A new year, a new beta update: "The fun in dysfunction"
Dedicated to the fine engineers at Apple - specifically the ones which
pay such an incredible amount of attention to detail, it's almost magic,
but certainly amaaaazing.
Major changes:
* Support for GStreamer camerabin2 video capture & recording: Once a
future release of the GStreamer-SDK catches up in functionality, this
will enable video capture and recording on 64-Bit OSX and for 64-Bit
Matlab on MS-Windows. Currently it only works on Linux, or for 32-Bit
Matlab on Windows or for OSX either with 32-Bit Matlab or with a
self-compiled GStreamer installation from Homebrew.
* Support for a new builtin imaging pipeline panel fitter: This allows
to open onscreen windows whose drawing area size is bigger or smaller
than the area of the screen occupied by the window. A new optional
parameter 'clientRect' for Screen('Openwindow',...); allows to specify
the "virtual size" of the window, whereas the old 'rect' parameter still
specifies the true size of the window on the screen. Psychtoolbox will
pretend your window is only the size of 'clientRect', but at
Screen('Flip') time it will upscale or downscale your stimulus image to
the true size of the window, applying linear or better filtering for
this zoom.
Why would you want to do that? Typical flat panels and projectors only
work optimally at their native resolution. All other resolutions usually
cause at least funny timing problems or sometimes visual artifacts with
dynamic stimuli. This option allows you to transfer a script written for
visual stimulation on, e.g., a CRT monitor or a panel on a certain
resolution, to a different experimental setup which has a different
native display resolution with minimal modifications to the script in
case your script made assumptions about the display resolution in use.
It would be especially useful if you worked in a place where some
equipment is badly maintained and tends to break down in the middle of
data collection, so it needs to be swapped with incompatible equipment
of different native resolution.
This release still misses the high-level setup code in PsychImaging() to
access and configure this feature in convenient ways, so the panel
fitter may not work well or need some tinkering when used together with
some advanced features, e.g., some stereo display modes or high
precision display modes which use a unusual display geometry themselves.
This will be fixed in a future release. For standard bread & butter
rescaling it works well.
Of course it is still better to choose your resolutions wisely or write
your code resolution independent if possible, as the scaling comes at
some cost of another millisecond or two spent on rescaling in the gpu.
* Improved Linux support for timestamping, graphics problem detection
and desktop compositor handling. This further increases the advantage
Linux users already had over OSX and Windows users wrt. timing
precision. This beta has been tested with different common desktop GUI's
for Linux wrt. to correct stimulus display, proper timing and
timestamping and display synchronization across dual-display setups,
e.g., for stereoscopic presentation, with a NVidia GeForce 8800 card
under Ubuntu Linux 12.04.1-LTS, the latest NVidia binary drivers and a
Datapixx for external measurement of stimulus onset timing and timestamping.
It works perfectly well on single-display and on dual-display setups
under KDE-4 with KWin desktop manager, GNOME-3/Gnome-Shell with Mutter
dektop manager, LXDE with OpenBox, GNOME-2 classic without desktop
compositor, and XFCE-4 with/without compositor.
It works perfectly well on single display setups, but not on
multi-display setups with desktop GUI's based on Compiz: Unity and
GNOME-2 work well single display but not dual-display. This must be a
bug in Compiz introduced by the latest software updates to Ubuntu 12.04
LTS sometime December 2012. Canonical applied some optimizations to
improve video game performance, which worked out well on single display
setups, but apparently went wrong for dual-display setups.
It doesn't work well at all with the Unity-2D desktop, something is
broken there.
If you run data collection i would recommend running a desktop GUI
without compositor, e.g., by creating a separate "experiment" user
account on your machine where data collection is run and a pure 2d
desktop is installed, e.g., XFCE-4 with desktop composition turned off,
or LXDE+OpenBox. This gives you that extra piece of mind that even if
there would be compositor related bugs in Psychtoolbox or your system,
no compositor could interfere, because something that doesn't exist
can't interfere. It may also give you a few percent extra performance.
However, the desktop GUIs mentioned above are well behaved even if a
compositor is active, as PTB can temporarily turn off the compositor, or
the compositors switch themselves to standby during an experiment
session. It's a tradeoff you have to make between paranoia vs.
convenience and bling.
* Improved support for handling of MS-Windows-8 dektop compositor:
Psychtoolbox is better at detecting if a desktop compositor on Windows
Vista/7/8 interferes with timing and can warn you in most cases.
