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_sample_read(sx, AFMT_##SIGN##BIT##_##ENDIAN); \
475 y = pcm_sample_read(sy, AFMT_##SIGN##BIT##_##ENDIAN); \
476 x = Z_LINEAR_INTERPOLATE_##BIT(z, x, y); \
477 pcm_sample_write(dst, x, AFMT_##SIGN##BIT##_##ENDIAN); \
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 /*
506 * Sine Cardinal (SINC) Interpolation. Scaling is done in 64 bit, so
507 * there's no point to hold the plate any longer. All samples will be
508 * shifted to a full 32 bit, scaled and restored during write for
509 * maximum dynamic range (only for downsampling).
510 */
511 #define _Z_SINC_ACCUMULATE(SIGN, BIT, ENDIAN, adv) \
512 c += z >> Z_SHIFT; \
513 z &= Z_MASK; \
514 coeff = Z_COEFF_INTERPOLATE(z, z_coeff[c], z_dcoeff[c]); \
515 x = pcm_sample_read(p, AFMT_##SIGN##BIT##_##ENDIAN); \
516 v += Z_NORM_##BIT((intpcm64_t)x * coeff); \
517 z += info->z_dy; \
518 p adv##= info->channels * PCM_##BIT##_BPS
519
520 /*
521 * XXX GCC4 optimization is such a !@#$%, need manual unrolling.
522 */
523 #if defined(__GNUC__) && __GNUC__ >= 4
524 #define Z_SINC_ACCUMULATE(...) do { \
525 _Z_SINC_ACCUMULATE(__VA_ARGS__); \
526 _Z_SINC_ACCUMULATE(__VA_ARGS__); \
527 } while (0)
528 #define Z_SINC_ACCUMULATE_DECR 2
529 #else
530 #define Z_SINC_ACCUMULATE(...) do { \
531 _Z_SINC_ACCUMULATE(__VA_ARGS__); \
532 } while (0)
533 #define Z_SINC_ACCUMULATE_DECR 1
534 #endif
535
536 #define Z_DECLARE_SINC(SIGN, BIT, ENDIAN) \
537 static void \
538 z_feed_sinc_##SIGN##BIT##ENDIAN(struct z_info *info, uint8_t *dst) \
539 { \
540 intpcm64_t v; \
541 intpcm_t x; \
542 uint8_t *p; \
543 int32_t coeff, z, *z_coeff, *z_dcoeff; \
544 uint32_t c, center, ch, i; \
545 \
546 z_coeff = info->z_coeff; \
547 z_dcoeff = info->z_dcoeff; \
548 center = z_prev(info, info->z_start, info->z_size); \
549 ch = info->channels * PCM_##BIT##_BPS; \
550 dst += ch; \
551 \
552 do { \
553 dst -= PCM_##BIT##_BPS; \
554 ch -= PCM_##BIT##_BPS; \
555 v = 0; \
556 z = info->z_alpha * info->z_dx; \
557 c = 0; \
558 p = info->z_delay + (z_next(info, center, 1) * \
559 info->channels * PCM_##BIT##_BPS) + ch; \
560 for (i = info->z_size; i != 0; i -= Z_SINC_ACCUMULATE_DECR) \
561 Z_SINC_ACCUMULATE(SIGN, BIT, ENDIAN, +); \
562 z = info->z_dy - (info->z_alpha * info->z_dx); \
563 c = 0; \
564 p = info->z_delay + (center * info->channels * \
565 PCM_##BIT##_BPS) + ch; \
566 for (i = info->z_size; i != 0; i -= Z_SINC_ACCUMULATE_DECR) \
567 Z_SINC_ACCUMULATE(SIGN, BIT, ENDIAN, -); \
568 if (info->z_scale != Z_ONE) \
569 v = Z_SCALE_##BIT(v, info->z_scale); \
570 else \
571 v >>= Z_COEFF_SHIFT - Z_GUARD_BIT_##BIT; \
572 Z_CLIP_CHECK(v, BIT); \
573 pcm_sample_write(dst, pcm_clamp(v, AFMT_##SIGN##BIT##_##ENDIAN),\
574 AFMT_##SIGN##BIT##_##ENDIAN); \
575 } while (ch != 0); \
576 }
577
578 #define Z_DECLARE_SINC_POLYPHASE(SIGN, BIT, ENDIAN) \
579 static void \
580 z_feed_sinc_polyphase_##SIGN##BIT##ENDIAN(struct z_info *info, uint8_t *dst) \
581 { \
582 intpcm64_t v; \
583 intpcm_t x; \
584 uint8_t *p; \
585 int32_t ch, i, start, *z_pcoeff; \
586 \
587 ch = info->channels * PCM_##BIT##_BPS; \
588 dst += ch; \
589 start = z_prev(info, info->z_start, (info->z_size << 1) - 1) * ch; \
590 \
591 do { \
592 dst -= PCM_##BIT##_BPS; \
593 ch -= PCM_##BIT##_BPS; \
594 v = 0; \
595 p = info->z_delay + start + ch; \
596 z_pcoeff = info->z_pcoeff + \
597 ((info->z_alpha * info->z_size) << 1); \
598 for (i = info->z_size; i != 0; i--) { \
599 x = pcm_sample_read(p, AFMT_##SIGN##BIT##_##ENDIAN); \
600 v += Z_NORM_##BIT((intpcm64_t)x * *z_pcoeff); \
601 z_pcoeff++; \
602 p += info->channels * PCM_##BIT##_BPS; \
603 x = pcm_sample_read(p, AFMT_##SIGN##BIT##_##ENDIAN); \
604 v += Z_NORM_##BIT((intpcm64_t)x * *z_pcoeff); \
605 z_pcoeff++; \
606 p += info->channels * PCM_##BIT##_BPS; \
607 } \
