>Psychtoolbox>PsychTests

VRRTest([test=’sine’][, n=2000][, maxFlipDelta=0.2][, hwmeasurement=0][, testImage][, saveplots=0][, screenNumber=max][, usevulkan=0][, jitterFrac=0])

Test accuracy of VRR stimulation with variable timing, aka “FreeSync”,
“DisplayPort Adaptive Sync”, “HDMI VRR” or “G-Sync”.

The test exercises VRR on a suitable system, see “help VRRSupport” for setup
instructions, general info and caveats.

It submits OpenGL bufferswaps / flips of varying ‘delay’ between successive
flips and measures and plots how well the hw can follow the requested timing.

Usage:

All parameters are optional.

The ‘test’ parameter selects the test pattern:

‘sine’ Sine wave, smoothly changing. Exceeds VRR range to test low framerate
compensation (lfc). ‘sine’ is the default pattern if parameter is omitted.

‘random’ Randomized from flip to flip, within VRR range.

‘maxrandom’ Randomized from flip to flip, extending outside VRR range.

‘upsweep’ Linear increasing duration.

‘downsweep’ Linear decreasing duration.

‘upstep’ Stepwise increasing duration every 60 flips.

‘downstep’ Stepwise decreasing duration every 60 flips.

‘const’ Run at some constant frame duration.

‘n’ Number of flips / trials to run. 2000 by default.

‘maxFlipDelta’ Maximum frame duration in seconds. 0.2 secs = 200 msecs by default.

‘hwmeasurement’ Use external measurement hardware to get timing ground-truth:

0 = Off [Default]. This display a static ‘testImage’ to allow to check how much
the VRR display flickers under different stimulation timing regimes.

1 = VPixx DataPixx or similar VPixx device.

2 = VideoSwitcher + RTBox TTL pulse input port. Works with RTBox, but also with
emulated RTBox of the CRS Bits#. Takes TTL pulse input from the VideoSwitcher.
Selecting this will execute a enable/disable sequence for the VideoSwitcher
at start and end of a measurement session.

3 = CRS Bits# via a loopback cable from trigger out to trigger in BNC port.
Careful: Uses T-Lock for signalling/triggering at stimulus onset and therefore
a rather deficient (mis)design. This works as long as low framerate compensation
does not get triggered, otherwise the reference timestamps from the CRS hardware
will be completely useless and bogus trash!

4 = Like 2 – RtBox pulse input, but for use with a ColorCal2 or photo-diode that
sends a TTL pulse to the RtBox / Bits# BNC trigger input instead of VideoSwitcher.

5 = Measure via a supported photo-diode via PsychPhotodiode().

6 = Produce light-flash pattern to drive external photo-diode. Don’t record yourself.

‘testImage’ Either the name of an image file, or a numeric m x n or m x n x 3
matric with color values. The image read from the image file, or given image
matrix, will be displayed covering the full window in hwmeasurement == 0 mode.
If the parameter is omitted, one of Psychtoolbox default demo images will be
displayed. The purpose of this static image display is to test how much the
display device flickers under different VRR stimulation timings.

‘saveplots’ Should plots with results be saved to filesystem? Defaults to
0 for ‘No’, 1 = ‘Yes’.

‘screenNumber’ Number of screen to test on. Maximum X-Screen by default.

‘usevulkan’ If set to 1, use Vulkan display backend instead of OpenGL backend.
Default is 0 = Use standard OpenGL backend.

‘jitterFrac’ Amount of random timing noise to introduce into delay before
flip submission. 0 - 1 makes sense, 0 is the default.

You can abort the test earlier by pressing the ESC key.

The main plot figure will plot actual measured delay between successive flips
(blue) against requested optimal duration (red). It will also show the difference
in green and median error in red at the bottom of the graph. Median/Average error
and standard deviation is printed to the console last. The plot is also saved
to the current working directory as PDF file.

Preliminary results with a patch-set targeted at Linux 5.2:

Testing on a Linux 5.3 kernel showed good behaviour on tested AMD
DCE-8 (Sea Islands), DCE-11 (Polaris) and DCN-1 (Raven) gpu’s within the VRR
range of the monitor. Also pretty good behaviour on DCN-1 for flip rates below
the minimum VRR rate by use of low framerate compensation. lfc performance was
mixed on the DCE-8 and DCE-11 gpu’s. For ‘upsweep’, ‘upstep’ and some ‘const’
duration it was pretty reasonable or as good as DCN-1. For ‘downsweep’ and
‘downstep’ and some ‘const’ durations, lfc didn’t do a great job, with actual
frame durations often deviating by dozens of msecs from requested ones.