dcn30_cm_common.c

  1/*
  2 * Copyright 2020 Advanced Micro Devices, Inc.
  3 *
  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:
 10 *
 11 * The above copyright notice and this permission notice shall be included in
 12 * all copies or substantial portions of the Software.
 13 *
 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.
 21 *
 22 * Authors: AMD
 23 *
 24 */
 25
 26#include "dm_services.h"
 27#include "core_types.h"
 28#include "reg_helper.h"
 29#include "dcn30/dcn30_dpp.h"
 30#include "basics/conversion.h"
 31#include "dcn30/dcn30_cm_common.h"
 32#include "custom_float.h"
 33
 34#define REG(reg) reg
 35
 36#define CTX \
 37	ctx //dpp->base.ctx
 38
 39#undef FN
 40#define FN(reg_name, field_name) \
 41	reg->shifts.field_name, reg->masks.field_name
 42
 43void 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)
 47{
 48	uint32_t reg_region_cur;
 49	unsigned int i = 0;
 50
 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);
 60
 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);
 67
 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);
 74
 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);
 84
 85	for (reg_region_cur = reg->region_start;
 86		reg_region_cur <= reg->region_end;
 87		reg_region_cur++) {
 88
 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]);
 91
 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);
 97
 98		i++;
 99	}
100}
101
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
107
108#define DC_LOGGER \
109		ctx->logger
110
111bool cm3_helper_translate_curve_to_hw_format(struct dc_context *ctx,
112					     const struct dc_transfer_func *output_tf,
113					     struct pwl_params *lut_params, bool fixpoint)
114{
115	struct curve_points3 *corner_points;
116	struct pwl_result_data *rgb_resulted;
117	struct pwl_result_data *rgb;
118	struct pwl_result_data *rgb_plus_1;
119	struct pwl_result_data *rgb_minus_1;
120
121	int32_t region_start, region_end;
122	int32_t i;
123	uint32_t j, k, seg_distr[MAX_REGIONS_NUMBER], increment, start_index, hw_points;
124
125	if (output_tf == NULL || lut_params == NULL || output_tf->type == TF_TYPE_BYPASS)
126		return false;
127
128	corner_points = lut_params->corner_points;
129	rgb_resulted = lut_params->rgb_resulted;
130	hw_points = 0;
131
132	memset(lut_params, 0, sizeof(struct pwl_params));
133	memset(seg_distr, 0, sizeof(seg_distr));
134
135	if (output_tf->tf == TRANSFER_FUNCTION_PQ || output_tf->tf == TRANSFER_FUNCTION_GAMMA22 ||
136		output_tf->tf == TRANSFER_FUNCTION_HLG) {
137		/* 32 segments
138		 * segments are from 2^-25 to 2^7
139		 */
140		for (i = 0; i < NUMBER_REGIONS ; i++)
141			seg_distr[i] = 3;
142
143		region_start = -MAX_LOW_POINT;
144		region_end   = NUMBER_REGIONS - MAX_LOW_POINT;
145	} else {
146		/* 13 segments
147		 * segment is from 2^-12 to 2^0
148		 * There are less than 256 points, for optimization
149		 */
150		const uint8_t SEG_COUNT = 12;
151
152		for (i = 0; i < SEG_COUNT; i++)
153			seg_distr[i] = 4;
154
155		seg_distr[SEG_COUNT] = 1;
156
157		region_start = -SEG_COUNT;
158		region_end = 1;
159	}
160
161	for (i = region_end - region_start; i < MAX_REGIONS_NUMBER ; i++)
162		seg_distr[i] = -1;
163
164	for (k = 0; k < MAX_REGIONS_NUMBER; k++) {
165		if (seg_distr[k] != -1)
166			hw_points += (1 << seg_distr[k]);
167	}
168
169	// DCN3+ have 257 pts in lieu of no separate slope registers
