Re: [PATCH v1 2/6] units: Add value of π * 10⁹

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

On Tue, 28 Oct 2025 10:08:04 +0200
Andy Shevchenko <andriy.shevchenko@linux.intel.com> wrote:

> On Mon, Oct 27, 2025 at 07:30:33PM +0000, David Laight wrote:
> > On Mon, 27 Oct 2025 15:34:51 +0100
> > Andy Shevchenko <andriy.shevchenko@linux.intel.com> wrote:
> >   
> > > There are a few drivers that want to have this value, and at least one
> > > known to come soon. Let's define a value for them.  
> 
> > > +/* Value of π * 10⁹ */

Thinks - should non-ascii characters be allowed in comments.
They can cause grief.

> > > +#define PI	3141592653LL  
> > 
> > Is that the right value?
> > IIRC the next digits are 58979 (I used to know the next few as well)
> > which means it should be rounded up.  
> 
> Right, today I have the same thought that actually ChromeOS driver has a
> off-by-one issue there.

Not as though it is likely to make a difference.
If any code needs anything more accurate than 3.1416 it probably cares
whether the value is rounded down or up.

> Btw, do you know if we can have compile-time divisions that can make 32-bit
> constants out of the 64-bit input? DIV_ROUND_CLOSEST_ULL() doesn't seem allow
> that.

Not sure, I did wonder how much thought had gone into the LL suffix.
The value used will fit in u32 - but then any maths becomes unsigned.
OTOH using LL forces 64bit maths on 32bit - not good.

I suspect it would be better to drop a digit so the value fits in a
signed 32bit variable.

The other issue is whether scaling by a power of 10 is even right.
If the code has to rescale the value using divides you really want
to divide by a big power of 2 not 10.
So multiplying by 2**29 may be more useful.

	David