Closed tehn closed 4 years ago
Just finished long stress soak at 75% sgynth load, 53% matron load (best guess 1000 messages per sec). 0 xruns and no increase of memory usage over a 4.5 hour run at 64fpp/2ppb. The test exercises metros, front-panel controls, cairo, OSC send. Several terminals open to norns over wifi during test (emacs session, top etc).
I think the remaining small 'leaks' detected by valgrind after https://github.com/monome/norns/pull/774 are things which are allocated only once and never freed, so only a problem in so much as they obfuscate valgrind results.
Anyone think an object pool will definitely give measurably better performance? Personally wouldn't like to bet either way!
Anyone think an object pool will definitely give measurably better performance? Personally wouldn't like to bet either way.
i'm willing to make only a very small bet, and the key word is "measurably." i just want to make the change b/c malloc in a realtime loop is decidedly gross.