From: Steinar H. G. <sgu...@bi...> - 2012-11-15 23:38:59
|
On Thu, Nov 15, 2012 at 03:31:28PM -0800, Dan Dennedy wrote: > Now that I just made a new MLT release, I will take a serious look at > this and the other glsl contribution to MLT. I do not plan to use > movit's filter chain (the other glsl contribution reused MLT's), and I > plan to borrow and discard liberally, in general. IOW I do not plan to > integrate movit as independent lib/project. Fair enough. Fitting into MLT's filter chain was not a goal for Movit; rather, I wanted a “clean break” to see what could be done. I'm actually rather glad I did it, since compiling the shaders together instead of bouncing back and forth through memory seems to give quite sizable performance gains (less memory bandwidth, more optimizations available for the GLSL compiler). I do believe integrating Movit into MLT would be possible, and would have some advantages over the GLSL/MLT code that I've seen so far, but it would require MLT recognizing two or more Movit filters in an MLT chain and compiling them together (anything else would probably give sub-par performance). I also realize that this requires quite a big effort on the MLT side, though, and I can't really see anyone stepping up to do it if you don't want to. In any case, if the design can give you the inspiration you need for a good GLSL implementation in MLT, I guess that's already a success. :-) > I will also take a closer look at adopting gtest as that looks interesting. It was pretty much a no-brainer for me, since I already use it almost every day at work. It's relatively hard to unit test graphics software, but it was certainly worth it in my case. /* Steinar */ -- Homepage: http://www.sesse.net/ |