.com]

> Do you need the JavaFX 8 graphics capabilities?  If not, can you run it with graphics disabled?

I do need the graphics capabilities (otherwise I wouldn't use FX ;-) so I cannot disable the graphics.

> On the chance that there might be useful information in them, here are two links for java I have from my days of noodling around with the raspi: ...

Thanks for the links but unfortunately they don't help. I have a Raspi and I can also run my JavaFX applications on it without problems. But they are rather slow on the Raspi so I wanted to do some performance tests with the BBB hardware.

[rodrigo...@gmail.com]

Hi littlf...
Did you find how to get jdk8 and javafx running on BBB?
I'm trying to do the same (RasPi was to slow, so I want to see if BBB does better).
Also, how do you use sound in JavaFx+ RPi ??? (AudioClip and all media doesnt work on RPi)
Thanks
(It's been 2 days fighting with BBB, any herlp will be appreciated)
@rodrinauta

[gwhit...@gmail.com]

Can any one post some updated info on getting JavaFX running on BBB with the latest Debian.  I'm new building kernels in Linux.  I'm trying to get JavaFX running on a BBB through the HDMI port and if possible a touchscreen like the BB-VIEW or 4D.  Any help would be greatly appreciated, even if its just a link to an article I may have missed in my searh.

Thanks,

Greg

[Nick Apperley]

Managed to get OpenJFX (open source version of JavaFX) working on the BBB, however it is running on Ubuntu. Got a sample JavaFX program (Ensemble) working by accident on X11 by specifying a renderer that isn't supported by OpenJFX. Key part here is that you must specify the unsupported renderer otherwise no JavaFX program will start, and some errors will appear in the terminal. Automatic renderer detection with OpenJFX is currently broken.

Big mystery as to what renderer is being used since it isn't listed in the Monocle wiki page on the OpenJDK website. X11 render doesn't work just in case anyone was wondering. Note that there are no existing builds of OpenJFX available to use on the BBB. A custom build must be created which is only compatible with Java SE 8 Embedded and Java SE 8.

[Nick Apperley]

You're not going to like this suggestion. Far better option to get JavaFX and OpenJFX working and performing well is to run JavaFX programs on a Wandboard (eg the Dual version) since all the Vivante GPUs are supported, even graphics acceleration via both GPUs. Sadly the GPU on the BBB will NEVER be fully supported as long as Imagitech refuse to completely release the specifications for all of its GPUs.

Plenty of people are in the same situation with trying to get the SGX GPU working on the BBB but to no avail. So much time spent trying to get the GPU supported. To make matters worse the Beaglebone Foundation's upcoming Beagleboard X15 also contains a SGX GPU. Extremely FRUSTRATING!

Much better off spending the money on a Wandboard to get JavaFX programs running easily. Ubuntu/Debian have a package containing the open source Vivante GPU driver which covers most of the Vivante GPUs (2D and 3D ones). Easy as pie to install the GPU driver via apt-get (single step). Wish it was the same with the BBB.

Why does the Beaglebone Foundation always insist on using a SGX GPU in every board that gets released?

[Robert Nelson]

On Sun, Nov 23, 2014 at 2:06 PM, Nick Apperley <napp...@gmail.com> wrote:
> Managed to get OpenJFX (open source version of JavaFX) working on the BBB,
> however it is running on Ubuntu. Got a sample JavaFX program (Ensemble)
> working by accident on X11 by specifying a renderer that isn't supported by
> OpenJFX. Key part here is that you must specify the unsupported renderer
> otherwise no JavaFX program will start, and some errors will appear in the
> terminal. Automatic renderer detection with OpenJFX is currently broken.

