Re: [PATCH RFC RFT 3/8] drm/amd/display: Interpolate when reading PWM back as a percentage

[Ray Wu <ray.wu@amd.com>]

Reviewed-by: Ray Wu <ray.wu@amd.com>

On 11/4/2026 5:55 PM, Hugo Sales wrote:
> backlight_pwm_to_millipercent_legacy() finds the closest backlight_lut
> entry and returns its whole percent. That was the exact inverse while
> backlight_millipercent_to_pwm_legacy() only returned LUT entries. Now
> that it interpolates, a request of 1.5% programs a PWM halfway between
> two entries, which reads back as 1% or 2%, and setting the value read
> back moves the backlight by up to half a LUT step
>
> Interpolate between the two entries around the PWM instead. A PWM equal
> to a LUT entry still reads back as that entry's whole percent
>
> Signed-off-by: Hugo Sales <hugo@hsal.es>
> ---
> Converting a percentage to PWM, back, and to PWM again now moves the
> PWM by about a two-thousandth of the LUT step around it, plus rounding.
> On the GV301QH LUT, whose steps are at most 1799, that keeps every PWM
> within 1 of where it started. A curve with much wider steps can move
> further, but by far less than the half step it could before
>
> KUnit: dm_test_pwm_lut_readback from patch 6 fails without this patch
> and passes with it
> ---
>   .../gpu/drm/amd/display/modules/power/power_abm.c  | 33 ++++++++--------------
>   1 file changed, 11 insertions(+), 22 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/display/modules/power/power_abm.c b/drivers/gpu/drm/amd/display/modules/power/power_abm.c
> index d9c1f1401b604..83d79c20281ec 100644
> --- a/drivers/gpu/drm/amd/display/modules/power/power_abm.c
> +++ b/drivers/gpu/drm/amd/display/modules/power/power_abm.c
> @@ -409,34 +409,23 @@ static unsigned int backlight_pwm_to_millipercent_legacy(
>   			break;
>   	}
>   
> -	/* In this case, exact match is not found. Check if mid/min/max
> -	 * value is actually closer.
> -	 */
> +	/* No exact match, so on a sorted LUT lut[max] < pwm < lut[min] */
>   	if (max < min) {
> -		unsigned int min_delta;
> -		unsigned int mid_delta;
> -		unsigned int max_delta;
> -
> -		min_delta = (core_power->bl_prop[inst].backlight_lut[min] > pwm) ?
> -				core_power->bl_prop[inst].backlight_lut[min] - pwm :
> -				pwm - core_power->bl_prop[inst].backlight_lut[min];
> -
> -		mid_delta = (core_power->bl_prop[inst].backlight_lut[mid] > pwm) ?
> -				core_power->bl_prop[inst].backlight_lut[mid] - pwm :
> -				pwm - core_power->bl_prop[inst].backlight_lut[mid];
> +		unsigned int lower = core_power->bl_prop[inst].backlight_lut[max];
> +		unsigned int upper = core_power->bl_prop[inst].backlight_lut[min];
> +		unsigned long long span, den;
>   
> -		max_delta = (core_power->bl_prop[inst].backlight_lut[max] > pwm) ?
> -				core_power->bl_prop[inst].backlight_lut[max] - pwm :
> -				pwm - core_power->bl_prop[inst].backlight_lut[max];
> +		if (pwm <= lower || pwm >= upper)
> +			return BACKLIGHT_MILLIPERCENT_MAX * mid / max_index;
>   
> -		if ((min_delta < mid_delta) && (min_delta < max_delta))
> -			mid = min;
> +		span = upper - lower;
> +		den = span * max_index;
>   
> -		if ((max_delta < mid_delta) && (max_delta < min_delta))
> -			mid = max;
> +		/* LUT entries fit in 16 bits, so this stays far below 2^64 */
> +		return (unsigned int)div64_u64((max * span + pwm - lower) *
> +					       BACKLIGHT_MILLIPERCENT_MAX + den / 2, den);
>   	}
>   
> -	/* No interpolation, just take closest index */
>   	millipercent = BACKLIGHT_MILLIPERCENT_MAX * mid / max_index;
>   
>   	return millipercent;
>