1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 2005-2009 Ariff Abdullah <ariff@FreeBSD.org>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 /*
30 * feeder_rate: (Codename: Z Resampler), which means any effort to create
31 * future replacement for this resampler are simply absurd unless
32 * the world decide to add new alphabet after Z.
33 *
34 * FreeBSD bandlimited sinc interpolator, technically based on
35 * "Digital Audio Resampling" by Julius O. Smith III
36 * - http://ccrma.stanford.edu/~jos/resample/
37 *
38 * The Good:
39 * + all out fixed point integer operations, no soft-float or anything like
40 * that.
41 * + classic polyphase converters with high quality coefficient's polynomial
42 * interpolators.
43 * + fast, faster, or the fastest of its kind.
44 * + compile time configurable.
45 * + etc etc..
46 *
47 * The Bad:
48 * - The z, z_, and Z_ . Due to mental block (or maybe just 0x7a69), I
49 * couldn't think of anything simpler than that (feeder_rate_xxx is just
50 * too long). Expect possible clashes with other zitizens (any?).
51 */
52
53 #ifdef _KERNEL
54 #ifdef HAVE_KERNEL_OPTION_HEADERS
55 #include "opt_snd.h"
56 #endif
57 #include <dev/sound/pcm/sound.h>
58 #include <dev/sound/pcm/pcm.h>
59 #include "feeder_if.h"
60
61 #define SND_USE_FXDIV
62 #include "snd_fxdiv_gen.h"
63 #endif
64
65 #include "feeder_rate_gen.h"
66
67 #if !defined(_KERNEL) && defined(SND_DIAGNOSTIC)
68 #undef Z_DIAGNOSTIC
69 #define Z_DIAGNOSTIC 1
70 #elif defined(_KERNEL)
71 #undef Z_DIAGNOSTIC
72 #endif
73
74 #ifndef Z_QUALITY_DEFAULT
75 #define Z_QUALITY_DEFAULT Z_QUALITY_LINEAR
76 #endif
77
78 #define Z_RESERVOIR 2048
79 #define Z_RESERVOIR_MAX 131072
80
81 #define Z_SINC_MAX 0x3fffff
82 #define Z_SINC_DOWNMAX 48 /* 384000 / 8000 */
83
84 #ifdef _KERNEL
85 #define Z_POLYPHASE_MAX 183040 /* 286 taps, 640 phases */
86 #else
87 #define Z_POLYPHASE_MAX 1464320 /* 286 taps, 5120 phases */
88 #endif
89
90 #define Z_RATE_DEFAULT 48000
91
92 #define Z_RATE_MIN FEEDRATE_RATEMIN
93 #define Z_RATE_MAX FEEDRATE_RATEMAX
94 #define Z_ROUNDHZ FEEDRATE_ROUNDHZ
95 #define Z_ROUNDHZ_MIN FEEDRATE_ROUNDHZ_MIN
96 #define Z_ROUNDHZ_MAX FEEDRATE_ROUNDHZ_MAX
97
98 #define Z_RATE_SRC FEEDRATE_SRC
99 #define Z_RATE_DST FEEDRATE_DST
100 #define Z_RATE_QUALITY FEEDRATE_QUALITY
101 #define Z_RATE_CHANNELS FEEDRATE_CHANNELS
102
103 #define Z_PARANOID 1
104
105 #define Z_MULTIFORMAT 1
106
107 #ifdef _KERNEL
108 #undef Z_USE_ALPHADRIFT
109 #define Z_USE_ALPHADRIFT 1
110 #endif
111
112 #define Z_FACTOR_MIN 1
113 #define Z_FACTOR_MAX Z_MASK
114 #define Z_FACTOR_SAFE(v) (!((v) < Z_FACTOR_MIN || (v) > Z_FACTOR_MAX))
115
116 struct z_info;
117
118 typedef void (*z_resampler_t)(struct z_info *, uint8_t *);
119
120 struct z_info {
121 int32_t rsrc, rdst; /* original source / destination rates */
122 int32_t src, dst; /* rounded source / destination rates */
123 int32_t channels; /* total channels */
124 int32_t bps; /* bytes-per-sample */
125 int32_t quality; /* resampling quality */
126
127 int32_t z_gx, z_gy; /* interpolation / decimation ratio */
128 int32_t z_alpha; /* output sample time phase / drift */
129 uint8_t *z_delay; /* FIR delay line / linear buffer */
130 int32_t *z_coeff; /* FIR coefficients */
131 int32_t *z_dcoeff; /* FIR coefficients differences */
132 int32_t *z_pcoeff; /* FIR polyphase coefficients */
133 int32_t z_scale; /* output scaling */
134 int32_t z_dx; /* input sample drift increment */
135 int32_t z_dy; /* output sample drift increment */
136 #ifdef Z_USE_ALPHADRIFT
137 int32_t z_alphadrift; /* alpha drift rate */
138 int32_t z_startdrift; /* buffer start position drift rate */
139 #endif
140 int32_t z_mask; /* delay line full length mask */
141 int32_t z_size; /* half width of FIR taps */
142 int32_t z_full; /* full size of delay line */
143 int32_t z_alloc; /* largest allocated full size of delay line */
144 int32_t z_start; /* buffer processing start position */
145 int32_t z_pos; /* current position for the next feed */
146 #ifdef Z_DIAGNOSTIC
147 uint32_t z_cycle; /* output cycle, purely for statistical */
148 #endif
149 int32_t z_maxfeed; /* maximum feed to avoid 32bit overflow */
150
151 z_resampler_t z_resample;
152 };
153
154 int feeder_rate_min = Z_RATE_MIN;
155 int feeder_rate_max = Z_RATE_MAX;
156 int feeder_rate_round = Z_ROUNDHZ;
157 int feeder_rate_quality = Z_QUALITY_DEFAULT;
158
159 static int feeder_rate_polyphase_max = Z_POLYPHASE_MAX;
160
161 #ifdef _KERNEL
162 static char feeder_rate_presets[] = FEEDER_RATE_PRESETS;
163 SYSCTL_STRING(_hw_snd, OID_AUTO, feeder_rate_presets, CTLFLAG_RD,
164 &feeder_rate_presets, 0, "compile-time rate presets");
165 SYSCTL_INT(_hw_snd, OID_AUTO, feeder_rate_polyphase_max, CTLFLAG_RWTUN,
166 &feeder_rate_polyphase_max, 0, "maximum allowable polyphase entries");
167
168 static int
sysctl_hw_snd_feeder_rate_min(SYSCTL_HANDLER_ARGS)169 sysctl_hw_snd_feeder_rate_min(SYSCTL_HANDLER_ARGS)
170 {
171 int err, val;
172
173 val = feeder_rate_min;
174 err = sysctl_handle_int(oidp, &val, 0, req);
175
176 if (err != 0 || req->newptr == NULL || val == feeder_rate_min)
177 return (err);
178
179 if (!(Z_FACTOR_SAFE(val) && val < feeder_rate_max))
180 return (EINVAL);
181
182 feeder_rate_min = val;
183
184 return (0);
185 }
186 SYSCTL_PROC(_hw_snd, OID_AUTO, feeder_rate_min,
187 CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, 0, sizeof(int),
188 sysctl_hw_snd_feeder_rate_min, "I",
189 "minimum allowable rate");
190
191 static int
sysctl_hw_snd_feeder_rate_max(SYSCTL_HANDLER_ARGS)192 sysctl_hw_snd_feeder_rate_max(SYSCTL_HANDLER_ARGS)
193 {
194 int err, val;
195
196 val = feeder_rate_max;
197 err = sysctl_handle_int(oidp, &val, 0, req);
198
199 if (err != 0 || req->newptr == NULL || val == feeder_rate_max)
200 return (err);
201
202 if (!(Z_FACTOR_SAFE(val) && val > feeder_rate_min))
203 return (EINVAL);
204
205 feeder_rate_max = val;
206
207 return (0);
208 }
209 SYSCTL_PROC(_hw_snd, OID_AUTO, feeder_rate_max,
210 CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, 0, sizeof(int),
211 sysctl_hw_snd_feeder_rate_max, "I",
212 "maximum allowable rate");
213
214 static int
sysctl_hw_snd_feeder_rate_round(SYSCTL_HANDLER_ARGS)215 sysctl_hw_snd_feeder_rate_round(SYSCTL_HANDLER_ARGS)
216 {
217 int err, val;
218
219 val = feeder_rate_round;
220 err = sysctl_handle_int(oidp, &val, 0, req);
221
222 if (err != 0 || req->newptr == NULL || val == feeder_rate_round)
223 return (err);
224
225 if (val < Z_ROUNDHZ_MIN || val > Z_ROUNDHZ_MAX)
226 return (EINVAL);
227
228 feeder_rate_round = val - (val % Z_ROUNDHZ);
229
230 return (0);
231 }
232 SYSCTL_PROC(_hw_snd, OID_AUTO, feeder_rate_round,
233 CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, 0, sizeof(int),
234 sysctl_hw_snd_feeder_rate_round, "I",
235 "sample rate converter rounding threshold");
236
237 static int
sysctl_hw_snd_feeder_rate_quality(SYSCTL_HANDLER_ARGS)238 sysctl_hw_snd_feeder_rate_quality(SYSCTL_HANDLER_ARGS)
239 {
240 struct snddev_info *d;
241 struct pcm_channel *c;
242 struct pcm_feeder *f;
243 int i, err, val;
244
245 val = feeder_rate_quality;
246 err = sysctl_handle_int(oidp, &val, 0, req);
247
248 if (err != 0 || req->newptr == NULL || val == feeder_rate_quality)
249 return (err);
250
251 if (val < Z_QUALITY_MIN || val > Z_QUALITY_MAX)
252 return (EINVAL);
253
254 feeder_rate_quality = val;
255
256 /*
257 * Traverse all available channels on each device and try to
258 * set resampler quality if and only if it is exist as
259 * part of feeder chains and the channel is idle.
