[FFmpeg-devel] [PATCH] Common code for Indeo interactive

Kostya kostya.shishkov
Mon Mar 23 07:05:50 CET 2009


On Sun, Mar 22, 2009 at 10:28:28PM +0100, Maxim wrote:
[...]
> My Question (Jason, Loren, Michael, others?): Is there any possibility
> to measure the precision of the slant transform's output mathematically?
> The input coeffs have 12bits precision. I'm afraid I'm not skilled
> enough to do that. Examine the outputs can take long time... Any suggestion?

I think you can just operate with domains, i.e.:
input - 12-bit

if we have operation A*X its range is increased by log2(X) bits
if we have operation (A + B) or (A - B) its range is increased by one bit
if we have operation A >> N its range is decreased by N bits

so if we have transform

 (4*A + 4*B) >> 5 its range will be  (12 + 2 + 1) - 5 = 10 bits
 
> Regards
> Maxim



More information about the ffmpeg-devel mailing list