[mythtv] mythtv-vid branch needs testing
dburr at veritel.com.au
Fri Aug 10 12:26:30 UTC 2007
On Thu, 2007-08-09 at 22:57 +0100, Andrew Lyon wrote:
> On 8/9/07, Matt <skd5aner at gmail.com> wrote:
> On 8/9/07, Mark Kendall <mark.kendall at gmail.com> wrote:
> > At some point in the next few months, I'll finish the code I
> > for bicubic filtering which will be the first step towards
> > 'upscaling'. This is likely to require sifnificant gpu
> > though.
> WHOO HOO!
> Is that multi threaded? I notice the core 2 quad 2.4ghz is good value,
> i can afford to upgrade my c2d 2.4 to that at least.. I want to anyway
> for when ffmpeg has multi threaded h.264 mbaff.
I don't believe that upscaling/filtering will parallelize very well because processing each line/block of output data requires accessing data from n lines above/below. This is in contrast to decoding where each line/block can be processed in parallel because they are independent.
A hack could be to start one core processing from the beginning of the frame and the other from the end and let them meet up in the middle.
However, I believe this is all moot because Mark mentioned "GPU horsepower" implying that the filtering will be performed by the video card. Have you considered using NVidia's Cg (http://developer.nvidia.com/page/cg_main.html) to implement it in vertex shaders? I realise that it wouldn't be portable to VIA and Intel chips but it would seriously rock.
More information about the mythtv-dev