260 */
261 for (i = 0; pcm_devclass != NULL &&
262 i < devclass_get_maxunit(pcm_devclass); i++) {
263 d = devclass_get_softc(pcm_devclass, i);
264 if (!PCM_REGISTERED(d))
265 continue;
266 PCM_LOCK(d);
267 PCM_WAIT(d);
268 PCM_ACQUIRE(d);
269 CHN_FOREACH(c, d, channels.pcm) {
270 CHN_LOCK(c);
271 f = feeder_find(c, FEEDER_RATE);
272 if (f == NULL || f->data == NULL || CHN_STARTED(c)) {
273 CHN_UNLOCK(c);
274 continue;
275 }
276 (void)FEEDER_SET(f, FEEDRATE_QUALITY, val);
277 CHN_UNLOCK(c);
278 }
279 PCM_RELEASE(d);
280 PCM_UNLOCK(d);
281 }
282
283 return (0);
284 }
285 SYSCTL_PROC(_hw_snd, OID_AUTO, feeder_rate_quality,
286 CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, 0, sizeof(int),
287 sysctl_hw_snd_feeder_rate_quality, "I",
288 "sample rate converter quality ("__XSTRING(Z_QUALITY_MIN)"=low .. "
289 __XSTRING(Z_QUALITY_MAX)"=high)");
290 #endif /* _KERNEL */
291
292 /*
293 * Resampler type.
294 */
295 #define Z_IS_ZOH(i) ((i)->quality == Z_QUALITY_ZOH)
296 #define Z_IS_LINEAR(i) ((i)->quality == Z_QUALITY_LINEAR)
297 #define Z_IS_SINC(i) ((i)->quality > Z_QUALITY_LINEAR)
298
299 /*
300 * Macroses for accurate sample time drift calculations.
301 *
302 * gy2gx : given the amount of output, return the _exact_ required amount of
303 * input.
304 * gx2gy : given the amount of input, return the _maximum_ amount of output
305 * that will be generated.
306 * drift : given the amount of input and output, return the elapsed
307 * sample-time.
308 */
309 #define _Z_GCAST(x) ((uint64_t)(x))
310
311 #if defined(__i386__)
312 /*
313 * This is where i386 being beaten to a pulp. Fortunately this function is
314 * rarely being called and if it is, it will decide the best (hopefully)
315 * fastest way to do the division. If we can ensure that everything is dword
316 * aligned, letting the compiler to call udivdi3 to do the division can be
317 * faster compared to this.
318 *
319 * amd64 is the clear winner here, no question about it.
320 */
321 static __inline uint32_t
Z_DIV(uint64_t v,uint32_t d)322 Z_DIV(uint64_t v, uint32_t d)
323 {
324 uint32_t hi, lo, quo, rem;
325
326 hi = v >> 32;
327 lo = v & 0xffffffff;
328
329 /*
330 * As much as we can, try to avoid long division like a plague.
331 */
332 if (hi == 0)
333 quo = lo / d;
334 else
335 __asm("divl %2"
336 : "=a" (quo), "=d" (rem)
337 : "r" (d), "0" (lo), "1" (hi));
338
339 return (quo);
340 }
341 #else
342 #define Z_DIV(x, y) ((x) / (y))
343 #endif
344
345 #define _Z_GY2GX(i, a, v) \
346 Z_DIV(((_Z_GCAST((i)->z_gx) * (v)) + ((i)->z_gy - (a) - 1)), \
347 (i)->z_gy)
348
349 #define _Z_GX2GY(i, a, v) \
350 Z_DIV(((_Z_GCAST((i)->z_gy) * (v)) + (a)), (i)->z_gx)
351
352 #define _Z_DRIFT(i, x, y) \
353 ((_Z_GCAST((i)->z_gy) * (x)) - (_Z_GCAST((i)->z_gx) * (y)))
354
355 #define z_gy2gx(i, v) _Z_GY2GX(i, (i)->z_alpha, v)
356 #define z_gx2gy(i, v) _Z_GX2GY(i, (i)->z_alpha, v)
357 #define z_drift(i, x, y) _Z_DRIFT(i, x, y)
358
359 /*
360 * Macroses for SINC coefficients table manipulations.. whatever.
361 */
362 #define Z_SINC_COEFF_IDX(i) ((i)->quality - Z_QUALITY_LINEAR - 1)
363
364 #define Z_SINC_LEN(i) \
365 ((int32_t)(((uint64_t)z_coeff_tab[Z_SINC_COEFF_IDX(i)].len << \
366 Z_SHIFT) / (i)->z_dy))
367
368 #define Z_SINC_BASE_LEN(i) \
369 ((z_coeff_tab[Z_SINC_COEFF_IDX(i)].len - 1) >> (Z_DRIFT_SHIFT - 1))
370
371 /*
372 * Macroses for linear delay buffer operations. Alignment is not
373 * really necessary since we're not using true circular buffer, but it
374 * will help us guard against possible trespasser. To be honest,
375 * the linear block operations does not need guarding at all due to
376 * accurate drifting!
377 */
378 #define z_align(i, v) ((v) & (i)->z_mask)
379 #define z_next(i, o, v) z_align(i, (o) + (v))
380 #define z_prev(i, o, v) z_align(i, (o) - (v))
381 #define z_fetched(i) (z_align(i, (i)->z_pos - (i)->z_start) - 1)
382 #define z_free(i) ((i)->z_full - (i)->z_pos)
383
384 /*
385 * Macroses for Bla Bla .. :)
386 */
387 #define z_copy(src, dst, sz) (void)memcpy(dst, src, sz)
388 #define z_feed(...) FEEDER_FEED(__VA_ARGS__)
389
390 static __inline uint32_t
z_min(uint32_t x,uint32_t y)391 z_min(uint32_t x, uint32_t y)
392 {
393
394 return ((x < y) ? x : y);
395 }
396
397 static int32_t
z_gcd(int32_t x,int32_t y)398 z_gcd(int32_t x, int32_t y)
399 {
400 int32_t w;
401
402 while (y != 0) {
403 w = x % y;
404 x = y;
405 y = w;
406 }
407
408 return (x);
409 }
410
411 static int32_t
z_roundpow2(int32_t v)412 z_roundpow2(int32_t v)
413 {
414 int32_t i;
415
416 i = 1;
417
418 /*
419 * Let it overflow at will..
420 */
421 while (i > 0 && i < v)
422 i <<= 1;
423
424 return (i);
425 }
426
427 /*
428 * Zero Order Hold, the worst of the worst, an insult against quality,
429 * but super fast.
430 */
431 static void
z_feed_zoh(struct z_info * info,uint8_t * dst)432 z_feed_zoh(struct z_info *info, uint8_t *dst)
433 {
434 uint32_t cnt;
435 uint8_t *src;
436
437 cnt = info->channels * info->bps;
438 src = info->z_delay + (info->z_start * cnt);
439
440 /*
441 * This is a bit faster than doing bcopy() since we're dealing
442 * with possible unaligned samples.
443 */
444 do {
445 *dst++ = *src++;
446 } while (--cnt != 0);
447 }
448
449 /*
450 * Linear Interpolation. This at least sounds better (perceptually) and fast,
451 * but without any proper filtering which means aliasing still exist and
452 * could become worst with a right sample. Interpolation centered within
453 * Z_LINEAR_ONE between the present and previous sample and everything is
454 * done with simple 32bit scaling arithmetic.
455 */
456 #define Z_DECLARE_LINEAR(SIGN, BIT, ENDIAN) \
457 static void \
458 z_feed_linear_##SIGN##BIT##ENDIAN(struct z_info *info, uint8_t *dst) \
459 { \
460 int32_t z; \
461 intpcm_t x, y; \
462 uint32_t ch; \
463 uint8_t *sx, *sy; \
464 \
465 z = ((uint32_t)info->z_alpha * info->z_dx) >> Z_LINEAR_UNSHIFT; \
466 \
467 sx = info->z_delay + (info->z_start * info->channels * \
468 PCM_##BIT##_BPS); \
469 sy = sx - (info->channels * PCM_##BIT##_BPS); \
470 \
471 ch = info->channels; \
472 \
473 do { \
474 x = _PCM_READ_##SIGN##BIT##_##ENDIAN(sx); \
475 y = _PCM_READ_##SIGN##BIT##_##ENDIAN(sy); \
476 x = Z_LINEAR_INTERPOLATE_##BIT(z, x, y); \
477 _PCM_WRITE_##SIGN##BIT##_##ENDIAN(dst, x); \
478 sx += PCM_##BIT##_BPS; \
479 sy += PCM_##BIT##_BPS; \
480 dst += PCM_##BIT##_BPS; \
481 } while (--ch != 0); \
482 }
483
484 /*
485 * Userland clipping diagnostic check, not enabled in kernel compilation.
486 * While doing sinc interpolation, unrealistic samples like full scale sine
487 * wav will clip, but for other things this will not make any noise at all.
488 * Everybody should learn how to normalized perceived loudness of their own
489 * music/sounds/samples (hint: ReplayGain).
