drm/amd/display: Add support for Freesync 2 HDR and Content to Display Mapping

[Why]
Freesync 2 HDR and support for HDR content
outside the range of the HDR display
require implementation on Dal 3 to better match
Dal2.

[How]
Add support for Freesync HDR and mapping
of source content to display ranges for better
representation of HDR content.
Signed-off-by: SivapiriyanKumarasamy <sivapiriyan.kumarasamy@amd.com>
Reviewed-by: Anthony Koo <Anthony.Koo@amd.com>
Acked-by: Bhawanpreet Lakha <Bhawanpreet.Lakha@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c

@@ -164,7 +164,7 @@ int amdgpu_dm_set_regamma_lut(struct dm_crtc_state *crtc)
 	 */
 	stream->out_transfer_func->type = TF_TYPE_DISTRIBUTED_POINTS;
 	ret = mod_color_calculate_regamma_params(stream->out_transfer_func,
-						 gamma, true, adev->asic_type <= CHIP_RAVEN);
+						 gamma, true, adev->asic_type <= CHIP_RAVEN, NULL);
 	dc_gamma_release(&gamma);
 	if (!ret) {
 		stream->out_transfer_func->type = old_type;

drivers/gpu/drm/amd/display/dc/dcn10/dcn10_cm_common.c

@@ -268,7 +268,7 @@ bool cm_helper_translate_curve_to_hw_format(
 	memset(lut_params, 0, sizeof(struct pwl_params));
 	memset(seg_distr, 0, sizeof(seg_distr));
 
-	if (output_tf->tf == TRANSFER_FUNCTION_PQ) {
+	if (output_tf->tf == TRANSFER_FUNCTION_PQ || output_tf->tf == TRANSFER_FUNCTION_GAMMA22) {
 		/* 32 segments
 		 * segments are from 2^-25 to 2^7
 		 */

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

@@ -306,6 +306,18 @@ static struct fixed31_32 translate_from_linear_space(
 			a1);
 }
 
+static struct fixed31_32 calculate_gamma22(struct fixed31_32 arg)
+{
+	struct fixed31_32 gamma = dc_fixpt_from_fraction(22, 10);
+
+	return translate_from_linear_space(arg,
+			dc_fixpt_zero,
+			dc_fixpt_zero,
+			dc_fixpt_zero,
+			dc_fixpt_zero,
+			gamma);
+}
+
 static struct fixed31_32 translate_to_linear_space(
 	struct fixed31_32 arg,
 	struct fixed31_32 a0,
@@ -709,6 +721,160 @@ static void build_regamma(struct pwl_float_data_ex *rgb_regamma,
 	}
 }
 