Hey Nick,

I'd love to see your setup that got it working on ubuntu. I can setup
a *.deb package on: http://repos.rcn-ee.net/

Regards,

--
Robert Nelson
http://www.rcn-ee.com/

[Robert Nelson]

On Sun, Nov 23, 2014 at 2:39 PM, Nick Apperley <napp...@gmail.com> wrote:
> You're not going to like this suggestion. Far better option to get JavaFX
> and OpenJFX working and performing well is to run JavaFX programs on a
> Wandboard (eg the Dual version) since all the Vivante GPUs are supported,
> even graphics acceleration via both GPUs. Sadly the GPU on the BBB will
> NEVER be fully supported as long as Imagitech refuse to completely release
> the specifications for all of its GPUs.
>
> Plenty of people are in the same situation with trying to get the SGX GPU
> working on the BBB but to no avail. So much time spent trying to get the GPU
> supported. To make matters worse the Beaglebone Foundation's upcoming
> Beagleboard X15 also contains a SGX GPU. Extremely FRUSTRATING!
>
> Much better off spending the money on a Wandboard to get JavaFX programs
> running easily. Ubuntu/Debian have a package containing the open source
> Vivante GPU driver which covers most of the Vivante GPUs (2D and 3D ones).
> Easy as pie to install the GPU driver via apt-get (single step). Wish it was
> the same with the BBB.

For reference, which package is that? The x15's got a 2D engine from
Vivante.. Does it still rely on a blob from vivante to build the
final package?

[Nick Apperley]

Found a closed source Vivante GPU driver package (gpu-viv-bin-mx6q) however it is heavily tied into a Freescale ARM CPU. Will see if I can the name of the open source one. Saw it somewhere on the Phoenix (Linux gamer/performance website).

[Nick Apperley]

Open source GPU driver is called etna_viv. Phoronix have an article covering etna_viv. No package for the open source driver yet. Momentarily got confused with the gpu-viv-acc-fb package which only contains the closed source Vivante GPU driver, not the open source one that I thought the package had.

[Robert Nelson]

On Sun, Nov 23, 2014 at 3:13 PM, Nick Apperley <napp...@gmail.com> wrote:
> Open source GPU driver is called etna_viv. Phoronix have an article covering
> etna_viv. No package for the open source driver yet. Momentarily got
> confused with the gpu-viv-acc-fb package which only contains the closed
> source Vivante GPU driver, not the open source one that I thought the
> package had.

Darn! Thanks for the update, I've been loosely following etna_viv, was
hoping they got passed the freescale *.bin requirement..

[Nick Apperley]

@Robert - Are you referring to OpenJFX for a Deb package. See that there is still no OpenJDK 8 package for Ubuntu/Debian, how annoying. OpenJFX (ARM) is reported to work on OpenJDK 8 (ARM).

Building OpenJDK is not for the faint of heart. Attempting to do so would likely result in a brain dump. Also a super computer would be required to do the build in quick time, which I don't have.

Luckily doing a build of OpenJFX in desktop Linux is straightforward and can be done in around 30 min on a Intel Core 2 Duo laptop for example.

[Robert Nelson]

On Sun, Nov 23, 2014 at 3:25 PM, Nick Apperley <napp...@gmail.com> wrote:
> @Robert - Are you referring to OpenJFX for a Deb package. See that there is
> still no OpenJDK 8 package for Ubuntu/Debian, how annoying. OpenJFX (ARM) is
> reported to work on OpenJDK 8 (ARM).
>
> Building OpenJDK is not for the faint of heart. Attempting to do so would
> likely result in a brain dump. Also a super computer would be required to do
> the build in quick time, which I don't have.
>
> Luckily doing a build of OpenJFX in desktop Linux is straightforward and can
> be done in around 30 min on a Intel Core 2 Duo laptop for example.

Yeap... deb package for arm for ubuntu/debian.. ;)

Up until the last few release's i've been building chromium on my
imx6/2gb & omap5/2gb hardware, but even that's oom'ing now (even with
lpae enabled on the omap5..)... I'm hoping an a80/4gb that's coming
will fix that.

[Nick Apperley]

Sounds as crazy like building DirectFB 1.6 on the BBB running Debian. Amazing considering the BBB only has 512 MB of RAM and swap isn't enabled.

