drm/amd/display: Add user_regamma to color module

Signed-off-by: Krunoslav Kovac <Krunoslav.Kovac@amd.com>
Reviewed-by: Anthony Koo <Anthony.Koo@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/display/modules/color/color_gamma.c

@@ -691,7 +691,7 @@ static void build_degamma(struct pwl_float_data_ex *curve,
 	}
 }
 
-static bool scale_gamma(struct pwl_float_data *pwl_rgb,
+static void scale_gamma(struct pwl_float_data *pwl_rgb,
 		const struct dc_gamma *ramp,
 		struct dividers dividers)
 {
@@ -752,11 +752,9 @@ static bool scale_gamma(struct pwl_float_data *pwl_rgb,
 			dividers.divider3);
 	rgb->b = dal_fixed31_32_mul(rgb_last->b,
 			dividers.divider3);
-
-	return true;
 }
 
-static bool scale_gamma_dx(struct pwl_float_data *pwl_rgb,
+static void scale_gamma_dx(struct pwl_float_data *pwl_rgb,
 		const struct dc_gamma *ramp,
 		struct dividers dividers)
 {
@@ -818,8 +816,71 @@ static bool scale_gamma_dx(struct pwl_float_data *pwl_rgb,
 				pwl_rgb[i-1].g, 2), pwl_rgb[i-2].g);
 	pwl_rgb[i].b =  dal_fixed31_32_sub(dal_fixed31_32_mul_int(
 				pwl_rgb[i-1].b, 2), pwl_rgb[i-2].b);
+}
 
-	return true;
+/* todo: all these scale_gamma functions are inherently the same but
+ *  take different structures as params or different format for ramp
+ *  values. We could probably implement it in a more generic fashion
+ */
+static void scale_user_regamma_ramp(struct pwl_float_data *pwl_rgb,
+		const struct regamma_ramp *ramp,
+		struct dividers dividers)
+{
+	unsigned short max_driver = 0xFFFF;
+	unsigned short max_os = 0xFF00;
+	unsigned short scaler = max_os;
+	uint32_t i;
+	struct pwl_float_data *rgb = pwl_rgb;
+	struct pwl_float_data *rgb_last = rgb + GAMMA_RGB_256_ENTRIES - 1;
+
+	i = 0;
+	do {
+		if (ramp->gamma[i] > max_os ||
+				ramp->gamma[i + 256] > max_os ||
+				ramp->gamma[i + 512] > max_os) {
+			scaler = max_driver;
+			break;
+		}
+		i++;
+	} while (i != GAMMA_RGB_256_ENTRIES);
+
+	i = 0;
+	do {
+		rgb->r = dal_fixed31_32_from_fraction(
+				ramp->gamma[i], scaler);
+		rgb->g = dal_fixed31_32_from_fraction(
+				ramp->gamma[i + 256], scaler);
+		rgb->b = dal_fixed31_32_from_fraction(
+				ramp->gamma[i + 512], scaler);
+
+		++rgb;
+		++i;
+	} while (i != GAMMA_RGB_256_ENTRIES);
+
+	rgb->r = dal_fixed31_32_mul(rgb_last->r,
+			dividers.divider1);
+	rgb->g = dal_fixed31_32_mul(rgb_last->g,
+			dividers.divider1);
+	rgb->b = dal_fixed31_32_mul(rgb_last->b,
+			dividers.divider1);
+
+	++rgb;
+
+	rgb->r = dal_fixed31_32_mul(rgb_last->r,
+			dividers.divider2);
+	rgb->g = dal_fixed31_32_mul(rgb_last->g,
+			dividers.divider2);
+	rgb->b = dal_fixed31_32_mul(rgb_last->b,
+			dividers.divider2);
+
+	++rgb;
+
+	rgb->r = dal_fixed31_32_mul(rgb_last->r,
+			dividers.divider3);
+	rgb->g = dal_fixed31_32_mul(rgb_last->g,
+			dividers.divider3);
+	rgb->b = dal_fixed31_32_mul(rgb_last->b,
+			dividers.divider3);
 }
 
 /*
@@ -949,7 +1010,7 @@ static inline void copy_rgb_regamma_to_coordinates_x(
 	uint32_t i = 0;
 	const struct pwl_float_data_ex *rgb_regamma = rgb_ex;
 
