RE: [PATCH] media: atomisp: Use clamp_t() in ia_css_eed1_8_vmem_encode()

[David Laight <David.Laight@ACULAB.COM>]

From: Mauro Carvalho Chehab
> Sent: 06 September 2024 07:16
> 
> Em Sat, 27 Jul 2024 14:51:56 +0200
> Christophe JAILLET <christophe.jaillet@wanadoo.fr> escreveu:
> 
> > Using clamp_t() instead of min_t(max_t()) is easier to read.
> >
> > It also reduces the size of the preprocessed files by ~ 193 ko.
> > (see [1] for a discussion about it)
> >
> > $ ls -l ia_css_eed1_8.host*.i
> >  4829993 27 juil. 14:36 ia_css_eed1_8.host.old.i
> >  4636649 27 juil. 14:42 ia_css_eed1_8.host.new.i
> >
> > [1]: https://lore.kernel.org/all/23bdb6fc8d884ceebeb6e8b8653b8cfe@AcuMS.aculab.com/
> >
> > Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
> > ---
> >  .../isp/kernels/eed1_8/ia_css_eed1_8.host.c   | 24 +++++++++----------
> >  1 file changed, 12 insertions(+), 12 deletions(-)
> >
> > diff --git a/drivers/staging/media/atomisp/pci/isp/kernels/eed1_8/ia_css_eed1_8.host.c
> b/drivers/staging/media/atomisp/pci/isp/kernels/eed1_8/ia_css_eed1_8.host.c
> > index e4fc90f88e24..96c13ebc4331 100644
> > --- a/drivers/staging/media/atomisp/pci/isp/kernels/eed1_8/ia_css_eed1_8.host.c
> > +++ b/drivers/staging/media/atomisp/pci/isp/kernels/eed1_8/ia_css_eed1_8.host.c
> > @@ -172,25 +172,25 @@ ia_css_eed1_8_vmem_encode(
> >  		base = shuffle_block * i;
> >
> >  		for (j = 0; j < IA_CSS_NUMBER_OF_DEW_ENHANCE_SEGMENTS; j++) {
> > -			to->e_dew_enh_x[0][base + j] = min_t(int, max_t(int,
> > -							     from->dew_enhance_seg_x[j], 0),
> > -							     8191);
> > -			to->e_dew_enh_y[0][base + j] = min_t(int, max_t(int,
> > -							     from->dew_enhance_seg_y[j], -8192),
> > -							     8191);
> > +			to->e_dew_enh_x[0][base + j] = clamp_t(int,
> > +							       from->dew_enhance_seg_x[j],
> > +							       0, 8191);
> > +			to->e_dew_enh_y[0][base + j] = clamp_t(int,
> > +							       from->dew_enhance_seg_y[j],
> > +							       -8192, 8191);
> 
> Such change introduces two warnings on smatch:
> 
> drivers/staging/media/atomisp/pci/isp/kernels/eed1_8/ia_css_eed1_8.host.c:
> drivers/staging/media/atomisp/pci/isp/kernels/eed1_8/ia_css_eed1_8.host.c:177
> ia_css_eed1_8_vmem_encode() warn: assigning (-8192) to unsigned variable 'to->e_dew_enh_y[0][base +
> j]'
> drivers/staging/media/atomisp/pci/isp/kernels/eed1_8/ia_css_eed1_8.host.c:
> drivers/staging/media/atomisp/pci/isp/kernels/eed1_8/ia_css_eed1_8.host.c:182
> ia_css_eed1_8_vmem_encode() warn: assigning (-8192) to unsigned variable 'to->e_dew_enh_a[0][base +
> j]'
> 
> Should dew_enhance_seg_x and dew_enhance_seg_y be converted to signed?

Someone clearly needs to read the code and work out what it is doing.

First stage is to use clamp() (not clamp_t) to get warnings from the
compiler for the RHS.
The snippet implies that the _x values are unsigned but the _y ones
can be negative.
Holding negative values in unsigned variables is a recipe for disaster.

	David


> 
> 
> Thanks,
> Mauro

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)