27 Feb
2003
27 Feb
'03
11:48 a.m.
From the 2.5.63 changelog:
o POSIX clocks & timers
Phat :)
--
Bob Ham <rah(a)bash.sh>
27 Feb
27 Feb
1:04 p.m.
From the 2.5.63
changelog:
o POSIX clocks & timers
Phat :)
2:21 p.m.
On Thu, 2003-02-27 at 17:00, Paul Davis wrote:
i just hope that this change reflects the whole
enchilada
(reprogramming the APIC) rather than just the POSIX clock API. is
there any sign of this?
I believe so. From this thread in the high-res-timers archives:
http://sourceforge.net/mailarchive/message.php?msg_id=2707604
comes this snippet of a patch:
+/*
+ * Set the APIC timer for a one shot expiry in nanoseconds.
+ * This is called from the posix-timers code.
+ */
+int ns2clock;
+void set_APIC_timer(int ns)
+{
Wooha! :)
--
Bob Ham <rah(a)bash.sh>
28 Feb
28 Feb
10:58 a.m.
Paul Davis writes:
Super-Callifragilistic-Expealido-phat-cious!!
finally, we can do real MIDI sequencing with Linux:
while (1) {
...
usleep (usecs_till_next_event);
/* we're know its the right time,
instead of being rounded to
the default HZ
*/
deliver_events ();
}
Is this really new ? I was porting some C++ thread classes (originally
developed for Solaris) to 2.4.19 this week. The ITC mechanism uses
pthread_cond_timedwait(), which takes a struct timespec * referring to
an absolute time, the same you can get with clock_gettime(). I had the
impression this gave me almost microsecond resolution timing (but I did
not look closesly at the precision at the time).
--
Fons Adriaensen
ALCATEL SPACE
From the 2.5.63 changelog:
o POSIX clocks & timers
Phat :) 
11:34 a.m.
On Fri, Feb 28, 2003 at 03:49:59 +0100, Fons Adriaensen wrote:
Is this really new ? I was porting some C++ thread
classes (originally
developed for Solaris) to 2.4.19 this week. The ITC mechanism uses
pthread_cond_timedwait(), which takes a struct timespec * referring to
an absolute time, the same you can get with clock_gettime(). I had the
impression this gave me almost microsecond resolution timing (but I did
not look closesly at the precision at the time).
There is a bif difference between how my resolution you can specify and
how accurate it is.
- Steve
12:04 p.m.
Steve Harris writes:
On Fri, Feb 28, 2003 at 03:49:59 +0100, Fons
Adriaensen wrote:
I know... but when I ran the unit tests I didn't look to closely at
the accuracy, only to happy that it was working at all :-).
I'll very this next week.
--
FA
Is this really new ? I was porting some C++
thread classes (originally
developed for Solaris) to 2.4.19 this week. The ITC mechanism uses
pthread_cond_timedwait(), which takes a struct timespec * referring to
an absolute time, the same you can get with clock_gettime(). I had the
impression this gave me almost microsecond resolution timing (but I did
not look closesly at the precision at the time).
There is a bif difference between how my resolution you can specify and
how accurate it is.
3 Mar
3 Mar
12:39 p.m.
while
(1) {
...
usleep (usecs_till_next_event);
/* we're know its the right time,
instead of being rounded to
the default HZ
*/
deliver_events ();
}
Is this really new ? I was porting some C++ thread classes (originally
developed for Solaris) to 2.4.19 this week. The ITC mechanism uses
pthread_cond_timedwait(), which takes a struct timespec * referring to
an absolute time, the same you can get with clock_gettime(). I had the
impression this gave me almost microsecond resolution timing (but I did
not look closesly at the precision at the time).
4 Mar
4 Mar
5:03 a.m.
Paul Davis writes:
the kernel gives non-SCHED_FIFO/RR threads roughly
1/HZ
resolution.
Seems to be 20 ms on my unpatched standard SuSE 2.4.19.
I imagine this can be lowered by modifying the kernel config ?
SCHED_FIFO/RR threads can get better than that if the
delay is very small, but its implemented by spinning on the
processor.
Yes, in my first test I used something like 150 microseconds,
and the machine froze for 15 seconds (100000 iterations).
the new code will provide usec resolution without
spinning. it
reprograms the APIC timer to interrupt the CPU at the right time. i
believe that you will still need capabilities, euid=0 and/or
SCHED_FIFO/RR to use it fully.
No problem with that.
TNX for the info.
--
FA
7723
days inactive
7728
days old
linux-audio-dev@lists.linuxaudio.org
7 comments
4 participants
participants (4)
-
Bob Ham -
Fons Adriaensen -
Paul Davis -
Steve Harris