-	while (i <= hw_points_num) {
+	while (i <= hw_points_num + 1) {
 		coords->regamma_y_red = rgb_regamma->r;
 		coords->regamma_y_green = rgb_regamma->g;
 		coords->regamma_y_blue = rgb_regamma->b;
@@ -1002,6 +1063,102 @@ static bool calculate_interpolated_hardware_curve(
 	return true;
 }
 
+/* The "old" interpolation uses a complicated scheme to build an array of
+ * coefficients while also using an array of 0-255 normalized to 0-1
+ * Then there's another loop using both of the above + new scaled user ramp
+ * and we concatenate them. It also searches for points of interpolation and
+ * uses enums for positions.
+ *
+ * This function uses a different approach:
+ * user ramp is always applied on X with 0/255, 1/255, 2/255, ..., 255/255
+ * To find index for hwX , we notice the following:
+ * i/255 <= hwX < (i+1)/255  <=> i <= 255*hwX < i+1
+ * See apply_lut_1d which is the same principle, but on 4K entry 1D LUT
+ *
+ * Once the index is known, combined Y is simply:
+ * user_ramp(index) + (hwX-index/255)*(user_ramp(index+1) - user_ramp(index)
+ *
+ * We should switch to this method in all cases, it's simpler and faster
+ * ToDo one day - for now this only applies to ADL regamma to avoid regression
+ * for regular use cases (sRGB and PQ)
+ */
+static void interpolate_user_regamma(uint32_t hw_points_num,
+		struct pwl_float_data *rgb_user,
+		bool apply_degamma,
+		struct dc_transfer_func_distributed_points *tf_pts)
+{
+	uint32_t i;
+	uint32_t color = 0;
+	int32_t index;
+	int32_t index_next;
+	struct fixed31_32 *tf_point;
+	struct fixed31_32 hw_x;
+	struct fixed31_32 norm_factor =
+			dal_fixed31_32_from_int_nonconst(255);
+	struct fixed31_32 norm_x;
+	struct fixed31_32 index_f;
+	struct fixed31_32 lut1;
+	struct fixed31_32 lut2;
+	struct fixed31_32 delta_lut;
+	struct fixed31_32 delta_index;
+
+	i = 0;
+	/* fixed_pt library has problems handling too small values */
+	while (i != 32) {
+		tf_pts->red[i] = dal_fixed31_32_zero;
+		tf_pts->green[i] = dal_fixed31_32_zero;
+		tf_pts->blue[i] = dal_fixed31_32_zero;
+		++i;
+	}
+	while (i <= hw_points_num + 1) {
+		for (color = 0; color < 3; color++) {
+			if (color == 0)
+				tf_point = &tf_pts->red[i];
+			else if (color == 1)
+				tf_point = &tf_pts->green[i];
+			else
+				tf_point = &tf_pts->blue[i];
+
+			if (apply_degamma) {
+				if (color == 0)
+					hw_x = coordinates_x[i].regamma_y_red;
+				else if (color == 1)
+					hw_x = coordinates_x[i].regamma_y_green;
+				else
+					hw_x = coordinates_x[i].regamma_y_blue;
+			} else
+				hw_x = coordinates_x[i].x;
+
+			norm_x = dal_fixed31_32_mul(norm_factor, hw_x);
+			index = dal_fixed31_32_floor(norm_x);
+			if (index < 0 || index > 255)
+				continue;
+
+			index_f = dal_fixed31_32_from_int_nonconst(index);
+			index_next = (index == 255) ? index : index + 1;
+
+			if (color == 0) {
+				lut1 = rgb_user[index].r;
+				lut2 = rgb_user[index_next].r;
+			} else if (color == 1) {
+				lut1 = rgb_user[index].g;
+				lut2 = rgb_user[index_next].g;
+			} else {
+				lut1 = rgb_user[index].b;
+				lut2 = rgb_user[index_next].b;
+			}
+
+			// we have everything now, so interpolate
+			delta_lut = dal_fixed31_32_sub(lut2, lut1);
+			delta_index = dal_fixed31_32_sub(norm_x, index_f);
+
+			*tf_point = dal_fixed31_32_add(lut1,
+				dal_fixed31_32_mul(delta_index, delta_lut));
+		}
+		++i;
+	}
+}
+
 static void build_new_custom_resulted_curve(
 	uint32_t hw_points_num,
 	struct dc_transfer_func_distributed_points *tf_pts)
@@ -1025,6 +1182,29 @@ static void build_new_custom_resulted_curve(
 	}
 }
 