490 */
491 #ifdef Z_DIAGNOSTIC
492 #define Z_CLIP_CHECK(v, BIT) do { \
493 if ((v) > PCM_S##BIT##_MAX) { \
494 fprintf(stderr, "Overflow: v=%jd, max=%jd\n", \
495 (intmax_t)(v), (intmax_t)PCM_S##BIT##_MAX); \
496 } else if ((v) < PCM_S##BIT##_MIN) { \
497 fprintf(stderr, "Underflow: v=%jd, min=%jd\n", \
498 (intmax_t)(v), (intmax_t)PCM_S##BIT##_MIN); \
499 } \
500 } while (0)
501 #else
502 #define Z_CLIP_CHECK(...)
503 #endif
504
505 #define Z_CLAMP(v, BIT) \
506 (((v) > PCM_S##BIT##_MAX) ? PCM_S##BIT##_MAX : \
507 (((v) < PCM_S##BIT##_MIN) ? PCM_S##BIT##_MIN : (v)))
508
509 /*
510 * Sine Cardinal (SINC) Interpolation. Scaling is done in 64 bit, so
511 * there's no point to hold the plate any longer. All samples will be
512 * shifted to a full 32 bit, scaled and restored during write for
513 * maximum dynamic range (only for downsampling).
514 */
515 #define _Z_SINC_ACCUMULATE(SIGN, BIT, ENDIAN, adv) \
516 c += z >> Z_SHIFT; \
517 z &= Z_MASK; \
518 coeff = Z_COEFF_INTERPOLATE(z, z_coeff[c], z_dcoeff[c]); \
519 x = _PCM_READ_##SIGN##BIT##_##ENDIAN(p); \
520 v += Z_NORM_##BIT((intpcm64_t)x * coeff); \
521 z += info->z_dy; \
522 p adv##= info->channels * PCM_##BIT##_BPS
523
524 /*
525 * XXX GCC4 optimization is such a !@#$%, need manual unrolling.
526 */
527 #if defined(__GNUC__) && __GNUC__ >= 4
528 #define Z_SINC_ACCUMULATE(...) do { \
529 _Z_SINC_ACCUMULATE(__VA_ARGS__); \
530 _Z_SINC_ACCUMULATE(__VA_ARGS__); \
531 } while (0)
532 #define Z_SINC_ACCUMULATE_DECR 2
533 #else
534 #define Z_SINC_ACCUMULATE(...) do { \
535 _Z_SINC_ACCUMULATE(__VA_ARGS__); \
536 } while (0)
537 #define Z_SINC_ACCUMULATE_DECR 1
538 #endif
539
540 #define Z_DECLARE_SINC(SIGN, BIT, ENDIAN) \
541 static void \
542 z_feed_sinc_##SIGN##BIT##ENDIAN(struct z_info *info, uint8_t *dst) \
543 { \
544 intpcm64_t v; \
545 intpcm_t x; \
546 uint8_t *p; \
547 int32_t coeff, z, *z_coeff, *z_dcoeff; \
548 uint32_t c, center, ch, i; \
549 \
550 z_coeff = info->z_coeff; \
551 z_dcoeff = info->z_dcoeff; \
552 center = z_prev(info, info->z_start, info->z_size); \
553 ch = info->channels * PCM_##BIT##_BPS; \
554 dst += ch; \
555 \
556 do { \
557 dst -= PCM_##BIT##_BPS; \
558 ch -= PCM_##BIT##_BPS; \
559 v = 0; \
560 z = info->z_alpha * info->z_dx; \
561 c = 0; \
562 p = info->z_delay + (z_next(info, center, 1) * \
563 info->channels * PCM_##BIT##_BPS) + ch; \
564 for (i = info->z_size; i != 0; i -= Z_SINC_ACCUMULATE_DECR) \
565 Z_SINC_ACCUMULATE(SIGN, BIT, ENDIAN, +); \
566 z = info->z_dy - (info->z_alpha * info->z_dx); \
567 c = 0; \
568 p = info->z_delay + (center * info->channels * \
569 PCM_##BIT##_BPS) + ch; \
570 for (i = info->z_size; i != 0; i -= Z_SINC_ACCUMULATE_DECR) \
571 Z_SINC_ACCUMULATE(SIGN, BIT, ENDIAN, -); \
572 if (info->z_scale != Z_ONE) \
573 v = Z_SCALE_##BIT(v, info->z_scale); \
574 else \
575 v >>= Z_COEFF_SHIFT - Z_GUARD_BIT_##BIT; \
576 Z_CLIP_CHECK(v, BIT); \
577 _PCM_WRITE_##SIGN##BIT##_##ENDIAN(dst, Z_CLAMP(v, BIT)); \
578 } while (ch != 0); \
579 }
580
581 #define Z_DECLARE_SINC_POLYPHASE(SIGN, BIT, ENDIAN) \
582 static void \
583 z_feed_sinc_polyphase_##SIGN##BIT##ENDIAN(struct z_info *info, uint8_t *dst) \
584 { \
585 intpcm64_t v; \
586 intpcm_t x; \
587 uint8_t *p; \
588 int32_t ch, i, start, *z_pcoeff; \
589 \
590 ch = info->channels * PCM_##BIT##_BPS; \
591 dst += ch; \
592 start = z_prev(info, info->z_start, (info->z_size << 1) - 1) * ch; \
593 \
594 do { \
595 dst -= PCM_##BIT##_BPS; \
596 ch -= PCM_##BIT##_BPS; \
597 v = 0; \
598 p = info->z_delay + start + ch; \
599 z_pcoeff = info->z_pcoeff + \
600 ((info->z_alpha * info->z_size) << 1); \
601 for (i = info->z_size; i != 0; i--) { \
602 x = _PCM_READ_##SIGN##BIT##_##ENDIAN(p); \
603 v += Z_NORM_##BIT((intpcm64_t)x * *z_pcoeff); \
604 z_pcoeff++; \
605 p += info->channels * PCM_##BIT##_BPS; \
606 x = _PCM_READ_##SIGN##BIT##_##ENDIAN(p); \
607 v += Z_NORM_##BIT((intpcm64_t)x * *z_pcoeff); \
608 z_pcoeff++; \
609 p += info->channels * PCM_##BIT##_BPS; \
610 } \
611 if (info->z_scale != Z_ONE) \
612 v = Z_SCALE_##BIT(v, info->z_scale); \
613 else \
614 v >>= Z_COEFF_SHIFT - Z_GUARD_BIT_##BIT; \
615 Z_CLIP_CHECK(v, BIT); \
616 _PCM_WRITE_##SIGN##BIT##_##ENDIAN(dst, Z_CLAMP(v, BIT)); \
617 } while (ch != 0); \
618 }
619
620 #define Z_DECLARE(SIGN, BIT, ENDIAN) \
621 Z_DECLARE_LINEAR(SIGN, BIT, ENDIAN) \
622 Z_DECLARE_SINC(SIGN, BIT, ENDIAN) \
623 Z_DECLARE_SINC_POLYPHASE(SIGN, BIT, ENDIAN)
624
625 #if BYTE_ORDER == LITTLE_ENDIAN || defined(SND_FEEDER_MULTIFORMAT)
626 Z_DECLARE(S, 16, LE)
627 Z_DECLARE(S, 32, LE)
628 #endif
629 #if BYTE_ORDER == BIG_ENDIAN || defined(SND_FEEDER_MULTIFORMAT)
630 Z_DECLARE(S, 16, BE)
631 Z_DECLARE(S, 32, BE)
632 #endif
633 #ifdef SND_FEEDER_MULTIFORMAT
634 Z_DECLARE(S, 8, NE)
635 Z_DECLARE(S, 24, LE)
636 Z_DECLARE(S, 24, BE)
637 Z_DECLARE(U, 8, NE)
638 Z_DECLARE(U, 16, LE)
639 Z_DECLARE(U, 24, LE)
640 Z_DECLARE(U, 32, LE)
641 Z_DECLARE(U, 16, BE)
642 Z_DECLARE(U, 24, BE)
643 Z_DECLARE(U, 32, BE)
644 #endif
645
646 enum {
647 Z_RESAMPLER_ZOH,
648 Z_RESAMPLER_LINEAR,
649 Z_RESAMPLER_SINC,
650 Z_RESAMPLER_SINC_POLYPHASE,
651 Z_RESAMPLER_LAST
652 };
653
654 #define Z_RESAMPLER_IDX(i) \
655 (Z_IS_SINC(i) ? Z_RESAMPLER_SINC : (i)->quality)
656
657 #define Z_RESAMPLER_ENTRY(SIGN, BIT, ENDIAN) \
658 { \
659 AFMT_##SIGN##BIT##_##ENDIAN, \
660 { \
661 [Z_RESAMPLER_ZOH] = z_feed_zoh, \
662 [Z_RESAMPLER_LINEAR] = z_feed_linear_##SIGN##BIT##ENDIAN, \
663 [Z_RESAMPLER_SINC] = z_feed_sinc_##SIGN##BIT##ENDIAN, \
664 [Z_RESAMPLER_SINC_POLYPHASE] = \
665 z_feed_sinc_polyphase_##SIGN##BIT##ENDIAN \
666 } \
667 }
668
669 static const struct {
670 uint32_t format;
671 z_resampler_t resampler[Z_RESAMPLER_LAST];
672 } z_resampler_tab[] = {
673 #if BYTE_ORDER == LITTLE_ENDIAN || defined(SND_FEEDER_MULTIFORMAT)
674 Z_RESAMPLER_ENTRY(S, 16, LE),
675 Z_RESAMPLER_ENTRY(S, 32, LE),
676 #endif
677 #if BYTE_ORDER == BIG_ENDIAN || defined(SND_FEEDER_MULTIFORMAT)
678 Z_RESAMPLER_ENTRY(S, 16, BE),
679 Z_RESAMPLER_ENTRY(S, 32, BE),
680 #endif
681 #ifdef SND_FEEDER_MULTIFORMAT
682 Z_RESAMPLER_ENTRY(S, 8, NE),