608 if (info->z_scale != Z_ONE) \
609 v = Z_SCALE_##BIT(v, info->z_scale); \
610 else \
611 v >>= Z_COEFF_SHIFT - Z_GUARD_BIT_##BIT; \
612 Z_CLIP_CHECK(v, BIT); \
613 pcm_sample_write(dst, pcm_clamp(v, AFMT_##SIGN##BIT##_##ENDIAN),\
614 AFMT_##SIGN##BIT##_##ENDIAN); \
615 } while (ch != 0); \
616 }
617
618 #define Z_DECLARE(SIGN, BIT, ENDIAN) \
619 Z_DECLARE_LINEAR(SIGN, BIT, ENDIAN) \
620 Z_DECLARE_SINC(SIGN, BIT, ENDIAN) \
621 Z_DECLARE_SINC_POLYPHASE(SIGN, BIT, ENDIAN)
622
623 #if BYTE_ORDER == LITTLE_ENDIAN || defined(SND_FEEDER_MULTIFORMAT)
624 Z_DECLARE(S, 16, LE)
625 Z_DECLARE(S, 32, LE)
626 #endif
627 #if BYTE_ORDER == BIG_ENDIAN || defined(SND_FEEDER_MULTIFORMAT)
628 Z_DECLARE(S, 16, BE)
629 Z_DECLARE(S, 32, BE)
630 #endif
631 #ifdef SND_FEEDER_MULTIFORMAT
632 Z_DECLARE(S, 8, NE)
633 Z_DECLARE(S, 24, LE)
634 Z_DECLARE(S, 24, BE)
635 Z_DECLARE(U, 8, NE)
636 Z_DECLARE(U, 16, LE)
637 Z_DECLARE(U, 24, LE)
638 Z_DECLARE(U, 32, LE)
639 Z_DECLARE(U, 16, BE)
640 Z_DECLARE(U, 24, BE)
641 Z_DECLARE(U, 32, BE)
642 #endif
643
644 enum {
645 Z_RESAMPLER_ZOH,
646 Z_RESAMPLER_LINEAR,
647 Z_RESAMPLER_SINC,
648 Z_RESAMPLER_SINC_POLYPHASE,
649 Z_RESAMPLER_LAST
650 };
651
652 #define Z_RESAMPLER_IDX(i) \
653 (Z_IS_SINC(i) ? Z_RESAMPLER_SINC : (i)->quality)
654
655 #define Z_RESAMPLER_ENTRY(SIGN, BIT, ENDIAN) \
656 { \
657 AFMT_##SIGN##BIT##_##ENDIAN, \
658 { \
659 [Z_RESAMPLER_ZOH] = z_feed_zoh, \
660 [Z_RESAMPLER_LINEAR] = z_feed_linear_##SIGN##BIT##ENDIAN, \
661 [Z_RESAMPLER_SINC] = z_feed_sinc_##SIGN##BIT##ENDIAN, \
662 [Z_RESAMPLER_SINC_POLYPHASE] = \
663 z_feed_sinc_polyphase_##SIGN##BIT##ENDIAN \
664 } \
665 }
666
667 static const struct {
668 uint32_t format;
669 z_resampler_t resampler[Z_RESAMPLER_LAST];
670 } z_resampler_tab[] = {
671 #if BYTE_ORDER == LITTLE_ENDIAN || defined(SND_FEEDER_MULTIFORMAT)
672 Z_RESAMPLER_ENTRY(S, 16, LE),
673 Z_RESAMPLER_ENTRY(S, 32, LE),
674 #endif
675 #if BYTE_ORDER == BIG_ENDIAN || defined(SND_FEEDER_MULTIFORMAT)
676 Z_RESAMPLER_ENTRY(S, 16, BE),
677 Z_RESAMPLER_ENTRY(S, 32, BE),
678 #endif
679 #ifdef SND_FEEDER_MULTIFORMAT
680 Z_RESAMPLER_ENTRY(S, 8, NE),
681 Z_RESAMPLER_ENTRY(S, 24, LE),
682 Z_RESAMPLER_ENTRY(S, 24, BE),
683 Z_RESAMPLER_ENTRY(U, 8, NE),
684 Z_RESAMPLER_ENTRY(U, 16, LE),
685 Z_RESAMPLER_ENTRY(U, 24, LE),
686 Z_RESAMPLER_ENTRY(U, 32, LE),
687 Z_RESAMPLER_ENTRY(U, 16, BE),
688 Z_RESAMPLER_ENTRY(U, 24, BE),
689 Z_RESAMPLER_ENTRY(U, 32, BE),
690 #endif
691 };
692
693 #define Z_RESAMPLER_TAB_SIZE \
694 ((int32_t)(sizeof(z_resampler_tab) / sizeof(z_resampler_tab[0])))
695
696 static void
z_resampler_reset(struct z_info * info)697 z_resampler_reset(struct z_info *info)
698 {
699
700 info->src = info->rsrc - (info->rsrc % ((feeder_rate_round > 0 &&
701 info->rsrc > feeder_rate_round) ? feeder_rate_round : 1));
702 info->dst = info->rdst - (info->rdst % ((feeder_rate_round > 0 &&
703 info->rdst > feeder_rate_round) ? feeder_rate_round : 1));
704 info->z_gx = 1;
705 info->z_gy = 1;
706 info->z_alpha = 0;
707 info->z_resample = NULL;
708 info->z_size = 1;
709 info->z_coeff = NULL;
710 info->z_dcoeff = NULL;
711 if (info->z_pcoeff != NULL) {
712 free(info->z_pcoeff, M_DEVBUF);
713 info->z_pcoeff = NULL;
714 }
715 info->z_scale = Z_ONE;
716 info->z_dx = Z_FULL_ONE;
717 info->z_dy = Z_FULL_ONE;
718 #ifdef Z_DIAGNOSTIC
719 info->z_cycle = 0;
720 #endif
721 if (info->quality < Z_QUALITY_MIN)
722 info->quality = Z_QUALITY_MIN;
723 else if (info->quality > Z_QUALITY_MAX)
724 info->quality = Z_QUALITY_MAX;
725 }
726
727 #ifdef Z_PARANOID
728 static int32_t
z_resampler_sinc_len(struct z_info * info)729 z_resampler_sinc_len(struct z_info *info)
730 {
731 int32_t c, z, len, lmax;
732
733 if (!Z_IS_SINC(info))
734 return (1);
735
736 /*
737 * A rather careful (or useless) way to calculate filter length.