[Nick Apperley]

Below are some handy links for creating custom builds of OpenJDK and OpenJFX.

- Building OpenJFX

- Building OpenJFX For ARM

- Official Debian GIT For OpenJFX (GIT Hub)

- How To Build And Package OpenJDK 8 With OpenJFX On Linux (x86 only)

Hmmmm.... Looks as though Debian officially has some packaging for OpenJFX, however they would need to also have packaging for OpenJDK 8 too in order for OpenJFX to work. Wonder if the packaging targets x86 only or if it extends to ARM as well. Robert - If you have time can you please look into this one.

On Monday, 24 November 2014 09:44:00 UTC+13, RobertCNelson wrote:

[Nick Apperley]

Wonder if the OpenJFX package (package name?) has been uploaded to the Debian Unstable repository yet.

[Robert Nelson]

On Mon, Nov 24, 2014 at 3:08 PM, Nick Apperley <napp...@gmail.com> wrote:
> Wonder if the OpenJFX package (package name?) has been uploaded to the
> Debian Unstable repository yet.

It's there:

https://packages.qa.debian.org/o/openjfx.html

Just a few build errors:

https://buildd.debian.org/status/package.php?p=openjfx&suite=unstable

[Nick Apperley]

Thanks for the update. Damm, oh well there is the option to do a custom OpenJFX build. Hardest part is to do a custom build of OpenJDK 8 for ARM. Having great difficulty in tracking down good documentation in this area, specifically for the ARM side. Might be worth looking at the CACAO and JamVM JREs (Java 8 compatible?) and seeing if they will be compatible with the OpenJFX runtime.

[Robert Nelson]

On Mon, Nov 24, 2014 at 3:25 PM, Nick Apperley <napp...@gmail.com> wrote:
> Thanks for the update. Damm, oh well there is the option to do a custom
> OpenJFX build. Hardest part is to do a custom build of OpenJDK 8 for ARM.
> Having great difficulty in tracking down good documentation in this area,
> specifically for the ARM side. Might be worth looking at the CACAO and JamVM
> JREs (Java 8 compatible?) and seeing if they will be compatible with the
> OpenJFX runtime.

well less errors on armhf than openjfx. ;)

https://buildd.debian.org/status/package.php?p=openjdk-8&suite=unstable

[Nick Apperley]

Is there a successful build of OpenJDK 8u20 for armhf?

[Robert Nelson]

On Mon, Nov 24, 2014 at 3:57 PM, Nick Apperley <napp...@gmail.com> wrote:
> Is there a successful build of OpenJDK 8u20 for armhf?

I'll fire up a build and see. ;)

[Nick Apperley]

Thanks for that. Discovered that JamVM isn't an actual JRE since it relies on the class libraries which are available in OpenJDK. JamVM is only a JVM (Java Virtual Machine). From the outset JamVM might be able to run JavaFX programs since it is compatible with OpenJDK 8. IcedTea (OpenJDK derivative) uses JamVM as its JVM.

Originally JamVM was designed to perform well on very resource constrained embedded devices. Now JamVM can be used on Linux PCs (x86/AMD64).

[Nick Apperley]

Hopefully if the custom build of OpenJDK 8 works then that can be adjusted to work on the stable Debian/Ubuntu images for the BBB.

On Tuesday, 25 November 2014 10:59:26 UTC+13, RobertCNelson wrote:

[Robert Nelson]

On Mon, Nov 24, 2014 at 4:21 PM, Nick Apperley <napp...@gmail.com> wrote:
> Hopefully if the custom build of OpenJDK 8 works then that can be adjusted
> to work on the stable Debian/Ubuntu images for the BBB.

If the test build goes thru, i'll just add it to:

https://github.com/rcn-ee/repos

So it'll end up on repos.rcn-ee.net :)

[Nick Apperley]

Excellent, highly appreciated :)

[Nick Apperley]

Can't see the OpenJDK 8 build. I take it the build failed?

Have some fantastic news on the Ubuntu front. Various OpenJDK 8 packages via an unofficial PPA are available for Ubuntu 12.04 (Precise) and 14.04 (Trusty).

