1 /* Copyright (c) 2013 The Chromium OS Authors. All rights reserved.
2  * Use of this source code is governed by a BSD-style license that can be
3  * found in the LICENSE file.
4  */
5 
6 /* Copyright (C) 2011 Google Inc. All rights reserved.
7  * Use of this source code is governed by a BSD-style license that can be
8  * found in the LICENSE.WEBKIT file.
9  */
10 
11 #include <stdio.h>
12 #include <stdlib.h>
13 #include <string.h>
14 
15 #include "drc_math.h"
16 #include "drc_kernel.h"
17 
18 #define MAX_PRE_DELAY_FRAMES 1024U
19 #define MAX_PRE_DELAY_FRAMES_MASK (MAX_PRE_DELAY_FRAMES - 1)
20 #define DEFAULT_PRE_DELAY_FRAMES 256U
21 #define DIVISION_FRAMES 32U
22 #define DIVISION_FRAMES_MASK (DIVISION_FRAMES - 1)
23 
24 #define assert_on_compile(e) ((void)sizeof(char[1 - 2 * !(e)]))
25 #define assert_on_compile_is_power_of_2(n) \
26 	assert_on_compile((n) != 0 && (((n) & ((n) - 1)) == 0))
27 
28 const float uninitialized_value = -1;
29 static int drc_math_initialized;
30 
dk_init(struct drc_kernel * dk,float sample_rate)31 void dk_init(struct drc_kernel *dk, float sample_rate)
32 {
33 	unsigned int i;
34 
35 	if (!drc_math_initialized) {
36 		drc_math_initialized = 1;
37 		drc_math_init();
38 	}
39 
40 	dk->sample_rate = sample_rate;
41 	dk->detector_average = 0;
42 	dk->compressor_gain = 1;
43 	dk->enabled = 0;
44 	dk->processed = 0;
45 	dk->last_pre_delay_frames = DEFAULT_PRE_DELAY_FRAMES;
46 	dk->pre_delay_read_index = 0;
47 	dk->pre_delay_write_index = DEFAULT_PRE_DELAY_FRAMES;
48 	dk->max_attack_compression_diff_db = -INFINITY;
49 	dk->ratio = uninitialized_value;
50 	dk->slope = uninitialized_value;
51 	dk->linear_threshold = uninitialized_value;
52 	dk->db_threshold = uninitialized_value;
53 	dk->db_knee = uninitialized_value;
54 	dk->knee_threshold = uninitialized_value;
55 	dk->ratio_base = uninitialized_value;
56 	dk->K = uninitialized_value;
57 
58 	assert_on_compile_is_power_of_2(DIVISION_FRAMES);
59 	assert_on_compile(DIVISION_FRAMES % 4 == 0);
60 	/* Allocate predelay buffers */
61 	assert_on_compile_is_power_of_2(MAX_PRE_DELAY_FRAMES);
62 	for (i = 0; i < DRC_NUM_CHANNELS; i++) {
63 		size_t size = sizeof(float) * MAX_PRE_DELAY_FRAMES;
64 		dk->pre_delay_buffers[i] = (float *)calloc(1, size);
65 	}
66 }
67 
dk_free(struct drc_kernel * dk)68 void dk_free(struct drc_kernel *dk)
69 {
70 	unsigned int i;
71 	for (i = 0; i < DRC_NUM_CHANNELS; ++i)
72 		free(dk->pre_delay_buffers[i]);
73 }
74 
75 /* Sets the pre-delay (lookahead) buffer size */
set_pre_delay_time(struct drc_kernel * dk,float pre_delay_time)76 static void set_pre_delay_time(struct drc_kernel *dk, float pre_delay_time)
77 {
78 	unsigned int i;
79 	/* Re-configure look-ahead section pre-delay if delay time has
80 	 * changed. */
81 	unsigned pre_delay_frames = pre_delay_time * dk->sample_rate;
82 	pre_delay_frames = min(pre_delay_frames, MAX_PRE_DELAY_FRAMES - 1);
83 
84 	/* Make pre_delay_frames multiplies of DIVISION_FRAMES. This way we
85 	 * won't split a division of samples into two blocks of memory, so it is
86 	 * easier to process. This may make the actual delay time slightly less
87 	 * than the specified value, but the difference is less than 1ms. */
88 	pre_delay_frames &= ~DIVISION_FRAMES_MASK;
89 
90 	/* We need at least one division buffer, so the incoming data won't
91 	 * overwrite the output data */
92 	pre_delay_frames = max(pre_delay_frames, DIVISION_FRAMES);
93 
94 	if (dk->last_pre_delay_frames != pre_delay_frames) {
95 		dk->last_pre_delay_frames = pre_delay_frames;
96 		for (i = 0; i < DRC_NUM_CHANNELS; ++i) {
97 			size_t size = sizeof(float) * MAX_PRE_DELAY_FRAMES;
98 			memset(dk->pre_delay_buffers[i], 0, size);
99 		}
100 
101 		dk->pre_delay_read_index = 0;
102 		dk->pre_delay_write_index = pre_delay_frames;
103 	}
104 }
105 
106 /* Exponential curve for the knee.  It is 1st derivative matched at
107  * dk->linear_threshold and asymptotically approaches the value
108  * dk->linear_threshold + 1 / k.
109  *
110  * This is used only when calculating the static curve, not used when actually
111  * compress the input data (knee_curveK below is used instead).
112  */
knee_curve(struct drc_kernel * dk,float x,float k)113 static float knee_curve(struct drc_kernel *dk, float x, float k)
114 {
115 	/* Linear up to threshold. */
116 	if (x < dk->linear_threshold)
117 		return x;
118 
119 	return dk->linear_threshold +
120 		(1 - knee_expf(-k * (x - dk->linear_threshold))) / k;
121 }
122 
123 /* Approximate 1st derivative with input and output expressed in dB.  This slope
124  * is equal to the inverse of the compression "ratio".  In other words, a
125  * compression ratio of 20 would be a slope of 1/20.