738 * Z_SINC_LEN() itself is accurate enough to do its job. Extra
739 * sanity checking is not going to hurt though..
740 */
741 c = 0;
742 z = info->z_dy;
743 len = 0;
744 lmax = z_coeff_tab[Z_SINC_COEFF_IDX(info)].len;
745
746 do {
747 c += z >> Z_SHIFT;
748 z &= Z_MASK;
749 z += info->z_dy;
750 } while (c < lmax && ++len > 0);
751
752 if (len != Z_SINC_LEN(info)) {
753 #ifdef _KERNEL
754 printf("%s(): sinc l=%d != Z_SINC_LEN=%d\n",
755 __func__, len, Z_SINC_LEN(info));
756 #else
757 fprintf(stderr, "%s(): sinc l=%d != Z_SINC_LEN=%d\n",
758 __func__, len, Z_SINC_LEN(info));
759 return (-1);
760 #endif
761 }
762
763 return (len);
764 }
765 #else
766 #define z_resampler_sinc_len(i) (Z_IS_SINC(i) ? Z_SINC_LEN(i) : 1)
767 #endif
768
769 #define Z_POLYPHASE_COEFF_SHIFT 0
770
771 /*
772 * Pick suitable polynomial interpolators based on filter oversampled ratio
773 * (2 ^ Z_DRIFT_SHIFT).
774 */
775 #if !(defined(Z_COEFF_INTERP_ZOH) || defined(Z_COEFF_INTERP_LINEAR) || \
776 defined(Z_COEFF_INTERP_QUADRATIC) || defined(Z_COEFF_INTERP_HERMITE) || \
777 defined(Z_COEFF_INTER_BSPLINE) || defined(Z_COEFF_INTERP_OPT32X) || \
778 defined(Z_COEFF_INTERP_OPT16X) || defined(Z_COEFF_INTERP_OPT8X) || \
779 defined(Z_COEFF_INTERP_OPT4X) || defined(Z_COEFF_INTERP_OPT2X))
780 #if Z_DRIFT_SHIFT >= 6
781 #define Z_COEFF_INTERP_BSPLINE 1
782 #elif Z_DRIFT_SHIFT >= 5
783 #define Z_COEFF_INTERP_OPT32X 1
784 #elif Z_DRIFT_SHIFT == 4
785 #define Z_COEFF_INTERP_OPT16X 1
786 #elif Z_DRIFT_SHIFT == 3
787 #define Z_COEFF_INTERP_OPT8X 1
788 #elif Z_DRIFT_SHIFT == 2
789 #define Z_COEFF_INTERP_OPT4X 1
790 #elif Z_DRIFT_SHIFT == 1
791 #define Z_COEFF_INTERP_OPT2X 1
792 #else
793 #error "Z_DRIFT_SHIFT screwed!"
794 #endif
795 #endif
796
797 /*
798 * In classic polyphase mode, the actual coefficients for each phases need to
799 * be calculated based on default prototype filters. For highly oversampled
800 * filter, linear or quadradatic interpolator should be enough. Anything less
801 * than that require 'special' interpolators to reduce interpolation errors.
802 *
803 * "Polynomial Interpolators for High-Quality Resampling of Oversampled Audio"
804 * by Olli Niemitalo
805 * - http://www.student.oulu.fi/~oniemita/dsp/deip.pdf
806 *
807 */
808 static int32_t
z_coeff_interpolate(int32_t z,int32_t * z_coeff)809 z_coeff_interpolate(int32_t z, int32_t *z_coeff)
810 {
811 int32_t coeff;
812 #if defined(Z_COEFF_INTERP_ZOH)
813
814 /* 1-point, 0th-order (Zero Order Hold) */
815 z = z;
816 coeff = z_coeff[0];
817 #elif defined(Z_COEFF_INTERP_LINEAR)
818 int32_t zl0, zl1;
819
820 /* 2-point, 1st-order Linear */
821 zl0 = z_coeff[0];
822 zl1 = z_coeff[1] - z_coeff[0];
823
824 coeff = Z_RSHIFT((int64_t)zl1 * z, Z_SHIFT) + zl0;
825 #elif defined(Z_COEFF_INTERP_QUADRATIC)
826 int32_t zq0, zq1, zq2;
827
828 /* 3-point, 2nd-order Quadratic */
829 zq0 = z_coeff[0];
830 zq1 = z_coeff[1] - z_coeff[-1];
831 zq2 = z_coeff[1] + z_coeff[-1] - (z_coeff[0] << 1);
832
833 coeff = Z_RSHIFT((Z_RSHIFT((int64_t)zq2 * z, Z_SHIFT) +
834 zq1) * z, Z_SHIFT + 1) + zq0;
835 #elif defined(Z_COEFF_INTERP_HERMITE)
836 int32_t zh0, zh1, zh2, zh3;
837
838 /* 4-point, 3rd-order Hermite */
839 zh0 = z_coeff[0];
840 zh1 = z_coeff[1] - z_coeff[-1];
841 zh2 = (z_coeff[-1] << 1) - (z_coeff[0] * 5) + (z_coeff[1] << 2) -
842 z_coeff[2];
843 zh3 = z_coeff[2] - z_coeff[-1] + ((z_coeff[0] - z_coeff[1]) * 3);
844
845 coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((int64_t)zh3 * z, Z_SHIFT) +
