Re: [bmp280 v1 2/6] iio: pressure: bmp280: reduce overhead on read with MODE_FORCED

[ASSI <Stromeko@nexgo.de>]

Jonathan Cameron writes:
> On Sun,  3 Aug 2025 16:07:58 +0200
> Achim Gratz <Achim.Gratz@Stromeko.DE> wrote:
>
>> When measuring with MODE_FORCED, each read through sysfs triggers a
>> new measurement cycle through aLL channels with the current channel
>> configuration, even though we can only access a single channel.
>> Reduce the incurred overhead (especially for higher oversampling_ratio
>> settings) by temporarily switching off the unused channels.  This
>> savea about a third of the acquisition time when reading all three
>> channels in succession:
>> 
>> | oversampling | max/full | max/skip | time/full | time/skip |
>> |              |     [ms] |     [ms] |      [ms] |      [ms] |
>> |--------------+----------+----------+-----------+-----------|
>> |           16 |      339 |      120 |       315 |       127 |
>> |            8 |      174 |       65 |       166 |        76 |
>> |            4 |       90 |       38 |        94 |        53 |
>> |            2 |       49 |       24 |        59 |        41 |
>> |            1 |       28 |       17 |        36 |        33 |
>> 
>> The results are from an I²C connected sensor at 400kHz, so there is
>> considerable overhead from the changing the channel configuration,
>> most noticeably with low oversampling_ratio values.  Faster
>> communication will reduce this overhead further; and since there is
>> still a net reduction in acquisition time even for
>> oversampling_ratio=1 switching off the channels is always done.
>> 
>> Note: The IIR filters will process a slightly noisier input signal.
>> 
>> Signed-off-by: Achim Gratz <Achim.Gratz@Stromeko.DE>
> Hi Achim.
>
> I'm not really sure what the algorithm implemented here is and what the
> various local variables actually mean as state.  Please add some
> more comments to the code.

Based on yours and Andys comment I'm currently inclined to abandon this
part of the series and just let anybody still using the MODE_FORCED
operation suffer from the overly long acquisition cycles when using the
sysfs interface.

> Maybe worth factoring out he contents of the switch case that has goto restore
> into a separate function so that the scope of the restore is same as the code
> that is calling it.  (i.e. avoid a goto jumping out of a switch.

I'll think about that when I have decided whether to keep that
feature in the code.


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+