Re: [PATCH v6 05/13] iio: afe: rescale: add INT_PLUS_{MICRO,NANO} support

["Liam Beguin" <liambeguin@gmail.com>]

On Wed Jul 28, 2021 at 3:19 AM EDT, Peter Rosin wrote:
> On 2021-07-28 02:21, Liam Beguin wrote:
> > On Fri Jul 23, 2021 at 5:16 PM EDT, Peter Rosin wrote:
> >> On 2021-07-21 05:06, Liam Beguin wrote:
> >>> From: Liam Beguin <lvb@xiphos.com>
> >>>
> >>> Some ADCs use IIO_VAL_INT_PLUS_{NANO,MICRO} scale types.
> >>> Add support for these to allow using the iio-rescaler with them.
> >>>
> >>> Signed-off-by: Liam Beguin <lvb@xiphos.com>
> >>> ---
> >>>  drivers/iio/afe/iio-rescale.c | 14 ++++++++++++++
> >>>  1 file changed, 14 insertions(+)
> >>>
> >>> diff --git a/drivers/iio/afe/iio-rescale.c b/drivers/iio/afe/iio-rescale.c
> >>> index d0669fd8eac5..2b73047365cc 100644
> >>> --- a/drivers/iio/afe/iio-rescale.c
> >>> +++ b/drivers/iio/afe/iio-rescale.c
> >>> @@ -41,6 +41,20 @@ int rescale_process_scale(struct rescale *rescale, int scale_type,
> >>>  		do_div(tmp, 1000000000LL);
> >>>  		*val = tmp;
> >>>  		return scale_type;
> >>> +	case IIO_VAL_INT_PLUS_NANO:
> >>> +		tmp = ((s64)*val * 1000000000LL + *val2) * rescale->numerator;
> >>> +		tmp = div_s64(tmp, rescale->denominator);
> >>> +
> >>> +		*val = div_s64(tmp, 1000000000LL);
> >>> +		*val2 = tmp - *val * 1000000000LL;
> >>> +		return scale_type;
> > 
> > Hi Peter,
> > 
> >>
> >> Hi!
> >>
> >> My objection from v5 still stands. Did you forget or did you simply send
> >> the
> >> wrong patch?
> > 
> > Apologies, again I didn't mean to make it seem like I ignored your comments.
> > I tried your suggestion, but had issues when *val2 would overflow into
> > the integer part.

Hi Peter,

>
> Not saying anything about it not working does indeed make it seem like
> you
> ignored it :-) Or did I just miss where you said this? Anyway, no
> problem,
> it can be a mess dealing with a string of commits when there are
> numerous
> things to take care of between each iteration. And it's very easy to
> burn
> out and just back away. Please don't do that!

It was my mistake. Thanks for the encouragement :-)

>
> > Even though what I has was more prone to integer overflow with the first
> > multiplication, I thought it was still a valid solution as it passed the
> > tests.
>
> I did state that you'd need to add overflow handling from the fraction
> calculation and handling for negative values, so it was no surprise that
> my original sketchy suggestion didn't work as-is.
>
> > 
> >>
> >> Untested suggestion, this time handling negative values and
> >> canonicalizing any
> >> overflow from the fraction calculation.
> >>
> >> neg = *val < 0 || *val2 < 0;
> >> tmp = (s64)abs(*val) * rescale->numerator;
> >> rem = do_div(tmp, rescale->denominator);
> >> *val = tmp;
> >> tmp = rem * 1000000000LL + (s64)abs(*val2) * rescale->numerator;
> >> do_div(tmp, rescale->denominator);
> >> *val2 = do_div(tmp, 1000000000LL);
> >> *val += tmp;
> >> if (neg) {
> >> if (*val < 0)
> >> *val = -*val;
> >> else
> >> *val2 = -*val;
>
> This last line should of course be *val2 = -*val2;
> Sorry.
>
> > 
> > I'll look into this suggestion.
>
> Thanks!
>

Starting from what you suggested, here's what I came up with.
I also added a few test cases to cover corner cases.

	if (scale_type == IIO_VAL_INT_PLUS_NANO)
		mult = 1000000000LL;
	else
		mult = 1000000LL;
	/*
	 * For IIO_VAL_INT_PLUS_{MICRO,NANO} scale types if *val OR
	 * *val2 is negative the schan scale is negative
	 */
	neg = *val < 0 || *val2 < 0;

	tmp = (s64)abs(*val) * (s32)abs(rescale->numerator);
	*val = div_s64_rem(tmp, (s32)abs(rescale->denominator), &rem);

	tmp = (s64)rem * mult +
		(s64)abs(*val2) * (s32)abs(rescale->numerator);
	tmp = div_s64(tmp, (s32)abs(rescale->denominator));

	*val += div_s64_rem(tmp, mult, val2);

	/*
	 * If the schan scale or only one of the rescaler elements is
	 * negative, the combined scale is negative.
	 */
	if (neg || ((rescale->numerator < 0) ^ (rescale->denominator < 0)))
		*val = -*val;

	return scale_type;
> > 
> >> }
> >>
> >>> +	case IIO_VAL_INT_PLUS_MICRO:
> >>> +		tmp = ((s64)*val * 1000000LL + *val2) * rescale->numerator;
> >>> +		tmp = div_s64(tmp, rescale->denominator);
> >>> +
> >>> +		*val = div_s64(tmp, 1000000);
> >>
> >> Why do you not have the LL suffix here?
> > 
> > Doesnt' LL make it into a 64 bit integer?
> > I left it out because the second parameter of div_s64() should be s32.
>
> It just looked really odd with 1000000000LL for all instances, but then
> 1000000LL only for some. The lack of symmetry bothered me.
>
> To me, it seems as if we either need to support old/small crap with
> int being 16-bit, or we don't. If we don't need support for 16-bit,
> then we don't need any LL suffix, since 1000000000 fits just fine in
> 32-bit. If we do need 16-bit support, then we need LL (or something)
> all over since neither 1000000 nor 1000000000 fit in 16-bit.
>
> I think the compiler looks at the value of the constant and not the
> size of its type when selecting how big values the mul/add/whatever
> needs handle. So, adding LL feels like the safe option. Further, I
> guesstimate that the runtime cost of adding LL is zero and that the
> compile time cost is negligible.

Thanks for the explanation, I thought it might matter but I agree that
the asymmetry looks odd. I'll fix it.

Thanks,
Liam

>
> But maybe I'm missing something?
>
> Cheers,
> Peter
>
> > 
> > Thanks,
> > Liam
> > 
> >>
> >> Cheers,
> >> Peter
> >>
> >>> +		*val2 = tmp - *val * 1000000;
> >>> +		return scale_type;
> >>>  	default:
> >>>  		return -EOPNOTSUPP;
> >>>  	}
> >>>
> >