846 zh2) * z, Z_SHIFT) + zh1) * z, Z_SHIFT + 1) + zh0;
847 #elif defined(Z_COEFF_INTERP_BSPLINE)
848 int32_t zb0, zb1, zb2, zb3;
849
850 /* 4-point, 3rd-order B-Spline */
851 zb0 = Z_RSHIFT(0x15555555LL * (((int64_t)z_coeff[0] << 2) +
852 z_coeff[-1] + z_coeff[1]), 30);
853 zb1 = z_coeff[1] - z_coeff[-1];
854 zb2 = z_coeff[-1] + z_coeff[1] - (z_coeff[0] << 1);
855 zb3 = Z_RSHIFT(0x15555555LL * (((z_coeff[0] - z_coeff[1]) * 3) +
856 z_coeff[2] - z_coeff[-1]), 30);
857
858 coeff = (Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((int64_t)zb3 * z, Z_SHIFT) +
859 zb2) * z, Z_SHIFT) + zb1) * z, Z_SHIFT) + zb0 + 1) >> 1;
860 #elif defined(Z_COEFF_INTERP_OPT32X)
861 int32_t zoz, zoe1, zoe2, zoe3, zoo1, zoo2, zoo3;
862 int32_t zoc0, zoc1, zoc2, zoc3, zoc4, zoc5;
863
864 /* 6-point, 5th-order Optimal 32x */
865 zoz = z - (Z_ONE >> 1);
866 zoe1 = z_coeff[1] + z_coeff[0];
867 zoe2 = z_coeff[2] + z_coeff[-1];
868 zoe3 = z_coeff[3] + z_coeff[-2];
869 zoo1 = z_coeff[1] - z_coeff[0];
870 zoo2 = z_coeff[2] - z_coeff[-1];
871 zoo3 = z_coeff[3] - z_coeff[-2];
872
873 zoc0 = Z_RSHIFT((0x1ac2260dLL * zoe1) + (0x0526cdcaLL * zoe2) +
874 (0x00170c29LL * zoe3), 30);
875 zoc1 = Z_RSHIFT((0x14f8a49aLL * zoo1) + (0x0d6d1109LL * zoo2) +
876 (0x008cd4dcLL * zoo3), 30);
877 zoc2 = Z_RSHIFT((-0x0d3e94a4LL * zoe1) + (0x0bddded4LL * zoe2) +
878 (0x0160b5d0LL * zoe3), 30);
879 zoc3 = Z_RSHIFT((-0x0de10cc4LL * zoo1) + (0x019b2a7dLL * zoo2) +
880 (0x01cfe914LL * zoo3), 30);
881 zoc4 = Z_RSHIFT((0x02aa12d7LL * zoe1) + (-0x03ff1bb3LL * zoe2) +
882 (0x015508ddLL * zoe3), 30);
883 zoc5 = Z_RSHIFT((0x051d29e5LL * zoo1) + (-0x028e7647LL * zoo2) +
884 (0x0082d81aLL * zoo3), 30);
885
886 coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT(
887 (int64_t)zoc5 * zoz, Z_SHIFT) +
888 zoc4) * zoz, Z_SHIFT) + zoc3) * zoz, Z_SHIFT) +
889 zoc2) * zoz, Z_SHIFT) + zoc1) * zoz, Z_SHIFT) + zoc0;
890 #elif defined(Z_COEFF_INTERP_OPT16X)
891 int32_t zoz, zoe1, zoe2, zoe3, zoo1, zoo2, zoo3;
892 int32_t zoc0, zoc1, zoc2, zoc3, zoc4, zoc5;
893
894 /* 6-point, 5th-order Optimal 16x */
895 zoz = z - (Z_ONE >> 1);
896 zoe1 = z_coeff[1] + z_coeff[0];
897 zoe2 = z_coeff[2] + z_coeff[-1];
898 zoe3 = z_coeff[3] + z_coeff[-2];
899 zoo1 = z_coeff[1] - z_coeff[0];
900 zoo2 = z_coeff[2] - z_coeff[-1];
901 zoo3 = z_coeff[3] - z_coeff[-2];
902
903 zoc0 = Z_RSHIFT((0x1ac2260dLL * zoe1) + (0x0526cdcaLL * zoe2) +
904 (0x00170c29LL * zoe3), 30);
905 zoc1 = Z_RSHIFT((0x14f8a49aLL * zoo1) + (0x0d6d1109LL * zoo2) +
906 (0x008cd4dcLL * zoo3), 30);
907 zoc2 = Z_RSHIFT((-0x0d3e94a4LL * zoe1) + (0x0bddded4LL * zoe2) +
908 (0x0160b5d0LL * zoe3), 30);
909 zoc3 = Z_RSHIFT((-0x0de10cc4LL * zoo1) + (0x019b2a7dLL * zoo2) +
910 (0x01cfe914LL * zoo3), 30);
911 zoc4 = Z_RSHIFT((0x02aa12d7LL * zoe1) + (-0x03ff1bb3LL * zoe2) +
912 (0x015508ddLL * zoe3), 30);
913 zoc5 = Z_RSHIFT((0x051d29e5LL * zoo1) + (-0x028e7647LL * zoo2) +
914 (0x0082d81aLL * zoo3), 30);
915
916 coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT(
917 (int64_t)zoc5 * zoz, Z_SHIFT) +
918 zoc4) * zoz, Z_SHIFT) + zoc3) * zoz, Z_SHIFT) +
919 zoc2) * zoz, Z_SHIFT) + zoc1) * zoz, Z_SHIFT) + zoc0;
920 #elif defined(Z_COEFF_INTERP_OPT8X)
921 int32_t zoz, zoe1, zoe2, zoe3, zoo1, zoo2, zoo3;
922 int32_t zoc0, zoc1, zoc2, zoc3, zoc4, zoc5;
923
924 /* 6-point, 5th-order Optimal 8x */
925 zoz = z - (Z_ONE >> 1);
926 zoe1 = z_coeff[1] + z_coeff[0];
927 zoe2 = z_coeff[2] + z_coeff[-1];
928 zoe3 = z_coeff[3] + z_coeff[-2];
929 zoo1 = z_coeff[1] - z_coeff[0];
