Without the patch (i.MX8MP, all-white RGGB-12 full HD input from
the sensor, YUV NV12 output from ISP, full range, histogram Y mode).
HIST_STEPSIZE = 3 (lowest permitted), all weights are set to 1:

isp_hist_h_size: 383 (= 1920 / 5 - 1)
isp_hist_v_size: 215 (= 1080 / 5 - 1)
histogram_measurement_result[16]: 0 0 0 0  0 0 0 0  0 0 0 0  0 0 0 229401

Apparently the histogram is missing the last column (3-pixel wide,
though only single pixels count) and the last (same idea) row
of the input image: 1917 * 1077 / 3 / 3 = 229401.

E.g. without the patch, the pixels counted in lines are:
0, 3, ... 1914 (which makes 1917/3), and patched:
0, 3, ... 1914, 1917 (which makes 1920/3).
The same is true for rows.

With the patch applied:
isp_hist_h_size: 384 (= 1920 / 5)
isp_hist_v_size: 216 (= 1080 / 5)
histogram_measurement_result[16]: 0 0 0 0  0 0 0 0  0 0 0 0  0 0 0 230400

1920 * 1080 / 3 / 3 = 230400

Signed-off-by: Krzysztof Hałasa <khalasa@piap.pl>

diff --git a/drivers/media/platform/rockchip/rkisp1/rkisp1-params.c b/drivers/media/platform/rockchip/rkisp1/rkisp1-params.c
index b28f4140c8a3..ca9b3e711e5f 100644
--- a/drivers/media/platform/rockchip/rkisp1/rkisp1-params.c
+++ b/drivers/media/platform/rockchip/rkisp1/rkisp1-params.c
@@ -819,8 +819,8 @@ static void rkisp1_hst_config_v10(struct rkisp1_params *params,
 		     arg->meas_window.v_offs);
 
 	block_hsize = arg->meas_window.h_size /
-		      RKISP1_CIF_ISP_HIST_COLUMN_NUM_V10 - 1;
-	block_vsize = arg->meas_window.v_size / RKISP1_CIF_ISP_HIST_ROW_NUM_V10 - 1;
+		      RKISP1_CIF_ISP_HIST_COLUMN_NUM_V10;
+	block_vsize = arg->meas_window.v_size / RKISP1_CIF_ISP_HIST_ROW_NUM_V10;
 
 	rkisp1_write(params->rkisp1, RKISP1_CIF_ISP_HIST_H_SIZE_V10,
 		     block_hsize);

-- 
Krzysztof "Chris" Hałasa

Sieć Badawcza Łukasiewicz
Przemysłowy Instytut Automatyki i Pomiarów PIAP
Al. Jerozolimskie 202, 02-486 Warszawa

Hi Krzysztof,

Thanks for the new version.

My bad, I should've mentioned in the review of the previous version; I think
you should have a paragraph in the beginning briefly explaining the problem in
english.

"The histogram window dimensions are unnecessarily decremented. This can be
seen in the following example, where when all pixels are of maximum intensity
there are pixels that we expect to be counted but are missing from the highest
bin. Fix this by removing the unnecessary decrement."

I think would be good enough.

Quoting Krzysztof Hałasa (2025-05-23 18:10:47)
> Without the patch (i.MX8MP, all-white RGGB-12 full HD input from
> the sensor, YUV NV12 output from ISP, full range, histogram Y mode).
> HIST_STEPSIZE = 3 (lowest permitted), all weights are set to 1:
> 
> isp_hist_h_size: 383 (= 1920 / 5 - 1)
> isp_hist_v_size: 215 (= 1080 / 5 - 1)
> histogram_measurement_result[16]: 0 0 0 0  0 0 0 0  0 0 0 0  0 0 0 229401
> 
> Apparently the histogram is missing the last column (3-pixel wide,
> though only single pixels count) and the last (same idea) row
> of the input image: 1917 * 1077 / 3 / 3 = 229401.
> 
> E.g. without the patch, the pixels counted in lines are:
> 0, 3, ... 1914 (which makes 1917/3), and patched:
> 0, 3, ... 1914, 1917 (which makes 1920/3).
> The same is true for rows.

I like this clarification.


With the above paragraph added,

Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>

(I suppose if you approve it then it could simply be added when it's applied
and you wouldn't need to send a new version, but I don't have permissions for
that so uh...)

> 
> With the patch applied:
> isp_hist_h_size: 384 (= 1920 / 5)
> isp_hist_v_size: 216 (= 1080 / 5)
> histogram_measurement_result[16]: 0 0 0 0  0 0 0 0  0 0 0 0  0 0 0 230400
> 
> 1920 * 1080 / 3 / 3 = 230400
> 
> Signed-off-by: Krzysztof Hałasa <khalasa@piap.pl>
> 
> diff --git a/drivers/media/platform/rockchip/rkisp1/rkisp1-params.c b/drivers/media/platform/rockchip/rkisp1/rkisp1-params.c
> index b28f4140c8a3..ca9b3e711e5f 100644
> --- a/drivers/media/platform/rockchip/rkisp1/rkisp1-params.c
> +++ b/drivers/media/platform/rockchip/rkisp1/rkisp1-params.c
> @@ -819,8 +819,8 @@ static void rkisp1_hst_config_v10(struct rkisp1_params *params,
>                      arg->meas_window.v_offs);
>  
>         block_hsize = arg->meas_window.h_size /
> -                     RKISP1_CIF_ISP_HIST_COLUMN_NUM_V10 - 1;
> -       block_vsize = arg->meas_window.v_size / RKISP1_CIF_ISP_HIST_ROW_NUM_V10 - 1;
> +                     RKISP1_CIF_ISP_HIST_COLUMN_NUM_V10;
> +       block_vsize = arg->meas_window.v_size / RKISP1_CIF_ISP_HIST_ROW_NUM_V10;
>  
>         rkisp1_write(params->rkisp1, RKISP1_CIF_ISP_HIST_H_SIZE_V10,
>                      block_hsize);
> 
> -- 
> Krzysztof "Chris" Hałasa
> 
> Sieć Badawcza Łukasiewicz
> Przemysłowy Instytut Automatyki i Pomiarów PIAP
> Al. Jerozolimskie 202, 02-486 Warszawa