Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.arch.embedded > #26612 > unrolled thread

Sync to Android frame rate

Back to article view | Back to comp.arch.embedded

Contents

  Sync to Android frame rate John Speth <johnspeth@yahoo.com> - 2017-05-17 17:23 -0700
    Re: Sync to Android frame rate John Speth <johnspeth@yahoo.com> - 2017-05-18 11:59 -0700
      Re: Sync to Android frame rate rickman <gnuarm@gmail.com> - 2017-05-18 21:53 -0400
        Re: Sync to Android frame rate Theo Markettos <theom+news@chiark.greenend.org.uk> - 2017-05-19 18:28 +0100
          Re: Sync to Android frame rate rickman <gnuarm@gmail.com> - 2017-05-19 13:47 -0400
          Re: Sync to Android frame rate John Speth <johnspeth@yahoo.com> - 2017-05-19 12:47 -0700
      Re: Sync to Android frame rate John Speth <johnspeth@yahoo.com> - 2017-05-19 12:51 -0700
        Re: Sync to Android frame rate Theo Markettos <theom+news@chiark.greenend.org.uk> - 2017-05-22 11:00 +0100
          Re: Sync to Android frame rate John Speth <johnspeth@yahoo.com> - 2017-05-22 09:24 -0700
            Re: Sync to Android frame rate Theo Markettos <theom+news@chiark.greenend.org.uk> - 2017-05-22 22:30 +0100

#26612 — Sync to Android frame rate

Hi Folks (and pardon me if I'm not on topic)-

I'm looking for a way to synchronize some code activity to the Android 
display frame rate. It's a short bit of code that will simply set the 
color of a small display region, nothing more. Is there some sort of 
method or hook that I can exploit for this purpose?

I can use a timer running at very fast intervals successfully but the 
rendering appears to suffer from aliasing with the frame rate. For 
example, my frame rate is 60 Hz so if I set the timer interval to 1/60 
seconds, the appearance is non-uniform. I can't set the timer to exactly 
16.666 msec and even if I could, I'd still see aliasing.

Thanks - JJS

[toc] | [next] | [standalone]

#26618

> I'm looking for a way to synchronize some code activity to the Android
> display frame rate. It's a short bit of code that will simply set the
> color of a small display region, nothing more. Is there some sort of
> method or hook that I can exploit for this purpose?

I'll answer my own question, partially that is.  It appears one solution 
is to use the Choreographer class which provides a way to hook a 
callback function into the frame rate.  Below is the code.  It executes 
but the timing is sloppy.

I can time the calls to doFrame() using System.nanoTime().  A little 
testing shows me the timing of calls to doFrame() is not uniform 
(ranging from 12 msec to 20 msec, expecting about 16 msec).  The 
callback argument is rock steady (about 16 msec) which makes me think 
it's a value that is computed at initialization time.

It's a solution but it's not doing what I wanted because the callback is 
not timing precisely enough.  Still seeking advice, thank you.

JJS

///////////////////////////////////////////////////////////////

private FrameCallback frameCallback = null;
private boolean frameCallbackPending = false;

public void armVSyncHandler() {
     if(!frameCallbackPending) {
         frameCallbackPending = true;
         if(frameCallback == null)
         {
             frameCallback = new FrameCallback() {
                 @Override
                 public void doFrame(long frameTimeNanos) {
                     frameCallbackPending = false;

                     // Do some work here

                     armVSyncHandler();
                 }
             };
         }
         Choreographer.getInstance().postFrameCallback(frameCallback);
     }
}

[toc] | [prev] | [next] | [standalone]

#26619

On 5/18/2017 2:59 PM, John Speth wrote:
>> I'm looking for a way to synchronize some code activity to the Android
>> display frame rate. It's a short bit of code that will simply set the
>> color of a small display region, nothing more. Is there some sort of
>> method or hook that I can exploit for this purpose?
>
> I'll answer my own question, partially that is.  It appears one solution
> is to use the Choreographer class which provides a way to hook a
> callback function into the frame rate.  Below is the code.  It executes
> but the timing is sloppy.
>
> I can time the calls to doFrame() using System.nanoTime().  A little
> testing shows me the timing of calls to doFrame() is not uniform
> (ranging from 12 msec to 20 msec, expecting about 16 msec).  The
> callback argument is rock steady (about 16 msec) which makes me think
> it's a value that is computed at initialization time.
>
> It's a solution but it's not doing what I wanted because the callback is
> not timing precisely enough.  Still seeking advice, thank you.

