Hi everyone,
Fons, thanks to your essay "Using a DLL to filter time" [1], that you
mentionned
in a discussion we had in August, I've put together a small library, named
Pendule, for accurate timing within a real time context. It's a tiny piece of
code, but offers a very easy to use replacement for gettimeofday().
Plus, it includes some benchmarking/measuring tools, that produce that sort of
fancy graphics (this one using a PCI hda intel card):
http://www.samalyse.com/code/pendule/shortgraph.png
Usage is very simple. first create an instance:
#include <pendule.h>
Pendule *pendule = pendule_new(buffer_size / sample_rate, bandwidth);
Then, in the realtime thread, update the loop every process cycle:
pendule_cycle(pendule);
And use pendule_gettime() instead of gettimeofday():
double current_time_in_seconds = pendule_gettime(pendule);
As expected, looking at the graph above, the obtained time doesn't drift as the
audio time, and has a lower jitter than the system time. Well, at least, it
works for me.
However, although my measures are pretty encouraging, I am not 100% sure of my
DLL implementation. Could you please review it ? It's there:
http://svn.samalyse.com/pendule/trunk/src/pendule.c
hi
instead of gettimeofday() you could use
clock_gettime(CLOCK_MONOTONIC,..) where its available. Its a posix
function and it has the benefit of beeing monotonic, where
gettimeofday() is imho useless if you run ntp.
also in _new() you should probably use calloc, or make sure you init all
fields.
Stefan
You can grab everything using svn:
svn co
http://svn.samalyse.com/pendule/trunk pendule
I'd like to know how it works for others. You can easily test your
hardware/system and make graphs with:
./waf configure
./waf build
./measure
./graph
Check the README for more.
[1]
http://www.kokkinizita.net/papers/usingdll.pdf
Best regards,