170	// Prior HW had 256 base+slope pairs
171	// Shaper LUT (i.e. fixpoint == true) is still 256 bases and 256 deltas
172	hw_points = fixpoint ? (hw_points - 1) : hw_points;
173
174	j = 0;
175	for (k = 0; k < (region_end - region_start); k++) {
176		increment = NUMBER_SW_SEGMENTS / (1 << seg_distr[k]);
177		start_index = (region_start + k + MAX_LOW_POINT) *
178				NUMBER_SW_SEGMENTS;
179		for (i = start_index; i < start_index + NUMBER_SW_SEGMENTS;
180				i += increment) {
181			if (j == hw_points)
182				break;
183			if (i >= TRANSFER_FUNC_POINTS)
184				return false;
185			rgb_resulted[j].red = output_tf->tf_pts.red[i];
186			rgb_resulted[j].green = output_tf->tf_pts.green[i];
187			rgb_resulted[j].blue = output_tf->tf_pts.blue[i];
188			j++;
189		}
190	}
191
192	/* last point */
193	start_index = (region_end + MAX_LOW_POINT) * NUMBER_SW_SEGMENTS;
194	rgb_resulted[hw_points].red = output_tf->tf_pts.red[start_index];
195	rgb_resulted[hw_points].green = output_tf->tf_pts.green[start_index];
196	rgb_resulted[hw_points].blue = output_tf->tf_pts.blue[start_index];
197
198	rgb_resulted[hw_points+1].red = rgb_resulted[hw_points].red;
199	rgb_resulted[hw_points+1].green = rgb_resulted[hw_points].green;
200	rgb_resulted[hw_points+1].blue = rgb_resulted[hw_points].blue;
201
202	// All 3 color channels have same x
203	corner_points[0].red.x = dc_fixpt_pow(dc_fixpt_from_int(2),
204					     dc_fixpt_from_int(region_start));
205	corner_points[0].green.x = corner_points[0].red.x;
206	corner_points[0].blue.x = corner_points[0].red.x;
207
208	corner_points[1].red.x = dc_fixpt_pow(dc_fixpt_from_int(2),
209					     dc_fixpt_from_int(region_end));
210	corner_points[1].green.x = corner_points[1].red.x;
211	corner_points[1].blue.x = corner_points[1].red.x;
212
213	corner_points[0].red.y = rgb_resulted[0].red;
214	corner_points[0].green.y = rgb_resulted[0].green;
215	corner_points[0].blue.y = rgb_resulted[0].blue;
216
217	corner_points[0].red.slope = dc_fixpt_div(corner_points[0].red.y,
218			corner_points[0].red.x);
219	corner_points[0].green.slope = dc_fixpt_div(corner_points[0].green.y,
220			corner_points[0].green.x);
221	corner_points[0].blue.slope = dc_fixpt_div(corner_points[0].blue.y,
222			corner_points[0].blue.x);
223
224	/* see comment above, m_arrPoints[1].y should be the Y value for the
225	 * region end (m_numOfHwPoints), not last HW point(m_numOfHwPoints - 1)
226	 */
227	corner_points[1].red.y = rgb_resulted[hw_points].red;
228	corner_points[1].green.y = rgb_resulted[hw_points].green;
229	corner_points[1].blue.y = rgb_resulted[hw_points].blue;
230	corner_points[1].red.slope = dc_fixpt_zero;
231	corner_points[1].green.slope = dc_fixpt_zero;
232	corner_points[1].blue.slope = dc_fixpt_zero;
233
234	lut_params->hw_points_num = hw_points + 1;
235
236	k = 0;
237	for (i = 1; i < MAX_REGIONS_NUMBER; i++) {
238		if (seg_distr[k] != -1) {
239			lut_params->arr_curve_points[k].segments_num =
240					seg_distr[k];
241			lut_params->arr_curve_points[i].offset =
242					lut_params->arr_curve_points[k].offset + (1 << seg_distr[k]);
243		}
244		k++;
245	}
246
247	if (seg_distr[k] != -1)
248		lut_params->arr_curve_points[k].segments_num = seg_distr[k];
249
250	rgb = rgb_resulted;
251	rgb_plus_1 = rgb_resulted + 1;
252	rgb_minus_1 = rgb;
253
254	if (fixpoint == true) {
255		i = 1;
256		while (i != hw_points + 2) {
257			uint32_t red_clamp;
258			uint32_t green_clamp;
259			uint32_t blue_clamp;
260
261			if (i >= hw_points) {
262				if (dc_fixpt_lt(rgb_plus_1->red, rgb->red))
263					rgb_plus_1->red = dc_fixpt_add(rgb->red,
264							rgb_minus_1->delta_red);
265				if (dc_fixpt_lt(rgb_plus_1->green, rgb->green))
266					rgb_plus_1->green = dc_fixpt_add(rgb->green,
267							rgb_minus_1->delta_green);
268				if (dc_fixpt_lt(rgb_plus_1->blue, rgb->blue))
269					rgb_plus_1->blue = dc_fixpt_add(rgb->blue,
270							rgb_minus_1->delta_blue);
271			}
272
273			rgb->delta_red   = dc_fixpt_sub(rgb_plus_1->red,   rgb->red);
274			rgb->delta_green = dc_fixpt_sub(rgb_plus_1->green, rgb->green);
275			rgb->delta_blue  = dc_fixpt_sub(rgb_plus_1->blue,  rgb->blue);
276
277			red_clamp = dc_fixpt_clamp_u0d14(rgb->delta_red);
278			green_clamp = dc_fixpt_clamp_u0d14(rgb->delta_green);
279			blue_clamp = dc_fixpt_clamp_u0d14(rgb->delta_blue);
280
281			if (red_clamp >> 10 || green_clamp >> 10 || blue_clamp >> 10)
282				DC_LOG_ERROR("Losing delta precision while programming shaper LUT.");
283
284			rgb->delta_red_reg   = red_clamp & 0x3ff;
285			rgb->delta_green_reg = green_clamp & 0x3ff;
286			rgb->delta_blue_reg  = blue_clamp & 0x3ff;
287			rgb->red_reg         = dc_fixpt_clamp_u0d14(rgb->red);
288			rgb->green_reg       = dc_fixpt_clamp_u0d14(rgb->green);
289			rgb->blue_reg        = dc_fixpt_clamp_u0d14(rgb->blue);
290
291			++rgb_plus_1;
292			rgb_minus_1 = rgb;
293			++rgb;
294			++i;
295		}
296	}
297	cm3_helper_convert_to_custom_float(rgb_resulted,
298						lut_params->corner_points,
299						hw_points+1, fixpoint);
300
301	return true;
302}
303
304bool cm3_helper_convert_to_custom_float(
305		struct pwl_result_data *rgb_resulted,
306		struct curve_points3 *corner_points,
307		uint32_t hw_points_num,
308		bool fixpoint)
309{
310	struct custom_float_format fmt;
311
312	struct pwl_result_data *rgb = rgb_resulted;
313
314	uint32_t i = 0;
315
316	fmt.exponenta_bits = 6;
317	fmt.mantissa_bits = 12;
318	fmt.sign = false;
319
320	/* corner_points[0] - beginning base, slope offset for R,G,B
321	 * corner_points[1] - end base, slope offset for R,G,B
322	 */
323	if (!convert_to_custom_float_format(corner_points[0].red.x, &fmt,
324				&corner_points[0].red.custom_float_x)) {
325		BREAK_TO_DEBUGGER();
326		return false;
327	}
328	if (!convert_to_custom_float_format(corner_points[0].green.x, &fmt,
329				&corner_points[0].green.custom_float_x)) {
330		BREAK_TO_DEBUGGER();
331		return false;
332	}
333	if (!convert_to_custom_float_format(corner_points[0].blue.x, &fmt,
334				&corner_points[0].blue.custom_float_x)) {
335		BREAK_TO_DEBUGGER();
336		return false;
337	}
338
339	if (!convert_to_custom_float_format(corner_points[0].red.offset, &fmt,
340				&corner_points[0].red.custom_float_offset)) {
341		BREAK_TO_DEBUGGER();
342		return false;
343	}
344	if (!convert_to_custom_float_format(corner_points[0].green.offset, &fmt,
345				&corner_points[0].green.custom_float_offset)) {
346		BREAK_TO_DEBUGGER();
347		return false;
348	}
349	if (!convert_to_custom_float_format(corner_points[0].blue.offset, &fmt,
350				&corner_points[0].blue.custom_float_offset)) {
351		BREAK_TO_DEBUGGER();
352		return false;
353	}
354
355	if (!convert_to_custom_float_format(corner_points[0].red.slope, &fmt,
356				&corner_points[0].red.custom_float_slope)) {
357		BREAK_TO_DEBUGGER();
358		return false;