126  */
slope_at(struct drc_kernel * dk,float x,float k)127 static float slope_at(struct drc_kernel *dk, float x, float k)
128 {
129 	if (x < dk->linear_threshold)
130 		return 1;
131 
132 	float x2 = x * 1.001;
133 
134 	float x_db = linear_to_decibels(x);
135 	float x2Db = linear_to_decibels(x2);
136 
137 	float y_db = linear_to_decibels(knee_curve(dk, x, k));
138 	float y2Db = linear_to_decibels(knee_curve(dk, x2, k));
139 
140 	float m = (y2Db - y_db) / (x2Db - x_db);
141 
142 	return m;
143 }
144 
k_at_slope(struct drc_kernel * dk,float desired_slope)145 static float k_at_slope(struct drc_kernel *dk, float desired_slope)
146 {
147 	float x_db = dk->db_threshold + dk->db_knee;
148 	float x = decibels_to_linear(x_db);
149 
150 	/* Approximate k given initial values. */
151 	float minK = 0.1;
152 	float maxK = 10000;
153 	float k = 5;
154 	unsigned int i;
155 
156 	for (i = 0; i < 15; ++i) {
157 		/* A high value for k will more quickly asymptotically approach
158 		 * a slope of 0. */
159 		float slope = slope_at(dk, x, k);
160 
161 		if (slope < desired_slope) {
162 			/* k is too high. */
163 			maxK = k;
164 		} else {
165 			/* k is too low. */
166 			minK = k;
167 		}
168 
169 		/* Re-calculate based on geometric mean. */
170 		k = sqrtf(minK * maxK);
171 	}
172 
173 	return k;
174 }
175 
update_static_curve_parameters(struct drc_kernel * dk,float db_threshold,float db_knee,float ratio)176 static void update_static_curve_parameters(struct drc_kernel *dk,
177 					   float db_threshold,
178 					   float db_knee, float ratio)
179 {
180 	if (db_threshold != dk->db_threshold || db_knee != dk->db_knee ||
181 	    ratio != dk->ratio) {
182 		/* Threshold and knee. */
183 		dk->db_threshold = db_threshold;
184 		dk->linear_threshold = decibels_to_linear(db_threshold);
185 		dk->db_knee = db_knee;
186 
187 		/* Compute knee parameters. */
188 		dk->ratio = ratio;
189 		dk->slope = 1 / dk->ratio;
190 
191 		float k = k_at_slope(dk, 1 / dk->ratio);
192 		dk->K = k;
193 		/* See knee_curveK() for details */
194 		dk->knee_alpha = dk->linear_threshold + 1 / k;
195 		dk->knee_beta = -expf(k * dk->linear_threshold) / k;
196 
197 		dk->knee_threshold = decibels_to_linear(db_threshold + db_knee);
198 		/* See volume_gain() for details */
199 		float y0 = knee_curve(dk, dk->knee_threshold, k);
200 		dk->ratio_base = y0 * powf(dk->knee_threshold, -dk->slope);
201 	}
202 }
203 
204 /* This is the knee part of the compression curve. Returns the output level
205  * given the input level x. */
knee_curveK(struct drc_kernel * dk,float x)206 static float knee_curveK(struct drc_kernel *dk, float x)
207 {
208 	/* The formula in knee_curveK is dk->linear_threshold +
209 	 * (1 - expf(-k * (x - dk->linear_threshold))) / k
210 	 * which simplifies to (alpha + beta * expf(gamma))
211 	 * where alpha = dk->linear_threshold + 1 / k
212 	 *	 beta = -expf(k * dk->linear_threshold) / k
213 	 *	 gamma = -k * x
214 	 */
215 	return dk->knee_alpha + dk->knee_beta * knee_expf(-dk->K * x);
216 }
217 
218 /* Full compression curve with constant ratio after knee. Returns the ratio of
219  * output and input signal. */
volume_gain(struct drc_kernel * dk,float x)220 static float volume_gain(struct drc_kernel *dk, float x)
221 {
222 	float y;
223 
224 	if (x < dk->knee_threshold) {
225 		if (x < dk->linear_threshold)
226 			return 1;
227 		y = knee_curveK(dk, x) / x;
228 	} else {
229 		/* Constant ratio after knee.
230 		 * log(y/y0) = s * log(x/x0)
231 		 * => y = y0 * (x/x0)^s
232 		 * => y = [y0 * (1/x0)^s] * x^s
233 		 * => y = dk->ratio_base * x^s
234 		 * => y/x = dk->ratio_base * x^(s - 1)
235 		 * => y/x = dk->ratio_base * e^(log(x) * (s - 1))
236 		 */
237 		y = dk->ratio_base * knee_expf(logf(x) * (dk->slope - 1));
238 	}
239 
240 	return y;
241 }
242 
dk_set_parameters(struct drc_kernel * dk,float db_threshold,float db_knee,float ratio,float attack_time,float release_time,float pre_delay_time,float db_post_gain,float releaseZone1,float releaseZone2,float releaseZone3,float releaseZone4)243 void dk_set_parameters(struct drc_kernel *dk,
244 		       float db_threshold,
245 		       float db_knee,
246 		       float ratio,
247 		       float attack_time,
248 		       float release_time,
249 		       float pre_delay_time,
250 		       float db_post_gain,
251 		       float releaseZone1,
252 		       float releaseZone2,
253 		       float releaseZone3,
254 		       float releaseZone4)
255 {
256 	float sample_rate = dk->sample_rate;
257 
258 	update_static_curve_parameters(dk, db_threshold, db_knee, ratio);
259 
260 	/* Makeup gain. */
261 	float full_range_gain = volume_gain(dk, 1);
262 	float full_range_makeup_gain = 1 / full_range_gain;
263 
264 	/* Empirical/perceptual tuning. */
265 	full_range_makeup_gain = powf(full_range_makeup_gain, 0.6f);
266 
267 	dk->master_linear_gain = decibels_to_linear(db_post_gain) *
268 		full_range_makeup_gain;
269 
270 	/* Attack parameters. */
271 	attack_time = max(0.001f, attack_time);
272 	dk->attack_frames = attack_time * sample_rate;
273 
274 	/* Release parameters. */
275 	float release_frames = sample_rate * release_time;
276 
277 	/* Detector release time. */
278 	float sat_release_time = 0.0025f;
279 	float sat_release_frames = sat_release_time * sample_rate;
280 	dk->sat_release_frames_inv_neg = -1 / sat_release_frames;
281 	dk->sat_release_rate_at_neg_two_db =
282 		decibels_to_linear(-2 * dk->sat_release_frames_inv_neg) - 1;
283 
284 	/* Create a smooth function which passes through four points.