683 Z_RESAMPLER_ENTRY(S, 24, LE),
684 Z_RESAMPLER_ENTRY(S, 24, BE),
685 Z_RESAMPLER_ENTRY(U, 8, NE),
686 Z_RESAMPLER_ENTRY(U, 16, LE),
687 Z_RESAMPLER_ENTRY(U, 24, LE),
688 Z_RESAMPLER_ENTRY(U, 32, LE),
689 Z_RESAMPLER_ENTRY(U, 16, BE),
690 Z_RESAMPLER_ENTRY(U, 24, BE),
691 Z_RESAMPLER_ENTRY(U, 32, BE),
692 #endif
693 };
694
695 #define Z_RESAMPLER_TAB_SIZE \
696 ((int32_t)(sizeof(z_resampler_tab) / sizeof(z_resampler_tab[0])))
697
698 static void
z_resampler_reset(struct z_info * info)699 z_resampler_reset(struct z_info *info)
700 {
701
702 info->src = info->rsrc - (info->rsrc % ((feeder_rate_round > 0 &&
703 info->rsrc > feeder_rate_round) ? feeder_rate_round : 1));
704 info->dst = info->rdst - (info->rdst % ((feeder_rate_round > 0 &&
705 info->rdst > feeder_rate_round) ? feeder_rate_round : 1));
706 info->z_gx = 1;
707 info->z_gy = 1;
708 info->z_alpha = 0;
709 info->z_resample = NULL;
710 info->z_size = 1;
711 info->z_coeff = NULL;
712 info->z_dcoeff = NULL;
713 if (info->z_pcoeff != NULL) {
714 free(info->z_pcoeff, M_DEVBUF);
715 info->z_pcoeff = NULL;
716 }
717 info->z_scale = Z_ONE;
718 info->z_dx = Z_FULL_ONE;
719 info->z_dy = Z_FULL_ONE;
720 #ifdef Z_DIAGNOSTIC
721 info->z_cycle = 0;
722 #endif
723 if (info->quality < Z_QUALITY_MIN)
724 info->quality = Z_QUALITY_MIN;
725 else if (info->quality > Z_QUALITY_MAX)
726 info->quality = Z_QUALITY_MAX;
727 }
728
729 #ifdef Z_PARANOID
730 static int32_t
z_resampler_sinc_len(struct z_info * info)731 z_resampler_sinc_len(struct z_info *info)
732 {
733 int32_t c, z, len, lmax;
734
735 if (!Z_IS_SINC(info))
736 return (1);
737
738 /*
739 * A rather careful (or useless) way to calculate filter length.
740 * Z_SINC_LEN() itself is accurate enough to do its job. Extra
741 * sanity checking is not going to hurt though..
742 */
743 c = 0;
744 z = info->z_dy;
745 len = 0;
746 lmax = z_coeff_tab[Z_SINC_COEFF_IDX(info)].len;
747
748 do {
749 c += z >> Z_SHIFT;
750 z &= Z_MASK;
751 z += info->z_dy;
752 } while (c < lmax && ++len > 0);
753
754 if (len != Z_SINC_LEN(info)) {
755 #ifdef _KERNEL
756 printf("%s(): sinc l=%d != Z_SINC_LEN=%d\n",
757 __func__, len, Z_SINC_LEN(info));
758 #else
759 fprintf(stderr, "%s(): sinc l=%d != Z_SINC_LEN=%d\n",
760 __func__, len, Z_SINC_LEN(info));
761 return (-1);
762 #endif
763 }
764
765 return (len);
766 }
767 #else
768 #define z_resampler_sinc_len(i) (Z_IS_SINC(i) ? Z_SINC_LEN(i) : 1)
769 #endif
770
771 #define Z_POLYPHASE_COEFF_SHIFT 0
772
773 /*
774 * Pick suitable polynomial interpolators based on filter oversampled ratio
775 * (2 ^ Z_DRIFT_SHIFT).
776 */
777 #if !(defined(Z_COEFF_INTERP_ZOH) || defined(Z_COEFF_INTERP_LINEAR) || \
778 defined(Z_COEFF_INTERP_QUADRATIC) || defined(Z_COEFF_INTERP_HERMITE) || \
779 defined(Z_COEFF_INTER_BSPLINE) || defined(Z_COEFF_INTERP_OPT32X) || \
780 defined(Z_COEFF_INTERP_OPT16X) || defined(Z_COEFF_INTERP_OPT8X) || \
781 defined(Z_COEFF_INTERP_OPT4X) || defined(Z_COEFF_INTERP_OPT2X))
782 #if Z_DRIFT_SHIFT >= 6
783 #define Z_COEFF_INTERP_BSPLINE 1
784 #elif Z_DRIFT_SHIFT >= 5
785 #define Z_COEFF_INTERP_OPT32X 1
786 #elif Z_DRIFT_SHIFT == 4
787 #define Z_COEFF_INTERP_OPT16X 1
788 #elif Z_DRIFT_SHIFT == 3
789 #define Z_COEFF_INTERP_OPT8X 1
790 #elif Z_DRIFT_SHIFT == 2
791 #define Z_COEFF_INTERP_OPT4X 1
792 #elif Z_DRIFT_SHIFT == 1
793 #define Z_COEFF_INTERP_OPT2X 1
794 #else
795 #error "Z_DRIFT_SHIFT screwed!"
796 #endif
797 #endif
798
799 /*
800 * In classic polyphase mode, the actual coefficients for each phases need to
801 * be calculated based on default prototype filters. For highly oversampled
802 * filter, linear or quadradatic interpolator should be enough. Anything less
803 * than that require 'special' interpolators to reduce interpolation errors.
804 *
805 * "Polynomial Interpolators for High-Quality Resampling of Oversampled Audio"
806 * by Olli Niemitalo
807 * - http://www.student.oulu.fi/~oniemita/dsp/deip.pdf
808 *
809 */
810 static int32_t
z_coeff_interpolate(int32_t z,int32_t * z_coeff)811 z_coeff_interpolate(int32_t z, int32_t *z_coeff)
812 {
813 int32_t coeff;
814 #if defined(Z_COEFF_INTERP_ZOH)
815
816 /* 1-point, 0th-order (Zero Order Hold) */
817 z = z;
818 coeff = z_coeff[0];
819 #elif defined(Z_COEFF_INTERP_LINEAR)
820 int32_t zl0, zl1;
821
822 /* 2-point, 1st-order Linear */
823 zl0 = z_coeff[0];
824 zl1 = z_coeff[1] - z_coeff[0];
825
826 coeff = Z_RSHIFT((int64_t)zl1 * z, Z_SHIFT) + zl0;
827 #elif defined(Z_COEFF_INTERP_QUADRATIC)
828 int32_t zq0, zq1, zq2;
829
830 /* 3-point, 2nd-order Quadratic */
831 zq0 = z_coeff[0];
832 zq1 = z_coeff[1] - z_coeff[-1];
833 zq2 = z_coeff[1] + z_coeff[-1] - (z_coeff[0] << 1);
834
835 coeff = Z_RSHIFT((Z_RSHIFT((int64_t)zq2 * z, Z_SHIFT) +
836 zq1) * z, Z_SHIFT + 1) + zq0;
837 #elif defined(Z_COEFF_INTERP_HERMITE)
838 int32_t zh0, zh1, zh2, zh3;
839
840 /* 4-point, 3rd-order Hermite */
841 zh0 = z_coeff[0];
842 zh1 = z_coeff[1] - z_coeff[-1];
843 zh2 = (z_coeff[-1] << 1) - (z_coeff[0] * 5) + (z_coeff[1] << 2) -
844 z_coeff[2];
845 zh3 = z_coeff[2] - z_coeff[-1] + ((z_coeff[0] - z_coeff[1]) * 3);
846
847 coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((int64_t)zh3 * z, Z_SHIFT) +
848 zh2) * z, Z_SHIFT) + zh1) * z, Z_SHIFT + 1) + zh0;
849 #elif defined(Z_COEFF_INTERP_BSPLINE)
850 int32_t zb0, zb1, zb2, zb3;
851
852 /* 4-point, 3rd-order B-Spline */
853 zb0 = Z_RSHIFT(0x15555555LL * (((int64_t)z_coeff[0] << 2) +
854 z_coeff[-1] + z_coeff[1]), 30);
855 zb1 = z_coeff[1] - z_coeff[-1];
856 zb2 = z_coeff[-1] + z_coeff[1] - (z_coeff[0] << 1);
857 zb3 = Z_RSHIFT(0x15555555LL * (((z_coeff[0] - z_coeff[1]) * 3) +
858 z_coeff[2] - z_coeff[-1]), 30);
859
860 coeff = (Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((int64_t)zb3 * z, Z_SHIFT) +
861 zb2) * z, Z_SHIFT) + zb1) * z, Z_SHIFT) + zb0 + 1) >> 1;
862 #elif defined(Z_COEFF_INTERP_OPT32X)
863 int32_t zoz, zoe1, zoe2, zoe3, zoo1, zoo2, zoo3;
864 int32_t zoc0, zoc1, zoc2, zoc3, zoc4, zoc5;
865
866 /* 6-point, 5th-order Optimal 32x */
867 zoz = z - (Z_ONE >> 1);
868 zoe1 = z_coeff[1] + z_coeff[0];
869 zoe2 = z_coeff[2] + z_coeff[-1];