However, Windows-8 does no longer allow to disable the compositor,
neither manually by you, nor automatically by us. This means you have to
completely rely on the compositor disabling itself for fullscreen window
displays, something that didn't work well at all in the past. Testing
with an up to date Windows-7 system showed that the compositor now
auto-disables for fullscreen windows on single-display setups, but fails
miserably on multi-display setups. I didn't test with Windows-8 as we
don't have or use that os. Maybe Windows-8 behaves better than
Windows-7. If it doesn't then it is completely unuseable for timed
stimulus presentation an anything but a single display setup, with no
second monitor even connected.
* Bugfixes / Workarounds for some MS-Windows bugs and many new MacOSX
10.7+ operating system bugs. If i had to describe Apples latest
operating systems in one word it would be "Dysfunctional". Especially
many fixes are related to use of high-precision display devices like
Datapixx/Viewpixx/Propixx/Bits+/Bits# etc.
* Improved workarounds for handling of slightly broken Intel-DDX
graphics drivers on Linux (pre 2.20.16 series). Intel video drivers
before version 2.20.16 but after June 2011 have a bug in their swap
scheduling. Psychtoolbox will detect this bug and enable a workaround
for such drivers since the previous beta. The detection code has been
improved. Of course it is best to update to a fixed driver (v2.20.16 or
later) asap.
* Small fixes to PsychKinectCore and to PsychHID.
* Various other fixes for small documentation and PTB bugs.
* Potential multi-threaded decoding improvements for GStreamer movie
playback. The multi-threaded high performance playback mode introduced
in the previous beta is so far only effective on Linux and on the
GStreamer packages provided by the Homebrew package manager on OSX, but
not with the GStreamer-SDK on OSX or Windows. This beta prepares higher
performance playback with a future version of the GStreamer-SDK.
* Support for a popular HDMI stereo frame packing format, side-by-side
horizontally compressed stereo. PsychImaging() has a function to enable
this, ImagingStereoDemo(102) demonstrates it.
* Datapixx and Eyelink improvements.
* Use of multisample anti-aliasing MSAA with imaging pipeline enabled
should now work better on deficient OSX graphics drivers, especially in
3d rendering mode due to addition of new/improved error handling and
fallback path.
-mario
Dedicated to the fine engineers at Apple - specifically the ones which
pay such an incredible amount of attention to detail, it's almost magic,
but certainly amaaaazing.
Major changes:
* Support for GStreamer camerabin2 video capture & recording: Once a
future release of the GStreamer-SDK catches up in functionality, this
will enable video capture and recording on 64-Bit OSX and for 64-Bit
Matlab on MS-Windows. Currently it only works on Linux, or for 32-Bit
Matlab on Windows or for OSX either with 32-Bit Matlab or with a
self-compiled GStreamer installation from Homebrew.
* Support for a new builtin imaging pipeline panel fitter: This allows
to open onscreen windows whose drawing area size is bigger or smaller
than the area of the screen occupied by the window. A new optional
parameter 'clientRect' for Screen('Openwindow',...); allows to specify
the "virtual size" of the window, whereas the old 'rect' parameter still
specifies the true size of the window on the screen. Psychtoolbox will
pretend your window is only the size of 'clientRect', but at
Screen('Flip') time it will upscale or downscale your stimulus image to
the true size of the window, applying linear or better filtering for
this zoom.
Why would you want to do that? Typical flat panels and projectors only
work optimally at their native resolution. All other resolutions usually
cause at least funny timing problems or sometimes visual artifacts with
dynamic stimuli. This option allows you to transfer a script written for
visual stimulation on, e.g., a CRT monitor or a panel on a certain
resolution, to a different experimental setup which has a different
native display resolution with minimal modifications to the script in
case your script made assumptions about the display resolution in use.
It would be especially useful if you worked in a place where some
equipment is badly maintained and tends to break down in the middle of
data collection, so it needs to be swapped with incompatible equipment
of different native resolution.
This release still misses the high-level setup code in PsychImaging() to
access and configure this feature in convenient ways, so the panel
fitter may not work well or need some tinkering when used together with
some advanced features, e.g., some stereo display modes or high
precision display modes which use a unusual display geometry themselves.
This will be fixed in a future release. For standard bread & butter
rescaling it works well.
Of course it is still better to choose your resolutions wisely or write
your code resolution independent if possible, as the scaling comes at
some cost of another millisecond or two spent on rescaling in the gpu.
