Thank you for the valuable information. I attach the result of heaptrack done with the system already started for 1-2 hours and with the system just rebooted (...cin.1185) [https://www.dropbox.com/s/kko33qk3gvnyni9/heaptrack.cin..tar.gz?dl=0]. I used a 28s video (h264) with 3 transitions. I simply started playback and when I reached the end of the track, I started reverse playback. I used Cache size=8192 and Preroll=2.0s. I attach an image of the heaptrack graph and the link to the files obtained from heaptrack. I'm not very good at interpreting them, except the attached graph which clearly shows the transitions problem. I have a PC with 32GB of Ram and a swap file of 4GB (I only need it to avoid problems with hibernation/suspension). - Do you recommend increasing my swap file? - From the heaptrack results of my test, do you see any memory issues? And if so, can they be fixed? - Do you have any suggestions on how to fluidify CinGG playback, especially with effects and transitions?
On Wednesday, May 4, 2022, Andrea paz via Cin <[email protected]> wrote:
Thank you for the valuable information. I attach the result of heaptrack done with the system already started for 1-2 hours and with the system just rebooted (...cin.1185) [https://www.dropbox.com/s/kko33qk3gvnyni9/heaptrack.cin..tar.gz?dl=0]. I used a 28s video (h264) with 3 transitions. I simply started playback and when I reached the end of the track, I started reverse playback. I used Cache size=8192 and Preroll=2.0s. I attach an image of the heaptrack graph and the link to the files obtained from heaptrack. I'm not very good at interpreting them, except the attached graph which clearly shows the transitions problem. I have a PC with 32GB of Ram and a swap file of 4GB (I only need it to avoid problems with hibernation/suspension).
- Do you recommend increasing my swap file? - From the heaptrack results of my test, do you see any memory issues? And if so, can they be fixed? - Do you have any suggestions on how to fluidify CinGG playback, especially with effects and transitions?
well, canonical suggestion is to use background rendering over selected region, I think? but in general you can try to add '-march=native' and other machine-specific cflags in adiition to -Ofast in cinelerra/Makefile (Ofast applied to our scaler functions). Also, time back Bill changed caching algo so it worked without crashing in 32-bit mode. May be we can conditionally revert those changes for 64-bit compile...
Interesting and quite helpful. Thanks.
While I have not tested the current state of the git repository, I thought I might add a few comments about testing performance in general from the perspective of a retired systems programmer. I am a fan of Cinelerra-GG and use it for burning bluray disks so I hope to be able to test the new code before long.
It will be interesting to hear your results of burning bluray disks -- especially if you have good speakers where you might be able to differentiate audio options of LPCM and tsMuxer results as I could not.
... Any settings within a given application, eg cache or pre-roll, will need to be measured in a controlled environment which means rebooting the computer to clear the file system file cache. Otherwise, any request for data that has been previously read may be satisfied from the operating systems file cache regardless of the settings in the application.
Measuring this in a controlled environment is very useful to a programmer to improve the code if possible. The problem is for a user, there is no such thing as a controlled environment and just how they "feel" is all that matters. So the best that can be done is for programmers to take advantage of the methods you outline. I do think that the "preroll" of 1.0 seconds as opposed to 0.5 seconds seems more consistent for me as a user -- it made me "feel" better!
... It goes without saying that any use of swap space will radically degrade performance unless that swap space is on a solid state drive and the swap partition is quite generous in size.
I gave up on swap space years ago -- if you have to swap, you may as well reboot.
participants (4)
-
Andrea paz -
Andrew Randrianasulu -
Phyllis Smith -
Richard Nolde