[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20241230165917.GA31768@pendragon.ideasonboard.com>
Date: Mon, 30 Dec 2024 18:59:17 +0200
From: Laurent Pinchart <laurent.pinchart@...asonboard.com>
To: Jai Luthra <jai.luthra@...asonboard.com>
Cc: Dafna Hirschfeld <dafna@...tmail.com>,
Mauro Carvalho Chehab <mchehab@...nel.org>,
Heiko Stuebner <heiko@...ech.de>, linux-media@...r.kernel.org,
linux-rockchip@...ts.infradead.org,
linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH RFC] media: rockchip: rkisp1: Add support for AWB64
On Mon, Dec 30, 2024 at 10:04:36PM +0530, Jai Luthra wrote:
> On Dec 30, 2024 at 17:30:37 +0200, Laurent Pinchart wrote:
> > On Mon, Dec 30, 2024 at 12:25:09PM +0530, Jai Luthra wrote:
> > > AWB64 is an advanced auto white-balance statistics block, present on the
> > > ISP in iMX8MP. This block can calculate color statistics for a maximum
> > > of 8 different (elliptical) regions, which can be used by the AWB
> > > algorithm to better tune the color gains.
> > >
> > > Signed-off-by: Jai Luthra <jai.luthra@...asonboard.com>
> > > ---
> > > This patch is only compile-tested, thus marked as an RFC.
> > >
> > > Some of the ellipse and color transformation parameters are sparsely
> > > documented, so the API might have to change on further testing.
> > > ---
> > > .../media/platform/rockchip/rkisp1/rkisp1-common.h | 9 ++
> > > .../media/platform/rockchip/rkisp1/rkisp1-dev.c | 3 +-
> > > .../media/platform/rockchip/rkisp1/rkisp1-params.c | 123 +++++++++++++++++++++
> > > .../media/platform/rockchip/rkisp1/rkisp1-regs.h | 41 +++++++
> > > .../media/platform/rockchip/rkisp1/rkisp1-stats.c | 31 +++++-
> > > include/uapi/linux/rkisp1-config.h | 105 ++++++++++++++++++
> > > 6 files changed, 310 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/media/platform/rockchip/rkisp1/rkisp1-common.h b/drivers/media/platform/rockchip/rkisp1/rkisp1-common.h
> > > index ca952fd0829ba7d923ad42fec92840ccd422b6e5..7dc9d101a1a01627f174cc4c454e4e57b38e1f8e 100644
> > > --- a/drivers/media/platform/rockchip/rkisp1/rkisp1-common.h
> > > +++ b/drivers/media/platform/rockchip/rkisp1/rkisp1-common.h
> > > @@ -118,6 +118,7 @@ enum rkisp1_isp_pad {
> > > * @RKISP1_FEATURE_DMA_34BIT: The ISP uses 34-bit DMA addresses
> > > * @RKISP1_FEATURE_BLS: The ISP has a dedicated BLS block
> > > * @RKISP1_FEATURE_COMPAND: The ISP has a companding block
> > > + * @RKISP1_FEATURE_AWB64: The ISP has an AWB64 block
> > > *
> > > * The ISP features are stored in a bitmask in &rkisp1_info.features and allow
> > > * the driver to implement support for features present in some ISP versions
> > > @@ -131,6 +132,7 @@ enum rkisp1_feature {
> > > RKISP1_FEATURE_DMA_34BIT = BIT(4),
> > > RKISP1_FEATURE_BLS = BIT(5),
> > > RKISP1_FEATURE_COMPAND = BIT(6),
> > > + RKISP1_FEATURE_AWB64 = BIT(7),
> > > };
> > >
> > > #define rkisp1_has_feature(rkisp1, feature) \
> > > @@ -345,6 +347,8 @@ struct rkisp1_stats;
> > > struct rkisp1_stats_ops {
> > > void (*get_awb_meas)(struct rkisp1_stats *stats,
> > > struct rkisp1_stat_buffer *pbuf);
> > > + void (*get_awb64_meas)(struct rkisp1_stats *stats,
> > > + struct rkisp1_stat_buffer *pbuf);
> > > void (*get_aec_meas)(struct rkisp1_stats *stats,
> > > struct rkisp1_stat_buffer *pbuf);
> > > void (*get_hst_meas)(struct rkisp1_stats *stats,
> > > @@ -381,6 +385,11 @@ struct rkisp1_params_ops {
> > > void (*awb_meas_enable)(struct rkisp1_params *params,
> > > const struct rkisp1_cif_isp_awb_meas_config *arg,
> > > bool en);
> > > + void (*awb64_meas_config)(struct rkisp1_params *params,
> > > + const struct rkisp1_cif_isp_awb64_meas_config *arg);
> > > + void (*awb64_meas_enable)(struct rkisp1_params *params,
> > > + const struct rkisp1_cif_isp_awb64_meas_config *arg,
> > > + bool en);
> > > void (*awb_gain_config)(struct rkisp1_params *params,