930 zoo2 = z_coeff[2] - z_coeff[-1];
931 zoo3 = z_coeff[3] - z_coeff[-2];
932
933 zoc0 = Z_RSHIFT((0x1aa9b47dLL * zoe1) + (0x053d9944LL * zoe2) +
934 (0x0018b23fLL * zoe3), 30);
935 zoc1 = Z_RSHIFT((0x14a104d1LL * zoo1) + (0x0d7d2504LL * zoo2) +
936 (0x0094b599LL * zoo3), 30);
937 zoc2 = Z_RSHIFT((-0x0d22530bLL * zoe1) + (0x0bb37a2cLL * zoe2) +
938 (0x016ed8e0LL * zoe3), 30);
939 zoc3 = Z_RSHIFT((-0x0d744b1cLL * zoo1) + (0x01649591LL * zoo2) +
940 (0x01dae93aLL * zoo3), 30);
941 zoc4 = Z_RSHIFT((0x02a7ee1bLL * zoe1) + (-0x03fbdb24LL * zoe2) +
942 (0x0153ed07LL * zoe3), 30);
943 zoc5 = Z_RSHIFT((0x04cf9b6cLL * zoo1) + (-0x0266b378LL * zoo2) +
944 (0x007a7c26LL * zoo3), 30);
945
946 coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT(
947 (int64_t)zoc5 * zoz, Z_SHIFT) +
948 zoc4) * zoz, Z_SHIFT) + zoc3) * zoz, Z_SHIFT) +
949 zoc2) * zoz, Z_SHIFT) + zoc1) * zoz, Z_SHIFT) + zoc0;
950 #elif defined(Z_COEFF_INTERP_OPT4X)
951 int32_t zoz, zoe1, zoe2, zoe3, zoo1, zoo2, zoo3;
952 int32_t zoc0, zoc1, zoc2, zoc3, zoc4, zoc5;
953
954 /* 6-point, 5th-order Optimal 4x */
955 zoz = z - (Z_ONE >> 1);
956 zoe1 = z_coeff[1] + z_coeff[0];
957 zoe2 = z_coeff[2] + z_coeff[-1];
958 zoe3 = z_coeff[3] + z_coeff[-2];
959 zoo1 = z_coeff[1] - z_coeff[0];
960 zoo2 = z_coeff[2] - z_coeff[-1];
961 zoo3 = z_coeff[3] - z_coeff[-2];
962
963 zoc0 = Z_RSHIFT((0x1a8eda43LL * zoe1) + (0x0556ee38LL * zoe2) +
964 (0x001a3784LL * zoe3), 30);
965 zoc1 = Z_RSHIFT((0x143d863eLL * zoo1) + (0x0d910e36LL * zoo2) +
966 (0x009ca889LL * zoo3), 30);
967 zoc2 = Z_RSHIFT((-0x0d026821LL * zoe1) + (0x0b837773LL * zoe2) +
968 (0x017ef0c6LL * zoe3), 30);
969 zoc3 = Z_RSHIFT((-0x0cef1502LL * zoo1) + (0x01207a8eLL * zoo2) +
970 (0x01e936dbLL * zoo3), 30);
971 zoc4 = Z_RSHIFT((0x029fe643LL * zoe1) + (-0x03ef3fc8LL * zoe2) +
972 (0x014f5923LL * zoe3), 30);
973 zoc5 = Z_RSHIFT((0x043a9d08LL * zoo1) + (-0x02154febLL * zoo2) +
974 (0x00670dbdLL * zoo3), 30);
975
976 coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT(
977 (int64_t)zoc5 * zoz, Z_SHIFT) +
978 zoc4) * zoz, Z_SHIFT) + zoc3) * zoz, Z_SHIFT) +
979 zoc2) * zoz, Z_SHIFT) + zoc1) * zoz, Z_SHIFT) + zoc0;
980 #elif defined(Z_COEFF_INTERP_OPT2X)
981 int32_t zoz, zoe1, zoe2, zoe3, zoo1, zoo2, zoo3;
982 int32_t zoc0, zoc1, zoc2, zoc3, zoc4, zoc5;
983
984 /* 6-point, 5th-order Optimal 2x */
985 zoz = z - (Z_ONE >> 1);
986 zoe1 = z_coeff[1] + z_coeff[0];
987 zoe2 = z_coeff[2] + z_coeff[-1];
988 zoe3 = z_coeff[3] + z_coeff[-2];
989 zoo1 = z_coeff[1] - z_coeff[0];
990 zoo2 = z_coeff[2] - z_coeff[-1];
991 zoo3 = z_coeff[3] - z_coeff[-2];
992
993 zoc0 = Z_RSHIFT((0x19edb6fdLL * zoe1) + (0x05ebd062LL * zoe2) +
994 (0x00267881LL * zoe3), 30);
995 zoc1 = Z_RSHIFT((0x1223af76LL * zoo1) + (0x0de3dd6bLL * zoo2) +
996 (0x00d683cdLL * zoo3), 30);
997 zoc2 = Z_RSHIFT((-0x0c3ee068LL * zoe1) + (0x0a5c3769LL * zoe2) +
998 (0x01e2aceaLL * zoe3), 30);
999 zoc3 = Z_RSHIFT((-0x0a8ab614LL * zoo1) + (-0x0019522eLL * zoo2) +
1000 (0x022cefc7LL * zoo3), 30);
1001 zoc4 = Z_RSHIFT((0x0276187dLL * zoe1) + (-0x03a801e8LL * zoe2) +
1002 (0x0131d935LL * zoe3), 30);
1003 zoc5 = Z_RSHIFT((0x02c373f5LL * zoo1) + (-0x01275f83LL * zoo2) +
1004 (0x0018ee79LL * zoo3), 30);
1005
1006 coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT(
1007 (int64_t)zoc5 * zoz, Z_SHIFT) +
1008 zoc4) * zoz, Z_SHIFT) + zoc3) * zoz, Z_SHIFT) +
1009 zoc2) * zoz, Z_SHIFT) + zoc1) * zoz, Z_SHIFT) + zoc0;
1010 #else
1011 #error "Interpolation type screwed!"