On Tuesday, 25 November 2014 11:23:20 UTC+13, RobertCNelson wrote:

[Robert Nelson]

On Tue, Nov 25, 2014 at 4:42 PM, Nick Apperley <napp...@gmail.com> wrote:
> Can't see the OpenJDK 8 build. I take it the build failed?
>
> Have some fantastic news on the Ubuntu front. Various OpenJDK 8 packages via
> an unofficial PPA are available for Ubuntu 12.04 (Precise) and 14.04
> (Trusty).

OH, it's still building.. Been at it almost 24 hours...

It's in the middle of the test suite:

compiler/java/jar running ...
Passed: tools/javac/annotations/repeatingAnnotations/combo/ReflectionTest.java
Passed: tools/javac/annotations/repeatingAnnotations/combo/RetentionAnnoCombo.java
Passed: tools/javac/annotations/repeatingAnnotations/combo/TargetAnnoCombo.java
Passed: tools/javac/annotations/repeatingAnnotations/BaseAnnoAsContainerAnno.java

Compare with:
https://buildd.debian.org/status/fetch.php?pkg=openjdk-8&arch=armhf&ver=8u40~b04-2.1&stamp=1413548257

which took over 5 days..

I think my omap5 is a little more powerful then debian's armhf builder..

[William Hermans]

OH, it's still building..  Been at it almost 24 hours...

It's in the middle of the test suite:

compiler/java/jar running ...
Passed: tools/javac/annotations/

repeatingAnnotations/combo/ReflectionTest.java
Passed: tools/javac/annotations/repeatingAnnotations/combo/RetentionAnnoCombo.java
Passed: tools/javac/annotations/repeatingAnnotations/combo/TargetAnnoCombo.java
Passed: tools/javac/annotations/repeatingAnnotations/BaseAnnoAsContainerAnno.java

Compare with:
https://buildd.debian.org/status/fetch.php?pkg=openjdk-8&arch=armhf&ver=8u40~b04-2.1&stamp=1413548257

which took over 5 days..

I think my omap5 is a little more powerful then debian's armhf builder..

My god, what on earth could possibly take that long to build on a quad with 2GB of ram ? I'm assuming you're using the wanderboard . . . but you mentioned omap5 above, which is what ? the X15 ? lol

If OpenJDK takes that long the kernel must take weeks . . . and yeah, mostly that was sarcasm.

[Robert Nelson]

For the heavy stuff i use an omap5 (Cortex-A15 dual core 1.5Ghz) it's got 2GB of ram just like the wand quad's. It's actually faster then the Quad A9's at 1Ghz.

The omap5 is two generations prior to the am57xx found on the x15..

Regards,

[Robert Nelson]

Just a quick random benchmark which had the omap5 @ 800Mhz..

http://www.cnx-software.com/2013/08/17/texas-instruments-omap5432-evm-benchmarked-against-odroid-u2-beaglebone-black-gk802-and-an-intel-core-i7-2600k-based-pc/

Regards,

[William Hermans]

LOL !!!

The oddity is the BeagleBone Black (BBB) which can outperform the quad core i.MX6 in all tests

I'll have to read that in fuller detail when im 100% sober . . .

 

--
For more options, visit http://beagleboard.org/discuss
---
You received this message because you are subscribed to the Google Groups "BeagleBoard" group.
To unsubscribe from this group and stop receiving emails from it, send an email to beagleboard...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Robert Nelson]

On Tue, Nov 25, 2014 at 7:18 PM, William Hermans <yyr...@gmail.com> wrote:

LOL !!!

The oddity is the BeagleBone Black (BBB) which can outperform the quad core i.MX6 in all tests

I'll have to read that in fuller detail when im 100% sober . . .

Since that was 2013, i wouldn't be surprised a tunned uni-processor arm kernel from that era was just more task optimized then an arm smp setup.