285 	 * Polynomial of the form y = a + b*x + c*x^2 + d*x^3 + e*x^4
286 	 */
287 	float y1 = release_frames * releaseZone1;
288 	float y2 = release_frames * releaseZone2;
289 	float y3 = release_frames * releaseZone3;
290 	float y4 = release_frames * releaseZone4;
291 
292 	/* All of these coefficients were derived for 4th order polynomial curve
293 	 * fitting where the y values match the evenly spaced x values as
294 	 * follows: (y1 : x == 0, y2 : x == 1, y3 : x == 2, y4 : x == 3)
295 	 */
296 	dk->kA = 0.9999999999999998f*y1 + 1.8432219684323923e-16f*y2
297 		- 1.9373394351676423e-16f*y3 + 8.824516011816245e-18f*y4;
298 	dk->kB = -1.5788320352845888f*y1 + 2.3305837032074286f*y2
299 		- 0.9141194204840429f*y3 + 0.1623677525612032f*y4;
300 	dk->kC = 0.5334142869106424f*y1 - 1.272736789213631f*y2
301 		+ 0.9258856042207512f*y3 - 0.18656310191776226f*y4;
302 	dk->kD = 0.08783463138207234f*y1 - 0.1694162967925622f*y2
303 		+ 0.08588057951595272f*y3 - 0.00429891410546283f*y4;
304 	dk->kE = -0.042416883008123074f*y1 + 0.1115693827987602f*y2
305 		- 0.09764676325265872f*y3 + 0.028494263462021576f*y4;
306 
307 	/* x ranges from 0 -> 3	      0	   1	2   3
308 	 *			     -15  -10  -5   0db
309 	 *
310 	 * y calculates adaptive release frames depending on the amount of
311 	 * compression.
312 	 */
313 	set_pre_delay_time(dk, pre_delay_time);
314 }
315 
dk_set_enabled(struct drc_kernel * dk,int enabled)316 void dk_set_enabled(struct drc_kernel *dk, int enabled)
317 {
318 	dk->enabled = enabled;
319 }
320 
321 /* Updates the envelope_rate used for the next division */
dk_update_envelope(struct drc_kernel * dk)322 static void dk_update_envelope(struct drc_kernel *dk)
323 {
324 	const float kA = dk->kA;
325 	const float kB = dk->kB;
326 	const float kC = dk->kC;
327 	const float kD = dk->kD;
328 	const float kE = dk->kE;
329 	const float attack_frames = dk->attack_frames;
330 
331 	/* Calculate desired gain */
332 	float desired_gain = dk->detector_average;
333 
334 	/* Pre-warp so we get desired_gain after sin() warp below. */
335 	float scaled_desired_gain = warp_asinf(desired_gain);
336 
337 	/* Deal with envelopes */
338 
339 	/* envelope_rate is the rate we slew from current compressor level to
340 	 * the desired level.  The exact rate depends on if we're attacking or
341 	 * releasing and by how much.
342 	 */
343 	float envelope_rate;
344 
345 	int is_releasing = scaled_desired_gain > dk->compressor_gain;
346 
347 	/* compression_diff_db is the difference between current compression
348 	 * level and the desired level. */
349 	float compression_diff_db = linear_to_decibels(
350 		dk->compressor_gain / scaled_desired_gain);
351 
352 	if (is_releasing) {
353 		/* Release mode - compression_diff_db should be negative dB */
354 		dk->max_attack_compression_diff_db = -INFINITY;
355 
356 		/* Fix gremlins. */
357 		if (isbadf(compression_diff_db))
358 			compression_diff_db = -1;
359 
360 		/* Adaptive release - higher compression (lower
361 		 * compression_diff_db) releases faster. Contain within range:
362 		 * -12 -> 0 then scale to go from 0 -> 3
363 		 */
364 		float x = compression_diff_db;
365 		x = max(-12.0f, x);
366 		x = min(0.0f, x);
367 		x = 0.25f * (x + 12);
368 
369 		/* Compute adaptive release curve using 4th order polynomial.
370 		 * Normal values for the polynomial coefficients would create a
371 		 * monotonically increasing function.
372 		 */
373 		float x2 = x * x;
374 		float x3 = x2 * x;
375 		float x4 = x2 * x2;
376 		float release_frames = kA + kB * x + kC * x2 + kD * x3 +
377 			kE * x4;
378 
379 #define kSpacingDb 5
380 		float db_per_frame = kSpacingDb / release_frames;
381 		envelope_rate = decibels_to_linear(db_per_frame);
382 	} else {
383 		/* Attack mode - compression_diff_db should be positive dB */
384 
385 		/* Fix gremlins. */
386 		if (isbadf(compression_diff_db))
387 			compression_diff_db = 1;
388 
389 		/* As long as we're still in attack mode, use a rate based off
390 		 * the largest compression_diff_db we've encountered so far.
