Hi!
On Thu, 14 Feb 2002, Joe Burks wrote:
> Anyway, if I read that code right the answer is that VIDIOCMCAPTURE ioctl
> must be called once for every frame you capture. If the frame in question is
> not yet available because it is spooling, the ioctl will block until the
> image is available.
Yes, VIDIOCMCAPTURE must be called for every frame.
No, VIDIOCMCAPTURE doesn't block. VIDIOCSYNC blocks.
To achieve the full capture rate the application usually has to call
VIDIOCMCAPTURE-for-frame-n+1 before frame n finishes, i.e. before the
VIDIOCSYNC-for-frame-n. To make this possible the driver must offer more
than one capture buffer and must be able to queue up several incoming
VIDIOCMCAPTURE calls - and dequeue them on every beginning frame
(interrupt-driven in case of capture cards).
A two-frame capture (frame n to buffer 0, frame n+1 to buffer 1) may look
like this (time axis points downwards):
Action
by app: .
- frame n-1
.
.
- VIDIOCMCAPTURE (buffer 0) // returns within nano/microseconds
- VIDIOCMCAPTURE (buffer 1) // returns within nano/microseconds
- VIDIOCSYNC (buffer 0) // return may last up to 40 ms (PAL)
.
.
- frame n // capturing to buffer 0 begins
.
.
.
. // VIDIOCSYNC returns
- frame n+1 // capturing to buffer 1 begins
- Process buffer 0
- VIDIOCSYNC (buffer 1) // return may last up to 20 ms (PAL)
.
.
. // VIDIOCSYNC returns
- frame n+2 // not captured
- Process buffer 1
.
Well, in this example - capturing ends after two frames - the processing
of buffer 0 may also take place after the second VIDIOCSYNC returns.
Bye,
Rolf