Regards,

[Robert Nelson]

On Tue, Nov 25, 2014 at 4:42 PM, Nick Apperley <napp...@gmail.com> wrote:
> Can't see the OpenJDK 8 build. I take it the build failed?
>
> Have some fantastic news on the Ubuntu front. Various OpenJDK 8 packages via
> an unofficial PPA are available for Ubuntu 12.04 (Precise) and 14.04
> (Trusty).

So it only took 38 hours 19 minutes. ;)

http://repos.rcn-ee.net/debian/pool/main/o/openjdk-8/

Built for jessie:

sudo apt-get update ; sudo apt-get install openjdk-8-jdk

[William Hermans]

Ah, the guy was talking specifically about the OpenSSL test specifically in that case. Reading that "article" today, seems like this person is not very good at benchmarking. As everything is done willy nilly with unfair advantages to some hardware. e.g. the Beaglebone was benchmarked @ 720p for some test, where one of the other boards was tested at 1080p, and directly compared.

that said, it should give an "out-of-the-box" feel for each board. That and this person did more than I did in this regard so . . .

[John Syn]

On 11/26/14, 7:59 AM, "Robert Nelson" <robert...@gmail.com> wrote:

>On Tue, Nov 25, 2014 at 4:42 PM, Nick Apperley <napp...@gmail.com>
>wrote:
>> Can't see the OpenJDK 8 build. I take it the build failed?
>>
>> Have some fantastic news on the Ubuntu front. Various OpenJDK 8
>>packages via
>> an unofficial PPA are available for Ubuntu 12.04 (Precise) and 14.04
>> (Trusty).
>
>So it only took 38 hours 19 minutes. ;)

So the Quad CortexA9 took 5 days (120 hours) and the OMAP5 took 38 hours
which is more then 3X performance. My bet is the build was running all
CPUs close to 100% for the entire build so this would be a good comparison.

Regards,
John

>
>http://repos.rcn-ee.net/debian/pool/main/o/openjdk-8/
>
>Built for jessie:
>
>sudo apt-get update ; sudo apt-get install openjdk-8-jdk
>
>Regards,
>
>--
>Robert Nelson
>http://www.rcn-ee.com/
>

[Nick Apperley]

Successfully built OpenJFX from the newest HG revision at the time of this post. OpenJFX build took just over 3 min on a Linux Mint 17 Intel Core 2 Duo laptop. Not sure why I thought the build was going to take around 30 min.

Must not be much in OpenJFX or the laptop is reasonably fast at compiling software (??).

[Nick Apperley]

Exact build time is 3 mins 4.834 secs.

[Nick Apperley]

Curious. Which ARM CPU does the OMAP5 SoC use?

[liyaoshi]

Cortex-A15 dual cores

[Nick Apperley]

No wonder the OMAP5 is so much more grunty.

[Nick Apperley]

Wonder what it would be like to create an OpenJDK 8 build using a lower end Beagleboard X15 (the one that is currently in the design phase).

[John Syn]

From: Nick Apperley <napp...@gmail.com>
Reply-To: "beagl...@googlegroups.com" <beagl...@googlegroups.com>
Date: Wednesday, November 26, 2014 at 6:39 PM
To: "beagl...@googlegroups.com" <beagl...@googlegroups.com>
Subject: Re: [beagleboard] Re: JavaFX 8 on BeagleBone Black (SGX driver problem?)

Wonder what it would be like to create an OpenJDK 8 build using a lower end Beagleboard X15 (the one that is currently in the design phase).

The ARM performance of the AM572x will have a similar performance to the OMAP5. Both are Dual CortexA15 running at 1.5GHz. However, the DSP on the AM572x is faster than the DSP on the OMAP5 and the AM572x has two DSPs. 

Regards,

John

--

[Nick Apperley]

Have uploaded a copy of the OpenJFX 1.8u40 runtime (GPL 2 license with class path exception). On the BBB Extract the lib directory from the 7z file and move the directory to the JRE installation directory.