I know nothing about Android, but it seems to me there should be a 
graphics library that provides this sort of functionality.  I would 
think your code would provide a buffer of data to be presented to the 
screen and a graphics call would hand it to the graphic processor which 
in turn copies that buffer in a highly synchronized manner.

Trying to run user code synchronized to the frame update would be 
difficult to say the least.

-- 

Rick C

[toc] | [prev] | [next] | [standalone]

#26623

rickman <gnuarm@gmail.com> wrote:
> 
> I know nothing about Android, but it seems to me there should be a 
> graphics library that provides this sort of functionality.  I would 
> think your code would provide a buffer of data to be presented to the 
> screen and a graphics call would hand it to the graphic processor which 
> in turn copies that buffer in a highly synchronized manner.

There is, it's called SurfaceFlinger.
Though you might want to use higher levels of the graphics stack because you
may need further layers of compositing (eg controls on top of the image):
https://source.android.com/devices/graphics/architecture

Theo

[toc] | [prev] | [next] | [standalone]

#26624

On 5/19/2017 1:28 PM, Theo Markettos wrote:
> rickman <gnuarm@gmail.com> wrote:
>>
>> I know nothing about Android, but it seems to me there should be a
>> graphics library that provides this sort of functionality.  I would
>> think your code would provide a buffer of data to be presented to the
>> screen and a graphics call would hand it to the graphic processor which
>> in turn copies that buffer in a highly synchronized manner.
>
> There is, it's called SurfaceFlinger.
> Though you might want to use higher levels of the graphics stack because you
> may need further layers of compositing (eg controls on top of the image):
> https://source.android.com/devices/graphics/architecture

Looks right, but complicated.  I didn't see anything that would provide 
the synchronization the OP is looking for but I can't believe it 
wouldn't be there.  That would be required for glitch-less display 
updates which we know happens.

-- 

Rick C

[toc] | [prev] | [next] | [standalone]

#26625

On 5/19/2017 10:28 AM, Theo Markettos wrote:
> rickman <gnuarm@gmail.com> wrote:
>>
>> I know nothing about Android, but it seems to me there should be a
>> graphics library that provides this sort of functionality.  I would
>> think your code would provide a buffer of data to be presented to the
>> screen and a graphics call would hand it to the graphic processor which
>> in turn copies that buffer in a highly synchronized manner.
>
> There is, it's called SurfaceFlinger.
> Though you might want to use higher levels of the graphics stack because you
> may need further layers of compositing (eg controls on top of the image):
> https://source.android.com/devices/graphics/architecture
>
> Theo
>

Thanks, Theo.  That looks like it's worth some experimentation.

JJS

[toc] | [prev] | [next] | [standalone]

