Hello Users,
I am new to Psychophysical Toolbox and am benchmarking its
performance ofor my machines, displays, and experimental tasks.
As a quality-assurance test, I tried to display a 32 bit, 1024x786
movie at 100hz on a PC. This shouldnt work too well; it is a lot of
memory to copy in 10ms. In fact, even rushed at maximum priority,
tearing is evident. I was looking for evidence of the failure in time
stamps from getSecs. Unfortunately, I did not find it; the time
stamps were consistent with no missed frames! This is unfortunate.
In my experiments, it is desirable to know when the system is not
behaving as specified.
To help troubleshoot the problem, I put getSecs commands around the
SCREEN('copywindow') command. It reported taking between 40 and 80
microsecnds to copy from an offscreen window to the onscreen one. I
suspect it takes this duration to issue the command to the video card
but that copying the window takes considerably longer. My guess is
that displaying a rushed loop is a parallel process in which Matlab
issues a command to copy the window to the video card but doesnt wait
for the actual copying to finish. Hence, all the time stamps are
determined by waitblanking and indicate no missed frames even though
the display is not what is specified.
My question is how can I find out, on a frame-by-frame basis, whether
the latency of copying the window took within the requisite time for
frame-accurate displays (in a rushed loop). I realize I can look at the
screen for certain stimuli and see tearing. But I would like a
frame-by-frame basis for assuring the quality of collected
experimental data.
I certainly increase the likelihood of frame-accurate displays by
using 8 bit displays, coding images in lookup tables, or copying
smaller amounts of memory. As a matter of experimental control, I am
seeking a frame-by-frame way of knowing whether the computer has
executed the commands as specified or not. Any help will be greatly
appreciated.
Thank you,
Jeff Rouder
ps. Please cc replies to me at rouderj@...
pps.Here is my beginner's quality code. timer1 and timer2 output should
be in milliseconds.
numtrials=4;
numImage = 5;
dur=[1,1,1,1,1];
bg=0;
white=255;
grey=127;
xc=1024/2;
yc=768/2;
rect=[xc-100,yc-300 ,xc+100,yc+300];
w=Screen(0,'OpenWindow',[],[],32);
HideCursor;
myscale=(0:255)'*ones(1,3);
myscale(2,1:3)=[255 0 0];
myscale(3,1:3)=[0 255 0];
SCREEN(w,'SetClut',myscale); %greyscale?
SCREEN(w,'FillRect',bg);
waitSecs(1); % Give display a moment to recover from the change of
pixelSize
for i=1:numImage
s(i)=SCREEN(w,'OpenOffscreenWindow',bg);
end
timer1=1:numImage;
timer2=1:numImage;
timer3=1:numImage;
priorityLevel=MaxPriority(0,'WaitBlanking');
SCREEN(s(1),'FillRect',bg,rect);
SCREEN(s(2),'FillRect',grey,rect);
SCREEN(s(3),'FillRect',white,rect);
SCREEN(s(4),'FillRect',grey,rect);
SCREEN(s(5),'FillRect',bg,rect);
for t=1:numtrials
loop={
'SCREEN(w,''WaitBlanking'');'
'for i=1:numImage;'
'timer1(i)=getSecs;'
'SCREEN(''CopyWindow'',s(i),w);'
'timer2(i)=getSecs;'
'timer3(i)=SCREEN(w,''WaitBlanking'',dur(i));'
'end;'
};
SCREEN('Screens');
RUSH(loop);
waitSecs(.3);
end;
Screen('CloseAll');
diff(timer1)*1000
diff(timer2)*1000
timer3
1000*(timer2-timer1)
I am new to Psychophysical Toolbox and am benchmarking its
performance ofor my machines, displays, and experimental tasks.
As a quality-assurance test, I tried to display a 32 bit, 1024x786
movie at 100hz on a PC. This shouldnt work too well; it is a lot of
memory to copy in 10ms. In fact, even rushed at maximum priority,
tearing is evident. I was looking for evidence of the failure in time
stamps from getSecs. Unfortunately, I did not find it; the time
stamps were consistent with no missed frames! This is unfortunate.
In my experiments, it is desirable to know when the system is not
behaving as specified.
To help troubleshoot the problem, I put getSecs commands around the
SCREEN('copywindow') command. It reported taking between 40 and 80
microsecnds to copy from an offscreen window to the onscreen one. I
suspect it takes this duration to issue the command to the video card
but that copying the window takes considerably longer. My guess is
that displaying a rushed loop is a parallel process in which Matlab
issues a command to copy the window to the video card but doesnt wait
for the actual copying to finish. Hence, all the time stamps are
determined by waitblanking and indicate no missed frames even though
the display is not what is specified.
My question is how can I find out, on a frame-by-frame basis, whether
the latency of copying the window took within the requisite time for
frame-accurate displays (in a rushed loop). I realize I can look at the
screen for certain stimuli and see tearing. But I would like a
frame-by-frame basis for assuring the quality of collected
experimental data.
I certainly increase the likelihood of frame-accurate displays by
using 8 bit displays, coding images in lookup tables, or copying
smaller amounts of memory. As a matter of experimental control, I am
seeking a frame-by-frame way of knowing whether the computer has
executed the commands as specified or not. Any help will be greatly
appreciated.
Thank you,
Jeff Rouder
ps. Please cc replies to me at rouderj@...
pps.Here is my beginner's quality code. timer1 and timer2 output should
be in milliseconds.
numtrials=4;
numImage = 5;
dur=[1,1,1,1,1];
bg=0;
white=255;
grey=127;
xc=1024/2;
yc=768/2;
rect=[xc-100,yc-300 ,xc+100,yc+300];
w=Screen(0,'OpenWindow',[],[],32);
HideCursor;
myscale=(0:255)'*ones(1,3);
myscale(2,1:3)=[255 0 0];
myscale(3,1:3)=[0 255 0];
SCREEN(w,'SetClut',myscale); %greyscale?
SCREEN(w,'FillRect',bg);
waitSecs(1); % Give display a moment to recover from the change of
pixelSize
for i=1:numImage
s(i)=SCREEN(w,'OpenOffscreenWindow',bg);
end
timer1=1:numImage;
timer2=1:numImage;
timer3=1:numImage;
priorityLevel=MaxPriority(0,'WaitBlanking');
SCREEN(s(1),'FillRect',bg,rect);
SCREEN(s(2),'FillRect',grey,rect);
SCREEN(s(3),'FillRect',white,rect);
SCREEN(s(4),'FillRect',grey,rect);
SCREEN(s(5),'FillRect',bg,rect);
for t=1:numtrials
loop={
'SCREEN(w,''WaitBlanking'');'
'for i=1:numImage;'
'timer1(i)=getSecs;'
'SCREEN(''CopyWindow'',s(i),w);'
'timer2(i)=getSecs;'
'timer3(i)=SCREEN(w,''WaitBlanking'',dur(i));'
'end;'
};
SCREEN('Screens');
RUSH(loop);
waitSecs(.3);
end;
Screen('CloseAll');
diff(timer1)*1000
diff(timer2)*1000
timer3
1000*(timer2-timer1)