21 Apr
2005
21 Apr
'05
3:12 a.m.
Hi all,
I need to know if gettimeofday (or one of the other standard time
functions) is okay to use in a realtime jack thread. I can't find any
information on the net about whether it's deterministic or not (or even
how expensive it is).
Problem is I have an event-based system that pushes events (from
OSC/Midi) into a dequeue, which the audio thread runs through to
execute. If you flood with events, too many need to be processed per
block, the audio thread takes too long and you get dropouts.
I'd like to be able to only process (non timestamped) events for x% of
the time available to prevent this from happening; so I need to be able
to determine how much time has been eaten by event processing so far.
Is calling gettimeofday this many times in the realtime thread a wise
idea?
Cheers,
-DR-
21 Apr
21 Apr
3:41 a.m.
Dave Robillard <drobilla(a)connect.carleton.ca> writes:
I need to know if gettimeofday (or one of the other
standard time
functions) is okay to use in a realtime jack thread. I can't find any
information on the net about whether it's deterministic or not (or even
how expensive it is).
On Linux it does not wait (AFAIK). But, it does involve the overhead
a system call. Plus, there's always some danger when taking a trip
through the system call exit path that the scheduler will screw things
up and decide to run some other thread.
The JACK ALSA backend calls it once per cycle, but we use the cycle
counter (on platforms that support it) to avoid paying this cost any
more than that.
Problem is I have an event-based system that pushes
events (from
OSC/Midi) into a dequeue, which the audio thread runs through to
execute. If you flood with events, too many need to be processed per
block, the audio thread takes too long and you get dropouts.
I'd like to be able to only process (non timestamped) events for x% of
the time available to prevent this from happening; so I need to be able
to determine how much time has been eaten by event processing so far.
Is calling gettimeofday this many times in the realtime thread a wise
idea?
How often? Once per MIDI event? That should be OK, I suppose.
--
joq
7:25 a.m.
On Thursday 21 April 2005 07.45, Jack O'Quin wrote:
[...]
I did a quick hack for testing some custom hardware on an embedded
system, where I just polled gettimeofday() to get some code to run at
10 kHz. (Didn't care about the occasional stalls - which were very
few anyway, as this was the only process actually doing anything at
the time.) That is, a bunch of gettimeofday() calls for each loop,
times 10 kHz. Worked just fine.
This was on a 300 MHz low power Pentium clone, which has about 10-15%
of the power of a 1 GHz P-III CPU. I'm not even sure if Linux is
using RDTSC for gettimeofday() on this thing...
//David Olofson - Programmer, Composer, Open Source Advocate
.- Audiality -----------------------------------------------.
| Free/Open Source audio engine for games and multimedia. |
| MIDI, modular synthesis, real time effects, scripting,... |
`-----------------------------------> http://audiality.org -'
--- http://olofson.net --- http://www.reologica.se ---
Is calling
gettimeofday this many times in the realtime thread a
wise idea?
How often? Once per MIDI event? That should be OK, I suppose. 
12:28 p.m.
On Thu, 2005-04-21 at 11:28 +0200, David Olofson wrote:
On Thursday 21 April 2005 07.45, Jack O'Quin
wrote:
[...]
I did a quick hack for testing some custom hardware on an embedded
system, where I just polled gettimeofday() to get some code to run at
10 kHz. (Didn't care about the occasional stalls - which were very
few anyway, as this was the only process actually doing anything at
the time.) That is, a bunch of gettimeofday() calls for each loop,
times 10 kHz. Worked just fine.
This was on a 300 MHz low power Pentium clone, which has about 10-15%
of the power of a 1 GHz P-III CPU. I'm not even sure if Linux is
using RDTSC for gettimeofday() on this thing...
Yes, Linux implements gettimeofday via rdtsc() on x86. I think it may
use the ACPI PM timer under some conditions also. Which is why if you
can get away with just using rdtsc directly for timing it's better than
gettimeofday.
Lee
Is
calling gettimeofday this many times in the realtime thread a
wise idea?
How often? Once per MIDI event? That should be OK, I suppose.
22 Apr
22 Apr
9:03 a.m.
On Thursday 21 April 2005 16.28, Lee Revell wrote:
[...]
Yes, Linux implements gettimeofday via rdtsc() on x86.
...except on some x86 clones that don't have a working RDTSC
implementation. Generally, all Pentium compatibles and higher should
have it, but some don't.
Also, RDTSC is essentially useless on some early systems with core
clock throttling. Finally, RDTSC sometimes means trouble on SMP
systems, where the TSCs are not guaranteed to be synchronized. AFAIK,
both of these issues have been dealt with in current generation CPUs
and chipsets, but it might be a good idea to check...
Which is why if you
can get away with just using rdtsc directly for timing it's better
than gettimeofday.
Yeah. It's probably ok to just assume it's there if you're dealing
with P3 generation intels and alternatives, but for maximum x86
portability (non-intel CPUs and pre-Pentium CPUs), and portability
beyond x86, one needs a fallback.
//David Olofson - Programmer, Composer, Open Source Advocate
.- Audiality -----------------------------------------------.
| Free/Open Source audio engine for games and multimedia. |
| MIDI, modular synthesis, real time effects, scripting,... |
`-----------------------------------> http://audiality.org -'
--- http://olofson.net --- http://www.reologica.se ---
7286
days inactive
7287
days old
linux-audio-dev@lists.linuxaudio.org
4 comments
4 participants
participants (4)
-
Dave Robillard -
David Olofson -
Jack O'Quin -
Lee Revell