On Thursday, 27 November 2014 14:38:44 UTC+13, Nick Apperley wrote:

[Nick Apperley]

Google Groups text editor has mysteriously decided to stop working. Here is the link to the OpenJFX runtime.

[Nick Apperley]

By default OpenJDK 8 uses the Zero JVM which doesn't provide a high level of performance. Highly recommended to use the high performance JamVM JVM instead, which is comparable performance wise to the JVM used in Oracle's Java SE Embedded 8. JamVM was originally designed to perform well on resource constrained embedded devices. Java 8 Lambda features can be used in a JavaFX program on an embedded device using JamVM, which the Oracle JVM doesn't support (in the case of Java SE Embedded 8).

[Nick Apperley]

Do be aware with OpenJFX that JavaFX programs on the BBB will only run in X11 with DirectFB being specified as the renderer. DirectFB on the BBB with the current revision of OpenJFX that I was using had the bizarre GTK requirement, which prevents the framebuffer (no X11) from being used. Also a custom version of DirectFB must be built and installed. There is an ancient version of DirectFB in the Debian/Ubuntu repository but it doesn't work with OpenJFX or JavaFX.

[Nick Apperley]

Some very useful info on changing JRE/JVMs with Debian/Ubuntu.

[Nick Apperley]

Basic overview of setting up a BBB running Ubuntu 14.04 to run a JavaFX program (rough draft):

1. Add unofficial OpenJDK 8 PPA

2. Update package list

3. Install OpenJDK 8 JamVM package

4. Make sure OpenJDK is installed by running java -version in a terminal (should see OpenJDK 1.8 being mentioned in the output)

5. Build DirectFB 1.6

6. Install DirectFB to /usr

7. Extract OpenJFX 1.8u40 runtime file (7z) using p7zip and copy the lib directory to the OpenJDK 8 JRE directory

8. Run Ensemble8 demo in X11 via LXDE to make sure that a JavaFX program will run properly:

java -Djavafx.platform=DirectFB -jamvm -jar /home/a_user/sample_dir/Ensemble8.jar

Would it be possible to have a JavaFX section in the BBB wiki on the ELinux website?

[Nick Apperley]

FYI OpenJFX 1.8u40 includes Dialog/Alert API. If you use the Oracle JDK then you will have to wait until around March 2015 to take advantage of the new API.

[Robert Nelson]

On Wed, Nov 26, 2014 at 2:38 PM, John Syn <john...@gmail.com> wrote:
>
> On 11/26/14, 7:59 AM, "Robert Nelson" <robert...@gmail.com> wrote:
>
>>On Tue, Nov 25, 2014 at 4:42 PM, Nick Apperley <napp...@gmail.com>
>>wrote:
>>> Can't see the OpenJDK 8 build. I take it the build failed?
>>>
>>> Have some fantastic news on the Ubuntu front. Various OpenJDK 8
>>>packages via
>>> an unofficial PPA are available for Ubuntu 12.04 (Precise) and 14.04
>>> (Trusty).
>>
>>So it only took 38 hours 19 minutes. ;)
> So the Quad CortexA9 took 5 days (120 hours) and the OMAP5 took 38 hours
> which is more then 3X performance. My bet is the build was running all
> CPUs close to 100% for the entire build so this would be a good comparison.

So over thanksgiving i let my quad wand do the same task:

omap5_uevm: (i was also building cloud9 during week, let's say 95% of
the time was openjdk)
Build needed 38:19:47, 6215244k disc space

wand quad: (i was out of state while this was running, so 100% core usage..)
Build needed 50:06:26, 6215460k disc space

The dual a15's come out to 30% faster then the quad a9's...

[Robert Nelson]

I'm thinking debian might still have been using a mx53loco based box
for that 120hour run..

[John Syn]

Interesting.

http://www.arm.com/products/processors/cortex-a/cortex-a15.php

In this article, ARM says the CortexA15 is 2X performance compared to the
CortexA9, so 2xA15 = 4xA9; however, the OMAP5432 is running at 1.5MHz and
the Wandboard running at 1MHz, so given that, I would expect the OMAP5 to
be about 50% faster. Clearly Debian are running much slower hardware.

