Performance: Improve rendering speed (by GPU)
Summary
Also, maybe we should benchmark versions of kdenlive built on different systems. It's possible that the improvements in compilers give us a significant performance boost. If this is the case, maybe we could consider giving an "old system version" aimed at running in a maximal number of systems, and a "performance optimized version" which would work only on really recent systems.
STEPS TO REPRODUCE
- To start some version comparison, I just uploaded appimages of several MLT versions here:
https://files.kde.org/kdenlive/unstable/mlt-appimage-compare/
This will/could be useful to check for regressions in rendering performance. Note that since MLT 0.8.4 used Qt4, the Qt modules (titler, qimage, qtblend transitions) are not included in this one. You can download the appimages and use them in a terminal to render a project file like the normal command line melt.
So if we can get a sample project with source clips, we could test rendering performance across these MLT versions.
- Add 'transform' effect, with some keyframes
OBSERVED RESULT
Render times: a 14mn video (3 tracks, few clips) took 1h14 to render. Switching from qtblend to composite track composition didn't help much, I stopped render after 30mn (around half file length).
ADDITIONAL INFORMATION
I solved my own problem: actually due to a bad laptop power supply the CPU was in powersave mode at 800MHz instead of 3.5GHz. Adding 'processor.ignore_ppc=1' to kernel boot options brings back x4 performance!
Not very scientifically, I switched at the same time from x264 to webm(vp8) with max encoding speed, using 6 cores, and rendered the video in about 10mn