1012 #endif
1013
1014 #if Z_POLYPHASE_COEFF_SHIFT > 0
1015 coeff = Z_RSHIFT(coeff, Z_POLYPHASE_COEFF_SHIFT);
1016 #endif
1017 return (coeff);
1018 }
1019
1020 static int
z_resampler_build_polyphase(struct z_info * info)1021 z_resampler_build_polyphase(struct z_info *info)
1022 {
1023 int32_t alpha, c, i, z, idx;
1024
1025 /* Let this be here first. */
1026 if (info->z_pcoeff != NULL) {
1027 free(info->z_pcoeff, M_DEVBUF);
1028 info->z_pcoeff = NULL;
1029 }
1030
1031 if (feeder_rate_polyphase_max < 1)
1032 return (ENOTSUP);
1033
1034 if (((int64_t)info->z_size * info->z_gy * 2) >
1035 feeder_rate_polyphase_max) {
1036 #ifndef _KERNEL
1037 fprintf(stderr, "Polyphase entries exceed: [%d/%d] %jd > %d\n",
1038 info->z_gx, info->z_gy,
1039 (intmax_t)info->z_size * info->z_gy * 2,
1040 feeder_rate_polyphase_max);
1041 #endif
1042 return (E2BIG);
1043 }
1044
1045 info->z_pcoeff = malloc(sizeof(int32_t) *
1046 info->z_size * info->z_gy * 2, M_DEVBUF, M_NOWAIT | M_ZERO);
1047 if (info->z_pcoeff == NULL)
1048 return (ENOMEM);
1049
1050 for (alpha = 0; alpha < info->z_gy; alpha++) {
1051 z = alpha * info->z_dx;
1052 c = 0;
1053 for (i = info->z_size; i != 0; i--) {
1054 c += z >> Z_SHIFT;
1055 z &= Z_MASK;
1056 idx = (alpha * info->z_size * 2) +
1057 (info->z_size * 2) - i;
1058 info->z_pcoeff[idx] =
1059 z_coeff_interpolate(z, info->z_coeff + c);
1060 z += info->z_dy;
1061 }
1062 z = info->z_dy - (alpha * info->z_dx);
1063 c = 0;
1064 for (i = info->z_size; i != 0; i--) {
1065 c += z >> Z_SHIFT;
1066 z &= Z_MASK;
1067 idx = (alpha * info->z_size * 2) + i - 1;
1068 info->z_pcoeff[idx] =
1069 z_coeff_interpolate(z, info->z_coeff + c);
1070 z += info->z_dy;
1071 }
1072 }
1073
1074 #ifndef _KERNEL
1075 fprintf(stderr, "Polyphase: [%d/%d] %d entries\n",
1076 info->z_gx, info->z_gy, info->z_size * info->z_gy * 2);
1077 #endif
1078
1079 return (0);
1080 }
1081
1082 static int
z_resampler_setup(struct pcm_feeder * f)1083 z_resampler_setup(struct pcm_feeder *f)
1084 {
1085 struct z_info *info;
1086 int64_t gy2gx_max, gx2gy_max;
1087 uint32_t format;
1088 int32_t align, i, z_scale;
1089 int adaptive;
1090
1091 info = f->data;
1092 z_resampler_reset(info);
1093
1094 if (info->src == info->dst)
1095 return (0);
1096
1097 /* Shrink by greatest common divisor. */
1098 i = z_gcd(info->src, info->dst);
1099 info->z_gx = info->src / i;
1100 info->z_gy = info->dst / i;
1101
1102 /* Too big, or too small. Bail out. */
1103 if (!(Z_FACTOR_SAFE(info->z_gx) && Z_FACTOR_SAFE(info->z_gy)))
1104 return (EINVAL);
1105
1106 format = f->desc->in;
1107 adaptive = 0;
1108 z_scale = 0;
1109
1110 /*
1111 * Setup everything: filter length, conversion factor, etc.
1112 */
1113 if (Z_IS_SINC(info)) {
1114 /*
1115 * Downsampling, or upsampling scaling factor. As long as the
1116 * factor can be represented by a fraction of 1 << Z_SHIFT,
1117 * we're pretty much in business. Scaling is not needed for
1118 * upsampling, so we just slap Z_ONE there.
1119 */
1120 if (info->z_gx > info->z_gy)
1121 /*
1122 * If the downsampling ratio is beyond sanity,
1123 * enable semi-adaptive mode. Although handling
1124 * extreme ratio is possible, the result of the
1125 * conversion is just pointless, unworthy,
1126 * nonsensical noises, etc.
1127 */
1128 if ((info->z_gx / info->z_gy) > Z_SINC_DOWNMAX)
1129 z_scale = Z_ONE / Z_SINC_DOWNMAX;
1130 else
1131 z_scale = ((uint64_t)info->z_gy << Z_SHIFT) /
1132 info->z_gx;
1133 else
1134 z_scale = Z_ONE;
1135
1136 /*
1137 * This is actually impossible, unless anything above
1138 * overflow.
1139 */
1140 if (z_scale < 1)
1141 return (E2BIG);
1142
1143 /*
1144 * Calculate sample time/coefficients index drift. It is
1145 * a constant for upsampling, but downsampling require
1146 * heavy duty filtering with possible too long filters.
1147 * If anything goes wrong, revisit again and enable
1148 * adaptive mode.
1149 */
1150 z_setup_adaptive_sinc:
1151 if (info->z_pcoeff != NULL) {
1152 free(info->z_pcoeff, M_DEVBUF);
1153 info->z_pcoeff = NULL;
1154 }
1155
1156 if (adaptive == 0) {
1157 info->z_dy = z_scale << Z_DRIFT_SHIFT;
1158 if (info->z_dy < 1)
1159 return (E2BIG);
1160 info->z_scale = z_scale;
1161 } else {
1162 info->z_dy = Z_FULL_ONE;
1163 info->z_scale = Z_ONE;
1164 }
1165
1166 /* Smallest drift increment. */
1167 info->z_dx = info->z_dy / info->z_gy;
1168
1169 /*
1170 * Overflow or underflow. Try adaptive, let it continue and
1171 * retry.
1172 */
1173 if (info->z_dx < 1) {
1174 if (adaptive == 0) {
1175 adaptive = 1;
1176 goto z_setup_adaptive_sinc;
1177 }
1178 return (E2BIG);
1179 }
1180
1181 /*
1182 * Round back output drift.