391 		 */
392 		dk->max_attack_compression_diff_db = max(
393 			dk->max_attack_compression_diff_db,
394 			compression_diff_db);
395 
396 		float eff_atten_diff_db =
397 			max(0.5f, dk->max_attack_compression_diff_db);
398 
399 		float x = 0.25f / eff_atten_diff_db;
400 		envelope_rate = 1 - powf(x, 1 / attack_frames);
401 	}
402 
403 	dk->envelope_rate = envelope_rate;
404 	dk->scaled_desired_gain = scaled_desired_gain;
405 }
406 
407 /* For a division of frames, take the absolute values of left channel and right
408  * channel, store the maximum of them in output. */
409 #ifdef __ARM_NEON__
410 #include <arm_neon.h>
max_abs_division(float * output,float * data0,float * data1)411 static inline void max_abs_division(float *output, float *data0, float *data1)
412 {
413 	float32x4_t x, y;
414 	int count = DIVISION_FRAMES / 4;
415 
416 	__asm__ __volatile__(
417 		"1:                                     \n"
418 		"vld1.32 {%e[x],%f[x]}, [%[data0]]!     \n"
419 		"vld1.32 {%e[y],%f[y]}, [%[data1]]!     \n"
420 		"vabs.f32 %q[x], %q[x]                  \n"
421 		"vabs.f32 %q[y], %q[y]                  \n"
422 		"vmax.f32 %q[x], %q[y]                  \n"
423 		"vst1.32 {%e[x],%f[x]}, [%[output]]!    \n"
424 		"subs %[count], #1                      \n"
425 		"bne 1b                                 \n"
426 		: /* output */
427 		  "=r"(data0),
428 		  "=r"(data1),
429 		  "=r"(output),
430 		  "=r"(count),
431 		  [x]"=&w"(x),
432 		  [y]"=&w"(y)
433 		: /* input */
434 		  [data0]"0"(data0),
435 		  [data1]"1"(data1),
436 		  [output]"2"(output),
437 		  [count]"3"(count)
438 		: /* clobber */
439 		  "memory", "cc"
440 		);
441 }
442 #elif defined(__SSE3__)
443 #include <emmintrin.h>
max_abs_division(float * output,float * data0,float * data1)444 static inline void max_abs_division(float *output, float *data0, float *data1)
445 {
446 	__m128 x, y;
447 	int count = DIVISION_FRAMES / 4;
448 
449 	__asm__ __volatile__(
450 		"1:                                     \n"
451 		"lddqu (%[data0]), %[x]                 \n"
452 		"lddqu (%[data1]), %[y]                 \n"
453 		"andps %[mask], %[x]                    \n"
454 		"andps %[mask], %[y]                    \n"
455 		"maxps %[y], %[x]                       \n"
456 		"movdqu %[x], (%[output])               \n"
457 		"add $16, %[data0]                      \n"
458 		"add $16, %[data1]                      \n"
459 		"add $16, %[output]                     \n"
460 		"sub $1, %[count]                       \n"
461 		"jnz 1b                                 \n"
462 		: /* output */
463 		  [data0]"+r"(data0),
464 		  [data1]"+r"(data1),
465 		  [output]"+r"(output),
466 		  [count]"+r"(count),
467 		  [x]"=&x"(x),
468 		  [y]"=&x"(y)
469 		: /* input */
470 		  [mask]"x"(_mm_set1_epi32(0x7fffffff))
471 		: /* clobber */
472 		  "memory", "cc"
473 		);
474 }
475 #else
max_abs_division(float * output,float * data0,float * data1)476 static inline void max_abs_division(float *output, float *data0, float *data1)
477 {
478 	unsigned int i;
479 	for (i = 0; i < DIVISION_FRAMES; i++)
480 		output[i] = fmaxf(fabsf(data0[i]), fabsf(data1[i]));
481 }
482 #endif
483 
484 /* Update detector_average from the last input division. */
dk_update_detector_average(struct drc_kernel * dk)485 static void dk_update_detector_average(struct drc_kernel *dk)
486 {
487 	float abs_input_array[DIVISION_FRAMES];
488 	const float sat_release_frames_inv_neg = dk->sat_release_frames_inv_neg;
489 	const float sat_release_rate_at_neg_two_db =
490 		dk->sat_release_rate_at_neg_two_db;
491 	float detector_average = dk->detector_average;
492 	unsigned int div_start, i;
493 
494 	/* Calculate the start index of the last input division */
495 	if (dk->pre_delay_write_index == 0) {
496 		div_start = MAX_PRE_DELAY_FRAMES - DIVISION_FRAMES;
497 	} else {
498 		div_start = dk->pre_delay_write_index - DIVISION_FRAMES;
499 	}
500 
501 	/* The max abs value across all channels for this frame */
502 	max_abs_division(abs_input_array,
503 			 &dk->pre_delay_buffers[0][div_start],
504 			 &dk->pre_delay_buffers[1][div_start]);
505 
506 	for (i = 0; i < DIVISION_FRAMES; i++) {
507 		/* Compute compression amount from un-delayed signal */
508 		float abs_input = abs_input_array[i];
509 
510 		/* Calculate shaped power on undelayed input.  Put through
511 		 * shaping curve. This is linear up to the threshold, then
512 		 * enters a "knee" portion followed by the "ratio" portion. The
513 		 * transition from the threshold to the knee is smooth (1st
514 		 * derivative matched). The transition from the knee to the
515 		 * ratio portion is smooth (1st derivative matched).
