2 * Copyright 2020 Advanced Micro Devices, Inc.
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
26 #include "dm_services.h"
27 #include "core_types.h"
28 #include "reg_helper.h"
29 #include "dcn30_dpp.h"
30 #include "basics/conversion.h"
31 #include "dcn30_cm_common.h"
32 #include "custom_float.h"
40 #define FN(reg_name, field_name) \
41 reg->shifts.field_name, reg->masks.field_name
43 void cm_helper_program_gamcor_xfer_func(
44 struct dc_context *ctx,
45 const struct pwl_params *params,
46 const struct dcn3_xfer_func_reg *reg)
48 uint32_t reg_region_cur;
51 REG_SET_2(reg->start_cntl_b, 0,
52 exp_region_start, params->corner_points[0].blue.custom_float_x,
53 exp_resion_start_segment, 0);
54 REG_SET_2(reg->start_cntl_g, 0,
55 exp_region_start, params->corner_points[0].green.custom_float_x,
56 exp_resion_start_segment, 0);
57 REG_SET_2(reg->start_cntl_r, 0,
58 exp_region_start, params->corner_points[0].red.custom_float_x,
59 exp_resion_start_segment, 0);
61 REG_SET(reg->start_slope_cntl_b, 0, //linear slope at start of curve
62 field_region_linear_slope, params->corner_points[0].blue.custom_float_slope);
63 REG_SET(reg->start_slope_cntl_g, 0,
64 field_region_linear_slope, params->corner_points[0].green.custom_float_slope);
65 REG_SET(reg->start_slope_cntl_r, 0,
66 field_region_linear_slope, params->corner_points[0].red.custom_float_slope);
68 REG_SET(reg->start_end_cntl1_b, 0,
69 field_region_end_base, params->corner_points[1].blue.custom_float_y);
70 REG_SET(reg->start_end_cntl1_g, 0,
71 field_region_end_base, params->corner_points[1].green.custom_float_y);
72 REG_SET(reg->start_end_cntl1_r, 0,
73 field_region_end_base, params->corner_points[1].red.custom_float_y);
75 REG_SET_2(reg->start_end_cntl2_b, 0,
76 field_region_end_slope, params->corner_points[1].blue.custom_float_slope,
77 field_region_end, params->corner_points[1].blue.custom_float_x);
78 REG_SET_2(reg->start_end_cntl2_g, 0,
79 field_region_end_slope, params->corner_points[1].green.custom_float_slope,
80 field_region_end, params->corner_points[1].green.custom_float_x);
81 REG_SET_2(reg->start_end_cntl2_r, 0,
82 field_region_end_slope, params->corner_points[1].red.custom_float_slope,
83 field_region_end, params->corner_points[1].red.custom_float_x);
85 for (reg_region_cur = reg->region_start;
86 reg_region_cur <= reg->region_end;
89 const struct gamma_curve *curve0 = &(params->arr_curve_points[2 * i]);
90 const struct gamma_curve *curve1 = &(params->arr_curve_points[(2 * i) + 1]);
92 REG_SET_4(reg_region_cur, 0,
93 exp_region0_lut_offset, curve0->offset,
94 exp_region0_num_segments, curve0->segments_num,
95 exp_region1_lut_offset, curve1->offset,
96 exp_region1_num_segments, curve1->segments_num);
102 /* driver uses 32 regions or less, but DCN HW has 34, extra 2 are set to 0 */
103 #define MAX_REGIONS_NUMBER 34
104 #define MAX_LOW_POINT 25
105 #define NUMBER_REGIONS 32
106 #define NUMBER_SW_SEGMENTS 16
108 bool cm3_helper_translate_curve_to_hw_format(
109 const struct dc_transfer_func *output_tf,
110 struct pwl_params *lut_params, bool fixpoint)
112 struct curve_points3 *corner_points;
113 struct pwl_result_data *rgb_resulted;
114 struct pwl_result_data *rgb;