> > > const struct rkisp1_cif_isp_awb_gain_config *arg);
> > > void (*aec_config)(struct rkisp1_params *params,
> > > diff --git a/drivers/media/platform/rockchip/rkisp1/rkisp1-dev.c b/drivers/media/platform/rockchip/rkisp1/rkisp1-dev.c
> > > index 0100b9c3edbefbdc001e2e4d13049efa9493e3bd..20903823b8eec2dd6b2fda788a42ef7545158f8c 100644
> > > --- a/drivers/media/platform/rockchip/rkisp1/rkisp1-dev.c
> > > +++ b/drivers/media/platform/rockchip/rkisp1/rkisp1-dev.c
> > > @@ -557,7 +557,8 @@ static const struct rkisp1_info imx8mp_isp_info = {
> > > .isp_ver = RKISP1_V_IMX8MP,
> > > .features = RKISP1_FEATURE_MAIN_STRIDE
> > > | RKISP1_FEATURE_DMA_34BIT
> > > - | RKISP1_FEATURE_COMPAND,
> > > + | RKISP1_FEATURE_COMPAND
> > > + | RKISP1_FEATURE_AWB64,
> > > .max_width = 4096,
> > > .max_height = 3072,
> > > };
> > > diff --git a/drivers/media/platform/rockchip/rkisp1/rkisp1-params.c b/drivers/media/platform/rockchip/rkisp1/rkisp1-params.c
> > > index b28f4140c8a309a3231d44d825c6461e3ecb2a44..1ee5f07d461c5ad7aadff908c25f7a5bd2ff81f9 100644
> > > --- a/drivers/media/platform/rockchip/rkisp1/rkisp1-params.c
> > > +++ b/drivers/media/platform/rockchip/rkisp1/rkisp1-params.c
> > > @@ -60,6 +60,7 @@ union rkisp1_ext_params_config {
> > > struct rkisp1_ext_params_afc_config afc;
> > > struct rkisp1_ext_params_compand_bls_config compand_bls;
> > > struct rkisp1_ext_params_compand_curve_config compand_curve;
> > > + struct rkisp1_ext_params_awb64_meas_config awb64;
> > > };
> > >
> > > enum rkisp1_params_formats {
> > > @@ -674,6 +675,101 @@ rkisp1_awb_meas_enable_v12(struct rkisp1_params *params,
> > > }
> > > }
> > >
> > > +/* ISP White Balance Mode */
> > > +static void
> > > +rkisp1_awb64_meas_config_v10(struct rkisp1_params *params,
> > > + const struct rkisp1_cif_isp_awb64_meas_config *arg)
> >
> > v10 refers to the Rockchip ISP v1.0. As the AWB64 block is found in the
> > VSI ISP8000 Nano (v18.02 if I'm not mistaken), which we refer to in this
> > driver as "RKISP1_V_IMX8MP", I would name this function
> > rkisp1_awb64_meas_config_imx8mp() or just rkisp1_awb64_meas_config().
> > Same for rkisp1_awb64_meas_enable_v10().
> >
>
> Will fix.
>
> > > +{
> > > + /* window offset */
> >
> > /* Window offset and size */
> >
> > > + rkisp1_write(params->rkisp1, RKISP1_CIF_ISP_AWB64_V_OFFS,
> > > + arg->awb_wnd.v_offs);
> > > + rkisp1_write(params->rkisp1, RKISP1_CIF_ISP_AWB64_V_OFFS,
> > > + arg->awb_wnd.v_offs);
> >
> > You're writing the same register twice.
> >
>
> Oops, will fix.
>
> > > + /* AWB window size */
> >
> > Drop this.
> >
> > > + rkisp1_write(params->rkisp1, RKISP1_CIF_ISP_AWB64_V_SIZE,
> > > + arg->awb_wnd.v_size);
> > > + rkisp1_write(params->rkisp1, RKISP1_CIF_ISP_AWB64_H_SIZE,
> > > + arg->awb_wnd.h_size);
> >
> > A blank line here would make code easier to read.
> >
> > > + /* RGB thresholds for measurement */
> > > + rkisp1_write(params->rkisp1, RKISP1_CIF_ISP_AWB64_R_MIN_MAX,
> > > + arg->max_r << 8 | arg->min_r);
> >
> > To simplify the calculations in the kernel, would it make sense for
> > userspace to compute this and store it in a single 16-bit field ?
> >
>
> Indeed, will fix.
>
> > > + rkisp1_write(params->rkisp1, RKISP1_CIF_ISP_AWB64_G_MIN_MAX,
> > > + arg->max_g << 8 | arg->min_g);
> > > + rkisp1_write(params->rkisp1, RKISP1_CIF_ISP_AWB64_B_MIN_MAX,
> > > + arg->max_b << 8 | arg->min_b);
> >
> > Same here.
> >
> > > + /* Minimum input divider threshold */
> > > + rkisp1_write(params->rkisp1, RKISP1_CIF_ISP_AWB64_MIN_DIVIDER,
> > > + arg->min_div & RKISP1_CIF_ISP_AWB64_MIN_DIV_MASK);
> >
> > And here.
> >
> > > + /* Colorspace matrix coefficients */
> > > + for (int i = 0; i < 3; i++)
> > > + for (int j = 0; j < 3; j++)
> > > + rkisp1_write(params->rkisp1,
> > > + RKISP1_CIF_ISP_AWB64_CC_COEFF(i * 3 + j),
> > > + arg->cc_coeff[i][j]);
> >
> > Use curly braces for outer loop.