516 		 */
517 		float gain = volume_gain(dk, abs_input);
518 		int is_release = (gain > detector_average);
519 		if (is_release) {
520 			if (gain > NEG_TWO_DB) {
521 				detector_average += (gain - detector_average) *
522 					sat_release_rate_at_neg_two_db;
523 			} else {
524 				float gain_db = linear_to_decibels(gain);
525 				float db_per_frame = gain_db *
526 					sat_release_frames_inv_neg;
527 				float sat_release_rate =
528 					decibels_to_linear(db_per_frame) - 1;
529 				detector_average += (gain - detector_average) *
530 					sat_release_rate;
531 			}
532 		} else {
533 			detector_average = gain;
534 		}
535 
536 		/* Fix gremlins. */
537 		if (isbadf(detector_average))
538 			detector_average = 1.0f;
539 		else
540 			detector_average = min(detector_average, 1.0f);
541 	}
542 
543 	dk->detector_average = detector_average;
544 }
545 
546 /* Calculate compress_gain from the envelope and apply total_gain to compress
547  * the next output division. */
548 #if defined(__ARM_NEON__)
549 #include <arm_neon.h>
dk_compress_output(struct drc_kernel * dk)550 static void dk_compress_output(struct drc_kernel *dk)
551 {
552 	const float master_linear_gain = dk->master_linear_gain;
553 	const float envelope_rate = dk->envelope_rate;
554 	const float scaled_desired_gain = dk->scaled_desired_gain;
555 	const float compressor_gain = dk->compressor_gain;
556 	unsigned const int div_start = dk->pre_delay_read_index;
557 	float *ptr_left = &dk->pre_delay_buffers[0][div_start];
558 	float *ptr_right = &dk->pre_delay_buffers[1][div_start];
559 	int count = DIVISION_FRAMES / 4;
560 
561 	/* See warp_sinf() for the details for the constants. */
562 	const float32x4_t A7 = vdupq_n_f32(-4.3330336920917034149169921875e-3f);
563 	const float32x4_t A5 = vdupq_n_f32(7.9434238374233245849609375e-2f);
564 	const float32x4_t A3 = vdupq_n_f32(-0.645892798900604248046875f);
565 	const float32x4_t A1 = vdupq_n_f32(1.5707910060882568359375f);
566 
567 	/* Exponential approach to desired gain. */
568 	if (envelope_rate < 1) {
569 		float c = compressor_gain - scaled_desired_gain;
570 		float r = 1 - envelope_rate;
571 		float32x4_t x0 = {c*r, c*r*r, c*r*r*r, c*r*r*r*r};
572 		float32x4_t x, x2, x4, left, right, tmp1, tmp2;
573 
574 		__asm__ __volatile(
575 			"b 2f                                               \n"
576 			"1:                                                 \n"
577 			"vmul.f32 %q[x0], %q[r4]                            \n"
578 			"2:                                                 \n"
579 			"vld1.32 {%e[left],%f[left]}, [%[ptr_left]]         \n"
580 			"vld1.32 {%e[right],%f[right]}, [%[ptr_right]]      \n"
581 			"vadd.f32 %q[x], %q[x0], %q[base]                   \n"
582 			/* Calculate warp_sin() for four values in x. */
583 			"vmul.f32 %q[x2], %q[x], %q[x]                      \n"
584 			"vmov.f32 %q[tmp1], %q[A5]                          \n"
585 			"vmov.f32 %q[tmp2], %q[A1]                          \n"
586 			"vmul.f32 %q[x4], %q[x2], %q[x2]                    \n"
587 			"vmla.f32 %q[tmp1], %q[A7], %q[x2]                  \n"
588 			"vmla.f32 %q[tmp2], %q[A3], %q[x2]                  \n"
589 			"vmla.f32 %q[tmp2], %q[tmp1], %q[x4]                \n"
590 			"vmul.f32 %q[tmp2], %q[tmp2], %q[x]                 \n"
591 			/* Now tmp2 contains the result of warp_sin(). */
592 			"vmul.f32 %q[tmp2], %q[tmp2], %q[g]                 \n"
593 			"vmul.f32 %q[left], %q[tmp2]                        \n"
594 			"vmul.f32 %q[right], %q[tmp2]                       \n"
595 			"vst1.32 {%e[left],%f[left]}, [%[ptr_left]]!        \n"
596 			"vst1.32 {%e[right],%f[right]}, [%[ptr_right]]!     \n"
597 			"subs %[count], #1                                  \n"
598 			"bne 1b                                             \n"
599 			: /* output */
600 			  "=r"(count),
601 			  "=r"(ptr_left),
602 			  "=r"(ptr_right),
603 			  "=w"(x0),
604 			  [x]"=&w"(x),
605 			  [x2]"=&w"(x2),
606 			  [x4]"=&w"(x4),
607 			  [left]"=&w"(left),
608 			  [right]"=&w"(right),
609 			  [tmp1]"=&w"(tmp1),
610 			  [tmp2]"=&w"(tmp2)
611 			: /* input */
612 			  [count]"0"(count),
613 			  [ptr_left]"1"(ptr_left),
614 			  [ptr_right]"2"(ptr_right),
615 			  [x0]"3"(x0),
616 			  [A1]"w"(A1),
617 			  [A3]"w"(A3),
618 			  [A5]"w"(A5),
619 			  [A7]"w"(A7),
620 			  [base]"w"(vdupq_n_f32(scaled_desired_gain)),
621 			  [r4]"w"(vdupq_n_f32(r*r*r*r)),
622 			  [g]"w"(vdupq_n_f32(master_linear_gain))
623 			: /* clobber */
624 			  "memory", "cc"
625 			);
626 		dk->compressor_gain = x[3];
627 	} else {
628 		float c = compressor_gain;
629 		float r = envelope_rate;
630 		float32x4_t x = {c*r, c*r*r, c*r*r*r, c*r*r*r*r};
631 		float32x4_t x2, x4, left, right, tmp1, tmp2;
632 
633 		__asm__ __volatile(
634 			"b 2f                                               \n"
635 			"1:                                                 \n"
636 			"vmul.f32 %q[x], %q[r4]                             \n"
