MacFormat España 21/macformat_21.iso/Shareware/Programación/VideoToolbox/TimeVideo results

Text File | 1996-09-11 | 11.9 KB | 197 lines | [TEXT/ttxt]

This file reports the timing and accuracy of all your video screens, as
measured by TimeVideo, a component of the VideoToolbox. TimeVideo runs on all
Macs, requiring only System 6.05 or better. To quickly test a large number of
computers, run TimeVideo from a floppy disk; all the results will accumulate
in a single results file. TimeVideo is free, and may be freely distributed.
You can get the latest version from:
ftp://mirror.apple.com/mirrors/info-mac/cfg/time-video-376.hqx

THE VIDEO TOOLBOX
The VideoToolbox is a collection of two hundred C subroutines and several demo
and utility programs that I and others have written to do visual psychophysics
with Macintosh computers. It is fully compatible with 680x0 and PowerPC Macs
and with Metrowerks CodeWarrior C and Symantec C compilers. It's free and may
not be sold without permission. It should be useful to anyone who wants to
present accurately specified visual stimuli or use the Mac for psychometric
experiments. The text file "Video synch" discusses all the ways of
synchronizing programs to video displays and the many pitfalls to avoid. The
TimeVideo application checks out the timing of all video devices in
anticipation of their use in critical real-time applications, e.g. movies or
lookup table animation. Low-level routines control video timing and lookup
tables, display real-time movies, and implement the luminance-control
algorithms suggested by Pelli and Zhang (1991). (D.G. Pelli and L. Zhang,
1991, "Accurate control of contrast on microcomputer displays." Vision
Research, 31, 1337-1350. Reprints are available.) In particular,
GetPixelsQuickly and SetPixelsQuickly peek and poke pixels in bitmaps and
pixmaps, CopyBitsQuickly and CopyWindows faithfully copy between bit/pixmaps
and the screen, WindowToEPS saves an image to disk, for later printing or
incorporation into a document, and SetEntriesQuickly and GDSetEntries load the
screen's color lookup table, all without any of QuickDraw's color
translations. High-level routines help analyze psychophysical experiments
(e.g. graphing or maximum-likelihood fitting of psychometric data). Assign.c
is a runtime C interpreter for C assignment statements, which is useful for
controlling experiments and sharing calibration data. Many of the routines are
Mac-specific, but some very useful routines, e.g. the luminance-control,
statistics, maximum-likelihood fitting algorithms, and the runtime interpreter
are written in Standard C and will work on any computer. Documentation is in
the source files themselves. This collection has been continually updated
since 1991. You can download the latest version of "video-toolbox" from a
public archive, e.g.:
ftp://sumex-aim.stanford.edu/info-mac/dev/lib/
ftp://mirror.apple.com/mirrors/info-mac/dev/lib/
ftp://mirrors.aol.com/pub/info-mac/dev/lib/
ftp://src.doc.ic.ac.uk/packages/mac/info-mac/dev/lib/
ftp://ftp.stolaf.edu/pub/macpsych/
CompuServe://MacDev forum/Library 4 "C and Pascal"/VIDEOT.SEA
Or you can request a file by email; for instructions send a query to
[e-mail].

To get future versions automatically, just send me your name and email
address. Each time I post a new version of the VideoToolbox to the Info-Mac
Archives, I email a copy to everyone on the subscription list; there are
currently over 170 subscribers. (Let me know if you prefer notification only,
without the 3 MB enclosure.)
Denis Pelli, [e-mail], Psychology Dept., New York University, 6
Washington Place, New York, NY 10003, USA


TIME VIDEO
Other than reading and writing pixels, all access to a video device normally
goes through the software video driver. (You can use VideoToolbox
SetEntriesQuickly.c to bypass the video driver of a few devices, but I don't
recommend that, except as a last resort, because your program will then only
work with the few video devices that SetEntriesQuickly.c supports.) The video
driver is normally supplied in ROM on the video card or, for built-in video,
in the computer's ROM. (Apparently, new PCI drivers can simply be tossed onto
the System Folder.) Many drivers have subtle bugs that TimeVideo has uncovered
and documented. A complete list of all known video driver bugs appears in the
"Video bugs" document included in the VideoToolbox archive described above.
Please send new bugs to [e-mail]