> >
> > > + /* Ellipse configuration */
> > > + for (int i = 0; i < RKISP1_CIF_ISP_AWB64_MAX_ELLIPSE; i++) {
> >
> > const struct rkisp1_cif_isp_awb64_ellip *ellip = &arg->ellip[i];
> >
> > > + /* Center */
> > > + rkisp1_write(params->rkisp1,
> > > + RKISP1_CIF_ISP_AWB64_ELLIP_CEN_X(i),
> > > + arg->ellip[i].cen_x &
> > > + RKISP1_CIF_ISP_AWB64_ELLIP_CEN_MASK);
> >
> > ellip->cen_x & RKISP1_CIF_ISP_AWB64_ELLIP_CEN_MASK);
> >
> > Same below.
> >
> > > + rkisp1_write(params->rkisp1,
> > > + RKISP1_CIF_ISP_AWB64_ELLIP_CEN_Y(i),
> > > + arg->ellip[i].cen_y &
> > > + RKISP1_CIF_ISP_AWB64_ELLIP_CEN_MASK);
> > > + /* Radius */
> > > + rkisp1_write(params->rkisp1,
> > > + RKISP1_CIF_ISP_AWB64_ELLIP_RMAX(i),
> > > + arg->ellip[i].rmax &
> > > + RKISP1_CIF_ISP_AWB64_ELLIP_RMAX_MASK);
> > > + /* CTM */
> > > + rkisp1_write(params->rkisp1,
> > > + RKISP1_CIF_ISP_AWB64_ELLIP_A1(i),
> > > + arg->ellip[i].ctm[0] &
> > > + RKISP1_CIF_ISP_AWB64_ELLIP_A1_A3_MASK);
> > > + rkisp1_write(params->rkisp1,
> > > + RKISP1_CIF_ISP_AWB64_ELLIP_A2(i),
> > > + arg->ellip[i].ctm[1] &
> > > + RKISP1_CIF_ISP_AWB64_ELLIP_A2_A4_MASK);
> > > + rkisp1_write(params->rkisp1,
> > > + RKISP1_CIF_ISP_AWB64_ELLIP_A3(i),
> > > + arg->ellip[i].ctm[2] &
> > > + RKISP1_CIF_ISP_AWB64_ELLIP_A1_A3_MASK);
> > > + rkisp1_write(params->rkisp1,
> > > + RKISP1_CIF_ISP_AWB64_ELLIP_A4(i),
> > > + arg->ellip[i].ctm[3] &
> > > + RKISP1_CIF_ISP_AWB64_ELLIP_A2_A4_MASK);
> > > + }
> > > +}
> > > +
> > > +static void
> > > +rkisp1_awb64_meas_enable_v10(struct rkisp1_params *params,
> > > + const struct rkisp1_cif_isp_awb64_meas_config *arg,
> > > + bool en)
> > > +{
> > > + u32 reg_val =
> > > + rkisp1_read(params->rkisp1, RKISP1_CIF_ISP_AWB64_MEAS_MODE);
> > > +
> > > + if (en) {
> > > + if (arg->enable_median_filter)
> > > + reg_val |= RKISP1_CIF_ISP_AWB64_PRE_FILTER_EN;
> > > +
> > > + if (arg->chrom_switch)
> > > + reg_val |= RKISP1_CIF_ISP_AWB64_CHROM_SWITCH;
> > > +
> > > + reg_val |= (arg->ellip_unite & RKISP1_CIF_ISP_AWB64_UNITE_MASK)
> > > + << RKISP1_CIF_ISP_AWB64_UNITE_SHIFT;
> >
> > You never clear any of those fields.
>
> Ah I completely missed that, will fix.
>
> > Is it intentional that you make it impossible to change those parameters
> > without disabling and re-enabling the block ?
>
> I was not exactly sure what all parameters were okay to toggle
> mid-stream, so I kept everything in the MEAS_MODE register together in
> this routine. For now I will move everything except the module enable
> bit to the awb64_meas_config() routine.
>
> But I guess more experimentation from the userspace side is needed, as I
> see in the AWB block we only allow changing the colorspace mode when
> enabling/disabling the block, and I am not sure if similar restrictions
> are present here.
None of the AWB64 are marked as having shadow registers, so I think
there's conceptually no difference. We should either allow programming
them all while the module is enabled, or forbid them all. I'd start by
allowing them all for now.
> > > +
> > > + reg_val |= RKISP1_CIF_ISP_AWB64_MEAS_EN |
> > > + RKISP1_CIF_ISP_AWB64_IRQ_EN;
> > > + } else {
> > > + reg_val &= ~RKISP1_CIF_ISP_AWB64_MEAS_EN;
> > > + }
> > > +
> > > + rkisp1_write(params->rkisp1, RKISP1_CIF_ISP_AWB64_MEAS_MODE,
> > > + reg_val);
> > > +}
> > > +
> > > static void
> > > rkisp1_awb_gain_config_v10(struct rkisp1_params *params,
> > > const struct rkisp1_cif_isp_awb_gain_config *arg)
> > > @@ -2005,6 +2101,25 @@ static void rkisp1_ext_params_compand_compress(struct rkisp1_params *params,
> > > RKISP1_CIF_ISP_COMPAND_CTRL_COMPRESS_ENABLE);
> > > }
> > >
> > > +static void rkisp1_ext_params_awb64(struct rkisp1_params *params,
> > > + const union rkisp1_ext_params_config *block)
> > > +{
> > > + const struct rkisp1_ext_params_awb64_meas_config *awb64 = &block->awb64;
> > > +
> > > + if (awb64->header.flags & RKISP1_EXT_PARAMS_FL_BLOCK_DISABLE) {
> > > + params->ops->awb64_meas_enable(params, &awb64->config,
> > > + false);
> >
> > Wrong indentation.
