> Am 05.05.2018 um 20:41 schrieb Rostislav Pehlivanov <atomnu...@gmail.com>: > > On 5 May 2018 at 18:40, Daniel Oberhoff <danieloberh...@googlemail.com> > wrote: > >> >>> >>> As a last note, you can find my GLSL(+Vulkan) implementations of the >>> algorithm changes described above, as well as all of the related >>> color-management code and various decision logic for what values to >>> infer/default here: >>> https://github.com/haasn/libplacebo/blob/master/src/shaders/colorspace.c >> >> Again I am wondering if there is interest of having gl and/or Vulkan >> filters in ffmpeg, along with direct transfer support where the >> source/target is on gpu, as with many of the hwaccel targets? >> >> That is what we do in our pipeline to get high throughput, and i‘d be >> happy in pushing some of that back into ffmpeg. >> >> Best, >> >> Daniel >> _______________________________________________ >> ffmpeg-devel mailing list >> ffmpeg-devel@ffmpeg.org >> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel >> > > There is already a Vulkan patchset alongside a bunch of filters. I posted > it 2 weeks ago. Yes, it does accept GLSL using libshaderc. Yes, it supports > mapping VAAPI and DRM frames. Yes, its much faster than opencl on all > systems I've tested. No, there will not be an opengl-based filtering > system, only vulkan.
That makes a lot of sense, gl is dying. We are just reluctant right now to switch (yet) for various reasons. Is there any chance for that to land? Can I have a look? :) Best _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
