Hi,
I’ve got a JOGL project that is displaying an animated, flickering box. Unfortunately, sometimes (typically about 30 seconds into the program) the flickering stops for a second or so (it should change about every 1/8 of a second) before picking up again.
specifying -verbose:gc indicates that it isn’t the garbage collection. What else could cause this? How can I check these causes?
Many thanks!
You’re giving us very little info 
Basically the animation is interrupted, and then continues, and it’s not the GC… oh well!
Maybe you’re using System.nanoTime(), which is known to go back in time every once in a while, due to CPUs throttling their frequency.
Other than that, we need much much more information, like a self contained (!) testcase, so that we can run it, or if you’re lucky: look through it and point you are potential bugs!
Fair enough.
Interestingly, your nanoTime comment might be it! I was using it because I wanted accuracy better than currentTimeMillis, but if it’s an issue, maybe that’s a good trade-off (no flicker for worse accuracy). Is there a better timer to use to get millisecond accuracy (not just precision) than nanoTime?
Anyway, here is the basic loop for the image presentation:
:
while( !m_stop_running && stimulus_iterator.hasNext() )
{
TimedStimulus item = stimulus_iterator.next();
// Load image into ByteBuffer
//
loadStimulus( item );
// Determine how long to wait before display
//
// Time to present - (elapsed time since last stimulus) - (typical time to render)
//
sleep_time = item.getTime() +
( m_start - System.nanoTime() ) -
( m_average_image_render_time );
// If we have plenty of time, then sleep until we're ready
//
if( sleep_time > 0 )
{
try
{
Thread.sleep( sleep_time / Constants.NANOSECONDS_PER_MILLISECOND );
}
catch( InterruptedException ex )
{
continue;
}
}
// Skip the image if it's over 5 seconds late
//
else if( sleep_time < -5 * Constants.NANOSECONDS_PER_SECOND )
{
log.warn( "Image display late " + sleep_time + " nsec. Skipping." );
continue;
}
// Present the stimulus
//
try
{
m_gl_canvas.display();
}
catch( GLException glx )
{
log.warn( "Error displaying image: " + glx.getLocalizedMessage() );
}
} // end loop over timed stimulus items in the task
:
The display method looks like this:
:
// Draw to the buffer as RGB bytes
// m_buffer populated in the loadImage() call in main loop
//
GL gl = drawable.getGL();
gl.glPushAttrib( GL.GL_COLOR_BUFFER_BIT );
gl.glDrawPixels( 800, 600, GL.GL_RGB, GL.GL_UNSIGNED_BYTE, m_buffer );
gl.glPopAttrib();
gl.glFlush();
:
Thanks for your help!
This might be your problem; the JVM shares CPU with the OS, so you should always call sleep() even it’s with a low value, otherwise the OS gets a backlog of tasks which may take a while to clear.
I disagree with you. Calling sleep even a vanishingly small time decreases the frame rate noticeably and is not necessary. I removed it in TUER and the frame rate is twice better. I have almost no problem of timing, I use currentTimeMillis for the moment.
I disagree with your disagreement! 
My tests showed a 10% performance drop without sleep() - maybe it’s because I’m using software rendering…
please sign some Non-Agreement Disclosure