> >
> > > + return;
> > > + }
> > > +
> > > + params->ops->awb64_meas_config(params, &awb64->config);
> > > +
> > > + if ((awb64->header.flags & RKISP1_EXT_PARAMS_FL_BLOCK_ENABLE) &&
> > > + !(params->enabled_blocks & BIT(awb64->header.type)))
> > > + params->ops->awb64_meas_enable(params, &awb64->config,
> > > + true);
> >
> > Same here.
> >
> > > +}
> > > +
> > > typedef void (*rkisp1_block_handler)(struct rkisp1_params *params,
> > > const union rkisp1_ext_params_config *config);
> > >
> > > @@ -2118,6 +2233,12 @@ static const struct rkisp1_ext_params_handler {
> > > .group = RKISP1_EXT_PARAMS_BLOCK_GROUP_OTHERS,
> > > .features = RKISP1_FEATURE_COMPAND,
> > > },
> > > + [RKISP1_EXT_PARAMS_BLOCK_TYPE_AWB64_MEAS] = {
> > > + .size = sizeof(struct rkisp1_ext_params_awb64_meas_config),
> > > + .handler = rkisp1_ext_params_awb64,
> > > + .group = RKISP1_EXT_PARAMS_BLOCK_GROUP_OTHERS,
> > > + .features = RKISP1_FEATURE_AWB64,
> > > + },
> > > };
> > >
> > > static void rkisp1_ext_params_config(struct rkisp1_params *params,
> > > @@ -2381,6 +2502,8 @@ static const struct rkisp1_params_ops rkisp1_v10_params_ops = {
> > > .goc_config = rkisp1_goc_config_v10,
> > > .awb_meas_config = rkisp1_awb_meas_config_v10,
> > > .awb_meas_enable = rkisp1_awb_meas_enable_v10,
> > > + .awb64_meas_config = rkisp1_awb64_meas_config_v10,
> > > + .awb64_meas_enable = rkisp1_awb64_meas_enable_v10,
> >
> > I think you can drop these two fields and call
> > rkisp1_awb64_meas_config() and rkisp1_awb64_meas_enable() directly.
>
> Thanks, will do.
>
> > > .awb_gain_config = rkisp1_awb_gain_config_v10,
> > > .aec_config = rkisp1_aec_config_v10,
> > > .hst_config = rkisp1_hst_config_v10,
> > > diff --git a/drivers/media/platform/rockchip/rkisp1/rkisp1-regs.h b/drivers/media/platform/rockchip/rkisp1/rkisp1-regs.h
> > > index bf0260600a1923eebde6b5fe233daf7d427362dd..bc292a5738198fff28715b2a7e55e26e8fbddc64 100644
> > > --- a/drivers/media/platform/rockchip/rkisp1/rkisp1-regs.h
> > > +++ b/drivers/media/platform/rockchip/rkisp1/rkisp1-regs.h
> > > @@ -516,6 +516,24 @@
> > > #define RKISP1_CIF_ISP_AWB_CBCR_MAX_REF 0x000000ff
> > > #define RKISP1_CIF_ISP_AWB_THRES_MAX_YC 0x000000ff
> > >
> > > +/* AWB64 */
> > > +/* ISP_AWB64_WHITE_CNT */
> > > +#define RKISP1_CIF_ISP_AWB64_GET_PIXEL_CNT(x) ((x) & 0xffffff)
> > > +/* ISP_AWB64_MEAS_MODE */
> > > +#define RKISP1_CIF_ISP_AWB64_MEAS_EN BIT(0)
> > > +#define RKISP1_CIF_ISP_AWB64_PRE_FILTER_EN BIT(1)
> > > +#define RKISP1_CIF_ISP_AWB64_IRQ_EN BIT(2)
> > > +#define RKISP1_CIF_ISP_AWB64_CHROM_SWITCH BIT(3)
> > > +#define RKISP1_CIF_ISP_AWB64_UNITE_MASK GENMASK(5, 0)
> > > +#define RKISP1_CIF_ISP_AWB64_UNITE_SHIFT 4
> > > +/* ISP_AWB64_DIVIDER_MIN */
> > > +#define RKISP1_CIF_ISP_AWB64_MIN_DIV_MASK GENMASK(9, 0)
> > > +/* ISP_AWB64_ELLIP */
> > > +#define RKISP1_CIF_ISP_AWB64_ELLIP_CEN_MASK GENMASK(9, 0)
> > > +#define RKISP1_CIF_ISP_AWB64_ELLIP_A1_A3_MASK GENMASK(11, 0)
> > > +#define RKISP1_CIF_ISP_AWB64_ELLIP_A2_A4_MASK GENMASK(8, 0)
> > > +#define RKISP1_CIF_ISP_AWB64_ELLIP_RMAX_MASK GENMASK(23, 0)
> > > +
> > > /* AE */
> > > /* ISP_EXP_CTRL */
> > > #define RKISP1_CIF_ISP_EXP_ENA BIT(0)
> > > @@ -1379,6 +1397,29 @@
> > > #define RKISP1_CIF_ISP_WDR_TONECURVE_YM_31_SHD (RKISP1_CIF_ISP_WDR_BASE + 0x0000012c)
> > > #define RKISP1_CIF_ISP_WDR_TONECURVE_YM_32_SHD (RKISP1_CIF_ISP_WDR_BASE + 0x00000130)