* Improved Linux support for timestamping, graphics problem detection
and desktop compositor handling. This further increases the advantage
Linux users already had over OSX and Windows users wrt. timing
precision. This beta has been tested with different common desktop GUI's
for Linux wrt. to correct stimulus display, proper timing and
timestamping and display synchronization across dual-display setups,
e.g., for stereoscopic presentation, with a NVidia GeForce 8800 card
under Ubuntu Linux 12.04.1-LTS, the latest NVidia binary drivers and a
Datapixx for external measurement of stimulus onset timing and timestamping.
It works perfectly well on single-display and on dual-display setups
under KDE-4 with KWin desktop manager, GNOME-3/Gnome-Shell with Mutter
dektop manager, LXDE with OpenBox, GNOME-2 classic without desktop
compositor, and XFCE-4 with/without compositor.
It works perfectly well on single display setups, but not on
multi-display setups with desktop GUI's based on Compiz: Unity and
GNOME-2 work well single display but not dual-display. This must be a
bug in Compiz introduced by the latest software updates to Ubuntu 12.04
LTS sometime December 2012. Canonical applied some optimizations to
improve video game performance, which worked out well on single display
setups, but apparently went wrong for dual-display setups.
It doesn't work well at all with the Unity-2D desktop, something is
broken there.
If you run data collection i would recommend running a desktop GUI
without compositor, e.g., by creating a separate "experiment" user
account on your machine where data collection is run and a pure 2d
desktop is installed, e.g., XFCE-4 with desktop composition turned off,
or LXDE+OpenBox. This gives you that extra piece of mind that even if
there would be compositor related bugs in Psychtoolbox or your system,
no compositor could interfere, because something that doesn't exist
can't interfere. It may also give you a few percent extra performance.
However, the desktop GUIs mentioned above are well behaved even if a
compositor is active, as PTB can temporarily turn off the compositor, or
the compositors switch themselves to standby during an experiment
session. It's a tradeoff you have to make between paranoia vs.
convenience and bling.
* Improved support for handling of MS-Windows-8 dektop compositor:
Psychtoolbox is better at detecting if a desktop compositor on Windows
Vista/7/8 interferes with timing and can warn you in most cases.
However, Windows-8 does no longer allow to disable the compositor,
neither manually by you, nor automatically by us. This means you have to
completely rely on the compositor disabling itself for fullscreen window
displays, something that didn't work well at all in the past. Testing
with an up to date Windows-7 system showed that the compositor now
auto-disables for fullscreen windows on single-display setups, but fails
miserably on multi-display setups. I didn't test with Windows-8 as we
don't have or use that os. Maybe Windows-8 behaves better than
Windows-7. If it doesn't then it is completely unuseable for timed
stimulus presentation an anything but a single display setup, with no
second monitor even connected.
* Bugfixes / Workarounds for some MS-Windows bugs and many new MacOSX
10.7+ operating system bugs. If i had to describe Apples latest
operating systems in one word it would be "Dysfunctional". Especially
many fixes are related to use of high-precision display devices like
Datapixx/Viewpixx/Propixx/Bits+/Bits# etc.
* Improved workarounds for handling of slightly broken Intel-DDX
graphics drivers on Linux (pre 2.20.16 series). Intel video drivers
before version 2.20.16 but after June 2011 have a bug in their swap
scheduling. Psychtoolbox will detect this bug and enable a workaround
for such drivers since the previous beta. The detection code has been
improved. Of course it is best to update to a fixed driver (v2.20.16 or
later) asap.
* Small fixes to PsychKinectCore and to PsychHID.
* Various other fixes for small documentation and PTB bugs.
* Potential multi-threaded decoding improvements for GStreamer movie
playback. The multi-threaded high performance playback mode introduced
in the previous beta is so far only effective on Linux and on the
GStreamer packages provided by the Homebrew package manager on OSX, but
not with the GStreamer-SDK on OSX or Windows. This beta prepares higher
performance playback with a future version of the GStreamer-SDK.
* Support for a popular HDMI stereo frame packing format, side-by-side
horizontally compressed stereo. PsychImaging() has a function to enable
this, ImagingStereoDemo(102) demonstrates it.
* Datapixx and Eyelink improvements.
* Use of multisample anti-aliasing MSAA with imaging pipeline enabled
should now work better on deficient OSX graphics drivers, especially in
3d rendering mode due to addition of new/improved error handling and
fallback path.
-mario