637 			"2:                                                 \n"
638 			"vld1.32 {%e[left],%f[left]}, [%[ptr_left]]         \n"
639 			"vld1.32 {%e[right],%f[right]}, [%[ptr_right]]      \n"
640 			"vmin.f32 %q[x], %q[one]                            \n"
641 			/* Calculate warp_sin() for four values in x. */
642 			"vmul.f32 %q[x2], %q[x], %q[x]                      \n"
643 			"vmov.f32 %q[tmp1], %q[A5]                          \n"
644 			"vmov.f32 %q[tmp2], %q[A1]                          \n"
645 			"vmul.f32 %q[x4], %q[x2], %q[x2]                    \n"
646 			"vmla.f32 %q[tmp1], %q[A7], %q[x2]                  \n"
647 			"vmla.f32 %q[tmp2], %q[A3], %q[x2]                  \n"
648 			"vmla.f32 %q[tmp2], %q[tmp1], %q[x4]                \n"
649 			"vmul.f32 %q[tmp2], %q[tmp2], %q[x]                 \n"
650 			/* Now tmp2 contains the result of warp_sin(). */
651 			"vmul.f32 %q[tmp2], %q[tmp2], %q[g]                 \n"
652 			"vmul.f32 %q[left], %q[tmp2]                        \n"
653 			"vmul.f32 %q[right], %q[tmp2]                       \n"
654 			"vst1.32 {%e[left],%f[left]}, [%[ptr_left]]!        \n"
655 			"vst1.32 {%e[right],%f[right]}, [%[ptr_right]]!     \n"
656 			"subs %[count], #1                                  \n"
657 			"bne 1b                                             \n"
658 			: /* output */
659 			  "=r"(count),
660 			  "=r"(ptr_left),
661 			  "=r"(ptr_right),
662 			  "=w"(x),
663 			  [x2]"=&w"(x2),
664 			  [x4]"=&w"(x4),
665 			  [left]"=&w"(left),
666 			  [right]"=&w"(right),
667 			  [tmp1]"=&w"(tmp1),
668 			  [tmp2]"=&w"(tmp2)
669 			: /* input */
670 			  [count]"0"(count),
671 			  [ptr_left]"1"(ptr_left),
672 			  [ptr_right]"2"(ptr_right),
673 			  [x]"3"(x),
674 			  [A1]"w"(A1),
675 			  [A3]"w"(A3),
676 			  [A5]"w"(A5),
677 			  [A7]"w"(A7),
678 			  [one]"w"(vdupq_n_f32(1)),
679 			  [r4]"w"(vdupq_n_f32(r*r*r*r)),
680 			  [g]"w"(vdupq_n_f32(master_linear_gain))
681 			: /* clobber */
682 			  "memory", "cc"
683 			);
684 		dk->compressor_gain = x[3];
685 	}
686 }
687 #elif defined(__SSE3__) && defined(__x86_64__)
688 #include <emmintrin.h>
dk_compress_output(struct drc_kernel * dk)689 static void dk_compress_output(struct drc_kernel *dk)
690 {
691 	const float master_linear_gain = dk->master_linear_gain;
692 	const float envelope_rate = dk->envelope_rate;
693 	const float scaled_desired_gain = dk->scaled_desired_gain;
694 	const float compressor_gain = dk->compressor_gain;
695 	const int div_start = dk->pre_delay_read_index;
696 	float *ptr_left = &dk->pre_delay_buffers[0][div_start];
697 	float *ptr_right = &dk->pre_delay_buffers[1][div_start];
698 	int count = DIVISION_FRAMES / 4;
699 
700 	/* See warp_sinf() for the details for the constants. */
701 	const __m128 A7 = _mm_set1_ps(-4.3330336920917034149169921875e-3f);
702 	const __m128 A5 = _mm_set1_ps(7.9434238374233245849609375e-2f);
703 	const __m128 A3 = _mm_set1_ps(-0.645892798900604248046875f);
704 	const __m128 A1 = _mm_set1_ps(1.5707910060882568359375f);
705 
706 	/* Exponential approach to desired gain. */
707 	if (envelope_rate < 1) {
708 		float c = compressor_gain - scaled_desired_gain;
709 		float r = 1 - envelope_rate;
710 		__m128 x0 = {c*r, c*r*r, c*r*r*r, c*r*r*r*r};
711 		__m128 x, x2, x4, left, right, tmp1, tmp2;
712 
713 		__asm__ __volatile(
714 			"jmp 2f                                     \n"
715 			"1:                                         \n"
716 			"mulps %[r4], %[x0]                         \n"
717 			"2:                                         \n"
718 			"lddqu (%[ptr_left]), %[left]               \n"
719 			"lddqu (%[ptr_right]), %[right]             \n"
720 			"movaps %[x0], %[x]                         \n"
721 			"addps %[base], %[x]                        \n"
722 			/* Calculate warp_sin() for four values in x. */
723 			"movaps %[x], %[x2]                         \n"
724 			"mulps %[x], %[x2]                          \n"
725 			"movaps %[x2], %[x4]                        \n"
726 			"movaps %[x2], %[tmp1]                      \n"
727 			"movaps %[x2], %[tmp2]                      \n"
728 			"mulps %[x2], %[x4]                         \n"
729 			"mulps %[A7], %[tmp1]                       \n"
730 			"mulps %[A3], %[tmp2]                       \n"
731 			"addps %[A5], %[tmp1]                       \n"
732 			"addps %[A1], %[tmp2]                       \n"
733 			"mulps %[x4], %[tmp1]                       \n"
734 			"addps %[tmp1], %[tmp2]                     \n"
735 			"mulps %[x], %[tmp2]                        \n"
736 			/* Now tmp2 contains the result of warp_sin(). */
737 			"mulps %[g], %[tmp2]                        \n"
738 			"mulps %[tmp2], %[left]                     \n"
739 			"mulps %[tmp2], %[right]                    \n"
740 			"movdqu %[left], (%[ptr_left])              \n"
741 			"movdqu %[right], (%[ptr_right])            \n"
742 			"add $16, %[ptr_left]                       \n"
743 			"add $16, %[ptr_right]                      \n"