870 zoe3 = z_coeff[3] + z_coeff[-2];
871 zoo1 = z_coeff[1] - z_coeff[0];
872 zoo2 = z_coeff[2] - z_coeff[-1];
873 zoo3 = z_coeff[3] - z_coeff[-2];
874
875 zoc0 = Z_RSHIFT((0x1ac2260dLL * zoe1) + (0x0526cdcaLL * zoe2) +
876 (0x00170c29LL * zoe3), 30);
877 zoc1 = Z_RSHIFT((0x14f8a49aLL * zoo1) + (0x0d6d1109LL * zoo2) +
878 (0x008cd4dcLL * zoo3), 30);
879 zoc2 = Z_RSHIFT((-0x0d3e94a4LL * zoe1) + (0x0bddded4LL * zoe2) +
880 (0x0160b5d0LL * zoe3), 30);
881 zoc3 = Z_RSHIFT((-0x0de10cc4LL * zoo1) + (0x019b2a7dLL * zoo2) +
882 (0x01cfe914LL * zoo3), 30);
883 zoc4 = Z_RSHIFT((0x02aa12d7LL * zoe1) + (-0x03ff1bb3LL * zoe2) +
884 (0x015508ddLL * zoe3), 30);
885 zoc5 = Z_RSHIFT((0x051d29e5LL * zoo1) + (-0x028e7647LL * zoo2) +
886 (0x0082d81aLL * zoo3), 30);
887
888 coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT(
889 (int64_t)zoc5 * zoz, Z_SHIFT) +
890 zoc4) * zoz, Z_SHIFT) + zoc3) * zoz, Z_SHIFT) +
891 zoc2) * zoz, Z_SHIFT) + zoc1) * zoz, Z_SHIFT) + zoc0;
892 #elif defined(Z_COEFF_INTERP_OPT16X)
893 int32_t zoz, zoe1, zoe2, zoe3, zoo1, zoo2, zoo3;
894 int32_t zoc0, zoc1, zoc2, zoc3, zoc4, zoc5;
895
896 /* 6-point, 5th-order Optimal 16x */
897 zoz = z - (Z_ONE >> 1);
898 zoe1 = z_coeff[1] + z_coeff[0];
899 zoe2 = z_coeff[2] + z_coeff[-1];
900 zoe3 = z_coeff[3] + z_coeff[-2];
901 zoo1 = z_coeff[1] - z_coeff[0];
902 zoo2 = z_coeff[2] - z_coeff[-1];
903 zoo3 = z_coeff[3] - z_coeff[-2];
904
905 zoc0 = Z_RSHIFT((0x1ac2260dLL * zoe1) + (0x0526cdcaLL * zoe2) +
906 (0x00170c29LL * zoe3), 30);
907 zoc1 = Z_RSHIFT((0x14f8a49aLL * zoo1) + (0x0d6d1109LL * zoo2) +
908 (0x008cd4dcLL * zoo3), 30);
909 zoc2 = Z_RSHIFT((-0x0d3e94a4LL * zoe1) + (0x0bddded4LL * zoe2) +
910 (0x0160b5d0LL * zoe3), 30);
911 zoc3 = Z_RSHIFT((-0x0de10cc4LL * zoo1) + (0x019b2a7dLL * zoo2) +
912 (0x01cfe914LL * zoo3), 30);
913 zoc4 = Z_RSHIFT((0x02aa12d7LL * zoe1) + (-0x03ff1bb3LL * zoe2) +
914 (0x015508ddLL * zoe3), 30);
915 zoc5 = Z_RSHIFT((0x051d29e5LL * zoo1) + (-0x028e7647LL * zoo2) +
916 (0x0082d81aLL * zoo3), 30);
917
918 coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT(
919 (int64_t)zoc5 * zoz, Z_SHIFT) +
920 zoc4) * zoz, Z_SHIFT) + zoc3) * zoz, Z_SHIFT) +
921 zoc2) * zoz, Z_SHIFT) + zoc1) * zoz, Z_SHIFT) + zoc0;
922 #elif defined(Z_COEFF_INTERP_OPT8X)
923 int32_t zoz, zoe1, zoe2, zoe3, zoo1, zoo2, zoo3;
924 int32_t zoc0, zoc1, zoc2, zoc3, zoc4, zoc5;
925
926 /* 6-point, 5th-order Optimal 8x */
927 zoz = z - (Z_ONE >> 1);
928 zoe1 = z_coeff[1] + z_coeff[0];
929 zoe2 = z_coeff[2] + z_coeff[-1];
930 zoe3 = z_coeff[3] + z_coeff[-2];
931 zoo1 = z_coeff[1] - z_coeff[0];
932 zoo2 = z_coeff[2] - z_coeff[-1];
933 zoo3 = z_coeff[3] - z_coeff[-2];
934
935 zoc0 = Z_RSHIFT((0x1aa9b47dLL * zoe1) + (0x053d9944LL * zoe2) +
936 (0x0018b23fLL * zoe3), 30);
937 zoc1 = Z_RSHIFT((0x14a104d1LL * zoo1) + (0x0d7d2504LL * zoo2) +
938 (0x0094b599LL * zoo3), 30);
939 zoc2 = Z_RSHIFT((-0x0d22530bLL * zoe1) + (0x0bb37a2cLL * zoe2) +
940 (0x016ed8e0LL * zoe3), 30);
941 zoc3 = Z_RSHIFT((-0x0d744b1cLL * zoo1) + (0x01649591LL * zoo2) +
942 (0x01dae93aLL * zoo3), 30);
943 zoc4 = Z_RSHIFT((0x02a7ee1bLL * zoe1) + (-0x03fbdb24LL * zoe2) +
944 (0x0153ed07LL * zoe3), 30);
945 zoc5 = Z_RSHIFT((0x04cf9b6cLL * zoo1) + (-0x0266b378LL * zoo2) +
946 (0x007a7c26LL * zoo3), 30);
947
948 coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT(
949 (int64_t)zoc5 * zoz, Z_SHIFT) +
950 zoc4) * zoz, Z_SHIFT) + zoc3) * zoz, Z_SHIFT) +
951 zoc2) * zoz, Z_SHIFT) + zoc1) * zoz, Z_SHIFT) + zoc0;
952 #elif defined(Z_COEFF_INTERP_OPT4X)
953 int32_t zoz, zoe1, zoe2, zoe3, zoo1, zoo2, zoo3;
954 int32_t zoc0, zoc1, zoc2, zoc3, zoc4, zoc5;
955
956 /* 6-point, 5th-order Optimal 4x */
957 zoz = z - (Z_ONE >> 1);
958 zoe1 = z_coeff[1] + z_coeff[0];
959 zoe2 = z_coeff[2] + z_coeff[-1];
960 zoe3 = z_coeff[3] + z_coeff[-2];
961 zoo1 = z_coeff[1] - z_coeff[0];
962 zoo2 = z_coeff[2] - z_coeff[-1];
963 zoo3 = z_coeff[3] - z_coeff[-2];
964
965 zoc0 = Z_RSHIFT((0x1a8eda43LL * zoe1) + (0x0556ee38LL * zoe2) +
966 (0x001a3784LL * zoe3), 30);
967 zoc1 = Z_RSHIFT((0x143d863eLL * zoo1) + (0x0d910e36LL * zoo2) +
968 (0x009ca889LL * zoo3), 30);
969 zoc2 = Z_RSHIFT((-0x0d026821LL * zoe1) + (0x0b837773LL * zoe2) +
970 (0x017ef0c6LL * zoe3), 30);
971 zoc3 = Z_RSHIFT((-0x0cef1502LL * zoo1) + (0x01207a8eLL * zoo2) +
972 (0x01e936dbLL * zoo3), 30);
973 zoc4 = Z_RSHIFT((0x029fe643LL * zoe1) + (-0x03ef3fc8LL * zoe2) +
974 (0x014f5923LL * zoe3), 30);
975 zoc5 = Z_RSHIFT((0x043a9d08LL * zoo1) + (-0x02154febLL * zoo2) +
976 (0x00670dbdLL * zoo3), 30);
977
978 coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT(
979 (int64_t)zoc5 * zoz, Z_SHIFT) +
980 zoc4) * zoz, Z_SHIFT) + zoc3) * zoz, Z_SHIFT) +
981 zoc2) * zoz, Z_SHIFT) + zoc1) * zoz, Z_SHIFT) + zoc0;
982 #elif defined(Z_COEFF_INTERP_OPT2X)
983 int32_t zoz, zoe1, zoe2, zoe3, zoo1, zoo2, zoo3;
984 int32_t zoc0, zoc1, zoc2, zoc3, zoc4, zoc5;
985
986 /* 6-point, 5th-order Optimal 2x */
987 zoz = z - (Z_ONE >> 1);
988 zoe1 = z_coeff[1] + z_coeff[0];
989 zoe2 = z_coeff[2] + z_coeff[-1];
990 zoe3 = z_coeff[3] + z_coeff[-2];
991 zoo1 = z_coeff[1] - z_coeff[0];
992 zoo2 = z_coeff[2] - z_coeff[-1];
993 zoo3 = z_coeff[3] - z_coeff[-2];
994
995 zoc0 = Z_RSHIFT((0x19edb6fdLL * zoe1) + (0x05ebd062LL * zoe2) +
996 (0x00267881LL * zoe3), 30);
997 zoc1 = Z_RSHIFT((0x1223af76LL * zoo1) + (0x0de3dd6bLL * zoo2) +
998 (0x00d683cdLL * zoo3), 30);
999 zoc2 = Z_RSHIFT((-0x0c3ee068LL * zoe1) + (0x0a5c3769LL * zoe2) +
1000 (0x01e2aceaLL * zoe3), 30);
1001 zoc3 = Z_RSHIFT((-0x0a8ab614LL * zoo1) + (-0x0019522eLL * zoo2) +
1002 (0x022cefc7LL * zoo3), 30);
1003 zoc4 = Z_RSHIFT((0x0276187dLL * zoe1) + (-0x03a801e8LL * zoe2) +
1004 (0x0131d935LL * zoe3), 30);
1005 zoc5 = Z_RSHIFT((0x02c373f5LL * zoo1) + (-0x01275f83LL * zoo2) +
1006 (0x0018ee79LL * zoo3), 30);
1007
1008 coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT(
1009 (int64_t)zoc5 * zoz, Z_SHIFT) +
1010 zoc4) * zoz, Z_SHIFT) + zoc3) * zoz, Z_SHIFT) +
1011 zoc2) * zoz, Z_SHIFT) + zoc1) * zoz, Z_SHIFT) + zoc0;
1012 #else
1013 #error "Interpolation type screwed!"