Regards,
John

>
>Regards,
>
>--
>Robert Nelson
>http://www.rcn-ee.com/
>

[Nick Apperley]

Robert - What is the performance like with a Dual Core Cortex A9 compared to a Quad Core Cortex A9?

[Robert Nelson]

On Nov 29, 2014 6:21 PM, "Nick Apperley" <napp...@gmail.com> wrote:
>
> Robert - What is the performance like with a Dual Core Cortex A9 compared to a Quad Core Cortex A9?

With the quad wand, you also get 2gb of ram over the dual which only has 1gb.. That alone is worth the price delta. For the price/performance it's really hard to beat the wand quad.

>
>
> On Sunday, 30 November 2014 04:51:10 UTC+13, RobertCNelson wrote:
>>
>> omap5_uevm: (i was also building cloud9 during week, let's say 95% of
>> the time was openjdk)
>> Build needed 38:19:47, 6215244k disc space
>>
>> wand quad: (i was out of state while this was running, so 100% core usage..)
>> Build needed 50:06:26, 6215460k disc space
>>
>> The dual a15's come out to 30% faster then the quad a9's...
>>
>> Regards,
>>
>> --
>> Robert Nelson
>> http://www.rcn-ee.com/
>

[Nick Apperley]

Robert - thanks for the advise. Would it make sense economically to go for the Wandboard Quad over the upcoming Beagleboard X15 when it comes to running GUI programs?

[Robert Nelson]

On Nov 29, 2014 7:22 PM, "Nick Apperley" <napp...@gmail.com> wrote:
>
> Robert - thanks for the advise. Would it make sense economically to go for the Wandboard Quad over the upcoming Beagleboard X15 when it comes to running GUI programs?

3d: vivante will be reversed engineered before powervr. Freescale hasn't released 'armhf' drivers whereas ti has..

The 'x15' will be supported by bb.org and ti.. Who really supports the wand? All the info I put out is just because I replaced my failing pandas with quad wands...

Regards,

[liyaoshi]

http://www.freescale.com/lgfiles/NMG/MAD/YOCTO/gpu-viv-bin-mx6q-3.10.17-1.0.0-hfp.bin

Do you mean this ?

--

[Nick Apperley]

Looks as though Freescale have made the closed source driver available for all the GPUs they use (Vivante). Have heard that JavaFX programs will run via the framebuffer (hardware rendering) just fine with both the closed (official) and open source (unofficial - reverse engineered) Vivante GPU drivers.

[Nick Apperley]

Presumably the closed source driver is an installer that is run on any Freescale imx6 board. If that is the case then TI's GPU support is **ahem, cough, cough** severely lacking.

[liyaoshi]

I have tested Ti gpu driver on omap5 uevm board based with GLSDK 6.04 

The gpu driver crash didn't be fixed for a very long time .

iMX6 GPU will crash also ,if high cpu load and very frequency interrupt case, you might get kernel panic on imx6-solo 

but on fsl QUAD cores  , Its more stable , maybe because of their gpu kernel  driver is based on work queue . .

But for FSL kernel driver . lack of I2C /SPI dma support , if you want use very high spi through put , like DMB/CMMB chip always do 

You will get poor performance .

[Nick Apperley]

Hang on a moment. The version of Yocto that is available for Wandboard targets the armhf CPU platform. Are the Panda's notoriously unreliable hardware wise?

[Robert Nelson]

On Sun, Nov 30, 2014 at 10:09 PM, Nick Apperley <napp...@gmail.com> wrote:
> Hang on a moment. The version of Yocto that is available for Wandboard
> targets the armhf CPU platform. Are the Panda's notoriously unreliable
> hardware wise?

i.mx6 builtin sata: wand <-> sata drive
panda no sata: panda <-> usb-sata <-> sata drive..

The smsc95xx used on the panda, just isn't 100% reliable, when
compared to the imx6's builtin sata...