> > >
> > > +#define RKISP1_CIF_ISP_AWB64_BASE 0x00002c00
> > > +#define RKISP1_CIF_ISP_AWB64_MEAS_MODE (RKISP1_CIF_ISP_AWB64_BASE + 0x00000000)
> > > +#define RKISP1_CIF_ISP_AWB64_H_OFFS (RKISP1_CIF_ISP_AWB64_BASE + 0x00000000)
> > > +#define RKISP1_CIF_ISP_AWB64_V_OFFS (RKISP1_CIF_ISP_AWB64_BASE + 0x00000000)
> > > +#define RKISP1_CIF_ISP_AWB64_H_SIZE (RKISP1_CIF_ISP_AWB64_BASE + 0x00000000)
> > > +#define RKISP1_CIF_ISP_AWB64_V_SIZE (RKISP1_CIF_ISP_AWB64_BASE + 0x00000000)
> > > +#define RKISP1_CIF_ISP_AWB64_R_MIN_MAX (RKISP1_CIF_ISP_AWB64_BASE + 0x00000000)
> > > +#define RKISP1_CIF_ISP_AWB64_G_MIN_MAX (RKISP1_CIF_ISP_AWB64_BASE + 0x00000000)
> > > +#define RKISP1_CIF_ISP_AWB64_B_MIN_MAX (RKISP1_CIF_ISP_AWB64_BASE + 0x00000000)
> > > +#define RKISP1_CIF_ISP_AWB64_MIN_DIVIDER (RKISP1_CIF_ISP_AWB64_BASE + 0x00000000)
> > > +#define RKISP1_CIF_ISP_AWB64_CC_COEFF(n) (RKISP1_CIF_ISP_AWB64_BASE + 0x00000000 + (n) * 4)
> >
> > All those registers seem to be at the same address. I think the next
> > version of this patch needs to be tested.
>
> Oops, will fix. I was testing the register writes through a separate
> shell script. Will test the kernel driver through libcamera before
> sending the next revision.
>
> > > +#define RKISP1_CIF_ISP_AWB64_ELLIP_CEN_X(n) (RKISP1_CIF_ISP_AWB64_BASE + 0x00000048 + (n) * 8)
> > > +#define RKISP1_CIF_ISP_AWB64_ELLIP_CEN_Y(n) (RKISP1_CIF_ISP_AWB64_BASE + 0x0000004c + (n) * 8)
> > > +#define RKISP1_CIF_ISP_AWB64_ELLIP_A1(n) (RKISP1_CIF_ISP_AWB64_BASE + 0x00000088 + (n) * 16)
> > > +#define RKISP1_CIF_ISP_AWB64_ELLIP_A2(n) (RKISP1_CIF_ISP_AWB64_BASE + 0x0000008c + (n) * 16)
> > > +#define RKISP1_CIF_ISP_AWB64_ELLIP_A3(n) (RKISP1_CIF_ISP_AWB64_BASE + 0x00000090 + (n) * 16)
> > > +#define RKISP1_CIF_ISP_AWB64_ELLIP_A4(n) (RKISP1_CIF_ISP_AWB64_BASE + 0x00000094 + (n) * 16)
> > > +#define RKISP1_CIF_ISP_AWB64_ELLIP_RMAX(n) (RKISP1_CIF_ISP_AWB64_BASE + 0x00000108 + (n) * 4)
> > > +#define RKISP1_CIF_ISP_AWB64_WHITE_CNT(n) (RKISP1_CIF_ISP_AWB64_BASE + 0x00000128 + (n) * 4)
> > > +#define RKISP1_CIF_ISP_AWB64_R_ACCU(n) (RKISP1_CIF_ISP_AWB64_BASE + 0x00000148 + (n) * 12)
> > > +#define RKISP1_CIF_ISP_AWB64_G_ACCU(n) (RKISP1_CIF_ISP_AWB64_BASE + 0x0000014c + (n) * 12)
> > > +#define RKISP1_CIF_ISP_AWB64_B_ACCU(n) (RKISP1_CIF_ISP_AWB64_BASE + 0x00000150 + (n) * 12)
> > > +
> > > #define RKISP1_CIF_ISP_HIST_BASE_V12 0x00002c00
> > > #define RKISP1_CIF_ISP_HIST_CTRL_V12 (RKISP1_CIF_ISP_HIST_BASE_V12 + 0x00000000)
> > > #define RKISP1_CIF_ISP_HIST_SIZE_V12 (RKISP1_CIF_ISP_HIST_BASE_V12 + 0x00000004)
> > > diff --git a/drivers/media/platform/rockchip/rkisp1/rkisp1-stats.c b/drivers/media/platform/rockchip/rkisp1/rkisp1-stats.c
> > > index d5fdb8f82dc78b0143f71d76f36817db389921b7..a5b82963e74dac2340a1b3b07cd8d99dd23b5d5d 100644
> > > --- a/drivers/media/platform/rockchip/rkisp1/rkisp1-stats.c
> > > +++ b/drivers/media/platform/rockchip/rkisp1/rkisp1-stats.c
> > > @@ -172,6 +172,31 @@ rkisp1_stats_init_vb2_queue(struct vb2_queue *q, struct rkisp1_stats *stats)
> > > return vb2_queue_init(q);
> > > }
> > >
> > > +static void rkisp1_stats_get_awb64_meas_v10(struct rkisp1_stats *stats,
> > > + struct rkisp1_stat_buffer *pbuf)
> > > +{
> > > + /* Protect against concurrent access from ISR? */
> >
> > If this is needed, let's address the issue. If it's not, let's drop the
> > comment.