1183 */
1184 info->z_dy = info->z_dx * info->z_gy;
1185
1186 for (i = 0; i < Z_COEFF_TAB_SIZE; i++) {
1187 if (Z_SINC_COEFF_IDX(info) != i)
1188 continue;
1189 /*
1190 * Calculate required filter length and guard
1191 * against possible abusive result. Note that
1192 * this represents only 1/2 of the entire filter
1193 * length.
1194 */
1195 info->z_size = z_resampler_sinc_len(info);
1196
1197 /*
1198 * Multiple of 2 rounding, for better accumulator
1199 * performance.
1200 */
1201 info->z_size &= ~1;
1202
1203 if (info->z_size < 2 || info->z_size > Z_SINC_MAX) {
1204 if (adaptive == 0) {
1205 adaptive = 1;
1206 goto z_setup_adaptive_sinc;
1207 }
1208 return (E2BIG);
1209 }
1210 info->z_coeff = z_coeff_tab[i].coeff + Z_COEFF_OFFSET;
1211 info->z_dcoeff = z_coeff_tab[i].dcoeff;
1212 break;
1213 }
1214
1215 if (info->z_coeff == NULL || info->z_dcoeff == NULL)
1216 return (EINVAL);
1217 } else if (Z_IS_LINEAR(info)) {
1218 /*
1219 * Don't put much effort if we're doing linear interpolation.
1220 * Just center the interpolation distance within Z_LINEAR_ONE,
1221 * and be happy about it.
1222 */
1223 info->z_dx = Z_LINEAR_FULL_ONE / info->z_gy;
1224 }
1225
1226 /*
1227 * We're safe for now, lets continue.. Look for our resampler
1228 * depending on configured format and quality.
1229 */
1230 for (i = 0; i < Z_RESAMPLER_TAB_SIZE; i++) {
1231 int ridx;
1232
1233 if (AFMT_ENCODING(format) != z_resampler_tab[i].format)
1234 continue;
1235 if (Z_IS_SINC(info) && adaptive == 0 &&
1236 z_resampler_build_polyphase(info) == 0)
1237 ridx = Z_RESAMPLER_SINC_POLYPHASE;
1238 else
1239 ridx = Z_RESAMPLER_IDX(info);
1240 info->z_resample = z_resampler_tab[i].resampler[ridx];
1241 break;
1242 }
1243
1244 if (info->z_resample == NULL)
1245 return (EINVAL);
1246
1247 info->bps = AFMT_BPS(format);
1248 align = info->channels * info->bps;
1249
1250 /*
1251 * Calculate largest value that can be fed into z_gy2gx() and
1252 * z_gx2gy() without causing (signed) 32bit overflow. z_gy2gx() will
1253 * be called early during feeding process to determine how much input
1254 * samples that is required to generate requested output, while
1255 * z_gx2gy() will be called just before samples filtering /
1256 * accumulation process based on available samples that has been
1257 * calculated using z_gx2gy().
1258 *
1259 * Now that is damn confusing, I guess ;-) .
1260 */
1261 gy2gx_max = (((uint64_t)info->z_gy * INT32_MAX) - info->z_gy + 1) /
1262 info->z_gx;
1263
1264 if ((gy2gx_max * align) > SND_FXDIV_MAX)
1265 gy2gx_max = SND_FXDIV_MAX / align;
1266
1267 if (gy2gx_max < 1)
1268 return (E2BIG);
1269
1270 gx2gy_max = (((uint64_t)info->z_gx * INT32_MAX) - info->z_gy) /
1271 info->z_gy;
1272
1273 if (gx2gy_max > INT32_MAX)
1274 gx2gy_max = INT32_MAX;
1275
1276 if (gx2gy_max < 1)
1277 return (E2BIG);
1278
1279 /*
1280 * Ensure that z_gy2gx() at its largest possible calculated value
1281 * (alpha = 0) will not cause overflow further late during z_gx2gy()
1282 * stage.
1283 */
1284 if (z_gy2gx(info, gy2gx_max) > _Z_GCAST(gx2gy_max))
1285 return (E2BIG);
1286
1287 info->z_maxfeed = gy2gx_max * align;
1288
1289 #ifdef Z_USE_ALPHADRIFT
1290 info->z_startdrift = z_gy2gx(info, 1);
1291 info->z_alphadrift = z_drift(info, info->z_startdrift, 1);
1292 #endif
1293
1294 i = z_gy2gx(info, 1);
1295 info->z_full = z_roundpow2((info->z_size << 1) + i);
1296
1297 /*
1298 * Too big to be true, and overflowing left and right like mad ..
1299 */
1300 if ((info->z_full * align) < 1) {
1301 if (adaptive == 0 && Z_IS_SINC(info)) {
1302 adaptive = 1;
1303 goto z_setup_adaptive_sinc;
1304 }
1305 return (E2BIG);
1306 }
1307
1308 /*
1309 * Increase full buffer size if its too small to reduce cyclic
1310 * buffer shifting in main conversion/feeder loop.
1311 */
1312 while (info->z_full < Z_RESERVOIR_MAX &&
1313 (info->z_full - (info->z_size << 1)) < Z_RESERVOIR)
1314 info->z_full <<= 1;
1315
1316 /* Initialize buffer position. */
1317 info->z_mask = info->z_full - 1;
1318 info->z_start = z_prev(info, info->z_size << 1, 1);
1319 info->z_pos = z_next(info, info->z_start, 1);
1320
1321 /*
1322 * Allocate or reuse delay line buffer, whichever makes sense.
1323 */
1324 i = info->z_full * align;
1325 if (i < 1)
1326 return (E2BIG);
1327
1328 if (info->z_delay == NULL || info->z_alloc < i ||
1329 i <= (info->z_alloc >> 1)) {
1330 if (info->z_delay != NULL)
1331 free(info->z_delay, M_DEVBUF);
1332 info->z_delay = malloc(i, M_DEVBUF, M_NOWAIT | M_ZERO);
1333 if (info->z_delay == NULL)
1334 return (ENOMEM);
1335 info->z_alloc = i;
1336 }
1337
1338 /*
1339 * Zero out head of buffer to avoid pops and clicks.