115 struct pwl_result_data *rgb_plus_1;
116 struct pwl_result_data *rgb_minus_1;
118 int32_t region_start, region_end;
120 uint32_t j, k, seg_distr[MAX_REGIONS_NUMBER], increment, start_index, hw_points;
122 if (output_tf == NULL || lut_params == NULL || output_tf->type == TF_TYPE_BYPASS)
125 corner_points = lut_params->corner_points;
126 rgb_resulted = lut_params->rgb_resulted;
129 memset(lut_params, 0, sizeof(struct pwl_params));
130 memset(seg_distr, 0, sizeof(seg_distr));
132 if (output_tf->tf == TRANSFER_FUNCTION_PQ || output_tf->tf == TRANSFER_FUNCTION_GAMMA22 ||
133 output_tf->tf == TRANSFER_FUNCTION_HLG) {
135 * segments are from 2^-25 to 2^7
137 for (i = 0; i < NUMBER_REGIONS ; i++)
140 region_start = -MAX_LOW_POINT;
141 region_end = NUMBER_REGIONS - MAX_LOW_POINT;
144 * segment is from 2^-10 to 2^0
145 * There are less than 256 points, for optimization
163 for (i = region_end - region_start; i < MAX_REGIONS_NUMBER ; i++)
166 for (k = 0; k < MAX_REGIONS_NUMBER; k++) {
167 if (seg_distr[k] != -1)
168 hw_points += (1 << seg_distr[k]);
172 for (k = 0; k < (region_end - region_start); k++) {
173 increment = NUMBER_SW_SEGMENTS / (1 << seg_distr[k]);
174 start_index = (region_start + k + MAX_LOW_POINT) *
176 for (i = start_index; i < start_index + NUMBER_SW_SEGMENTS;
180 rgb_resulted[j].red = output_tf->tf_pts.red[i];
181 rgb_resulted[j].green = output_tf->tf_pts.green[i];
182 rgb_resulted[j].blue = output_tf->tf_pts.blue[i];
188 start_index = (region_end + MAX_LOW_POINT) * NUMBER_SW_SEGMENTS;
189 rgb_resulted[hw_points].red = output_tf->tf_pts.red[start_index];
190 rgb_resulted[hw_points].green = output_tf->tf_pts.green[start_index];
191 rgb_resulted[hw_points].blue = output_tf->tf_pts.blue[start_index];
193 rgb_resulted[hw_points+1].red = rgb_resulted[hw_points].red;
194 rgb_resulted[hw_points+1].green = rgb_resulted[hw_points].green;
195 rgb_resulted[hw_points+1].blue = rgb_resulted[hw_points].blue;
197 // All 3 color channels have same x
198 corner_points[0].red.x = dc_fixpt_pow(dc_fixpt_from_int(2),
199 dc_fixpt_from_int(region_start));
200 corner_points[0].green.x = corner_points[0].red.x;
201 corner_points[0].blue.x = corner_points[0].red.x;
203 corner_points[1].red.x = dc_fixpt_pow(dc_fixpt_from_int(2),
204 dc_fixpt_from_int(region_end));
205 corner_points[1].green.x = corner_points[1].red.x;
206 corner_points[1].blue.x = corner_points[1].red.x;
208 corner_points[0].red.y = rgb_resulted[0].red;
209 corner_points[0].green.y = rgb_resulted[0].green;
210 corner_points[0].blue.y = rgb_resulted[0].blue;
212 corner_points[0].red.slope = dc_fixpt_div(corner_points[0].red.y,
213 corner_points[0].red.x);
214 corner_points[0].green.slope = dc_fixpt_div(corner_points[0].green.y,
215 corner_points[0].green.x);
216 corner_points[0].blue.slope = dc_fixpt_div(corner_points[0].blue.y,
217 corner_points[0].blue.x);
219 /* see comment above, m_arrPoints[1].y should be the Y value for the
220 * region end (m_numOfHwPoints), not last HW point(m_numOfHwPoints - 1)
222 corner_points[1].red.y = rgb_resulted[hw_points].red;
223 corner_points[1].green.y = rgb_resulted[hw_points].green;
224 corner_points[1].blue.y = rgb_resulted[hw_points].blue;