>
> Ack. I don't yet know if the same interrupt bit is used to signal stats
> ready for both AWB and AWB64 blocks. Will hopefully have a better idea
> once testing this with userspace.
>
> > > + struct rkisp1_device *rkisp1 = stats->rkisp1;
> > > + unsigned int i;
> > > + u32 white_cnt;
> > > +
> > > + pbuf->meas_type |= RKISP1_CIF_ISP_STAT_AWB64;
> > > +
> > > + for (i = 0; i < RKISP1_CIF_ISP_AWB64_MAX_ELLIPSE; i++) {
> >
> > for (unsigned int i = 0; i < RKISP1_CIF_ISP_AWB64_MAX_ELLIPSE; i++) {
> > struct rkisp1_cif_isp_awb64_meas *count =
> > &pbuf->params.awb64.count[i];
> >
> > and use count below.
> >
> > > + white_cnt = rkisp1_read(rkisp1,
> > > + RKISP1_CIF_ISP_AWB64_WHITE_CNT(i));
> > > + pbuf->params.awb64.count[i].cnt =
> > > + RKISP1_CIF_ISP_AWB64_GET_PIXEL_CNT(white_cnt);
> > > +
> > > + pbuf->params.awb64.count[i].accu_cr_or_r =
> > > + rkisp1_read(rkisp1, RKISP1_CIF_ISP_AWB64_R_ACCU(i));
> > > + pbuf->params.awb64.count[i].accu_y_or_g =
> > > + rkisp1_read(rkisp1, RKISP1_CIF_ISP_AWB64_G_ACCU(i));
> > > + pbuf->params.awb64.count[i].accu_cb_or_b =
> > > + rkisp1_read(rkisp1, RKISP1_CIF_ISP_AWB64_B_ACCU(i));
> > > + }
> > > +}
> > > +
> > > static void rkisp1_stats_get_awb_meas_v10(struct rkisp1_stats *stats,
> > > struct rkisp1_stat_buffer *pbuf)
> > > {
> > > @@ -325,6 +350,7 @@ static void rkisp1_stats_get_bls_meas(struct rkisp1_stats *stats,
> > >
> > > static const struct rkisp1_stats_ops rkisp1_v10_stats_ops = {
> > > .get_awb_meas = rkisp1_stats_get_awb_meas_v10,
> > > + .get_awb64_meas = rkisp1_stats_get_awb64_meas_v10,
> >
> > Not all ISP versions using rkisp1_v10_stats_ops support the AWB64 block.
> > You need to duplicate this structure, with a specific version for the
> > i.MX8MP.
> >
> > > .get_aec_meas = rkisp1_stats_get_aec_meas_v10,
> > > .get_hst_meas = rkisp1_stats_get_hst_meas_v10,
> > > };
> > > @@ -355,8 +381,11 @@ rkisp1_stats_send_measurement(struct rkisp1_stats *stats, u32 isp_ris)
> > >
> > > cur_stat_buf = (struct rkisp1_stat_buffer *)
> > > vb2_plane_vaddr(&cur_buf->vb.vb2_buf, 0);
> > > - if (isp_ris & RKISP1_CIF_ISP_AWB_DONE)
> > > + if (isp_ris & RKISP1_CIF_ISP_AWB_DONE) {
> > > stats->ops->get_awb_meas(stats, cur_stat_buf);
> > > + if (stats->ops->get_awb64_meas)
> > > + stats->ops->get_awb64_meas(stats, cur_stat_buf);
> >
> > As you need to create rkisp1_imx8mp_stats_ops as explained above, you
> > can combine awb and awb64 support in the .get_awb_meas() operation. One
> > option is to create a rkisp1_stats_get_awb_meas_imx8mp() that calls
> > rkisp1_stats_get_awb_meas_v10() and rkisp1_stats_get_awb64_meas().
>
> Makes sense, will fix.
>
> > > + }
> > >
> > > if (isp_ris & RKISP1_CIF_ISP_AFM_FIN)
> > > rkisp1_stats_get_afc_meas(stats, cur_stat_buf);
> > > diff --git a/include/uapi/linux/rkisp1-config.h b/include/uapi/linux/rkisp1-config.h
> > > index 430daceafac7056951be968f3b4d9cd50eb04e71..22243ab8e079925d58f739b3078efed04fd8acda 100644
> > > --- a/include/uapi/linux/rkisp1-config.h
> > > +++ b/include/uapi/linux/rkisp1-config.h
> > > @@ -45,6 +45,8 @@
> > > #define RKISP1_CIF_ISP_MODULE_DPF (1U << 16)
> > > /* Denoise Pre-Filter Strength */
> > > #define RKISP1_CIF_ISP_MODULE_DPF_STRENGTH (1U << 17)
> > > +/* Auto White Balancing 64 statistics configuration */
> > > +#define RKISP1_CIF_ISP_MODULE_AWB64 (1U << 18)
> >
> > This doesn't seem to be used, is that expected ?
