[Selene]

lmyllari, can I just ask about an earlier reference you mentioned when posting your solutions for grey-blacks and colours etc.

You had options 1 and 2 for those either using GPU or software rendering, then you had another option for better quality.

Is the third option far better across the board or just in certain scenarios? I'd prefer to use the GPU of my haswell i5 if possible, as that was kind of the whole point, but if I'm going to get a better (noticeably) result with another solution then I'd be interested in that also.

The third option just gets rid of some banding, most visible in smooth gradients. The difference in quality is not that big - most software players have this problem and very few people notice. If you're happy with the quality, and prefer hardware decoding, stay with that for now.

I know you're super-busy (being the driving-force behind the majority of these fixes) but I'd be interested on your take on the various solutions on offer. Unless of course, you think there'll be a new build coming soon enough with these fixes rolled in so those less experienced individuals (like me) can simply install and set a few settings.

Let me note that I've just analysed issues and filed bugs. Intel devs have fixed the issues, and fritsch is bringing the fixes to everyone.

For now, the patches going in just ensure proper colors and black levels.

To fix the banding issue, there are two ways: either limited range RGB all the way from colorspace conversion to display, or dithering in colorspace conversion. For the first, I already have a hack that works with software decoding (this is the third option referenced above). The second should be pretty easy with software decoding too. Unfortunately both are a little more difficult with hardware decoding.

I'll try to make an upstream acceptable version of my current hack. It will take some time though. If there is developer interest in that enough to get it done properly and merged, I will continue to implement the same for hardware decoding.

My goal is to have best possible quality to users with no special settings or just a single switch in xbmc.

---

I could be wrong, but for those asking, I think the fix will show up in Intel's "staging" branch first: http://cgit.freedesktop.org/vaapi/intel-...?h=staging

The fix is for kernel, not VA-API. I think http://cgit.freedesktop.org/~danvet/drm-intel/ is the repository to watch.

The patch is attached to bugzilla and could be pulled into OE from there.

https://github.com/OpenELEC/OpenELEC.tv/pull/2776