225 corner_points[1].red.slope = dc_fixpt_zero;
226 corner_points[1].green.slope = dc_fixpt_zero;
227 corner_points[1].blue.slope = dc_fixpt_zero;
229 // DCN3+ have 257 pts in lieu of no separate slope registers
230 // Prior HW had 256 base+slope pairs
231 lut_params->hw_points_num = hw_points + 1;
234 for (i = 1; i < MAX_REGIONS_NUMBER; i++) {
235 if (seg_distr[k] != -1) {
236 lut_params->arr_curve_points[k].segments_num =
238 lut_params->arr_curve_points[i].offset =
239 lut_params->arr_curve_points[k].offset + (1 << seg_distr[k]);
244 if (seg_distr[k] != -1)
245 lut_params->arr_curve_points[k].segments_num = seg_distr[k];
248 rgb_plus_1 = rgb_resulted + 1;
251 if (fixpoint == true) {
253 while (i != hw_points + 2) {
254 if (i >= hw_points) {
255 if (dc_fixpt_lt(rgb_plus_1->red, rgb->red))
256 rgb_plus_1->red = dc_fixpt_add(rgb->red,
257 rgb_minus_1->delta_red);
258 if (dc_fixpt_lt(rgb_plus_1->green, rgb->green))
259 rgb_plus_1->green = dc_fixpt_add(rgb->green,
260 rgb_minus_1->delta_green);
261 if (dc_fixpt_lt(rgb_plus_1->blue, rgb->blue))
262 rgb_plus_1->blue = dc_fixpt_add(rgb->blue,
263 rgb_minus_1->delta_blue);
266 rgb->delta_red_reg = dc_fixpt_clamp_u0d10(rgb->delta_red);
267 rgb->delta_green_reg = dc_fixpt_clamp_u0d10(rgb->delta_green);
268 rgb->delta_blue_reg = dc_fixpt_clamp_u0d10(rgb->delta_blue);
269 rgb->red_reg = dc_fixpt_clamp_u0d14(rgb->red);
270 rgb->green_reg = dc_fixpt_clamp_u0d14(rgb->green);
271 rgb->blue_reg = dc_fixpt_clamp_u0d14(rgb->blue);
279 cm3_helper_convert_to_custom_float(rgb_resulted,
280 lut_params->corner_points,
281 hw_points+1, fixpoint);
286 #define NUM_DEGAMMA_REGIONS 12
289 bool cm3_helper_translate_curve_to_degamma_hw_format(
290 const struct dc_transfer_func *output_tf,
291 struct pwl_params *lut_params)
293 struct curve_points3 *corner_points;
294 struct pwl_result_data *rgb_resulted;
295 struct pwl_result_data *rgb;
296 struct pwl_result_data *rgb_plus_1;
298 int32_t region_start, region_end;
300 uint32_t j, k, seg_distr[MAX_REGIONS_NUMBER], increment, start_index, hw_points;
302 if (output_tf == NULL || lut_params == NULL || output_tf->type == TF_TYPE_BYPASS)
305 corner_points = lut_params->corner_points;
306 rgb_resulted = lut_params->rgb_resulted;
309 memset(lut_params, 0, sizeof(struct pwl_params));
310 memset(seg_distr, 0, sizeof(seg_distr));
312 region_start = -NUM_DEGAMMA_REGIONS;
316 for (i = region_end - region_start; i < MAX_REGIONS_NUMBER ; i++)
319 * segments are from 2^-12 to 0
321 for (i = 0; i < NUM_DEGAMMA_REGIONS ; i++)
324 for (k = 0; k < MAX_REGIONS_NUMBER; k++) {
325 if (seg_distr[k] != -1)
326 hw_points += (1 << seg_distr[k]);
330 for (k = 0; k < (region_end - region_start); k++) {
331 increment = NUMBER_SW_SEGMENTS / (1 << seg_distr[k]);
332 start_index = (region_start + k + MAX_LOW_POINT) *
334 for (i = start_index; i < start_index + NUMBER_SW_SEGMENTS;
336 if (j == hw_points - 1)
338 rgb_resulted[j].red = output_tf->tf_pts.red[i];
339 rgb_resulted[j].green = output_tf->tf_pts.green[i];
340 rgb_resulted[j].blue = output_tf->tf_pts.blue[i];
346 start_index = (region_end + MAX_LOW_POINT) * NUMBER_SW_SEGMENTS;
347 rgb_resulted[hw_points - 1].red = output_tf->tf_pts.red[start_index];
