El 5 de octubre de 2011 18:28, Alexis Berlemont
<[email protected]>escribió:

> 2011/10/5 Fernando Herrero Carrón <[email protected]>:
> >
> >
> > 2011/10/5 Gilles Chanteperdrix <[email protected]>
> >>
> >> On 10/05/2011 09:48 AM, Julien Delange wrote:
> >> > On Tue, Oct 4, 2011 at 12:41 PM, Gilles Chanteperdrix
> >> > <[email protected]> wrote:
> >> >> Another mail in that thread seem to suggest that it could be long.
> Have
> >> >> you tried timing the various parts of your loop to see where the time
> >> >> is
> >> >> spent?
> >> >
> >> > No, sorry, my bad. Unfortunately, I cannot make more tests at this
> >> > time because I don't have the hardware.
> >> > I will do additional investigation soon, I would have access to the
> >> > acquisition card soon. My intent was to know if there was an obvious
> >> > error either in my code or in my kernel configuration. Apparently not,
> >> > so, I will try to figure out the root causes of these issues.
> >>
> >> It is my understaning when reading Fernando's answer to your mail that
> >> you are not using the proper interface for what you are trying to do.
> >>
> >
> > Yes, my suggestion is to use a different interface that will definitely
> > provide higher acquisition speeds. However I am wondering if using the
> > "instruction" interface the delay would be over the milliseconds. I'm
> > recalling some tests that I did using comedi in linux user-space, that is
> > without any real-time support software. I could acquire 8 channels and
> the
> > board's internal timer using the instruction list technique at 120us
> > (microseconds) intervals. However my board has 8 parallel ADCs (NI 6143),
> so
> > check if yours has parallel or serial converters. If it has to serialize
> > conversions then the overall acquisition period will increase.
> >
> > Back to your code, are you sure about  "insn_tab[i].data_size     = 8;"?
> Try
> > data_size = 1 instead.
> >
>
> The field data_size is supposed to hold the size of the acquisition in
> bytes.
>
> If you have an 8bits converter, 1 is OK; but according to the driver
> Julien contributed, the converter width is 16bits.
>

Oh, I see, thanks for pointing this out. On the code I had working with
comedi I used 1 for subdevices providing "sampl_t" samples (2 bytes) and 2
for subdevices providing "lsampl_t" samples (4 bytes) (timer/counter
subdevices, for instance). I really think that rationale is misleading, and
find working in terms of bytes much saner! In any case, using 8 as Julien
does seems to not be correct.

Best regards,
Fernando
_______________________________________________
Xenomai-help mailing list
[email protected]
https://mail.gna.org/listinfo/xenomai-help

Reply via email to