+static void apply_degamma_for_user_regamma(struct pwl_float_data_ex *rgb_regamma,
+		uint32_t hw_points_num)
+{
+	uint32_t i;
+
+	struct gamma_coefficients coeff;
+	struct pwl_float_data_ex *rgb = rgb_regamma;
+	const struct hw_x_point *coord_x = coordinates_x;
+
+	build_coefficients(&coeff, true);
+
+	i = 0;
+	while (i != hw_points_num + 1) {
+		rgb->r = translate_from_linear_space_ex(
+				coord_x->x, &coeff, 0);
+		rgb->g = rgb->r;
+		rgb->b = rgb->r;
+		++coord_x;
+		++rgb;
+		++i;
+	}
+}
+
 static bool map_regamma_hw_to_x_user(
 	const struct dc_gamma *ramp,
 	struct pixel_gamma_point *coeff128,
@@ -1062,6 +1242,7 @@ static bool map_regamma_hw_to_x_user(
 		}
 	}
 
+	/* this should be named differently, all it does is clamp to 0-1 */
 	build_new_custom_resulted_curve(hw_points_num, tf_pts);
 
 	return true;
@@ -1168,6 +1349,113 @@ bool mod_color_calculate_regamma_params(struct dc_transfer_func *output_tf,
 	return ret;
 }
 
+bool calculate_user_regamma_coeff(struct dc_transfer_func *output_tf,
+		const struct regamma_lut *regamma)
+{
+	struct gamma_coefficients coeff;
+	const struct hw_x_point *coord_x = coordinates_x;
+	uint32_t i = 0;
+
+	do {
+		coeff.a0[i] = dal_fixed31_32_from_fraction(
+				regamma->coeff.A0[i], 10000000);
+		coeff.a1[i] = dal_fixed31_32_from_fraction(
+				regamma->coeff.A1[i], 1000);
+		coeff.a2[i] = dal_fixed31_32_from_fraction(
+				regamma->coeff.A2[i], 1000);
+		coeff.a3[i] = dal_fixed31_32_from_fraction(
+				regamma->coeff.A3[i], 1000);
+		coeff.user_gamma[i] = dal_fixed31_32_from_fraction(
+				regamma->coeff.gamma[i], 1000);
+
+		++i;
+	} while (i != 3);
+
+	i = 0;
+	/* fixed_pt library has problems handling too small values */
+	while (i != 32) {
+		output_tf->tf_pts.red[i] = dal_fixed31_32_zero;
+		output_tf->tf_pts.green[i] = dal_fixed31_32_zero;
+		output_tf->tf_pts.blue[i] = dal_fixed31_32_zero;
+		++coord_x;
+		++i;
+	}
+	while (i != MAX_HW_POINTS + 1) {
+		output_tf->tf_pts.red[i] = translate_from_linear_space_ex(
+				coord_x->x, &coeff, 0);
+		output_tf->tf_pts.green[i] = translate_from_linear_space_ex(
+				coord_x->x, &coeff, 1);
+		output_tf->tf_pts.blue[i] = translate_from_linear_space_ex(
+				coord_x->x, &coeff, 2);
+		++coord_x;
+		++i;
+	}
+
+	// this function just clamps output to 0-1
+	build_new_custom_resulted_curve(MAX_HW_POINTS, &output_tf->tf_pts);
+	output_tf->type = TF_TYPE_DISTRIBUTED_POINTS;
+
+	return true;
+}
+
+bool calculate_user_regamma_ramp(struct dc_transfer_func *output_tf,
+		const struct regamma_lut *regamma)
+{
+	struct dc_transfer_func_distributed_points *tf_pts = &output_tf->tf_pts;
+	struct dividers dividers;
+
+	struct pwl_float_data *rgb_user = NULL;
+	struct pwl_float_data_ex *rgb_regamma = NULL;
+	bool ret = false;
+
+	if (regamma == NULL)
+		return false;
+
+	output_tf->type = TF_TYPE_DISTRIBUTED_POINTS;
+
+	rgb_user = kzalloc(sizeof(*rgb_user) * (GAMMA_RGB_256_ENTRIES + _EXTRA_POINTS),
+			GFP_KERNEL);
+	if (!rgb_user)
+		goto rgb_user_alloc_fail;
+
+	rgb_regamma = kzalloc(sizeof(*rgb_regamma) * (MAX_HW_POINTS + _EXTRA_POINTS),
+			GFP_KERNEL);
+	if (!rgb_regamma)
+		goto rgb_regamma_alloc_fail;
+
+	dividers.divider1 = dal_fixed31_32_from_fraction(3, 2);
+	dividers.divider2 = dal_fixed31_32_from_int(2);
+	dividers.divider3 = dal_fixed31_32_from_fraction(5, 2);
+
+	scale_user_regamma_ramp(rgb_user, &regamma->ramp, dividers);
+
+	if (regamma->flags.bits.applyDegamma == 1) {
+		apply_degamma_for_user_regamma(rgb_regamma, MAX_HW_POINTS);
+		copy_rgb_regamma_to_coordinates_x(coordinates_x,
+				MAX_HW_POINTS, rgb_regamma);
+	}
+
+	interpolate_user_regamma(MAX_HW_POINTS, rgb_user,
+			regamma->flags.bits.applyDegamma, tf_pts);
+
+	// no custom HDR curves!
+	tf_pts->end_exponent = 0;
+	tf_pts->x_point_at_y1_red = 1;
+	tf_pts->x_point_at_y1_green = 1;
+	tf_pts->x_point_at_y1_blue = 1;
+
+	// this function just clamps output to 0-1
+	build_new_custom_resulted_curve(MAX_HW_POINTS, tf_pts);
+
+	ret = true;
+
+	kfree(rgb_regamma);
+rgb_regamma_alloc_fail:
+	kfree(rgb_user);
+rgb_user_alloc_fail:
+	return ret;
+}
+
 bool mod_color_calculate_degamma_params(struct dc_transfer_func *input_tf,
 		const struct dc_gamma *ramp, bool mapUserRamp)
 {

