[MPlayer-dev-eng] [PATCH] fix for -srate bug

[D Richard Felker III]

On Wed, Oct 20, 2004 at 12:01:07AM +0100, Ed Wildgoose wrote:
> 
> >>I have to be honest, I can't see any tearing on either 50Hz or 60Hz 
> >>stuff.
> >>   
> >>
> >
> >look closer, it's definitely there. there's no way to avoid it,
> >regardless of what your refresh rate is, without waiting for vsync to
> >flip the page. with overlay this is done in hardware. otherwise you
> >need a vo driver that can do it in software, and afaik none of them
> >can since it usually requires root/kernel privs.
> > 
> >
> 
> Hmm, curious.  I know exactly what you are describing because I used to 
> see it before I tweaked the video modes.  It was a very clear break in 
> the picture (like a faint line) moving slowly up or down the screen.  
> However, everything is absolutely rock solid here on both 50 and 60Hz. 

no, it's not. it means that every so often (maybe one out of 100
frames) you see one frame on the top portion of the screen, and the
previous frame still on the bottom potion of the screen. if these
differ a lot it will look _very_ bad!! try watching a high action
scene. it's probably more apparent in animation because there are
often fewer frames drawn with more difference between consecutive
frames.

> At some point I intend to buy a big plasma or LCD (waiting until they 
> have dvi inputs that work with PCs...).  Then I will be much more 
> interested in playing with output quality tweaks.

tearing isn't an output quality tweaking issue. it look equally bad
regardless of mode, sharpness, display device, etc.

> >no, total opposite. playing dvd on some output that only support
> >44100.
> 
> Oh, curious.  What is your output that it only supports 44100?  Seems to 
> be an uncommon situation?  (the soundcard "locked" at 48Khz seems to be 
> the most common complaint?)

soundcard with max clock of 44100 (sb awe32). actually there's at
least one new card limited to 44100 too, though (possibly even locked
at 44100). locking at 44100 makes a lot more sense than locking at
48000, since one normally cares about music sounding perfect but can
get by with a small error in movie audio.

> However, as Michael has demonstrated on another post, the libsamplerate 
> code is perhaps half the speed, but works at 32 bit instead of 16 bit 
> arithmetic (which seems like a reasonable trade to me).

no, it works with float arithmetic instead of 16 bit integer
arithmetic. this will be 10x slower on my box, or worse, and the
improvement is not audible. (psnr is not very meaningful for audio
since perception is everything there...) if you want higher precision,
use 32bit integer arithmetic, but it really won't sound any different!

> I would be interested to see if it wasn't possible to get the quality of 
> the lavc resampler up to that of libsamplerate

of course it's possible, without much work most likely. but again i
don't think it's useful. the error isn't in the frequency domain, it's
just rounding error in the output sample values.

rich