OSXCompositorIdiocyTest - aleslab/Psychtoolbox-3-aleslab-fork GitHub Wiki

OSXCompositorIdiocyTest - Test for potential OSX compositor brokeness.

OSXCompositorIdiocyTest([testScreen=max][, distractorScreen]);

This test tests if the OSX desktop compositor is potentially
interfering with Screen('Flip') on a fullscreen window on a
display, especially in dual-display, dual-window configurations.

The test performs 300 black-white flips on 'testScreen'.
Then it optionally opens a fullscreen window on 'distractorScreen',
and repeats the 300 black-white flips on 'testScreen'. The 2nd
window does nothing.

What you should observe is black-white flicker in both "PASS I"
and "PASS II". If you observe flicker, that's a good sign, although
not a guarantee that everything is fine, timing-wise, on your system.

If you observe no flicker in "PASS I" then your systems flip
implementation is totally broken and not recommended for any
timing-sensitive visual stimulation.

If you observe flicker in "PASS I", but not in "PASS II" - when
the distractor window is open on 'distractorScreen', then your
system is possibly fine for single-display stimulation, but
broken for dual-display (e.g., binocular, stereoscopic, stereoMode 10)
stimulation wrt. presentation timing and timestamping.

What this test does is it alternates between the two buffers
of the doublebuffered window. One buffer is filled black, the
other white.

• If Screen('Flip') is executed by your operating system
  and graphics hardware via so called page-flipping, then you should
  observe flicker and the timing and timestamping of visual stimuli
  should be research grade.

• However, there is still some chance of observing flicker even
  if page-flipping isn't used, but the desktop compositor is active
  in a triple-buffer configuration. Therefore a successfull run of
  this test is not 100% proof that you are safe from timing woes.

If you don't observe flickering then this means a desktop compositor
or copy-swapping is active and the presentation timing and timestamps
are not to be trusted!

So far we know that at least OSX 10.8.5 with AMD graphics hardware
works properly for single-window / single-display visual stimulation,
but *not* in dual-window stimulation!
There is some indication that this dual-display stimulation OSX bug
can be possibly resolved or worked around via the applying the
Screen('Preference','ConserveVRAM', x); setting with x = 16384 or
x = 8192. Results look promising but are not 100% certain.

So far we also know that at least OSX 10.9 - 10.10 with NVidia
graphics hardware do not work properly at all in any configuration!