drivers/gpu/drm/amd/display/modules/color/color_gamma.h

@@ -32,6 +32,47 @@ struct dc_transfer_func_distributed_points;
 struct dc_rgb_fixed;
 enum dc_transfer_func_predefined;
 
+/* For SetRegamma ADL interface support
+ * Must match escape type
+ */
+union regamma_flags {
+	unsigned int raw;
+	struct {
+		unsigned int gammaRampArray       :1;    // RegammaRamp is in use
+		unsigned int gammaFromEdid        :1;    //gamma from edid is in use
+		unsigned int gammaFromEdidEx      :1;    //gamma from edid is in use , but only for Display Id 1.2
+		unsigned int gammaFromUser        :1;    //user custom gamma is used
+		unsigned int coeffFromUser        :1;    //coeff. A0-A3 from user is in use
+		unsigned int coeffFromEdid        :1;    //coeff. A0-A3 from edid is in use
+		unsigned int applyDegamma         :1;    //flag for additional degamma correction in driver
+		unsigned int gammaPredefinedSRGB  :1;    //flag for SRGB gamma
+		unsigned int gammaPredefinedPQ    :1;    //flag for PQ gamma
+		unsigned int gammaPredefinedPQ2084Interim :1;    //flag for PQ gamma, lower max nits
+		unsigned int gammaPredefined36    :1;    //flag for 3.6 gamma
+		unsigned int gammaPredefinedReset :1;    //flag to return to previous gamma
+	} bits;
+};
+
+struct regamma_ramp {
+	unsigned short gamma[256*3];  // gamma ramp packed  in same way as OS windows ,r , g & b
+};
+
+struct regamma_coeff {
+	int    gamma[3];
+	int    A0[3];
+	int    A1[3];
+	int    A2[3];
+	int    A3[3];
+};
+
+struct regamma_lut {
+	union regamma_flags flags;
+	union {
+		struct regamma_ramp ramp;
+		struct regamma_coeff coeff;
+	};
+};
+
 void setup_x_points_distribution(void);
 void precompute_pq(void);
 void precompute_de_pq(void);
@@ -48,6 +89,11 @@ bool mod_color_calculate_curve(enum dc_transfer_func_predefined  trans,
 bool mod_color_calculate_degamma_curve(enum dc_transfer_func_predefined trans,
 				struct dc_transfer_func_distributed_points *points);
 
+bool calculate_user_regamma_coeff(struct dc_transfer_func *output_tf,
+		const struct regamma_lut *regamma);
+
+bool calculate_user_regamma_ramp(struct dc_transfer_func *output_tf,
+		const struct regamma_lut *regamma);
 
 
 #endif /* COLOR_MOD_COLOR_GAMMA_H_ */