>
> Ah I don't think this is needed for blocks that are only used with
> extended params API. Will drop after double checking.
>
> > >
> > > #define RKISP1_CIF_ISP_CTK_COEFF_MAX 0x100
> > > #define RKISP1_CIF_ISP_CTK_OFFSET_MAX 0x800
> > > @@ -88,6 +90,11 @@
> > > #define RKISP1_CIF_ISP_AWB_MAX_GRID 1
> > > #define RKISP1_CIF_ISP_AWB_MAX_FRAMES 7
> > >
> > > +/*
> > > + * Automatic white balance extended block (AWB64)
> > > + */
> > > +#define RKISP1_CIF_ISP_AWB64_MAX_ELLIPSE 8
> > > +
> > > /*
> > > * Gamma out
> > > */
> > > @@ -176,6 +183,7 @@
> > > #define RKISP1_CIF_ISP_STAT_AUTOEXP (1U << 1)
> > > #define RKISP1_CIF_ISP_STAT_AFM (1U << 2)
> > > #define RKISP1_CIF_ISP_STAT_HIST (1U << 3)
> > > +#define RKISP1_CIF_ISP_STAT_AWB64 (1U << 4)
> > >
> > > /**
> > > * enum rkisp1_cif_isp_version - ISP variants
> > > @@ -519,6 +527,55 @@ struct rkisp1_cif_isp_awb_gain_config {
> > > __u16 gain_green_b;
> > > };
> > >
> > > +/**
> > > + * struct rkisp1_cif_isp_awb64_ellipse - Ellipse configuration for AWB64 measurement
> > > + *
> > > + * @rmax: Points within rmax (24 bits) distance from center are considered white point
> >
> > What's the unit ? A maximum radius of 16M pixels doesn't seem to make
> > sense. Is this the square of the maximum radius ? Or a fixed-point
> > integer value with a decimal part ?
>
> The documentation just says 24 bits for RMax. My guess is it might be
> some fixed point integer value, but I will have to experiment more to be
> sure.
>
> Same case for the color conversion and ellipse CTM co-efficients. Those
> are most likely fixed point, but the documentation does not mention
> that, and if it is how many bits are the fractional part.
Experimentation is needed. The reason why I mentioned the square of the
maximum radius is that it's easier to compute the square of euclidian
distances in hardware than the distance itself. It's just a guess
though, there may be an optimized way to perform the calculation in the
ISP that would use the distance for comparison.
Could you design a test procedure to try and understand the different
parameters ? We can then discuss how to implement it.
> > > + * @cen_x: X co-ordinate of the center of ellipse (10 bits)
> >
> > s/co-ordinate/coordinate/
> >
> > Same below.
> >
> > > + * @cen_y: Y co-ordinate of the center of ellipse (10 bits)
> > > + * @ctm: Co-ordinate transformation matrix
> > > + * ctm[0] and ctm[2] are 12 bits, ctm[1] and ctm[3] are 9 bits
> > > + */
> > > +struct rkisp1_cif_isp_awb64_ellip {
> > > + __u32 rmax;
> > > + __u16 cen_x;
> > > + __u16 cen_y;
> > > + __u16 ctm[4];
> > > +};
> > > +
> > > +/**
> > > + * struct rkisp1_cif_isp_awb64_meas_config - Configuration for the AWB64 stats
> > > + *
> > > + * @awb_wnd: White balance measurement window (in pixels)
> > > + * @ellip: Ellipse regions used for measurement
> > > + * @cc_coeff: Colorspace matrix (all coefficients are 11 bits)
> > > + * @min_div: Minimum divider, if input is less than this don't do division (10 bits)
> > > + * @max_r: Only pixels with R < max_r contribute to measurement
> > > + * @min_r: Only pixels with R > min_r contribute to measurement
> > > + * @max_g: Only pixels with G < max_g contribute to measurement
> > > + * @min_g: Only pixels with G > min_g contribute to measurement
> > > + * @max_b: Only pixels with B < max_b contribute to measurement
> > > + * @min_b: Only pixels with B > min_b contribute to measurement
> > > + * @ellip_unite: Bitmask to select which regions should be combined for measurement
> > > + * bits 0:2 to combine ellipse 0 with ellipse 1,2,3
> > > + * bits 3:5 to combine ellipse 4 with ellipse 5,6,7
> > > + */
> > > +struct rkisp1_cif_isp_awb64_meas_config {
> > > + struct rkisp1_cif_isp_window awb_wnd;
> > > + struct rkisp1_cif_isp_awb64_ellip ellip[RKISP1_CIF_ISP_AWB64_MAX_ELLIPSE];
> > > + __u16 cc_coeff[3][3];
> > > + __u16 min_div;
> > > + __u8 max_r;
> > > + __u8 min_r;
> > > + __u8 max_g;
> > > + __u8 min_g;
> > > + __u8 max_b;
> > > + __u8 min_b;
> > > + __u8 enable_median_filter;
> >
> > This field is not documented.
> >
> > > + __u8 ellip_unite;
> > > + __u8 chrom_switch;
> >
> > Same here. Please compile-test the documentation.
