5 Nov
2010
5 Nov
'10
4:54 p.m.
On Friday 05 November 2010 13:53:56 Darrin Thompson wrote:
Has anyone managed to get good results running
valgrind vs. jackd or
apps for it?
My experience with Valgrind at work is you run it and it magically
tells you the exact line of code where you have a serious bug, which
is a good thing. But it has really high overhead. So I've been
considering running it against say, Rosegarden, which I crashed
repeatedly, but I'll beg off if someone has already been there and
done that.
I did manage to run jackd (with the firewire backend) inside valgrind --
tool=callgrind. Its not suitable to get any audio output but it gives some
numbers and showed me the main processing time is spent in the locks for
phase-locked-loop of the streaming-clock... Might be biased because the
streaming clocks of pc and device where running completely different because of
the valgrind overhead. But experiments without valgrind and without the locks
gave reason that really a lot of time is spent waiting there. Unfortunately
the locks are needed as it will crash without them...
But its possible and does give some numbers and hints. The overhead of the
memcheck tool is lower then the overhead of callgrind btw.
Have fun,
Arnold