348 rgb_resulted[hw_points - 1].green = output_tf->tf_pts.green[start_index];
349 rgb_resulted[hw_points - 1].blue = output_tf->tf_pts.blue[start_index];
351 corner_points[0].red.x = dc_fixpt_pow(dc_fixpt_from_int(2),
352 dc_fixpt_from_int(region_start));
353 corner_points[0].green.x = corner_points[0].red.x;
354 corner_points[0].blue.x = corner_points[0].red.x;
355 corner_points[1].red.x = dc_fixpt_pow(dc_fixpt_from_int(2),
356 dc_fixpt_from_int(region_end));
357 corner_points[1].green.x = corner_points[1].red.x;
358 corner_points[1].blue.x = corner_points[1].red.x;
360 corner_points[0].red.y = rgb_resulted[0].red;
361 corner_points[0].green.y = rgb_resulted[0].green;
362 corner_points[0].blue.y = rgb_resulted[0].blue;
364 /* see comment above, m_arrPoints[1].y should be the Y value for the
365 * region end (m_numOfHwPoints), not last HW point(m_numOfHwPoints - 1)
367 corner_points[1].red.y = rgb_resulted[hw_points - 1].red;
368 corner_points[1].green.y = rgb_resulted[hw_points - 1].green;
369 corner_points[1].blue.y = rgb_resulted[hw_points - 1].blue;
370 corner_points[1].red.slope = dc_fixpt_zero;
371 corner_points[1].green.slope = dc_fixpt_zero;
372 corner_points[1].blue.slope = dc_fixpt_zero;
374 if (output_tf->tf == TRANSFER_FUNCTION_PQ) {
375 /* for PQ, we want to have a straight line from last HW X point,
376 * and the slope to be such that we hit 1.0 at 10000 nits.
378 const struct fixed31_32 end_value =
379 dc_fixpt_from_int(125);
381 corner_points[1].red.slope = dc_fixpt_div(
382 dc_fixpt_sub(dc_fixpt_one, corner_points[1].red.y),
383 dc_fixpt_sub(end_value, corner_points[1].red.x));
384 corner_points[1].green.slope = dc_fixpt_div(
385 dc_fixpt_sub(dc_fixpt_one, corner_points[1].green.y),
386 dc_fixpt_sub(end_value, corner_points[1].green.x));
387 corner_points[1].blue.slope = dc_fixpt_div(
388 dc_fixpt_sub(dc_fixpt_one, corner_points[1].blue.y),
389 dc_fixpt_sub(end_value, corner_points[1].blue.x));
392 lut_params->hw_points_num = hw_points;
395 for (i = 1; i < MAX_REGIONS_NUMBER; i++) {
396 if (seg_distr[k] != -1) {
397 lut_params->arr_curve_points[k].segments_num =
399 lut_params->arr_curve_points[i].offset =
400 lut_params->arr_curve_points[k].offset + (1 << seg_distr[k]);
405 if (seg_distr[k] != -1)
406 lut_params->arr_curve_points[k].segments_num = seg_distr[k];
409 rgb_plus_1 = rgb_resulted + 1;
412 while (i != hw_points + 1) {
413 if (dc_fixpt_lt(rgb_plus_1->red, rgb->red))
414 rgb_plus_1->red = rgb->red;
415 if (dc_fixpt_lt(rgb_plus_1->green, rgb->green))
416 rgb_plus_1->green = rgb->green;
417 if (dc_fixpt_lt(rgb_plus_1->blue, rgb->blue))
418 rgb_plus_1->blue = rgb->blue;
420 rgb->delta_red = dc_fixpt_sub(rgb_plus_1->red, rgb->red);
421 rgb->delta_green = dc_fixpt_sub(rgb_plus_1->green, rgb->green);
422 rgb->delta_blue = dc_fixpt_sub(rgb_plus_1->blue, rgb->blue);
428 cm3_helper_convert_to_custom_float(rgb_resulted,
429 lut_params->corner_points,
435 bool cm3_helper_convert_to_custom_float(
436 struct pwl_result_data *rgb_resulted,
437 struct curve_points3 *corner_points,
438 uint32_t hw_points_num,
441 struct custom_float_format fmt;
443 struct pwl_result_data *rgb = rgb_resulted;
447 fmt.exponenta_bits = 6;
448 fmt.mantissa_bits = 12;
