[FFmpeg-devel] [PATCH] Fast half-float to float conversion

Jimmy Christensen jimmy
Wed Jul 1 14:49:44 CEST 2009

On 2009-07-01 13:30, Michael Niedermayer wrote:
> On Wed, Jul 01, 2009 at 01:16:51PM +0200, Jimmy Christensen wrote:
>> On 2009-07-01 13:04, Michael Niedermayer wrote:
>>> On Wed, Jul 01, 2009 at 07:12:53AM +0200, Jimmy Christensen wrote:
>>> [...]
>>> what seems to be missed in this thread is that RGB in ffmpeg and pretty
>>> much
>>> everywhere else is not linear in terms of number of photons vs. value, it
>>> rather depends on the gamma value amongth other things which we did
>>> neglegt
>>> slightly but they can be stored in AVColorTransferCharacteristic.
>>> in that sense 16bit floats given sensible ordering of the sign,exponent
>>> and
>>> mantisse bits could be considered RGB48 with a funny
>>> AVColorTransferCharacteristic ...
>> Is that supported by eg. swscaler?
> no but patch welcome ...

The problem I see however is that then the approach would be different 
for 16-bit and 32-bit since there is no 32-bit per color frame format 
yet in ffmpeg.

More information about the ffmpeg-devel mailing list