[PATCH 21/33] drm/amd/display: Fix max brightness pixel accuracy
Aurabindo Pillai
aurabindo.pillai at amd.com
Wed Oct 21 14:22:45 UTC 2020
From: Clark <felclark at amd.com>
[WHY]
It was detected in some Freesync HDR tests that displays were not
reaching their maximum nominal brightness.
[HOW]
The Multi-plane combiner (MPC) Output Gamma (OGAM) block builds a
discrete Lookup Table (LUT). When the display's maximum brightness
falls in between two values, having to be linearly interpolated by
the hardware, rounding issues might occur that will cause the
display to never reach its maximum brightness.
The fix involves doing the calculations backwards, ensuring that
the interpolation in the maximum brightness values translates to an
output of 1.0.
Signed-off-by: Felipe Clark <Felipe.Clark at amd.com>
Acked-by: Aurabindo Pillai <aurabindo.pillai at amd.com>
---
.../amd/display/modules/color/color_gamma.c | 110 ++++++++++++++----
1 file changed, 89 insertions(+), 21 deletions(-)
diff --git a/drivers/gpu/drm/amd/display/modules/color/color_gamma.c b/drivers/gpu/drm/amd/display/modules/color/color_gamma.c
index b8695660b480..e866da639637 100644
--- a/drivers/gpu/drm/amd/display/modules/color/color_gamma.c
+++ b/drivers/gpu/drm/amd/display/modules/color/color_gamma.c
@@ -151,7 +151,7 @@ static void compute_de_pq(struct fixed31_32 in_x, struct fixed31_32 *out_y)
div = dc_fixpt_sub(c2, dc_fixpt_mul(c3, l_pow_m1));
base2 = dc_fixpt_div(base, div);
- //avoid complex numbers
+ // avoid complex numbers
if (dc_fixpt_lt(base2, dc_fixpt_zero))
base2 = dc_fixpt_sub(dc_fixpt_zero, base2);
@@ -161,7 +161,7 @@ static void compute_de_pq(struct fixed31_32 in_x, struct fixed31_32 *out_y)
}
-/*de gamma, none linear to linear*/
+/* de gamma, non-linear to linear */
static void compute_hlg_eotf(struct fixed31_32 in_x,
struct fixed31_32 *out_y,
uint32_t sdr_white_level, uint32_t max_luminance_nits)
@@ -193,7 +193,7 @@ static void compute_hlg_eotf(struct fixed31_32 in_x,
}
-/*re gamma, linear to none linear*/
+/* re gamma, linear to non-linear */
static void compute_hlg_oetf(struct fixed31_32 in_x, struct fixed31_32 *out_y,
uint32_t sdr_white_level, uint32_t max_luminance_nits)
{
@@ -830,7 +830,7 @@ static bool build_regamma(struct pwl_float_data_ex *rgb_regamma,
i = 0;
while (i <= hw_points_num) {
- /*TODO use y vs r,g,b*/
+ /* TODO use y vs r,g,b */
rgb->r = translate_from_linear_space_ex(
coord_x->x, coeff, 0, cal_buffer);
rgb->g = rgb->r;
@@ -937,6 +937,7 @@ static bool build_freesync_hdr(struct pwl_float_data_ex *rgb_regamma,
uint32_t i;
struct pwl_float_data_ex *rgb = rgb_regamma;
const struct hw_x_point *coord_x = coordinate_x;
+ const struct hw_x_point *prv_coord_x = coord_x;
struct fixed31_32 scaledX = dc_fixpt_zero;
struct fixed31_32 scaledX1 = dc_fixpt_zero;
struct fixed31_32 max_display;
@@ -947,6 +948,9 @@ static bool build_freesync_hdr(struct pwl_float_data_ex *rgb_regamma,
bool use_eetf = false;
bool is_clipped = false;
struct fixed31_32 sdr_white_level;
+ struct fixed31_32 coordX_diff;
+ struct fixed31_32 out_dist_max;
+ struct fixed31_32 bright_norm;
if (fs_params->max_content == 0 ||
fs_params->max_display == 0)
@@ -975,7 +979,7 @@ static bool build_freesync_hdr(struct pwl_float_data_ex *rgb_regamma,
for (i = 32; i <= hw_points_num; i++) {
if (!is_clipped) {
if (use_eetf) {