451 /* corner_points[0] - beginning base, slope offset for R,G,B
452 * corner_points[1] - end base, slope offset for R,G,B
454 if (!convert_to_custom_float_format(corner_points[0].red.x, &fmt,
455 &corner_points[0].red.custom_float_x)) {
459 if (!convert_to_custom_float_format(corner_points[0].green.x, &fmt,
460 &corner_points[0].green.custom_float_x)) {
464 if (!convert_to_custom_float_format(corner_points[0].blue.x, &fmt,
465 &corner_points[0].blue.custom_float_x)) {
470 if (!convert_to_custom_float_format(corner_points[0].red.offset, &fmt,
471 &corner_points[0].red.custom_float_offset)) {
475 if (!convert_to_custom_float_format(corner_points[0].green.offset, &fmt,
476 &corner_points[0].green.custom_float_offset)) {
480 if (!convert_to_custom_float_format(corner_points[0].blue.offset, &fmt,
481 &corner_points[0].blue.custom_float_offset)) {
486 if (!convert_to_custom_float_format(corner_points[0].red.slope, &fmt,
487 &corner_points[0].red.custom_float_slope)) {
491 if (!convert_to_custom_float_format(corner_points[0].green.slope, &fmt,
492 &corner_points[0].green.custom_float_slope)) {
496 if (!convert_to_custom_float_format(corner_points[0].blue.slope, &fmt,
497 &corner_points[0].blue.custom_float_slope)) {
502 if (fixpoint == true) {
503 corner_points[1].red.custom_float_y =
504 dc_fixpt_clamp_u0d14(corner_points[1].red.y);
505 corner_points[1].green.custom_float_y =
506 dc_fixpt_clamp_u0d14(corner_points[1].green.y);
507 corner_points[1].blue.custom_float_y =
508 dc_fixpt_clamp_u0d14(corner_points[1].blue.y);
510 if (!convert_to_custom_float_format(corner_points[1].red.y,
511 &fmt, &corner_points[1].red.custom_float_y)) {
515 if (!convert_to_custom_float_format(corner_points[1].green.y,
516 &fmt, &corner_points[1].green.custom_float_y)) {
520 if (!convert_to_custom_float_format(corner_points[1].blue.y,
521 &fmt, &corner_points[1].blue.custom_float_y)) {
527 fmt.mantissa_bits = 10;
530 if (!convert_to_custom_float_format(corner_points[1].red.x, &fmt,
531 &corner_points[1].red.custom_float_x)) {
535 if (!convert_to_custom_float_format(corner_points[1].green.x, &fmt,
536 &corner_points[1].green.custom_float_x)) {
540 if (!convert_to_custom_float_format(corner_points[1].blue.x, &fmt,
541 &corner_points[1].blue.custom_float_x)) {
546 if (!convert_to_custom_float_format(corner_points[1].red.slope, &fmt,
547 &corner_points[1].red.custom_float_slope)) {
551 if (!convert_to_custom_float_format(corner_points[1].green.slope, &fmt,
552 &corner_points[1].green.custom_float_slope)) {
556 if (!convert_to_custom_float_format(corner_points[1].blue.slope, &fmt,
557 &corner_points[1].blue.custom_float_slope)) {
562 if (hw_points_num == 0 || rgb_resulted == NULL || fixpoint == true)
565 fmt.mantissa_bits = 12;
567 while (i != hw_points_num) {
568 if (!convert_to_custom_float_format(rgb->red, &fmt,
574 if (!convert_to_custom_float_format(rgb->green, &fmt,
580 if (!convert_to_custom_float_format(rgb->blue, &fmt,
593 bool is_rgb_equal(const struct pwl_result_data *rgb, uint32_t num)
598 for (i = 0 ; i < num; i++) {
599 if (rgb[i].red_reg != rgb[i].green_reg ||
600 rgb[i].blue_reg != rgb[i].red_reg ||
601 rgb[i].blue_reg != rgb[i].green_reg) {