1014 #endif
1015
1016 #if Z_POLYPHASE_COEFF_SHIFT > 0
1017 coeff = Z_RSHIFT(coeff, Z_POLYPHASE_COEFF_SHIFT);
1018 #endif
1019 return (coeff);
1020 }
1021
1022 static int
z_resampler_build_polyphase(struct z_info * info)1023 z_resampler_build_polyphase(struct z_info *info)
1024 {
1025 int32_t alpha, c, i, z, idx;
1026
1027 /* Let this be here first. */
1028 if (info->z_pcoeff != NULL) {
1029 free(info->z_pcoeff, M_DEVBUF);
1030 info->z_pcoeff = NULL;
1031 }
1032
1033 if (feeder_rate_polyphase_max < 1)
1034 return (ENOTSUP);
1035
1036 if (((int64_t)info->z_size * info->z_gy * 2) >
1037 feeder_rate_polyphase_max) {
1038 #ifndef _KERNEL
1039 fprintf(stderr, "Polyphase entries exceed: [%d/%d] %jd > %d\n",
1040 info->z_gx, info->z_gy,
1041 (intmax_t)info->z_size * info->z_gy * 2,
1042 feeder_rate_polyphase_max);
1043 #endif
1044 return (E2BIG);
1045 }
1046
1047 info->z_pcoeff = malloc(sizeof(int32_t) *
1048 info->z_size * info->z_gy * 2, M_DEVBUF, M_NOWAIT | M_ZERO);
1049 if (info->z_pcoeff == NULL)
1050 return (ENOMEM);
1051
1052 for (alpha = 0; alpha < info->z_gy; alpha++) {
1053 z = alpha * info->z_dx;
1054 c = 0;
1055 for (i = info->z_size; i != 0; i--) {
1056 c += z >> Z_SHIFT;
1057 z &= Z_MASK;
1058 idx = (alpha * info->z_size * 2) +
1059 (info->z_size * 2) - i;
1060 info->z_pcoeff[idx] =
1061 z_coeff_interpolate(z, info->z_coeff + c);
1062 z += info->z_dy;
1063 }
1064 z = info->z_dy - (alpha * info->z_dx);
1065 c = 0;
1066 for (i = info->z_size; i != 0; i--) {
1067 c += z >> Z_SHIFT;
1068 z &= Z_MASK;
1069 idx = (alpha * info->z_size * 2) + i - 1;
1070 info->z_pcoeff[idx] =
1071 z_coeff_interpolate(z, info->z_coeff + c);
1072 z += info->z_dy;
1073 }
1074 }
1075
1076 #ifndef _KERNEL
1077 fprintf(stderr, "Polyphase: [%d/%d] %d entries\n",
1078 info->z_gx, info->z_gy, info->z_size * info->z_gy * 2);
1079 #endif
1080
1081 return (0);
1082 }
1083
1084 static int
z_resampler_setup(struct pcm_feeder * f)1085 z_resampler_setup(struct pcm_feeder *f)
1086 {
1087 struct z_info *info;
1088 int64_t gy2gx_max, gx2gy_max;
1089 uint32_t format;
1090 int32_t align, i, z_scale;
1091 int adaptive;
1092
1093 info = f->data;
1094 z_resampler_reset(info);
1095
1096 if (info->src == info->dst)
1097 return (0);
1098
1099 /* Shrink by greatest common divisor. */
1100 i = z_gcd(info->src, info->dst);
1101 info->z_gx = info->src / i;
1102 info->z_gy = info->dst / i;
1103
1104 /* Too big, or too small. Bail out. */
1105 if (!(Z_FACTOR_SAFE(info->z_gx) && Z_FACTOR_SAFE(info->z_gy)))
1106 return (EINVAL);
1107
1108 format = f->desc->in;
1109 adaptive = 0;
1110 z_scale = 0;
1111
1112 /*
1113 * Setup everything: filter length, conversion factor, etc.
1114 */
1115 if (Z_IS_SINC(info)) {
1116 /*
1117 * Downsampling, or upsampling scaling factor. As long as the
1118 * factor can be represented by a fraction of 1 << Z_SHIFT,
1119 * we're pretty much in business. Scaling is not needed for
1120 * upsampling, so we just slap Z_ONE there.
1121 */
1122 if (info->z_gx > info->z_gy)
1123 /*
1124 * If the downsampling ratio is beyond sanity,
1125 * enable semi-adaptive mode. Although handling
1126 * extreme ratio is possible, the result of the
1127 * conversion is just pointless, unworthy,
1128 * nonsensical noises, etc.
1129 */
1130 if ((info->z_gx / info->z_gy) > Z_SINC_DOWNMAX)
1131 z_scale = Z_ONE / Z_SINC_DOWNMAX;
1132 else
1133 z_scale = ((uint64_t)info->z_gy << Z_SHIFT) /
1134 info->z_gx;
1135 else
1136 z_scale = Z_ONE;
1137
1138 /*
1139 * This is actually impossible, unless anything above
1140 * overflow.
1141 */
1142 if (z_scale < 1)
1143 return (E2BIG);
1144
1145 /*
1146 * Calculate sample time/coefficients index drift. It is
1147 * a constant for upsampling, but downsampling require
1148 * heavy duty filtering with possible too long filters.
1149 * If anything goes wrong, revisit again and enable
1150 * adaptive mode.
1151 */
1152 z_setup_adaptive_sinc:
1153 if (info->z_pcoeff != NULL) {
1154 free(info->z_pcoeff, M_DEVBUF);
1155 info->z_pcoeff = NULL;
1156 }
1157
1158 if (adaptive == 0) {
1159 info->z_dy = z_scale << Z_DRIFT_SHIFT;
1160 if (info->z_dy < 1)
1161 return (E2BIG);
1162 info->z_scale = z_scale;
1163 } else {
1164 info->z_dy = Z_FULL_ONE;
1165 info->z_scale = Z_ONE;
1166 }
1167
1168 /* Smallest drift increment. */
1169 info->z_dx = info->z_dy / info->z_gy;
1170
1171 /*
1172 * Overflow or underflow. Try adaptive, let it continue and
1173 * retry.
1174 */
1175 if (info->z_dx < 1) {
1176 if (adaptive == 0) {
1177 adaptive = 1;
1178 goto z_setup_adaptive_sinc;
1179 }
1180 return (E2BIG);
1181 }
1182
1183 /*
1184 * Round back output drift.
1185 */
1186 info->z_dy = info->z_dx * info->z_gy;
1187
1188 for (i = 0; i < Z_COEFF_TAB_SIZE; i++) {
1189 if (Z_SINC_COEFF_IDX(info) != i)
1190 continue;
1191 /*
1192 * Calculate required filter length and guard
1193 * against possible abusive result. Note that
1194 * this represents only 1/2 of the entire filter
1195 * length.
1196 */
1197 info->z_size = z_resampler_sinc_len(info);
1198
1199 /*
1200 * Multiple of 2 rounding, for better accumulator
1201 * performance.
1202 */
1203 info->z_size &= ~1;
1204
1205 if (info->z_size < 2 || info->z_size > Z_SINC_MAX) {
1206 if (adaptive == 0) {
1207 adaptive = 1;
1208 goto z_setup_adaptive_sinc;
1209 }
1210 return (E2BIG);
1211 }
1212 info->z_coeff = z_coeff_tab[i].coeff + Z_COEFF_OFFSET;
1213 info->z_dcoeff = z_coeff_tab[i].dcoeff;
1214 break;
1215 }
1216
1217 if (info->z_coeff == NULL || info->z_dcoeff == NULL)
1218 return (EINVAL);
1219 } else if (Z_IS_LINEAR(info)) {
1220 /*
1221 * Don't put much effort if we're doing linear interpolation.
1222 * Just center the interpolation distance within Z_LINEAR_ONE,
1223 * and be happy about it.
1224 */
1225 info->z_dx = Z_LINEAR_FULL_ONE / info->z_gy;
1226 }
1227
1228 /*
1229 * We're safe for now, lets continue.. Look for our resampler
1230 * depending on configured format and quality.