- /*max content is equal 1 */
+ /* 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,
@@ -990,21 +994,65 @@ static bool build_freesync_hdr(struct pwl_float_data_ex *rgb_regamma,
else
output = calculate_gamma22(scaledX, use_eetf, cal_buffer);
+ // Ensure output respects reasonable boundaries
+ output = dc_fixpt_clamp(output, dc_fixpt_zero, dc_fixpt_one);
+
rgb->r = output;
rgb->g = output;
rgb->b = output;
} else {
+ /* Here clipping happens for the first time */
is_clipped = true;
- rgb->r = clip;
- rgb->g = clip;
- rgb->b = clip;
+
+ /* The next few lines implement the equation
+ * output = prev_out +
+ * (coord_x->x - prev_coord_x->x) *
+ * (1.0 - prev_out) /
+ * (maxDisp/sdr_white_level - prevCoordX)
+ *
+ * This equation interpolates the first point
+ * after max_display/80 so that the slope from
+ * hw_x_before_max and hw_x_after_max is such
+ * that we hit Y=1.0 at max_display/80.
+ */
+
+ coordX_diff = dc_fixpt_sub(coord_x->x, prv_coord_x->x);
+ out_dist_max = dc_fixpt_sub(dc_fixpt_one, output);
+ bright_norm = dc_fixpt_div(max_display, sdr_white_level);
+
+ output = dc_fixpt_add(
+ output, dc_fixpt_mul(
+ coordX_diff, dc_fixpt_div(
+ out_dist_max,
+ dc_fixpt_sub(bright_norm, prv_coord_x->x)
+ )
+ )
+ );
+
+ /* Relaxing the maximum boundary to 1.07 (instead of 1.0)
+ * because the last point in the curve must be such that
+ * the maximum display pixel brightness interpolates to
+ * exactly 1.0. The worst case scenario was calculated
+ * around 1.057, so the limit of 1.07 leaves some safety
+ * margin.
+ */
+ output = dc_fixpt_clamp(output, dc_fixpt_zero,
+ dc_fixpt_from_fraction(107, 100));
+
+ rgb->r = output;
+ rgb->g = output;
+ rgb->b = output;
}
} else {
+ /* Every other clipping after the first
+ * one is dealt with here
+ */
rgb->r = clip;
rgb->g = clip;
rgb->b = clip;
}
+ prv_coord_x = coord_x;
++coord_x;
++rgb;
}
@@ -1073,7 +1121,7 @@ static void build_hlg_degamma(struct pwl_float_data_ex *degamma,
const struct hw_x_point *coord_x = coordinate_x;
i = 0;
- //check when i == 434
+ // check when i == 434
while (i != hw_points_num + 1) {
compute_hlg_eotf(coord_x->x, &rgb->r, sdr_white_level, max_luminance_nits);
rgb->g = rgb->r;
@@ -1097,7 +1145,7 @@ static void build_hlg_regamma(struct pwl_float_data_ex *regamma,
i = 0;
- //when i == 471
+ // when i == 471
while (i != hw_points_num + 1) {
compute_hlg_oetf(coord_x->x, &rgb->r, sdr_white_level, max_luminance_nits);
rgb->g = rgb->r;
@@ -1331,6 +1379,8 @@ static void apply_lut_1d(
struct fixed31_32 lut1;
struct fixed31_32 lut2;
const int max_lut_index = 4095;
+ const struct fixed31_32 penult_lut_index_f =
+ dc_fixpt_from_int(max_lut_index-1);
const struct fixed31_32 max_lut_index_f =
dc_fixpt_from_int(max_lut_index);
int32_t index = 0, index_next = 0;
@@ -1355,10 +1405,21 @@ static void apply_lut_1d(
index = dc_fixpt_floor(norm_y);
index_f = dc_fixpt_from_int(index);
- if (index < 0 || index > max_lut_index)
+ if (index < 0)
continue;
- index_next = (index == max_lut_index) ? index : index+1;
+ if (index <= max_lut_index)
+ index_next = (index == max_lut_index) ? index : index+1;
+ else {
+ /* Here we are dealing with the last point in the curve,
+ * which in some cases might exceed the range given by
+ * max_lut_index. So we interpolate the value using
+ * max_lut_index and max_lut_index - 1.
