[lm-sensors] [PATCH 0/3] hwmon: More LM63 enhancements

[lm-sensors] [PATCH 0/3] hwmon: More LM63 enhancements

[Guenter Roeck]

Add support for unsigned temperatures on LM96163
Add support for writing remote T_CRIT limit
Add support for writing conversion rate register

Question:
  Calculations are currently along the line of "/ 125 * 32", causing loss of accuracy.
  Ok to change to "* 32 / 125" ?

v2 [unsigned temp patch]: Improved rounding for TEMP11U_TO_REG

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] [PATCH 0/3] hwmon: More LM63 enhancements

[Jean Delvare]

Hi Guenter,

On Tue, 22 Nov 2011 12:39:57 -0800, Guenter Roeck wrote:
> Add support for unsigned temperatures on LM96163
> Add support for writing remote T_CRIT limit
> Add support for writing conversion rate register
> 
> Question:
>   Calculations are currently along the line of "/ 125 * 32", causing loss of accuracy.
>   Ok to change to "* 32 / 125" ?

I think this is on purpose. Multiplying last guarantees that we don't
try to set bits which do not exist to 1. As long as proper rounding is
done before dividing, I don't quite see how we could be losing
accuracy. Do you have an actual example?

-- 
Jean Delvare

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] [PATCH 0/3] hwmon: More LM63 enhancements

[Guenter Roeck]

On Wed, Nov 23, 2011 at 03:34:29PM -0500, Jean Delvare wrote:
> Hi Guenter,
> 
> On Tue, 22 Nov 2011 12:39:57 -0800, Guenter Roeck wrote:
> > Add support for unsigned temperatures on LM96163
> > Add support for writing remote T_CRIT limit
> > Add support for writing conversion rate register
> > 
> > Question:
> >   Calculations are currently along the line of "/ 125 * 32", causing loss of accuracy.
> >   Ok to change to "* 32 / 125" ?
> 
> I think this is on purpose. Multiplying last guarantees that we don't
> try to set bits which do not exist to 1. As long as proper rounding is
> done before dividing, I don't quite see how we could be losing
> accuracy. Do you have an actual example?
> 
You are right. I didn't realize that "/ 125" reduces accuracy to 1/8 degree C, which is exactly
what the chips support, and that "* 32" simply shifts the value left by 5 bit to match the 11-bit
registers.

Sorry for the noise.

Guenter

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors