1 
2 /*
3  * Copyright 2017 Advanced Micro Devices, Inc.
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a
6  * copy of this software and associated documentation files (the "Software"),
7  * to deal in the Software without restriction, including without limitation
8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9  * and/or sell copies of the Software, and to permit persons to whom the
10  * Software is furnished to do so, subject to the following conditions:
11  *
12  * The above copyright notice and this permission notice shall be included in
13  * all copies or substantial portions of the Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
19  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
20  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
21  * OTHER DEALINGS IN THE SOFTWARE.
22  *
23  * Authors: AMD
24  *
25  */
26 #include <drm/drm_dsc.h>
27 
28 #include "os_types.h"
29 #include "rc_calc.h"
30 #include "qp_tables.h"
31 
32 #define table_hash(mode, bpc, max_min) ((mode << 16) | (bpc << 8) | max_min)
33 
34 #define MODE_SELECT(val444, val422, val420) \
35 	(cm == CM_444 || cm == CM_RGB) ? (val444) : (cm == CM_422 ? (val422) : (val420))
36 
37 
38 #define TABLE_CASE(mode, bpc, max)   case (table_hash(mode, BPC_##bpc, max)): \
39 	table = qp_table_##mode##_##bpc##bpc_##max; \
40 	table_size = sizeof(qp_table_##mode##_##bpc##bpc_##max)/sizeof(*qp_table_##mode##_##bpc##bpc_##max); \
41 	break
42 
43 
44 static void get_qp_set(qp_set qps, enum colour_mode cm, enum bits_per_comp bpc,
45 		       enum max_min max_min, float bpp)
46 {
47 	int mode = MODE_SELECT(444, 422, 420);
48 	int sel = table_hash(mode, bpc, max_min);
49 	int table_size = 0;
50 	int index;
51 	const struct qp_entry *table = 0L;
52 
53 	// alias enum
54 	enum { min = MM_MIN, max = MM_MAX };
55 	switch (sel) {
56 		TABLE_CASE(444,  8, max);
57 		TABLE_CASE(444,  8, min);
58 		TABLE_CASE(444, 10, max);
59 		TABLE_CASE(444, 10, min);
60 		TABLE_CASE(444, 12, max);
61 		TABLE_CASE(444, 12, min);
62 		TABLE_CASE(422,  8, max);
63 		TABLE_CASE(422,  8, min);
64 		TABLE_CASE(422, 10, max);
65 		TABLE_CASE(422, 10, min);
66 		TABLE_CASE(422, 12, max);
67 		TABLE_CASE(422, 12, min);
68 		TABLE_CASE(420,  8, max);
69 		TABLE_CASE(420,  8, min);
70 		TABLE_CASE(420, 10, max);
71 		TABLE_CASE(420, 10, min);
72 		TABLE_CASE(420, 12, max);
73 		TABLE_CASE(420, 12, min);
74 	}
75 
76 	if (table == 0)
77 		return;
78 
79 	index = (bpp - table[0].bpp) * 2;
80 
81 	/* requested size is bigger than the table */
82 	if (index >= table_size) {
83 		dm_error("ERROR: Requested rc_calc to find a bpp entry that exceeds the table size\n");
84 		return;
85 	}
86 
87 	memcpy(qps, table[index].qps, sizeof(qp_set));
88 }
89 
90 static double dsc_roundf(double num)
91 {
92 	if (num < 0.0)
93 		num = num - 0.5;
94 	else
95 		num = num + 0.5;
96 
97 	return (int)(num);
98 }
99 
100 static double dsc_ceil(double num)
101 {
102 	double retval = (int)num;
103 
104 	if (retval != num && num > 0)
105 		retval = num + 1;
106 
107 	return (int)retval;
108 }
109 
110 static void get_ofs_set(qp_set ofs, enum colour_mode mode, float bpp)
111 {
112 	int   *p = ofs;
113 
114 	if (mode == CM_444 || mode == CM_RGB) {
115 		*p++ = (bpp <=  6) ? (0) : ((((bpp >=  8) && (bpp <= 12))) ? (2) : ((bpp >= 15) ? (10) : ((((bpp > 6) && (bpp < 8))) ? (0 + dsc_roundf((bpp -  6) * (2 / 2.0))) : (2 + dsc_roundf((bpp - 12) * (8 / 3.0))))));
116 		*p++ = (bpp <=  6) ? (-2) : ((((bpp >=  8) && (bpp <= 12))) ? (0) : ((bpp >= 15) ? (8) : ((((bpp > 6) && (bpp < 8))) ? (-2 + dsc_roundf((bpp -  6) * (2 / 2.0))) : (0 + dsc_roundf((bpp - 12) * (8 / 3.0))))));
117 		*p++ = (bpp <=  6) ? (-2) : ((((bpp >=  8) && (bpp <= 12))) ? (0) : ((bpp >= 15) ? (6) : ((((bpp > 6) && (bpp < 8))) ? (-2 + dsc_roundf((bpp -  6) * (2 / 2.0))) : (0 + dsc_roundf((bpp - 12) * (6 / 3.0))))));
118 		*p++ = (bpp <=  6) ? (-4) : ((((bpp >=  8) && (bpp <= 12))) ? (-2) : ((bpp >= 15) ? (4) : ((((bpp > 6) && (bpp < 8))) ? (-4 + dsc_roundf((bpp -  6) * (2 / 2.0))) : (-2 + dsc_roundf((bpp - 12) * (6 / 3.0))))));
119 		*p++ = (bpp <=  6) ? (-6) : ((((bpp >=  8) && (bpp <= 12))) ? (-4) : ((bpp >= 15) ? (2) : ((((bpp > 6) && (bpp < 8))) ? (-6 + dsc_roundf((bpp -  6) * (2 / 2.0))) : (-4 + dsc_roundf((bpp - 12) * (6 / 3.0))))));
120 		*p++ = (bpp <= 12) ? (-6) : ((bpp >= 15) ? (0) : (-6 + dsc_roundf((bpp - 12) * (6 / 3.0))));
121 		*p++ = (bpp <= 12) ? (-8) : ((bpp >= 15) ? (-2) : (-8 + dsc_roundf((bpp - 12) * (6 / 3.0))));
122 		*p++ = (bpp <= 12) ? (-8) : ((bpp >= 15) ? (-4) : (-8 + dsc_roundf((bpp - 12) * (4 / 3.0))));
123 		*p++ = (bpp <= 12) ? (-8) : ((bpp >= 15) ? (-6) : (-8 + dsc_roundf((bpp - 12) * (2 / 3.0))));
124 		*p++ = (bpp <= 12) ? (-10) : ((bpp >= 15) ? (-8) : (-10 + dsc_roundf((bpp - 12) * (2 / 3.0))));
125 		*p++ = -10;
126 		*p++ = (bpp <=  6) ? (-12) : ((bpp >=  8) ? (-10) : (-12 + dsc_roundf((bpp -  6) * (2 / 2.0))));
127 		*p++ = -12;
128 		*p++ = -12;
129 		*p++ = -12;
130 	} else if (mode == CM_422) {
131 		*p++ = (bpp <=  8) ? (2) : ((bpp >= 10) ? (10) : (2 + dsc_roundf((bpp -  8) * (8 / 2.0))));
132 		*p++ = (bpp <=  8) ? (0) : ((bpp >= 10) ? (8) : (0 + dsc_roundf((bpp -  8) * (8 / 2.0))));
133 		*p++ = (bpp <=  8) ? (0) : ((bpp >= 10) ? (6) : (0 + dsc_roundf((bpp -  8) * (6 / 2.0))));
134 		*p++ = (bpp <=  8) ? (-2) : ((bpp >= 10) ? (4) : (-2 + dsc_roundf((bpp -  8) * (6 / 2.0))));
135 		*p++ = (bpp <=  8) ? (-4) : ((bpp >= 10) ? (2) : (-4 + dsc_roundf((bpp -  8) * (6 / 2.0))));
136 		*p++ = (bpp <=  8) ? (-6) : ((bpp >= 10) ? (0) : (-6 + dsc_roundf((bpp -  8) * (6 / 2.0))));
137 		*p++ = (bpp <=  8) ? (-8) : ((bpp >= 10) ? (-2) : (-8 + dsc_roundf((bpp -  8) * (6 / 2.0))));
138 		*p++ = (bpp <=  8) ? (-8) : ((bpp >= 10) ? (-4) : (-8 + dsc_roundf((bpp -  8) * (4 / 2.0))));
139 		*p++ = (bpp <=  8) ? (-8) : ((bpp >= 10) ? (-6) : (-8 + dsc_roundf((bpp -  8) * (2 / 2.0))));
140 		*p++ = (bpp <=  8) ? (-10) : ((bpp >= 10) ? (-8) : (-10 + dsc_roundf((bpp -  8) * (2 / 2.0))));
141 		*p++ = -10;
142 		*p++ = (bpp <=  6) ? (-12) : ((bpp >= 7) ? (-10) : (-12 + dsc_roundf((bpp -  6) * (2.0 / 1))));
143 		*p++ = -12;
144 		*p++ = -12;
145 		*p++ = -12;
146 	} else {
147 		*p++ = (bpp <=  6) ? (2) : ((bpp >=  8) ? (10) : (2 + dsc_roundf((bpp -  6) * (8 / 2.0))));
148 		*p++ = (bpp <=  6) ? (0) : ((bpp >=  8) ? (8) : (0 + dsc_roundf((bpp -  6) * (8 / 2.0))));
149 		*p++ = (bpp <=  6) ? (0) : ((bpp >=  8) ? (6) : (0 + dsc_roundf((bpp -  6) * (6 / 2.0))));
150 		*p++ = (bpp <=  6) ? (-2) : ((bpp >=  8) ? (4) : (-2 + dsc_roundf((bpp -  6) * (6 / 2.0))));
151 		*p++ = (bpp <=  6) ? (-4) : ((bpp >=  8) ? (2) : (-4 + dsc_roundf((bpp -  6) * (6 / 2.0))));
152 		*p++ = (bpp <=  6) ? (-6) : ((bpp >=  8) ? (0) : (-6 + dsc_roundf((bpp -  6) * (6 / 2.0))));
153 		*p++ = (bpp <=  6) ? (-8) : ((bpp >=  8) ? (-2) : (-8 + dsc_roundf((bpp -  6) * (6 / 2.0))));
154 		*p++ = (bpp <=  6) ? (-8) : ((bpp >=  8) ? (-4) : (-8 + dsc_roundf((bpp -  6) * (4 / 2.0))));
155 		*p++ = (bpp <=  6) ? (-8) : ((bpp >=  8) ? (-6) : (-8 + dsc_roundf((bpp -  6) * (2 / 2.0))));
156 		*p++ = (bpp <=  6) ? (-10) : ((bpp >=  8) ? (-8) : (-10 + dsc_roundf((bpp -  6) * (2 / 2.0))));
157 		*p++ = -10;
158 		*p++ = (bpp <=  4) ? (-12) : ((bpp >=  5) ? (-10) : (-12 + dsc_roundf((bpp -  4) * (2 / 1.0))));
159 		*p++ = -12;
160 		*p++ = -12;
161 		*p++ = -12;
162 	}
163 }
164 
165 static int median3(int a, int b, int c)
166 {
167 	if (a > b)
168 		swap(a, b);
169 	if (b > c)
170 		swap(b, c);
171 	if (a > b)
172 		swap(b, c);
173 
174 	return b;
175 }
176 
177 static void _do_calc_rc_params(struct rc_params *rc, enum colour_mode cm,
178 			       enum bits_per_comp bpc, u8 drm_bpp,
179 			       bool is_navite_422_or_420,
180 			       int slice_width, int slice_height,
181 			       int minor_version)
182 {
183 	float bpp;
184 	float bpp_group;
185 	float initial_xmit_delay_factor;
186 	int padding_pixels;
187 	int i;
188 
189 	bpp = ((float)drm_bpp / 16.0);
190 	/* in native_422 or native_420 modes, the bits_per_pixel is double the
191 	 * target bpp (the latter is what calc_rc_params expects)
192 	 */
193 	if (is_navite_422_or_420)
194 		bpp /= 2.0;
195 
196 	rc->rc_quant_incr_limit0 = ((bpc == BPC_8) ? 11 : (bpc == BPC_10 ? 15 : 19)) - ((minor_version == 1 && cm == CM_444) ? 1 : 0);
197 	rc->rc_quant_incr_limit1 = ((bpc == BPC_8) ? 11 : (bpc == BPC_10 ? 15 : 19)) - ((minor_version == 1 && cm == CM_444) ? 1 : 0);
198 
199 	bpp_group = MODE_SELECT(bpp, bpp * 2.0, bpp * 2.0);
200 
201 	switch (cm) {
202 	case CM_420:
203 		rc->initial_fullness_offset = (bpp >=  6) ? (2048) : ((bpp <=  4) ? (6144) : ((((bpp >  4) && (bpp <=  5))) ? (6144 - dsc_roundf((bpp - 4) * (512))) : (5632 - dsc_roundf((bpp -  5) * (3584)))));
204 		rc->first_line_bpg_offset   = median3(0, (12 + (int) (0.09 *  min(34, slice_height - 8))), (int)((3 * bpc * 3) - (3 * bpp_group)));
205 		rc->second_line_bpg_offset  = median3(0, 12, (int)((3 * bpc * 3) - (3 * bpp_group)));
206 		break;
207 	case CM_422:
208 		rc->initial_fullness_offset = (bpp >=  8) ? (2048) : ((bpp <=  7) ? (5632) : (5632 - dsc_roundf((bpp - 7) * (3584))));
209 		rc->first_line_bpg_offset   = median3(0, (12 + (int) (0.09 *  min(34, slice_height - 8))), (int)((3 * bpc * 4) - (3 * bpp_group)));
210 		rc->second_line_bpg_offset  = 0;
211 		break;
212 	case CM_444:
213 	case CM_RGB:
214 		rc->initial_fullness_offset = (bpp >= 12) ? (2048) : ((bpp <=  8) ? (6144) : ((((bpp >  8) && (bpp <= 10))) ? (6144 - dsc_roundf((bpp - 8) * (512 / 2))) : (5632 - dsc_roundf((bpp - 10) * (3584 / 2)))));
215 		rc->first_line_bpg_offset   = median3(0, (12 + (int) (0.09 *  min(34, slice_height - 8))), (int)(((3 * bpc + (cm == CM_444 ? 0 : 2)) * 3) - (3 * bpp_group)));
216 		rc->second_line_bpg_offset  = 0;
217 		break;
218 	}
219 
220 	initial_xmit_delay_factor = (cm == CM_444 || cm == CM_RGB) ? 1.0 : 2.0;
221 	rc->initial_xmit_delay = dsc_roundf(8192.0/2.0/bpp/initial_xmit_delay_factor);
222 
223 	if (cm == CM_422 || cm == CM_420)
224 		slice_width /= 2;
225 
226 	padding_pixels = ((slice_width % 3) != 0) ? (3 - (slice_width % 3)) * (rc->initial_xmit_delay / slice_width) : 0;
227 	if (3 * bpp_group >= (((rc->initial_xmit_delay + 2) / 3) * (3 + (cm == CM_422)))) {
228 		if ((rc->initial_xmit_delay + padding_pixels) % 3 == 1)
229 			rc->initial_xmit_delay++;
230 	}
231 
232 	rc->flatness_min_qp     = ((bpc == BPC_8) ?  (3) : ((bpc == BPC_10) ? (7)  : (11))) - ((minor_version == 1 && cm == CM_444) ? 1 : 0);
233 	rc->flatness_max_qp     = ((bpc == BPC_8) ? (12) : ((bpc == BPC_10) ? (16) : (20))) - ((minor_version == 1 && cm == CM_444) ? 1 : 0);
234 	rc->flatness_det_thresh = 2 << (bpc - 8);
235 
236 	get_qp_set(rc->qp_min, cm, bpc, MM_MIN, bpp);
237 	get_qp_set(rc->qp_max, cm, bpc, MM_MAX, bpp);
238 	if (cm == CM_444 && minor_version == 1) {
239 		for (i = 0; i < QP_SET_SIZE; ++i) {
240 			rc->qp_min[i] = rc->qp_min[i] > 0 ? rc->qp_min[i] - 1 : 0;
241 			rc->qp_max[i] = rc->qp_max[i] > 0 ? rc->qp_max[i] - 1 : 0;
242 		}
243 	}
244 	get_ofs_set(rc->ofs, cm, bpp);
245 
246 	/* fixed parameters */
247 	rc->rc_model_size    = 8192;
248 	rc->rc_edge_factor   = 6;
249 	rc->rc_tgt_offset_hi = 3;
250 	rc->rc_tgt_offset_lo = 3;
251 
252 	rc->rc_buf_thresh[0] = 896;
253 	rc->rc_buf_thresh[1] = 1792;
254 	rc->rc_buf_thresh[2] = 2688;
255 	rc->rc_buf_thresh[3] = 3584;
256 	rc->rc_buf_thresh[4] = 4480;
257 	rc->rc_buf_thresh[5] = 5376;
258 	rc->rc_buf_thresh[6] = 6272;
259 	rc->rc_buf_thresh[7] = 6720;
260 	rc->rc_buf_thresh[8] = 7168;
261 	rc->rc_buf_thresh[9] = 7616;
262 	rc->rc_buf_thresh[10] = 7744;
263 	rc->rc_buf_thresh[11] = 7872;
264 	rc->rc_buf_thresh[12] = 8000;
265 	rc->rc_buf_thresh[13] = 8064;
266 }
267 
268 static u32 _do_bytes_per_pixel_calc(int slice_width, u8 drm_bpp,
269 				    bool is_navite_422_or_420)
270 {
271 	float bpp;
272 	u32 bytes_per_pixel;
273 	double d_bytes_per_pixel;
274 
275 	bpp = ((float)drm_bpp / 16.0);
276 	d_bytes_per_pixel = dsc_ceil(bpp * slice_width / 8.0) / slice_width;
277 	// TODO: Make sure the formula for calculating this is precise (ceiling
278 	// vs. floor, and at what point they should be applied)
279 	if (is_navite_422_or_420)
280 		d_bytes_per_pixel /= 2;
281 
282 	bytes_per_pixel = (u32)dsc_ceil(d_bytes_per_pixel * 0x10000000);
283 
284 	return bytes_per_pixel;
285 }
286 
287 static u32 _do_calc_dsc_bpp_x16(u32 stream_bandwidth_kbps, u32 pix_clk_100hz,
288 				u32 bpp_increment_div)
289 {
290 	u32 dsc_target_bpp_x16;
291 	float f_dsc_target_bpp;
292 	float f_stream_bandwidth_100bps;
293 	// bpp_increment_div is actually precision
294 	u32 precision = bpp_increment_div;
295 
296 	f_stream_bandwidth_100bps = stream_bandwidth_kbps * 10.0f;
297 	f_dsc_target_bpp = f_stream_bandwidth_100bps / pix_clk_100hz;
298 
299 	// Round down to the nearest precision stop to bring it into DSC spec
300 	// range
301 	dsc_target_bpp_x16 = (u32)(f_dsc_target_bpp * precision);
302 	dsc_target_bpp_x16 = (dsc_target_bpp_x16 * 16) / precision;
303 
304 	return dsc_target_bpp_x16;
305 }
306 
307 /**
308  * calc_rc_params - reads the user's cmdline mode
309  * @rc: DC internal DSC parameters
310  * @pps: DRM struct with all required DSC values
311  *
312  * This function expects a drm_dsc_config data struct with all the required DSC
313  * values previously filled out by our driver and based on this information it
314  * computes some of the DSC values.
315  *
316  * @note This calculation requires float point operation, most of it executes
317  * under kernel_fpu_{begin,end}.
318  */
319 void calc_rc_params(struct rc_params *rc, const struct drm_dsc_config *pps)
320 {
321 	enum colour_mode mode;
322 	enum bits_per_comp bpc;
323 	bool is_navite_422_or_420;
324 	u8 drm_bpp = pps->bits_per_pixel;
325 	int slice_width  = pps->slice_width;
326 	int slice_height = pps->slice_height;
327 
328 	mode = pps->convert_rgb ? CM_RGB : (pps->simple_422  ? CM_444 :
329 					   (pps->native_422  ? CM_422 :
330 					    pps->native_420  ? CM_420 : CM_444));
331 	bpc = (pps->bits_per_component == 8) ? BPC_8 : (pps->bits_per_component == 10)
332 					     ? BPC_10 : BPC_12;
333 
334 	is_navite_422_or_420 = pps->native_422 || pps->native_420;
335 
336 	DC_FP_START();
337 	_do_calc_rc_params(rc, mode, bpc, drm_bpp, is_navite_422_or_420,
338 			   slice_width, slice_height,
339 			   pps->dsc_version_minor);
340 	DC_FP_END();
341 }
342 
343 /**
344  * calc_dsc_bytes_per_pixel - calculate bytes per pixel
345  * @pps: DRM struct with all required DSC values
346  *
347  * Based on the information inside drm_dsc_config, this function calculates the
348  * total of bytes per pixel.
349  *
350  * @note This calculation requires float point operation, most of it executes
351  * under kernel_fpu_{begin,end}.
352  *
353  * Return:
354  * Return the number of bytes per pixel
355  */
356 u32 calc_dsc_bytes_per_pixel(const struct drm_dsc_config *pps)
357 
358 {
359 	u32 ret;
360 	u8 drm_bpp = pps->bits_per_pixel;
361 	int slice_width  = pps->slice_width;
362 	bool is_navite_422_or_420 = pps->native_422 || pps->native_420;
363 
364 	DC_FP_START();
365 	ret = _do_bytes_per_pixel_calc(slice_width, drm_bpp,
366 				       is_navite_422_or_420);
367 	DC_FP_END();
368 	return ret;
369 }
370 
371 /**
372  * calc_dsc_bpp_x16 - retrieve the dsc bits per pixel
373  * @stream_bandwidth_kbps:
374  * @pix_clk_100hz:
375  * @bpp_increment_div:
376  *
377  * Calculate the total of bits per pixel for DSC configuration.
378  *
379  * @note This calculation requires float point operation, most of it executes
380  * under kernel_fpu_{begin,end}.
381  */
382 u32 calc_dsc_bpp_x16(u32 stream_bandwidth_kbps, u32 pix_clk_100hz,
383 		     u32 bpp_increment_div)
384 {
385 	u32 dsc_bpp;
386 
387 	DC_FP_START();
388 	dsc_bpp =  _do_calc_dsc_bpp_x16(stream_bandwidth_kbps, pix_clk_100hz,
389 					bpp_increment_div);
390 	DC_FP_END();
391 	return dsc_bpp;
392 }
393