+ */
+ index = max_lut_index - 1;
+ index_next = max_lut_index;
+ index_f = penult_lut_index_f;
+ }
if (color == 0) {
lut1 = ramp->entries.red[index];
@@ -1586,9 +1647,7 @@ static void build_new_custom_resulted_curve(
uint32_t hw_points_num,
struct dc_transfer_func_distributed_points *tf_pts)
{
- uint32_t i;
-
- i = 0;
+ uint32_t i = 0;
while (i != hw_points_num + 1) {
tf_pts->red[i] = dc_fixpt_clamp(
@@ -1637,7 +1696,8 @@ static bool map_regamma_hw_to_x_user(
const struct pwl_float_data_ex *rgb_regamma,
uint32_t hw_points_num,
struct dc_transfer_func_distributed_points *tf_pts,
- bool mapUserRamp)
+ bool mapUserRamp,
+ bool doClamping)
{
/* setup to spare calculated ideal regamma values */
@@ -1665,8 +1725,10 @@ 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);
+ if (doClamping) {
+ /* 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;
}
@@ -1914,11 +1976,12 @@ bool mod_color_calculate_degamma_params(struct dc_color_caps *dc_caps,
++i;
}
} else {
- //clamps to 0-1
+ // clamps to 0-1
map_regamma_hw_to_x_user(ramp, coeff, rgb_user,
coordinates_x, axis_x, curve,
MAX_HW_POINTS, tf_pts,
- mapUserRamp && ramp && ramp->type == GAMMA_RGB_256);
+ mapUserRamp && ramp && ramp->type == GAMMA_RGB_256,
+ true);
}
@@ -2034,6 +2097,7 @@ bool mod_color_calculate_regamma_params(struct dc_transfer_func *output_tf,
struct gamma_pixel *axis_x = NULL;
struct pixel_gamma_point *coeff = NULL;
enum dc_transfer_func_predefined tf = TRANSFER_FUNCTION_SRGB;
+ bool doClamping = true;
bool ret = false;
if (output_tf->type == TF_TYPE_BYPASS)
@@ -2100,11 +2164,15 @@ bool mod_color_calculate_regamma_params(struct dc_transfer_func *output_tf,
cal_buffer);
if (ret) {
+ doClamping = !(output_tf->tf == TRANSFER_FUNCTION_GAMMA22 &&
+ fs_params != NULL && fs_params->skip_tm == 0);
+
map_regamma_hw_to_x_user(ramp, coeff, rgb_user,
coordinates_x, axis_x, rgb_regamma,
MAX_HW_POINTS, tf_pts,
(mapUserRamp || (ramp && ramp->type != GAMMA_RGB_256)) &&
- (ramp && ramp->type != GAMMA_CS_TFM_1D));
+ (ramp && ramp->type != GAMMA_CS_TFM_1D),
+ doClamping);
if (ramp && ramp->type == GAMMA_CS_TFM_1D)
apply_lut_1d(ramp, MAX_HW_POINTS, tf_pts);
--
2.25.1
More information about the amd-gfx
mailing list