[FFmpeg-devel] Testing Inexact Data

[Mike Melanson]

Rich Felker wrote:
> The 99.99 is probably a dummy number since the actual PSNR should be
> infinity. No idea why it would be dummied out since infinity
> computations with floating point numbers work just fine... oh,
> tiny_psnr is all arcane fixed point code.. :) In that case, it'll take
> a little bit of reading to interpret the numbers, I think. PSNR is
> normally expressed in decibels but I think this is using some base-2
> or base-16 thing...?

Well, there is this note regarding infinity in the tiny_psnr.c source:

  psnr= 100*F-1; //floating point free infinity :)

Maybe that explains it.

-- 
	-Mike Melanson