Re: [PATCH v1 1/1] iio: common: scmi_sensors: Replace const_ilog2() with ilog2()

[David Laight <david.laight.linux@gmail.com>]

On Fri, 31 Oct 2025 11:54:30 +0200
Andy Shevchenko <andriy.shevchenko@linux.intel.com> wrote:

> On Fri, Oct 31, 2025 at 09:43:36AM +0000, David Laight wrote:
> > On Fri, 31 Oct 2025 08:45:00 +0100
> > Andy Shevchenko <andriy.shevchenko@linux.intel.com> wrote:  
> 
> ...
> 
> > >  		tstamp_scale = sensor->sensor_info->tstamp_scale +
> > > -			       const_ilog2(NSEC_PER_SEC) / const_ilog2(10);
> > > +			       ilog2(NSEC_PER_SEC) / ilog2(10);  
> > 
> > Is that just a strange way of writing 9 ?  
> 
> Why? It's correct way of writing log¹⁰(NSEC_PER_SEC), the problem here is that
> "i" people do not think about :-)

Even without the "i" the division could easily give 8.999999.
So you'd be relying on rounding to get the required integral value.

> But we have intlog10(), I completely forgot about it.

And it isn't the function the code is looking for.
(The result is shifted left 24 and it doesn't have an optimisation
for constants.)

> 
> > Mathematically log2(x)/log2(10) is log10(x) - which would be 9.
> > The code does seem to be 'in luck' though.
> > NSEC_PER_SEC is 10^9 or 0x3b9aca00, so ilog2(NSEC_PER_SEC) is 29.
> > ilog2(10) is 3, and 29/3 is 9.
> > 
> > Do the same for 10^10 and you get 11.  
> 
> That code looks like working by luck entirely, TBH. I just took the scope of
> the patch to start dropping const_ilog2() usages.

Something always crawls out of the woodwork...

	David