Re: [PATCH/RFC v10 08/19] leds: Add support for max77693 mfd flash cell

[Jacek Anaszewski <j.anaszewski@samsung.com>]

On 01/12/2015 02:25 PM, Pavel Machek wrote:
> Hi!
>
>>>> +struct max77693_sub_led {
>>>> +	/* related FLED output identifier */
>>>
>>> ->flash LED, about 4x.
>>>
>>>> +/* split composite current @i into two @iout according to @imax weights */
>>>> +static void __max77693_calc_iout(u32 iout[2], u32 i, u32 imax[2])
>>>> +{
>>>> +	u64 t = i;
>>>> +
>>>> +	t *= imax[1];
>>>> +	do_div(t, imax[0] + imax[1]);
>>>> +
>>>> +	iout[1] = (u32)t / FLASH_IOUT_STEP * FLASH_IOUT_STEP;
>>>> +	iout[0] = i - iout[1];
>>>> +}
>>>
>>> Is 64-bit arithmetics neccessary here? Could we do the FLASH_IOUT_STEP
>>> divisons before t *=, so that 64-bit division is not neccessary?
>>
>> It is required. All these operations allow for splitting the composite
>> current into both outputs according to weights given in the imax
>> array.
>
> I know.
>
> What about this?
>
> static void __max77693_calc_iout(u32 iout[2], u32 i, u32 imax[2])
> {
> 	u32 t = i;
>
> 	t *= imax[1] / FLASH_IOUT_STEP;

Let's consider following case:

t = 1000000
imax[1] = 1000000

multiplication of the above will give 10^12 - much more than
it is possible to encode on 32 bits.

> 	t = t / (imax[0] + imax[1]);
> 	t /= FLASH_IOUT_STEP
>
> 	iout[1] = (u32)t;
> 	iout[0] = i - iout[1];
> }
>
> Does it lack precision?
>
> Thanks,
> 									Pavel
>


-- 
Best Regards,
Jacek Anaszewski