#26626

On 5/18/2017 11:59 AM, John Speth wrote:
>> I'm looking for a way to synchronize some code activity to the Android
>> display frame rate. It's a short bit of code that will simply set the
>> color of a small display region, nothing more. Is there some sort of
>> method or hook that I can exploit for this purpose?
>
> I'll answer my own question, partially that is.  It appears one solution
> is to use the Choreographer class which provides a way to hook a
> callback function into the frame rate.  Below is the code.  It executes
> but the timing is sloppy.
>
> I can time the calls to doFrame() using System.nanoTime().  A little
> testing shows me the timing of calls to doFrame() is not uniform
> (ranging from 12 msec to 20 msec, expecting about 16 msec).  The
> callback argument is rock steady (about 16 msec) which makes me think
> it's a value that is computed at initialization time.
>
> It's a solution but it's not doing what I wanted because the callback is
> not timing precisely enough.  Still seeking advice, thank you.
>
> JJS
>
> ///////////////////////////////////////////////////////////////
>
> private FrameCallback frameCallback = null;
> private boolean frameCallbackPending = false;
>
> public void armVSyncHandler() {
>     if(!frameCallbackPending) {
>         frameCallbackPending = true;
>         if(frameCallback == null)
>         {
>             frameCallback = new FrameCallback() {
>                 @Override
>                 public void doFrame(long frameTimeNanos) {
>                     frameCallbackPending = false;
>
>                     // Do some work here
>
>                     armVSyncHandler();
>                 }
>             };
>         }
>         Choreographer.getInstance().postFrameCallback(frameCallback);
>     }
> }

Update: I ran this code on a newer and probably faster phone (Samsung 
Galaxy Grand Prime).  It seems to work like I wanted (no aliasing).  I 
conclude the original phone (ZTE X752C) was way short of compute capacity.

JJS

[toc] | [prev] | [next] | [standalone]

#26639

John Speth <johnspeth@yahoo.com> wrote:
> Update: I ran this code on a newer and probably faster phone (Samsung 
> Galaxy Grand Prime).  It seems to work like I wanted (no aliasing).  I 
> conclude the original phone (ZTE X752C) was way short of compute capacity.

If all you're doing is filling the screen with colour and you need to buy a
faster phone, it feels something is wrong.

For one thing, if that is the intended application (rather than a reduced
test case of some other application) you should be using the GPU to do it
and take basically no CPU time.

Theo

[toc] | [prev] | [next] | [standalone]

#26641

On 5/22/2017 3:00 AM, Theo Markettos wrote:
> John Speth <johnspeth@yahoo.com> wrote:
>> Update: I ran this code on a newer and probably faster phone (Samsung
>> Galaxy Grand Prime).  It seems to work like I wanted (no aliasing).  I
>> conclude the original phone (ZTE X752C) was way short of compute capacity.
>
> If all you're doing is filling the screen with colour and you need to buy a
> faster phone, it feels something is wrong.
>
> For one thing, if that is the intended application (rather than a reduced
> test case of some other application) you should be using the GPU to do it
> and take basically no CPU time.

You're right, Theo.  The application is a custom device that mates to a 
phone display and needs to receive signals, phone to custom device. 
Communication is one way and contains very little data (a few bytes per 
message).  I think I can signal a photo-transistor in the custom device 
by switching a region of the screen as the comm channel.  I'm weighing 
the trade-off of going deep and stepping around the Android graphics SW 
or playing nicely with it.  The alternative is a costlier but much more 
reliable BLE comm.

JJS

[toc] | [prev] | [next] | [standalone]

#26642

John Speth <johnspeth@yahoo.com> wrote:
> You're right, Theo.  The application is a custom device that mates to a 
> phone display and needs to receive signals, phone to custom device. 
> Communication is one way and contains very little data (a few bytes per 
> message).  I think I can signal a photo-transistor in the custom device 
> by switching a region of the screen as the comm channel.  I'm weighing 
> the trade-off of going deep and stepping around the Android graphics SW 
> or playing nicely with it.  The alternative is a costlier but much more 
> reliable BLE comm.

I'd have thought you make two buffers, one white the other black.
Then you alternately tell it to composite one or other.
You'll likely get a one-frame lag but, unless your phone is utterly ancient
(< Android 4.0), the GPU will render it for you.  Think of it as 'playing a
60Hz movie' but with only two images to choose from, that you never change
pixels inside.

If you're trying to colour in pixels from Java it's going to be a lot slower -
because each one has to reach down the graphics stack, and that's all CPU
side.

Theo

[toc] | [prev] | [standalone]

Back to top | Article view | comp.arch.embedded

csiph-web