[FFmpeg-devel] [RFC] support half-precision float in *_cuda filters

[Timo Rothenpieler]

On 22.10.2020 05:43, leozhang wrote:
> Half precision float is supported for those devices  sm capacity
>> =5.3. Theoretically, half float can achieve 2x performance speed up.
> So is there any plan to support half float in FFmpeg? Or any problem
> so that half float can't be met up?

I'm not aware of anyone planning to implement that, but feel free to 
send patches.
Performance improvements that do not impact quality are always welcome.