cflags I tried some time ago (on 32 bit build but with avx-enabled amd cpu fx-4300, with gcc 5.5.0)
https://github.com/Randrianasulu/CinelerraGG-slackbuild/blob/master/cinelerra-goodguy.SlackBuild "-O3 -ffast-math -mavx -march=native -mtune=native -fopenmp" not much use for openmp, but if you want some experiments you can try to put some #pragma omp simd" around hot source code blocks and see if it helps... On Monday, January 17, 2022, Andrew Randrianasulu <[email protected]> wrote: > > > On Monday, January 17, 2022, Mark Goldberg via Cin < > [email protected]> wrote: > >> >> On Mon, Jan 17, 2022, 3:32 AM mat <[email protected]> wrote: >> >>> >>> >>> @Mark, can you confirm vdpau works on your system, outside of CinGG? >>> Does vdpauinfo say it is OK? >>> Does "HW device" vdpau work in the compiled version without errors, >>> apart from the crash? >>> >> >> On RHEL8 the appimage for older distros is required as glibc is older. It >> does not work with vdpau, falling back to software. In the locally compiled >> version it does use vdpau, but I have 4k60 f-log videos, and adding the >> lut3d plugin slows things down such that it cannot play in real time. >> > > > you can try to hack makefiles of individual plugins (while lut3d is > ffmpeg's one, so not sure how to influence irs compile flags individually) > with > -march=native, -mavx (if you have avx instructions in your cpu), and > other suitable flags... CinGG for now mostly cpu/memory intensive, not > using OpenCL/cuda (unlike big commercial editors?) so setting correct cpu > flags may give relatively big boost. > > > >> @Andrew, I have compiled multiple times and changed versions of the >> Nvidia driver, same crash. >> > > from forum post I see case when gpu decoder utilization (as seen via > nvidia's utility) still not 0%, so it works, but probably pulling and > pushing frames between vram and ram via pci-e bus takes cpu, too (even if > dma should do it for free?). I remember long time ago open-source radeon > driver developers run into some underutilization of bus so they applied big > expensive logic analyzer (at pcie speeds!) for finding some bottlenecks.... > so I am afraid without much better developers poking graphics stack while > Cingg is running not much can be done.. :( > > @Andrea, for some reason I can't download attaches from forum (because I > have no login there?), can you add CMS.txt as attachment to email (again)? > > I recall older version of Final Cut (before X) has ColorSync plugin > (probably utilizing cpu at early 2000s), so may be making lcms2 plugin will > provide some workaround for lack of core color management? (like, you > switch slow colorcorrection plugin/effect for color work, turn it off for > edits... those transforms using monitor's profile tend to be math heavy if > executed on cpu only, even with simd... ) > >> >> @Andrea, I have seen the forum posting. There are issues with plugins >> slowing things down, but I will start a second thread for that. >> >> Latest compile is the multibit version. It's more complicated. If there >> is another video track on the timeline below what is playing, it does not >> crash and plays successfully. >> > > sorry, I can't test this due to lack of access to my main desktop... and > laptops here only have vaapi (may be I can recompile x86_32 version of > cingg... those are not *my* laptops, but I have live usb stick with > slackware /copy of my home system from ~2020) > >> >> Regards, >> >> Mark >> >>>
-- Cin mailing list [email protected] https://lists.cinelerra-gg.org/mailman/listinfo/cin