1231 */
1232 for (i = 0; i < Z_RESAMPLER_TAB_SIZE; i++) {
1233 int ridx;
1234
1235 if (AFMT_ENCODING(format) != z_resampler_tab[i].format)
1236 continue;
1237 if (Z_IS_SINC(info) && adaptive == 0 &&
1238 z_resampler_build_polyphase(info) == 0)
1239 ridx = Z_RESAMPLER_SINC_POLYPHASE;
1240 else
1241 ridx = Z_RESAMPLER_IDX(info);
1242 info->z_resample = z_resampler_tab[i].resampler[ridx];
1243 break;
1244 }
1245
1246 if (info->z_resample == NULL)
1247 return (EINVAL);
1248
1249 info->bps = AFMT_BPS(format);
1250 align = info->channels * info->bps;
1251
1252 /*
1253 * Calculate largest value that can be fed into z_gy2gx() and
1254 * z_gx2gy() without causing (signed) 32bit overflow. z_gy2gx() will
1255 * be called early during feeding process to determine how much input
1256 * samples that is required to generate requested output, while
1257 * z_gx2gy() will be called just before samples filtering /
1258 * accumulation process based on available samples that has been
1259 * calculated using z_gx2gy().
1260 *
1261 * Now that is damn confusing, I guess ;-) .
1262 */
1263 gy2gx_max = (((uint64_t)info->z_gy * INT32_MAX) - info->z_gy + 1) /
1264 info->z_gx;
1265
1266 if ((gy2gx_max * align) > SND_FXDIV_MAX)
1267 gy2gx_max = SND_FXDIV_MAX / align;
1268
1269 if (gy2gx_max < 1)
1270 return (E2BIG);
1271
1272 gx2gy_max = (((uint64_t)info->z_gx * INT32_MAX) - info->z_gy) /
1273 info->z_gy;
1274
1275 if (gx2gy_max > INT32_MAX)
1276 gx2gy_max = INT32_MAX;
1277
1278 if (gx2gy_max < 1)
1279 return (E2BIG);
1280
1281 /*
1282 * Ensure that z_gy2gx() at its largest possible calculated value
1283 * (alpha = 0) will not cause overflow further late during z_gx2gy()
1284 * stage.
1285 */
1286 if (z_gy2gx(info, gy2gx_max) > _Z_GCAST(gx2gy_max))
1287 return (E2BIG);
1288
1289 info->z_maxfeed = gy2gx_max * align;
1290
1291 #ifdef Z_USE_ALPHADRIFT
1292 info->z_startdrift = z_gy2gx(info, 1);
1293 info->z_alphadrift = z_drift(info, info->z_startdrift, 1);
1294 #endif
1295
1296 i = z_gy2gx(info, 1);
1297 info->z_full = z_roundpow2((info->z_size << 1) + i);
1298
1299 /*
1300 * Too big to be true, and overflowing left and right like mad ..
1301 */
1302 if ((info->z_full * align) < 1) {
1303 if (adaptive == 0 && Z_IS_SINC(info)) {
1304 adaptive = 1;
1305 goto z_setup_adaptive_sinc;
1306 }
1307 return (E2BIG);
1308 }
1309
1310 /*
1311 * Increase full buffer size if its too small to reduce cyclic
1312 * buffer shifting in main conversion/feeder loop.
1313 */
1314 while (info->z_full < Z_RESERVOIR_MAX &&
1315 (info->z_full - (info->z_size << 1)) < Z_RESERVOIR)
1316 info->z_full <<= 1;
1317
1318 /* Initialize buffer position. */
1319 info->z_mask = info->z_full - 1;
1320 info->z_start = z_prev(info, info->z_size << 1, 1);
1321 info->z_pos = z_next(info, info->z_start, 1);
1322
1323 /*
1324 * Allocate or reuse delay line buffer, whichever makes sense.
1325 */
1326 i = info->z_full * align;
1327 if (i < 1)
1328 return (E2BIG);
1329
1330 if (info->z_delay == NULL || info->z_alloc < i ||
1331 i <= (info->z_alloc >> 1)) {
1332 if (info->z_delay != NULL)
1333 free(info->z_delay, M_DEVBUF);
1334 info->z_delay = malloc(i, M_DEVBUF, M_NOWAIT | M_ZERO);
1335 if (info->z_delay == NULL)
1336 return (ENOMEM);
1337 info->z_alloc = i;
1338 }
1339
1340 /*
1341 * Zero out head of buffer to avoid pops and clicks.
1342 */
1343 memset(info->z_delay, sndbuf_zerodata(f->desc->out),
1344 info->z_pos * align);
1345
1346 #ifdef Z_DIAGNOSTIC
1347 /*
1348 * XXX Debuging mess !@#$%^
1349 */
1350 #define dumpz(x) fprintf(stderr, "\t%12s = %10u : %-11d\n", \
1351 "z_"__STRING(x), (uint32_t)info->z_##x, \
1352 (int32_t)info->z_##x)
1353 fprintf(stderr, "\n%s():\n", __func__);
1354 fprintf(stderr, "\tchannels=%d, bps=%d, format=0x%08x, quality=%d\n",
1355 info->channels, info->bps, format, info->quality);
1356 fprintf(stderr, "\t%d (%d) -> %d (%d), ",
1357 info->src, info->rsrc, info->dst, info->rdst);
1358 fprintf(stderr, "[%d/%d]\n", info->z_gx, info->z_gy);
1359 fprintf(stderr, "\tminreq=%d, ", z_gy2gx(info, 1));
1360 if (adaptive != 0)
1361 z_scale = Z_ONE;
1362 fprintf(stderr, "factor=0x%08x/0x%08x (%f)\n",
1363 z_scale, Z_ONE, (double)z_scale / Z_ONE);
1364 fprintf(stderr, "\tbase_length=%d, ", Z_SINC_BASE_LEN(info));
1365 fprintf(stderr, "adaptive=%s\n", (adaptive != 0) ? "YES" : "NO");
1366 dumpz(size);
1367 dumpz(alloc);
1368 if (info->z_alloc < 1024)
1369 fprintf(stderr, "\t%15s%10d Bytes\n",
1370 "", info->z_alloc);
1371 else if (info->z_alloc < (1024 << 10))
1372 fprintf(stderr, "\t%15s%10d KBytes\n",
1373 "", info->z_alloc >> 10);
1374 else if (info->z_alloc < (1024 << 20))
1375 fprintf(stderr, "\t%15s%10d MBytes\n",
1376 "", info->z_alloc >> 20);
1377 else
1378 fprintf(stderr, "\t%15s%10d GBytes\n",
1379 "", info->z_alloc >> 30);
1380 fprintf(stderr, "\t%12s %10d (min output samples)\n",
1381 "",
1382 (int32_t)z_gx2gy(info, info->z_full - (info->z_size << 1)));
1383 fprintf(stderr, "\t%12s %10d (min allocated output samples)\n",
1384 "",
1385 (int32_t)z_gx2gy(info, (info->z_alloc / align) -
1386 (info->z_size << 1)));
1387 fprintf(stderr, "\t%12s = %10d\n",
1388 "z_gy2gx()", (int32_t)z_gy2gx(info, 1));
1389 fprintf(stderr, "\t%12s = %10d -> z_gy2gx() -> %d\n",
1390 "Max", (int32_t)gy2gx_max, (int32_t)z_gy2gx(info, gy2gx_max));
1391 fprintf(stderr, "\t%12s = %10d\n",
1392 "z_gx2gy()", (int32_t)z_gx2gy(info, 1));
1393 fprintf(stderr, "\t%12s = %10d -> z_gx2gy() -> %d\n",
1394 "Max", (int32_t)gx2gy_max, (int32_t)z_gx2gy(info, gx2gy_max));
1395 dumpz(maxfeed);
1396 dumpz(full);
1397 dumpz(start);
1398 dumpz(pos);
1399 dumpz(scale);
1400 fprintf(stderr, "\t%12s %10f\n", "",
1401 (double)info->z_scale / Z_ONE);
1402 dumpz(dx);
1403 fprintf(stderr, "\t%12s %10f\n", "",
1404 (double)info->z_dx / info->z_dy);
1405 dumpz(dy);
1406 fprintf(stderr, "\t%12s %10d (drift step)\n", "",
1407 info->z_dy >> Z_SHIFT);
1408 fprintf(stderr, "\t%12s %10d (scaling differences)\n", "",
1409 (z_scale << Z_DRIFT_SHIFT) - info->z_dy);
1410 fprintf(stderr, "\t%12s = %u bytes\n",
1411 "intpcm32_t", sizeof(intpcm32_t));
1412 fprintf(stderr, "\t%12s = 0x%08x, smallest=%.16lf\n",
1413 "Z_ONE", Z_ONE, (double)1.0 / (double)Z_ONE);
1414 #endif
1415
1416 return (0);
1417 }
1418
1419 static int
z_resampler_set(struct pcm_feeder * f,int what,int32_t value)1420 z_resampler_set(struct pcm_feeder *f, int what, int32_t value)
1421 {
1422 struct z_info *info;
1423 int32_t oquality;
1424
1425 info = f->data;
1426
1427 switch (what) {
1428 case Z_RATE_SRC:
1429 if (value < feeder_rate_min || value > feeder_rate_max)
1430 return (E2BIG);
1431 if (value == info->rsrc)
1432 return (0);
1433 info->rsrc = value;
1434 break;
1435 case Z_RATE_DST:
1436 if (value < feeder_rate_min || value > feeder_rate_max)
1437 return (E2BIG);
1438 if (value == info->rdst)
1439 return (0);
1440 info->rdst = value;
1441 break;
1442 case Z_RATE_QUALITY:
1443 if (value < Z_QUALITY_MIN || value > Z_QUALITY_MAX)
1444 return (EINVAL);
1445 if (value == info->quality)
1446 return (0);
1447 /*
1448 * If we failed to set the requested quality, restore
1449 * the old one. We cannot afford leaving it broken since
1450 * passive feeder chains like vchans never reinitialize
1451 * itself.