744 			"sub $1, %[count]                           \n"
745 			"jne 1b                                     \n"
746 			: /* output */
747 			  "=r"(count),
748 			  "=r"(ptr_left),
749 			  "=r"(ptr_right),
750 			  "=x"(x0),
751 			  [x]"=&x"(x),
752 			  [x2]"=&x"(x2),
753 			  [x4]"=&x"(x4),
754 			  [left]"=&x"(left),
755 			  [right]"=&x"(right),
756 			  [tmp1]"=&x"(tmp1),
757 			  [tmp2]"=&x"(tmp2)
758 			: /* input */
759 			  [count]"0"(count),
760 			  [ptr_left]"1"(ptr_left),
761 			  [ptr_right]"2"(ptr_right),
762 			  [x0]"3"(x0),
763 			  [A1]"x"(A1),
764 			  [A3]"x"(A3),
765 			  [A5]"x"(A5),
766 			  [A7]"x"(A7),
767 			  [base]"x"(_mm_set1_ps(scaled_desired_gain)),
768 			  [r4]"x"(_mm_set1_ps(r*r*r*r)),
769 			  [g]"x"(_mm_set1_ps(master_linear_gain))
770 			: /* clobber */
771 			  "memory", "cc"
772 			);
773 		dk->compressor_gain = x[3];
774 	} else {
775 		/* See warp_sinf() for the details for the constants. */
776 		__m128 A7 = _mm_set1_ps(-4.3330336920917034149169921875e-3f);
777 		__m128 A5 = _mm_set1_ps(7.9434238374233245849609375e-2f);
778 		__m128 A3 = _mm_set1_ps(-0.645892798900604248046875f);
779 		__m128 A1 = _mm_set1_ps(1.5707910060882568359375f);
780 
781 		float c = compressor_gain;
782 		float r = envelope_rate;
783 		__m128 x = {c*r, c*r*r, c*r*r*r, c*r*r*r*r};
784 		__m128 x2, x4, left, right, tmp1, tmp2;
785 
786 		__asm__ __volatile(
787 			"jmp 2f                                     \n"
788 			"1:                                         \n"
789 			"mulps %[r4], %[x]                          \n"
790 			"2:                                         \n"
791 			"lddqu (%[ptr_left]), %[left]               \n"
792 			"lddqu (%[ptr_right]), %[right]             \n"
793 			"minps %[one], %[x]                         \n"
794 			/* Calculate warp_sin() for four values in x. */
795 			"movaps %[x], %[x2]                         \n"
796 			"mulps %[x], %[x2]                          \n"
797 			"movaps %[x2], %[x4]                        \n"
798 			"movaps %[x2], %[tmp1]                      \n"
799 			"movaps %[x2], %[tmp2]                      \n"
800 			"mulps %[x2], %[x4]                         \n"
801 			"mulps %[A7], %[tmp1]                       \n"
802 			"mulps %[A3], %[tmp2]                       \n"
803 			"addps %[A5], %[tmp1]                       \n"
804 			"addps %[A1], %[tmp2]                       \n"
805 			"mulps %[x4], %[tmp1]                       \n"
806 			"addps %[tmp1], %[tmp2]                     \n"
807 			"mulps %[x], %[tmp2]                        \n"
808 			/* Now tmp2 contains the result of warp_sin(). */
809 			"mulps %[g], %[tmp2]                        \n"
810 			"mulps %[tmp2], %[left]                     \n"
811 			"mulps %[tmp2], %[right]                    \n"
812 			"movdqu %[left], (%[ptr_left])              \n"
813 			"movdqu %[right], (%[ptr_right])            \n"
814 			"add $16, %[ptr_left]                       \n"
815 			"add $16, %[ptr_right]                      \n"
816 			"sub $1, %[count]                           \n"
817 			"jne 1b                                     \n"
818 			: /* output */
819 			  "=r"(count),
820 			  "=r"(ptr_left),
821 			  "=r"(ptr_right),
822 			  "=x"(x),
823 			  [x2]"=&x"(x2),
824 			  [x4]"=&x"(x4),
825 			  [left]"=&x"(left),
826 			  [right]"=&x"(right),
827 			  [tmp1]"=&x"(tmp1),
828 			  [tmp2]"=&x"(tmp2)
829 			: /* input */
830 			  [count]"0"(count),
831 			  [ptr_left]"1"(ptr_left),
832 			  [ptr_right]"2"(ptr_right),
833 			  [x]"3"(x),
834 			  [A1]"x"(A1),
835 			  [A3]"x"(A3),
836 			  [A5]"x"(A5),
837 			  [A7]"x"(A7),
838 			  [one]"x"(_mm_set1_ps(1)),
839 			  [r4]"x"(_mm_set1_ps(r*r*r*r)),
840 			  [g]"x"(_mm_set1_ps(master_linear_gain))
841 			: /* clobber */
842 			  "memory", "cc"
843 			);
844 		dk->compressor_gain = x[3];
845 	}
846 }
847 #else
dk_compress_output(struct drc_kernel * dk)848 static void dk_compress_output(struct drc_kernel *dk)
849 {
850 	const float master_linear_gain = dk->master_linear_gain;
851 	const float envelope_rate = dk->envelope_rate;
852 	const float scaled_desired_gain = dk->scaled_desired_gain;
853 	const float compressor_gain = dk->compressor_gain;
854 	const int div_start = dk->pre_delay_read_index;
855 	float *ptr_left = &dk->pre_delay_buffers[0][div_start];
856 	float *ptr_right = &dk->pre_delay_buffers[1][div_start];
857 	unsigned int count = DIVISION_FRAMES / 4;
858 
859 	unsigned int i, j;
860 
861 	/* Exponential approach to desired gain. */
862 	if (envelope_rate < 1) {
863 		/* Attack - reduce gain to desired. */
864 		float c = compressor_gain - scaled_desired_gain;
865 		float base = scaled_desired_gain;
866 		float r = 1 - envelope_rate;
867 		float x[4] = {c*r, c*r*r, c*r*r*r, c*r*r*r*r};
868 		float r4 = r*r*r*r;
869 
870 		i = 0;
871 		while (1) {
872 			for (j = 0; j < 4; j++) {
873 				/* Warp pre-compression gain to smooth out sharp
874 				 * exponential transition points.
