Re: rc3 - yuvscaler crash with -M BICUBIC

"Steven M. Schultz" <[email protected]> Thu, 6 Dec 2007 22:34:43 -0800 (PST)
Newsgroups gmane.comp.video.mjpeg.devel
Message-ID <[email protected]>
On Thu, 6 Dec 2007, herve.flores wrote:

> Hi this bug is not specific to rc3 (same behavior for rc2, I didn't =20
> test earlier versions)

	It would have been nice to hear about the problem a year ago
	with RC2 ;)

> seems to occurs with -M BICUBIC and too many input pixels
> command line:
> ffmpeg -v 0 -i ~/SANY0003.MP4 -aspect 16:9 -r 25:1 -s 1920x1080 -f =20
> yuv4mpegpipe - | yuvscaler -v 0 -M BICUBIC -O SIZE_720x576

	If anyone, perhaps the author of yuvscaler ;), needs to reproduce
	the problem without ffmpeg this can be used instead:

y4mblack -x 420jpeg  -n 10 -r 25:1 -w 1920 -h 1080 -a 1:1 -i b | \
	   yuvscaler -M BICUBIC -O SIZE_720x576 > /dev/null

> if I reduce size of input (before pipe to yuvscaler) =3D> succes

> I made some maths (with different sizes):
> 1920*1080 =3D 2073600 px =3D> failure :-(
> 1590*1080 =3D 1717200 px	:-(
> 1674*1028 =3D 1720872 px	:-(
> 
> 1580*1080 =3D 1706400 px =3D> succes :-)
> 1674*1026 =3D 1717524 px	:-)
> 1678*1024 =3D 1718272 px	:-)
	
	Interesting.

	However, the problem appears to be limited to Intel cpus running
	OSX 10.5.1

	OSX 10.4.11 PPC works fine
	OSX 10.5.1  PPC works fine

	I no longer have an Intel OSX 10.4.x system.

	Also, progressive input works, only interlaced input causes the
	crash.

> ----------------------------------------
> 
> The crash log:
	...

	Yep - I see the crash in the same place.

> ----------------------------------------
> 
> PS: if I remove -M BICUBIC, yuvscaler doesn't crash

	y4mscaler doesn't crash either ;)  (but a small patch is needed 
	until the author comes out with an updated version compatible
	with mjpegtools RC3).

> if you need more infos...

	You provided an excellent bug report.  Unfortunately I am not sure
	the problem can be fixed quickly - the code is hard to read/follow
	in places (i have a theory that the problem is once again due to
	malloc'd buffers being overrun).
	
> (sorry for my english)

	It is very good (and vastly better than my French :))

	Cheers,
	Steven Schultz


-------------------------------------------------------------------------
SF.Net email is sponsored by: 
Check out the new SourceForge.net Marketplace.
It's the best place to buy or sell services for
just about anything Open Source.
http://sourceforge.net/services/buy/index.php