359	}
360	if (!convert_to_custom_float_format(corner_points[0].green.slope, &fmt,
361				&corner_points[0].green.custom_float_slope)) {
362		BREAK_TO_DEBUGGER();
363		return false;
364	}
365	if (!convert_to_custom_float_format(corner_points[0].blue.slope, &fmt,
366				&corner_points[0].blue.custom_float_slope)) {
367		BREAK_TO_DEBUGGER();
368		return false;
369	}
370
371	if (fixpoint == true) {
372		corner_points[1].red.custom_float_y =
373				dc_fixpt_clamp_u0d14(corner_points[1].red.y);
374		corner_points[1].green.custom_float_y =
375				dc_fixpt_clamp_u0d14(corner_points[1].green.y);
376		corner_points[1].blue.custom_float_y =
377				dc_fixpt_clamp_u0d14(corner_points[1].blue.y);
378	} else {
379		if (!convert_to_custom_float_format(corner_points[1].red.y,
380				&fmt, &corner_points[1].red.custom_float_y)) {
381			BREAK_TO_DEBUGGER();
382			return false;
383		}
384		if (!convert_to_custom_float_format(corner_points[1].green.y,
385				&fmt, &corner_points[1].green.custom_float_y)) {
386			BREAK_TO_DEBUGGER();
387			return false;
388		}
389		if (!convert_to_custom_float_format(corner_points[1].blue.y,
390				&fmt, &corner_points[1].blue.custom_float_y)) {
391			BREAK_TO_DEBUGGER();
392			return false;
393		}
394	}
395
396	fmt.mantissa_bits = 10;
397	fmt.sign = false;
398
399	if (!convert_to_custom_float_format(corner_points[1].red.x, &fmt,
400				&corner_points[1].red.custom_float_x)) {
401		BREAK_TO_DEBUGGER();
402		return false;
403	}
404	if (!convert_to_custom_float_format(corner_points[1].green.x, &fmt,
405				&corner_points[1].green.custom_float_x)) {
406		BREAK_TO_DEBUGGER();
407		return false;
408	}
409	if (!convert_to_custom_float_format(corner_points[1].blue.x, &fmt,
410				&corner_points[1].blue.custom_float_x)) {
411		BREAK_TO_DEBUGGER();
412		return false;
413	}
414
415	if (!convert_to_custom_float_format(corner_points[1].red.slope, &fmt,
416				&corner_points[1].red.custom_float_slope)) {
417		BREAK_TO_DEBUGGER();
418		return false;
419	}
420	if (!convert_to_custom_float_format(corner_points[1].green.slope, &fmt,
421				&corner_points[1].green.custom_float_slope)) {
422		BREAK_TO_DEBUGGER();
423		return false;
424	}
425	if (!convert_to_custom_float_format(corner_points[1].blue.slope, &fmt,
426				&corner_points[1].blue.custom_float_slope)) {
427		BREAK_TO_DEBUGGER();
428		return false;
429	}
430
431	if (hw_points_num == 0 || rgb_resulted == NULL || fixpoint == true)
432		return true;
433
434	fmt.mantissa_bits = 12;
435
436	while (i != hw_points_num) {
437		if (!convert_to_custom_float_format(rgb->red, &fmt,
438						    &rgb->red_reg)) {
439			BREAK_TO_DEBUGGER();
440			return false;
441		}
442
443		if (!convert_to_custom_float_format(rgb->green, &fmt,
444						    &rgb->green_reg)) {
445			BREAK_TO_DEBUGGER();
446			return false;
447		}
448
449		if (!convert_to_custom_float_format(rgb->blue, &fmt,
450						    &rgb->blue_reg)) {
451			BREAK_TO_DEBUGGER();
452			return false;
453		}
454
455		++rgb;
456		++i;
457	}
458
459	return true;
460}
461
462bool is_rgb_equal(const struct pwl_result_data *rgb, uint32_t num)
463{
464	uint32_t i;
465	bool ret = true;
466
467	for (i = 0 ; i < num; i++) {
468		if (rgb[i].red_reg != rgb[i].green_reg ||
469		rgb[i].blue_reg != rgb[i].red_reg  ||
470		rgb[i].blue_reg != rgb[i].green_reg) {
471			ret = false;
472			break;
473		}
474	}
475	return ret;
476}
477