1452 */
1453 oquality = info->quality;
1454 info->quality = value;
1455 if (z_resampler_setup(f) == 0)
1456 return (0);
1457 info->quality = oquality;
1458 break;
1459 case Z_RATE_CHANNELS:
1460 if (value < SND_CHN_MIN || value > SND_CHN_MAX)
1461 return (EINVAL);
1462 if (value == info->channels)
1463 return (0);
1464 info->channels = value;
1465 break;
1466 default:
1467 return (EINVAL);
1468 break;
1469 }
1470
1471 return (z_resampler_setup(f));
1472 }
1473
1474 static int
z_resampler_get(struct pcm_feeder * f,int what)1475 z_resampler_get(struct pcm_feeder *f, int what)
1476 {
1477 struct z_info *info;
1478
1479 info = f->data;
1480
1481 switch (what) {
1482 case Z_RATE_SRC:
1483 return (info->rsrc);
1484 break;
1485 case Z_RATE_DST:
1486 return (info->rdst);
1487 break;
1488 case Z_RATE_QUALITY:
1489 return (info->quality);
1490 break;
1491 case Z_RATE_CHANNELS:
1492 return (info->channels);
1493 break;
1494 default:
1495 break;
1496 }
1497
1498 return (-1);
1499 }
1500
1501 static int
z_resampler_init(struct pcm_feeder * f)1502 z_resampler_init(struct pcm_feeder *f)
1503 {
1504 struct z_info *info;
1505 int ret;
1506
1507 if (f->desc->in != f->desc->out)
1508 return (EINVAL);
1509
1510 info = malloc(sizeof(*info), M_DEVBUF, M_NOWAIT | M_ZERO);
1511 if (info == NULL)
1512 return (ENOMEM);
1513
1514 info->rsrc = Z_RATE_DEFAULT;
1515 info->rdst = Z_RATE_DEFAULT;
1516 info->quality = feeder_rate_quality;
1517 info->channels = AFMT_CHANNEL(f->desc->in);
1518
1519 f->data = info;
1520
1521 ret = z_resampler_setup(f);
1522 if (ret != 0) {
1523 if (info->z_pcoeff != NULL)
1524 free(info->z_pcoeff, M_DEVBUF);
1525 if (info->z_delay != NULL)
1526 free(info->z_delay, M_DEVBUF);
1527 free(info, M_DEVBUF);
1528 f->data = NULL;
1529 }
1530
1531 return (ret);
1532 }
1533
1534 static int
z_resampler_free(struct pcm_feeder * f)1535 z_resampler_free(struct pcm_feeder *f)
1536 {
1537 struct z_info *info;
1538
1539 info = f->data;
1540 if (info != NULL) {
1541 if (info->z_pcoeff != NULL)
1542 free(info->z_pcoeff, M_DEVBUF);
1543 if (info->z_delay != NULL)
1544 free(info->z_delay, M_DEVBUF);
1545 free(info, M_DEVBUF);
1546 }
1547
1548 f->data = NULL;
1549
1550 return (0);
1551 }
1552
1553 static uint32_t
z_resampler_feed_internal(struct pcm_feeder * f,struct pcm_channel * c,uint8_t * b,uint32_t count,void * source)1554 z_resampler_feed_internal(struct pcm_feeder *f, struct pcm_channel *c,
1555 uint8_t *b, uint32_t count, void *source)
1556 {
1557 struct z_info *info;
1558 int32_t alphadrift, startdrift, reqout, ocount, reqin, align;
1559 int32_t fetch, fetched, start, cp;
1560 uint8_t *dst;
1561
1562 info = f->data;
1563 if (info->z_resample == NULL)
1564 return (z_feed(f->source, c, b, count, source));
1565
1566 /*
1567 * Calculate sample size alignment and amount of sample output.
1568 * We will do everything in sample domain, but at the end we
1569 * will jump back to byte domain.
1570 */
1571 align = info->channels * info->bps;
1572 ocount = SND_FXDIV(count, align);
1573 if (ocount == 0)
1574 return (0);
1575
1576 /*
1577 * Calculate amount of input samples that is needed to generate
1578 * exact amount of output.
1579 */
1580 reqin = z_gy2gx(info, ocount) - z_fetched(info);
1581
1582 #ifdef Z_USE_ALPHADRIFT
1583 startdrift = info->z_startdrift;
1584 alphadrift = info->z_alphadrift;
1585 #else
1586 startdrift = _Z_GY2GX(info, 0, 1);
1587 alphadrift = z_drift(info, startdrift, 1);
1588 #endif
1589
1590 dst = b;
1591
1592 do {
1593 if (reqin != 0) {
1594 fetch = z_min(z_free(info), reqin);
1595 if (fetch == 0) {
1596 /*
1597 * No more free spaces, so wind enough
1598 * samples back to the head of delay line
1599 * in byte domain.
1600 */
1601 fetched = z_fetched(info);
1602 start = z_prev(info, info->z_start,
1603 (info->z_size << 1) - 1);
1604 cp = (info->z_size << 1) + fetched;
1605 z_copy(info->z_delay + (start * align),
1606 info->z_delay, cp * align);
1607 info->z_start =
1608 z_prev(info, info->z_size << 1, 1);
1609 info->z_pos =
1610 z_next(info, info->z_start, fetched + 1);
1611 fetch = z_min(z_free(info), reqin);
1612 #ifdef Z_DIAGNOSTIC
1613 if (1) {
1614 static uint32_t kk = 0;
1615 fprintf(stderr,
1616 "Buffer Move: "
1617 "start=%d fetched=%d cp=%d "
1618 "cycle=%u [%u]\r",
1619 start, fetched, cp, info->z_cycle,
1620 ++kk);
1621 }
1622 info->z_cycle = 0;
1623 #endif
1624 }
1625 if (fetch != 0) {
1626 /*
1627 * Fetch in byte domain and jump back
1628 * to sample domain.
1629 */
1630 fetched = SND_FXDIV(z_feed(f->source, c,
1631 info->z_delay + (info->z_pos * align),
1632 fetch * align, source), align);
1633 /*
1634 * Prepare to convert fetched buffer,
1635 * or mark us done if we cannot fulfill
1636 * the request.
1637 */
1638 reqin -= fetched;
1639 info->z_pos += fetched;
1640 if (fetched != fetch)
1641 reqin = 0;
1642 }
1643 }
1644
1645 reqout = z_min(z_gx2gy(info, z_fetched(info)), ocount);
1646 if (reqout != 0) {
1647 ocount -= reqout;
1648
1649 /*
1650 * Drift.. drift.. drift..
1651 *
1652 * Notice that there are 2 methods of doing the drift
1653 * operations: The former is much cleaner (in a sense
1654 * of mathematical readings of my eyes), but slower
1655 * due to integer division in z_gy2gx(). Nevertheless,
1656 * both should give the same exact accurate drifting
1657 * results, so the later is favourable.
1658 */
1659 do {
1660 info->z_resample(info, dst);
1661 info->z_alpha += alphadrift;
1662 if (info->z_alpha < info->z_gy)
1663 info->z_start += startdrift;
1664 else {
1665 info->z_start += startdrift - 1;
1666 info->z_alpha -= info->z_gy;
1667 }
1668 dst += align;
1669 #ifdef Z_DIAGNOSTIC
1670 info->z_cycle++;
1671 #endif
1672 } while (--reqout != 0);
1673 }
1674 } while (reqin != 0 && ocount != 0);
1675
1676 /*
1677 * Back to byte domain..
1678 */
1679 return (dst - b);
1680 }
1681
1682 static int
z_resampler_feed(struct pcm_feeder * f,struct pcm_channel * c,uint8_t * b,uint32_t count,void * source)1683 z_resampler_feed(struct pcm_feeder *f, struct pcm_channel *c, uint8_t *b,
1684 uint32_t count, void *source)
1685 {
1686 uint32_t feed, maxfeed, left;
1687
1688 /*
1689 * Split count to smaller chunks to avoid possible 32bit overflow.
1690 */
1691 maxfeed = ((struct z_info *)(f->data))->z_maxfeed;
1692 left = count;
1693
1694 do {
1695 feed = z_resampler_feed_internal(f, c, b,
1696 z_min(maxfeed, left), source);
1697 b += feed;
1698 left -= feed;
1699 } while (left != 0 && feed != 0);
1700
1701 return (count - left);
1702 }
1703
1704 static struct pcm_feederdesc feeder_rate_desc[] = {
1705 { FEEDER_RATE, 0, 0, 0, 0 },
1706 { 0, 0, 0, 0, 0 },
1707 };
1708
1709 static kobj_method_t feeder_rate_methods[] = {
1710 KOBJMETHOD(feeder_init, z_resampler_init),
1711 KOBJMETHOD(feeder_free, z_resampler_free),
1712 KOBJMETHOD(feeder_set, z_resampler_set),
1713 KOBJMETHOD(feeder_get, z_resampler_get),
1714 KOBJMETHOD(feeder_feed, z_resampler_feed),
1715 KOBJMETHOD_END
1716 };
1717
1718 FEEDER_DECLARE(feeder_rate, NULL);
1719