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

[Hans de Goede <hdegoede@redhat.com>]

Hi,

On 9/6/24 10:05 AM, David Laight wrote:
> From: Hans de Goede
>> Sent: 06 September 2024 08:53
>>
>> Hi Mauro,
>>
>> On 9/6/24 8:15 AM, Mauro Carvalho Chehab wrote:
>>> 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?
>>
>> These already are s32, the problem is that e_dew_enh_a is of type t_vmem_elem which is:
>>
>> typedef u16 t_vmem_elem;
> 
> Ugg... :-)
> 
>>
>> And that type is used in a lot of places, so we cannot
>> just change that.
>>
>> I guess we could add a t_signed_vmem_elem (s16) and use that for these vmem-arrays ?
>>
>> I tried fixing it like this:

<snip>

>>  		/* enable group hold */
>> -		ret = cci_multi_reg_write(sensor->regmap, t4ka3_param_hold,
>> -					  ARRAY_SIZE(t4ka3_param_hold), NULL);
>> -		if (ret)
>> -			goto error_powerdown;
>> -
>> -		ret = cci_multi_reg_write(sensor->regmap, sensor->res->regs, sensor->res->regs_len, NULL);
>> +		cci_multi_reg_write(sensor->regmap, t4ka3_param_hold,
>> +				    ARRAY_SIZE(t4ka3_param_hold), &ret);
>> +		cci_multi_reg_write(sensor->regmap, sensor->res->regs,
>> +				    sensor->res->regs_len, &ret);
>>  		if (ret)
>>  			goto error_powerdown;
> 
> Isn't that unrelated?

Yes my bad.

>> 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 b79d78e5b77f..c9043d516192 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
>> @@ -172,21 +172,21 @@ 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] = clamp(from->dew_enhance_seg_x[j],
>> -							     0, 8191);
>> -			to->e_dew_enh_y[0][base + j] = clamp(from->dew_enhance_seg_y[j],
>> -							     -8192, 8191);
>> +			to->e_dew_enh_x[0][base + j] = (u16)clamp(from->dew_enhance_seg_x[j],
>> +								  0, 8191);
>> +			to->e_dew_enh_y[0][base + j] = (u16)clamp(from->dew_enhance_seg_y[j],
>> +								  -8192, 8191);
> 
> How about an explicit clamp(...) & 0xffffu?

Yes that should work, I tihnk.  I actually have changed the type of e_dew_enh_y
and e_dew_enh_f to s16 now and that does the trick of silencing smatch and seems
like a nicer fix.

I need to go and test the fix on actual hw to make sure nothing breaks and then
I'll submit it.

> 
>>
>>  		for (j = 0; j < (IA_CSS_NUMBER_OF_DEW_ENHANCE_SEGMENTS - 1); j++) {
>> -			to->e_dew_enh_a[0][base + j] = clamp(from->dew_enhance_seg_slope[j],
>> -							     -8192, 8191);
>> +			to->e_dew_enh_a[0][base + j] = (u16)clamp(from->dew_enhance_seg_slope[j],
>> +								  -8192, 8191);
>>  			/* Convert dew_enhance_seg_exp to flag:
>>  			 * 0 -> 0
>>  			 * 1...13 -> 1
>>  			 */
>> -			to->e_dew_enh_f[0][base + j] = clamp(from->dew_enhance_seg_exp[j],
>> -							     0, 13) > 0;
>> +			to->e_dew_enh_f[0][base + j] = (u16)clamp(from->dew_enhance_seg_exp[j],
>> +								  0, 13) > 0;
> 
> Isn't the RHS just from->dew_enhance_seg_exp[j] > 0 ?
> That shouldn't be generating any kind of warning anyway.

It indeed does not generate a warning I changed all the clamp() calls here
to keep things consistent.

Regards,

Hans