+static void hermite_spline_eetf(struct fixed31_32 input_x,
+				struct fixed31_32 max_display,
+				struct fixed31_32 min_display,
+				struct fixed31_32 max_content,
+				struct fixed31_32 *out_x)
+{
+	struct fixed31_32 min_lum_pq;
+	struct fixed31_32 max_lum_pq;
+	struct fixed31_32 max_content_pq;
+	struct fixed31_32 ks;
+	struct fixed31_32 E1;
+	struct fixed31_32 E2;
+	struct fixed31_32 E3;
+	struct fixed31_32 t;
+	struct fixed31_32 t2;
+	struct fixed31_32 t3;
+	struct fixed31_32 two;
+	struct fixed31_32 three;
+	struct fixed31_32 temp1;
+	struct fixed31_32 temp2;
+	struct fixed31_32 a = dc_fixpt_from_fraction(15, 10);
+	struct fixed31_32 b = dc_fixpt_from_fraction(5, 10);
+	struct fixed31_32 epsilon = dc_fixpt_from_fraction(1, 1000000); // dc_fixpt_epsilon is a bit too small
+
+	if (dc_fixpt_eq(max_content, dc_fixpt_zero)) {
+		*out_x = dc_fixpt_zero;
+		return;
+	}
+
+	compute_pq(input_x, &E1);
+	compute_pq(dc_fixpt_div(min_display, max_content), &min_lum_pq);
+	compute_pq(dc_fixpt_div(max_display, max_content), &max_lum_pq);
+	compute_pq(dc_fixpt_one, &max_content_pq); // always 1? DAL2 code is weird
+	a = dc_fixpt_div(dc_fixpt_add(dc_fixpt_one, b), max_content_pq); // (1+b)/maxContent
+	ks = dc_fixpt_sub(dc_fixpt_mul(a, max_lum_pq), b); // a * max_lum_pq - b
+
+	if (dc_fixpt_lt(E1, ks))
+		E2 = E1;
+	else if (dc_fixpt_le(ks, E1) && dc_fixpt_le(E1, dc_fixpt_one)) {
+		if (dc_fixpt_lt(epsilon, dc_fixpt_sub(dc_fixpt_one, ks)))
+			// t = (E1 - ks) / (1 - ks)
+			t = dc_fixpt_div(dc_fixpt_sub(E1, ks),
+					dc_fixpt_sub(dc_fixpt_one, ks));
+		else
+			t = dc_fixpt_zero;
+
+		two = dc_fixpt_from_int(2);
+		three = dc_fixpt_from_int(3);
+
+		t2 = dc_fixpt_mul(t, t);
+		t3 = dc_fixpt_mul(t2, t);
+		temp1 = dc_fixpt_mul(two, t3);
+		temp2 = dc_fixpt_mul(three, t2);
+
+		// (2t^3 - 3t^2 + 1) * ks
+		E2 = dc_fixpt_mul(ks, dc_fixpt_add(dc_fixpt_one,
+				dc_fixpt_sub(temp1, temp2)));
+
+		// (-2t^3 + 3t^2) * max_lum_pq
+		E2 = dc_fixpt_add(E2, dc_fixpt_mul(max_lum_pq,
+				dc_fixpt_sub(temp2, temp1)));
+
+		temp1 = dc_fixpt_mul(two, t2);
+		temp2 = dc_fixpt_sub(dc_fixpt_one, ks);
+
+		// (t^3 - 2t^2 + t) * (1-ks)
+		E2 = dc_fixpt_add(E2, dc_fixpt_mul(temp2,
+				dc_fixpt_add(t, dc_fixpt_sub(t3, temp1))));
+	}
+
+	temp1 = dc_fixpt_sub(dc_fixpt_one, E2);
+	temp2 = dc_fixpt_mul(temp1, temp1);
+	temp2 = dc_fixpt_mul(temp2, temp2);
+	// temp2 = (1-E2)^4
+
+	E3 =  dc_fixpt_add(E2, dc_fixpt_mul(min_lum_pq, temp2));
+	compute_de_pq(E3, out_x);
+
+	*out_x = dc_fixpt_div(*out_x, dc_fixpt_div(max_display, max_content));
+}
+
+static bool build_freesync_hdr(struct pwl_float_data_ex *rgb_regamma,
+		uint32_t hw_points_num,
+		const struct hw_x_point *coordinate_x,
+		const struct freesync_hdr_tf_params *fs_params)
+{
+	uint32_t i;
+	struct pwl_float_data_ex *rgb = rgb_regamma;
+	const struct hw_x_point *coord_x = coordinate_x;
+	struct fixed31_32 scaledX = dc_fixpt_zero;
+	struct fixed31_32 scaledX1 = dc_fixpt_zero;
+	struct fixed31_32 max_display = dc_fixpt_from_int(fs_params->max_display);
+	struct fixed31_32 min_display = dc_fixpt_from_fraction(fs_params->min_display, 10000);
+	struct fixed31_32 max_content = dc_fixpt_from_int(fs_params->max_content);
+	struct fixed31_32 min_content = dc_fixpt_from_fraction(fs_params->min_content, 10000);
+	struct fixed31_32 clip = dc_fixpt_one;
+	struct fixed31_32 output;
+	bool use_eetf = false;
+	struct fixed31_32 sdr_white_level = dc_fixpt_from_int(fs_params->sdr_white_level);
+
+	if (fs_params == NULL || fs_params->max_content == 0 ||
+			fs_params->max_display == 0)
+		return false;
+
+	if (fs_params->min_display > 1000) // cap at 0.1 at the bottom
+		min_display = dc_fixpt_from_fraction(1, 10);
+	if (fs_params->max_display < 100) // cap at 100 at the top
+		max_display = dc_fixpt_from_int(100);
+
+	if (fs_params->min_content < fs_params->min_display)
+		use_eetf = true;
+	else
+		min_content = min_display;
+
+	if (fs_params->max_content > fs_params->max_display)
+		use_eetf = true;
+	else
+		max_content = max_display;
+
+	rgb += 32; // first 32 points have problems with fixed point, too small
+	coord_x += 32;
+	for (i = 32; i <= hw_points_num; i++) {
+		if (use_eetf) {
+			/*max content is equal 1 */
+			scaledX1 = dc_fixpt_div(coord_x->x,
+					dc_fixpt_div(max_content, sdr_white_level));
+			hermite_spline_eetf(scaledX1, max_display, min_display,
+					max_content, &scaledX);
+		} else
+			scaledX = dc_fixpt_div(coord_x->x,
+					dc_fixpt_div(max_display, sdr_white_level));
+
+		if (dc_fixpt_lt(scaledX, clip)) {
+			if (dc_fixpt_lt(scaledX, dc_fixpt_zero))
+				output = dc_fixpt_zero;
+			else
+				output = calculate_gamma22(scaledX);
+
+			rgb->r = output;
+			rgb->g = output;
+			rgb->b = output;
+		} else {
+			rgb->r = clip;
+			rgb->g = clip;
+			rgb->b = clip;
+		}
+
+		++coord_x;
+		++rgb;
+	}
+
+	return true;
+}
+
 static void build_degamma(struct pwl_float_data_ex *curve,
 		uint32_t hw_points_num,
 		const struct hw_x_point *coordinate_x, bool is_2_4)
@@ -1356,7 +1522,8 @@ static bool map_regamma_hw_to_x_user(
 #define _EXTRA_POINTS 3
 