>
> Will do.
>
> > > +};
> > > +
> > > /**
> > > * struct rkisp1_cif_isp_flt_config - Configuration used by ISP filtering
> > > *
> > > @@ -1006,6 +1063,34 @@ struct rkisp1_cif_isp_hist_stat {
> > > __u32 hist_bins[RKISP1_CIF_ISP_HIST_BIN_N_MAX];
> > > };
> > >
> > > +/**
> > > + * struct rkisp1_cif_isp_awb64_meas - AWB64 measured values
> > > + *
> > > + * @cnt: White pixel count, number of "white pixels" found during last
> > > + * measurement
> > > + * @accu_y_or_g: Total value of Y within elliptical region,
> > > + * Green if RGB is selected.
> > > + * @accu_cb_or_b: Total value of Cb within elliptical region,
> > > + * Blue if RGB is selected.
> > > + * @accu_cr_or_r: Total value of Cr within elliptical region,
> > > + * Red if RGB is selected.
> > > + */
> > > +struct rkisp1_cif_isp_awb64_meas {
> > > + __u32 cnt;
> > > + __u32 accu_y_or_g;
> > > + __u32 accu_cb_or_b;
> > > + __u32 accu_cr_or_r;
> > > +};
> > > +
> > > +/**
> > > + * struct rkisp1_cif_isp_awb64_stat - statistics AWB64 data
> > > + *
> > > + * @count: Measured pixel accumulator data for elliptical regions
> > > + */
> > > +struct rkisp1_cif_isp_awb64_stat {
> > > + struct rkisp1_cif_isp_awb64_meas count[RKISP1_CIF_ISP_AWB64_MAX_ELLIPSE];
> > > +};
> > > +
> > > /**
> > > * struct rkisp1_cif_isp_stat - Rockchip ISP1 Statistics Data
> > > *
> > > @@ -1019,6 +1104,7 @@ struct rkisp1_cif_isp_stat {
> > > struct rkisp1_cif_isp_ae_stat ae;
> > > struct rkisp1_cif_isp_af_stat af;
> > > struct rkisp1_cif_isp_hist_stat hist;
> > > + struct rkisp1_cif_isp_awb64_stat awb64;
> > > };
> > >
> > > /**
> > > @@ -1059,6 +1145,7 @@ struct rkisp1_stat_buffer {
> > > * @RKISP1_EXT_PARAMS_BLOCK_TYPE_COMPAND_BLS: BLS in the compand block
> > > * @RKISP1_EXT_PARAMS_BLOCK_TYPE_COMPAND_EXPAND: Companding expand curve
> > > * @RKISP1_EXT_PARAMS_BLOCK_TYPE_COMPAND_COMPRESS: Companding compress curve
> > > + * @RKISP1_EXT_PARAMS_BLOCK_TYPE_AWB64_MEAS: Auto white balance 64 statistics
> > > */
> > > enum rkisp1_ext_params_block_type {
> > > RKISP1_EXT_PARAMS_BLOCK_TYPE_BLS,
> > > @@ -1081,6 +1168,7 @@ enum rkisp1_ext_params_block_type {
> > > RKISP1_EXT_PARAMS_BLOCK_TYPE_COMPAND_BLS,
> > > RKISP1_EXT_PARAMS_BLOCK_TYPE_COMPAND_EXPAND,
> > > RKISP1_EXT_PARAMS_BLOCK_TYPE_COMPAND_COMPRESS,
> > > + RKISP1_EXT_PARAMS_BLOCK_TYPE_AWB64_MEAS,
> > > };
> > >
> > > #define RKISP1_EXT_PARAMS_FL_BLOCK_DISABLE (1U << 0)
> > > @@ -1376,6 +1464,23 @@ struct rkisp1_ext_params_awb_meas_config {
> > > struct rkisp1_cif_isp_awb_meas_config config;
> > > } __attribute__((aligned(8)));
> > >
> > > +/**
> > > + * struct rkisp1_ext_params_awb64_meas_config - RkISP1 extensible params AWB64
> > > + * Meas config
> > > + *
> > > + * RkISP1 extensible parameters Auto-White Balance 64 Measurement configuration
> > > + * block. Identified by :c:type:`RKISP1_EXT_PARAMS_BLOCK_TYPE_AWB64_MEAS`.
> > > + *
> > > + * @header: The RkISP1 extensible parameters header, see
> > > + * :c:type:`rkisp1_ext_params_block_header`
> > > + * @config: Auto-White Balance 64 measure configuration, see
> > > + * :c:type:`rkisp1_cif_isp_awb64_meas_config`
> > > + */
> > > +struct rkisp1_ext_params_awb64_meas_config {
> > > + struct rkisp1_ext_params_block_header header;
> > > + struct rkisp1_cif_isp_awb64_meas_config config;
> > > +} __attribute__((aligned(8)));
> > > +
> > > /**
> > > * struct rkisp1_ext_params_hst_config - RkISP1 extensible params Histogram config
> > > *
> > >
> > > ---
> > > base-commit: 40384c840ea1944d7c5a392e8975ed088ecf0b37
> > > change-id: 20241230-awb64-79270098e15e
>
> Will fix all the whitespace/indent issues that I didn't explictly ack
> above.
--
Regards,
Laurent Pinchart
Powered by blists - more mailing lists