For each video card, TimeVideo measures the video frame rate, frequency of VBL
interrupts (ought to be one per frame), how long it takes to load the clut
(ought to be one frame or less), and how much of the screen you can fill with
a real-time one-image-per-frame movie shown by CopyBits() or
CopyBitsQuickly(). It then performs a random write-then-read test of the Color
Lookup Table (clut). This tests the clut memory hardware and the software used
to write and read the clut. We test writing by GDSetEntries(), which makes a
cscSetEntries call to the video driver, and, if possible, we also test writing
by SetEntriesQuickly(), which accesses the hardware directly.
(SetEntriesQuickly supports only a few video cards.) In either case, the clut
is read by GDGetEntries, which makes a cscGetEntries call to the video driver.
The testing is thorough; many video devices fail at least part of the test.
All the driver errors uncovered to date appear in the VideoToolbox "Video
bugs" text file, and have been reported to the video card's manufacturer. Add
your results by emailing this file to [e-mail]

Happily, we don't know of any cases of errors in cscSetEntries itself, and
we've never run across a hardware failure of the CLUT memory. Our experience
is that errors in the cscSetEntries/cscGetEntries write-then-read test of the
CLUT occur in two ways:
1. The driver does not implement cscSetGamma and the fixed gamma table is not
the identity transformation, i.e. "uncorrected gamma table". TimeVideo always
calls cscSetGamma, requesting an identity transform. In all drivers, the RGB
values supplied in a cscSetEntries call are transformed by the gamma table
before being loaded into the CLUT. The presence of a non-identity gamma table
is usually obvious from a glance at the color matrix, since the gains in the
matrix will exceed 1 by about twice the fractional rms error indicated by ±%.
2. The CLUT is set correctly, but the driver incorrectly implements
cscGetEntries. Bugs in cscGetEntries are usually obvious from a glance at the
color matrix, e.g. nonzero gains off the diagonal, or the pattern of errors.

The video driver on the PowerMac 7500 and 8500 supresses the VBL interrupt
while it's loading the CLUT, so any VBL-frame-counting task will miss an
interrupt each time you call cscSetEntries to load the clut. Calling
cscSetEntries once per frame to synchronize your code with the frame should
still work fine, but it'll be slightly tricky to check for overrun (i.e.
taking too long between cscSetEntries calls), since a simple
VBL-interrupt-based frame counter will run differently on different Macs.

Errors reported by TimeVideo are usually due to bugs in the video driver
software in the ROM of the video card (or built-in video device). If the bug
will interfere with your experiments, then to use the card (or built-in video)
you must either fix/replace the driver or bypass the driver, using
SetEntriesQuickly to access the hardware directly (if SetEntriesQuickly
supports it). I suggest that you try replacing the driver, because this will
keep your software hardware-independent. The VideoToolbox "Video synch"
document explains how to fix the the Mac IIci video driver, patching or
replacing the buggy version 0 .Display_Video_Apple_RBV1 driver by copying the
bug-free version 1 of the same driver from the Mac IIsi. It is very likely
that an analogous approach could be used to fix/replace the buggy version 0,
1, and 2 .Display_Video_Apple_DAFB drivers in the Quadra 700, 750, and 900, by
the bug-free version 3 or 5 of that driver in the Centris 650 or the LC 475.
And please report your bugs; the bug-free revised drivers are probably the
result of our past bug reports.