875 				 */
876 				float post_warp_compressor_gain =
877 					warp_sinf(x[j] + base);
878 
879 				/* Calculate total gain using master gain. */
880 				float total_gain = master_linear_gain *
881 					post_warp_compressor_gain;
882 
883 				/* Apply final gain. */
884 				*ptr_left++ *= total_gain;
885 				*ptr_right++ *= total_gain;
886 			}
887 
888 			if (++i == count)
889 				break;
890 
891 			for (j = 0; j < 4; j++)
892 				x[j] = x[j] * r4;
893 		}
894 
895 		dk->compressor_gain = x[3] + base;
896 	} else {
897 		/* Release - exponentially increase gain to 1.0 */
898 		float c = compressor_gain;
899 		float r = envelope_rate;
900 		float x[4] = {c*r, c*r*r, c*r*r*r, c*r*r*r*r};
901 		float r4 = r*r*r*r;
902 
903 		i = 0;
904 		while (1) {
905 			for (j = 0; j < 4; j++) {
906 				/* Warp pre-compression gain to smooth out sharp
907 				 * exponential transition points.
908 				 */
909 				float post_warp_compressor_gain =
910 					warp_sinf(x[j]);
911 
912 				/* Calculate total gain using master gain. */
913 				float total_gain = master_linear_gain *
914 					post_warp_compressor_gain;
915 
916 				/* Apply final gain. */
917 				*ptr_left++ *= total_gain;
918 				*ptr_right++ *= total_gain;
919 			}
920 
921 			if (++i == count)
922 				break;
923 
924 			for (j = 0; j < 4; j++)
925 				x[j] = min(1.0f, x[j] * r4);
926 		}
927 
928 		dk->compressor_gain = x[3];
929 	}
930 }
931 #endif
932 
933 /* After one complete divison of samples have been received (and one divison of
934  * samples have been output), we calculate shaped power average
935  * (detector_average) from the input division, update envelope parameters from
936  * detector_average, then prepare the next output division by applying the
937  * envelope to compress the samples.
938  */
dk_process_one_division(struct drc_kernel * dk)939 static void dk_process_one_division(struct drc_kernel *dk)
940 {
941 	dk_update_detector_average(dk);
942 	dk_update_envelope(dk);
943 	dk_compress_output(dk);
944 }
945 
946 /* Copy the input data to the pre-delay buffer, and copy the output data back to
947  * the input buffer */
dk_copy_fragment(struct drc_kernel * dk,float * data_channels[],unsigned frame_index,int frames_to_process)948 static void dk_copy_fragment(struct drc_kernel *dk, float *data_channels[],
949 			     unsigned frame_index, int frames_to_process)
950 {
951 	int write_index = dk->pre_delay_write_index;
952 	int read_index = dk->pre_delay_read_index;
953 	int j;
954 
955 	for (j = 0; j < DRC_NUM_CHANNELS; ++j) {
956 		memcpy(&dk->pre_delay_buffers[j][write_index],
957 		       &data_channels[j][frame_index],
958 		       frames_to_process * sizeof(float));
959 		memcpy(&data_channels[j][frame_index],
960 		       &dk->pre_delay_buffers[j][read_index],
961 		       frames_to_process * sizeof(float));
962 	}
963 
964 	dk->pre_delay_write_index = (write_index + frames_to_process) &
965 		MAX_PRE_DELAY_FRAMES_MASK;
966 	dk->pre_delay_read_index = (read_index + frames_to_process) &
967 		MAX_PRE_DELAY_FRAMES_MASK;
968 }
969 
970 /* Delay the input sample only and don't do other processing. This is used when
971  * the kernel is disabled. We want to do this to match the processing delay in
972  * kernels of other bands.
973  */
dk_process_delay_only(struct drc_kernel * dk,float * data_channels[],unsigned count)974 static void dk_process_delay_only(struct drc_kernel *dk, float *data_channels[],
975 				  unsigned count)
976 {
977 	int read_index = dk->pre_delay_read_index;
978 	int write_index = dk->pre_delay_write_index;
979 	unsigned int i = 0;
980 
981 	while (i < count) {
982 		unsigned int j;
983 		unsigned int small = min(read_index, write_index);
984 		unsigned int large = max(read_index, write_index);
985 		/* chunk is the minimum of readable samples in contiguous
986 		 * buffer, writable samples in contiguous buffer, and the
987 		 * available input samples. */
988 		unsigned int chunk = min(large - small,
989 					 MAX_PRE_DELAY_FRAMES - large);
990 		chunk = min(chunk, count - i);
991 		for (j = 0; j < DRC_NUM_CHANNELS; ++j) {
992 			memcpy(&dk->pre_delay_buffers[j][write_index],
993 			       &data_channels[j][i],
994 			       chunk * sizeof(float));
995 			memcpy(&data_channels[j][i],
996 			       &dk->pre_delay_buffers[j][read_index],
997 			       chunk * sizeof(float));
998 		}
999 		read_index = (read_index + chunk) & MAX_PRE_DELAY_FRAMES_MASK;
1000 		write_index = (write_index + chunk) & MAX_PRE_DELAY_FRAMES_MASK;
1001 		i += chunk;
1002 	}
1003 
1004 	dk->pre_delay_read_index = read_index;
1005 	dk->pre_delay_write_index = write_index;
1006 }
1007 
dk_process(struct drc_kernel * dk,float * data_channels[],unsigned count)1008 void dk_process(struct drc_kernel *dk, float *data_channels[], unsigned count)
1009 {
1010 	unsigned int i = 0;
1011 	int fragment;
1012 
1013 	if (!dk->enabled) {
1014 		dk_process_delay_only(dk, data_channels, count);
1015 		return;
1016 	}
1017 
1018 	if (!dk->processed) {
1019 		dk_update_envelope(dk);
1020 		dk_compress_output(dk);
1021 		dk->processed = 1;
1022 	}
1023 
1024 	int offset = dk->pre_delay_write_index & DIVISION_FRAMES_MASK;
1025 	while (i < count) {
1026 		fragment = min(DIVISION_FRAMES - offset, count - i);
1027 		dk_copy_fragment(dk, data_channels, i, fragment);
1028 		i += fragment;
1029 		offset = (offset + fragment) & DIVISION_FRAMES_MASK;
1030 
1031 		/* Process the input division (32 frames). */
1032 		if (offset == 0)
1033 			dk_process_one_division(dk);
1034 	}
1035 }
1036