Add some statistics about buffer processing time and realtime things to make them somewhat meaningfull#50
Add some statistics about buffer processing time and realtime things to make them somewhat meaningfull#50
Conversation
|
Cool, thanks. Having better statistics would be nice, although it should be included in the usual output in a TSV-like way so it can be used to plot or otherwise work with the data. The intent of this tool is to output something easy to get into pandas or gnuplot or whatever. The option should probably be removed so the new columns are just always there. |
… not jumping during measurement - add code and options to enable realtime operation while running the plugins - add code and option to measure individual buffers - collect minimum, average and maximum buffer runtime (to detect candidate realtime failures in plugins)
- rename options to make room :)
…specifying the -p option
|
I rebased and more or less rewrote this on https://github.com/lv2/lilv/tree/rt-stats The last commit adds a lot of platform-specific dependencies that would need to be tested for. I might merge everything up until the last commit and leave the realtime/mlock/buffer timings stuff on a branch for now. |
|
Hi! Thanks for the feedback! :) I'll have to find some more time soon, to take a closer look again! |
|
Merged as 72b6c2c, thanks. Aside from various cleanups, I removed the stuff that emitted results for every buffer, since that produced an output file that's not really usable. It could be added back as a separate option easily enough, but I'll leave that for future work. |
|
Cool! Thanks! |
With these changes you can do now:
-m for mlockall, -p 90 for SCHED_FIFO at priority 90 and -s for summary statistics.
Example output (somewhat cleaned):
Here we can identify triple_chorus and/or gx_cstb_#_cstb to maybe do something fishy.
I'm willing to clean up the code as well as the PR more if there's a chance of merging at all ;)