GLOSSARY
A video frame is a complete refresh of your video screen. (With interlacing it
takes two fields to do a complete refresh, i.e. a frame, but graphics displays
usually are not interlaced and have a single field per frame.) To show a
movie, you will want to reload the image once per frame; the table shows how
big that image can be, as a fraction of the screen area, and still be reloaded
once per frame. (Some video cards have multiple video "pages"--call
GDGetPageCnt()--that can be switched by calling GDSetPageDrawn() and
GDSetPageShown().) A "VBL" interrupt is produced by your video card, nominally
once per frame, but some video cards produce more, which is poor, but not
serious. (The VBLInstall.c program will deal with it.) Suppressed interrupts
during clut updates are bad. It means that the driver disables the VBL
interrupt for too long while it's loading the clut. This will throw off any
interrupt-based attempt to count frames. (The clut is the color lookup table
of your video card.) Lookup table animation, e.g. for temporal modulation of
contrast, requires that you reload the clut once per frame, so it's very
important that this be fast enough. The first call to SetEntriesQuickly() for
each device is slow--a cache is filled; the reported times are for subsequent
calls. Each video card can be in "Color" or "Gray" mode, as set by the Control
Panel:Monitors or the Macintosh Toolbox call SetDepth(). In "Gray" mode all
colors are transformed to luminance-equivalent grays. This is done by the
video driver, when loading the clut, but only if the pixelSize≤8. TimeVideo
does a linear regression on the results of write-then-read tests of the CLUT
to measure the driver's color transformation matrix. It reports the matrix if
it is other than the identity transformation:
(ROut) (1 0 0) (RIn)
(GOut)=(0 1 0)x(GIn)
(BOut) (0 0 1) (BIn)
The clut tests try loading the clut serially, one entry at a time, and all at
once; some video drivers fail the serial test. For more explanation see the
text file called "Video synch" on the VideoToolbox disk.

"cscSetEntries ... NAN ... frames"= the driver refuses to load the CLUT.
"ok"= passed all tests.
"!gray"= passed the color test, but is supposed to be in gray mode.
"!color"= passed the gray test, but is supposed to be in color mode.
"!serial"= passed when loaded all at once, but failed when loaded serially.
"bad"= read did not equal write and the error is reported explicitly.
We know that SetEntriesQuickly is "!serial" on the Quadra, alas.

This is a SimpleText document using the Monaco font to line up the
columns properly.

TimeVideo version 3.7.6
Wednesday, September 11, 1996.
Compiled by Metrowerks CodeWarrior C 9.00 for PowerPC, 4B int and 8B double.
Using RAM Doubler.
MF Colin Dalziel's PowerMac 6100/60, ppc601, System 7.5.3. GestaltMachineType
75.

PowerMac 6100/60 "Built-In DRAM Video" (.Display_Video_Apple_Sonora version 7)
8-bit dacs. 640x480 pixels.
pixel size                         1     2     4     8    16     bits
pages                              1     1     1     1     1
mode                              0x80  0x81  0x82  0x83  0x84
frame rate                        66.5  66.6  66.5  66.5  66.5   Hz
interrupts per frame               1.0   1.0   1.0   1.0   1.0
CopyBits movie size                4.55  2.50  1.27  0.58  0.22  screen
CopyBitsQuickly movie size         3.69  2.23  1.19  0.57  0.22  screen
CopyBits data rate                11.09 12.19 12.33 11.30  8.61  MB/s
CopyBitsQuickly data rate          9.00 10.88 11.63 11.20  8.57  MB/s
cscSetEntries duration             1.00  1.00  1.00  1.00  1.00  frames
cscSetEntries suppresses ints.for  0.0   0.0  -0.0   0.0  -0.0   frames
GDSetEntriesHighPriority duration  1.00  1.00  1.00  1.00  1.00  frames
color: cscSetEntries test            ok    ok    ok    ok    ok
gray: cscSetEntries test             ok    ok    ok    ok    ok
 (ROut± 0.4%) (0.30 0.59 0.11) (RIn)
 (GOut± 0.4%)=(0.30 0.59 0.11)x(GIn)
 (BOut± 0.4%) (0.30 0.59 0.11) (BIn)