 bool mod_color_calculate_regamma_params(struct dc_transfer_func *output_tf,
-		const struct dc_gamma *ramp, bool mapUserRamp, bool canRomBeUsed)
+		const struct dc_gamma *ramp, bool mapUserRamp, bool canRomBeUsed,
+		const struct freesync_hdr_tf_params *fs_params)
 {
 	struct dc_transfer_func_distributed_points *tf_pts = &output_tf->tf_pts;
 	struct dividers dividers;
@@ -1424,6 +1591,12 @@ bool mod_color_calculate_regamma_params(struct dc_transfer_func *output_tf,
 				MAX_HW_POINTS,
 				coordinates_x,
 				output_tf->sdr_ref_white_level);
+	} else if (tf == TRANSFER_FUNCTION_GAMMA22 &&
+			fs_params != NULL) {
+		build_freesync_hdr(rgb_regamma,
+				MAX_HW_POINTS,
+				coordinates_x,
+				fs_params);
 	} else {
 		tf_pts->end_exponent = 0;
 		tf_pts->x_point_at_y1_red = 1;

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

@@ -73,12 +73,21 @@ struct regamma_lut {
 	};
 };
 
+struct freesync_hdr_tf_params {
+	unsigned int sdr_white_level;
+	unsigned int min_content; // luminance in 1/10000 nits
+	unsigned int max_content; // luminance in nits
+	unsigned int min_display; // luminance in 1/10000 nits
+	unsigned int max_display; // luminance in nits
+};
+
 void setup_x_points_distribution(void);
 void precompute_pq(void);
 void precompute_de_pq(void);
 
 bool mod_color_calculate_regamma_params(struct dc_transfer_func *output_tf,
-		const struct dc_gamma *ramp, bool mapUserRamp, bool canRomBeUsed);
+		const struct dc_gamma *ramp, bool mapUserRamp, bool canRomBeUsed,
+		const struct freesync_hdr_tf_params *fs_params);
 
 bool mod_color_calculate_degamma_params(struct dc_transfer_func *output_tf,
 		const struct dc_gamma *ramp, bool mapUserRamp);