1340 */
1341 memset(info->z_delay, sndbuf_zerodata(f->desc->out),
1342 info->z_pos * align);
1343
1344 #ifdef Z_DIAGNOSTIC
1345 /*
1346 * XXX Debuging mess !@#$%^
1347 */
1348 #define dumpz(x) fprintf(stderr, "\t%12s = %10u : %-11d\n", \
1349 "z_"__STRING(x), (uint32_t)info->z_##x, \
1350 (int32_t)info->z_##x)
1351 fprintf(stderr, "\n%s():\n", __func__);
1352 fprintf(stderr, "\tchannels=%d, bps=%d, format=0x%08x, quality=%d\n",
1353 info->channels, info->bps, format, info->quality);
1354 fprintf(stderr, "\t%d (%d) -> %d (%d), ",
1355 info->src, info->rsrc, info->dst, info->rdst);
1356 fprintf(stderr, "[%d/%d]\n", info->z_gx, info->z_gy);
1357 fprintf(stderr, "\tminreq=%d, ", z_gy2gx(info, 1));
1358 if (adaptive != 0)
1359 z_scale = Z_ONE;
1360 fprintf(stderr, "factor=0x%08x/0x%08x (%f)\n",
1361 z_scale, Z_ONE, (double)z_scale / Z_ONE);
1362 fprintf(stderr, "\tbase_length=%d, ", Z_SINC_BASE_LEN(info));
1363 fprintf(stderr, "adaptive=%s\n", (adaptive != 0) ? "YES" : "NO");
1364 dumpz(size);
1365 dumpz(alloc);
1366 if (info->z_alloc < 1024)
1367 fprintf(stderr, "\t%15s%10d Bytes\n",
1368 "", info->z_alloc);
1369 else if (info->z_alloc < (1024 << 10))
1370 fprintf(stderr, "\t%15s%10d KBytes\n",
1371 "", info->z_alloc >> 10);
1372 else if (info->z_alloc < (1024 << 20))
1373 fprintf(stderr, "\t%15s%10d MBytes\n",
1374 "", info->z_alloc >> 20);
1375 else
1376 fprintf(stderr, "\t%15s%10d GBytes\n",
1377 "", info->z_alloc >> 30);
1378 fprintf(stderr, "\t%12s %10d (min output samples)\n",
1379 "",
1380 (int32_t)z_gx2gy(info, info->z_full - (info->z_size << 1)));
1381 fprintf(stderr, "\t%12s %10d (min allocated output samples)\n",
1382 "",
1383 (int32_t)z_gx2gy(info, (info->z_alloc / align) -
1384 (info->z_size << 1)));
1385 fprintf(stderr, "\t%12s = %10d\n",
1386 "z_gy2gx()", (int32_t)z_gy2gx(info, 1));
1387 fprintf(stderr, "\t%12s = %10d -> z_gy2gx() -> %d\n",
1388 "Max", (int32_t)gy2gx_max, (int32_t)z_gy2gx(info, gy2gx_max));
1389 fprintf(stderr, "\t%12s = %10d\n",
1390 "z_gx2gy()", (int32_t)z_gx2gy(info, 1));
1391 fprintf(stderr, "\t%12s = %10d -> z_gx2gy() -> %d\n",
1392 "Max", (int32_t)gx2gy_max, (int32_t)z_gx2gy(info, gx2gy_max));
1393 dumpz(maxfeed);
1394 dumpz(full);
1395 dumpz(start);
1396 dumpz(pos);
1397 dumpz(scale);
1398 fprintf(stderr, "\t%12s %10f\n", "",
1399 (double)info->z_scale / Z_ONE);
1400 dumpz(dx);
1401 fprintf(stderr, "\t%12s %10f\n", "",
1402 (double)info->z_dx / info->z_dy);
1403 dumpz(dy);
1404 fprintf(stderr, "\t%12s %10d (drift step)\n", "",
1405 info->z_dy >> Z_SHIFT);
1406 fprintf(stderr, "\t%12s %10d (scaling differences)\n", "",
1407 (z_scale << Z_DRIFT_SHIFT) - info->z_dy);
1408 fprintf(stderr, "\t%12s = %u bytes\n",
1409 "intpcm32_t", sizeof(intpcm32_t));
1410 fprintf(stderr, "\t%12s = 0x%08x, smallest=%.16lf\n",
1411 "Z_ONE", Z_ONE, (double)1.0 / (double)Z_ONE);
1412 #endif
1413
1414 return (0);
1415 }
1416
1417 static int
z_resampler_set(struct pcm_feeder * f,int what,int32_t value)1418 z_resampler_set(struct pcm_feeder *f, int what, int32_t value)
1419 {
1420 struct z_info *info;
1421 int32_t oquality;
1422
1423 info = f->data;
1424
1425 switch (what) {
1426 case Z_RATE_SRC:
1427 if (value < feeder_rate_min || value > feeder_rate_max)
1428 return (E2BIG);
1429 if (value == info->rsrc)
1430 return (0);
1431 info->rsrc = value;
1432 break;
1433 case Z_RATE_DST:
1434 if (value < feeder_rate_min || value > feeder_rate_max)
1435 return (E2BIG);
1436 if (value == info->rdst)
1437 return (0);
1438 info->rdst = value;
1439 break;
1440 case Z_RATE_QUALITY:
1441 if (value < Z_QUALITY_MIN || value > Z_QUALITY_MAX)
1442 return (EINVAL);
1443 if (value == info->quality)
1444 return (0);
1445 /*
1446 * If we failed to set the requested quality, restore
1447 * the old one. We cannot afford leaving it broken since
1448 * passive feeder chains like vchans never reinitialize
1449 * itself.
1450 */
1451 oquality = info->quality;
1452 info->quality = value;
1453 if (z_resampler_setup(f) == 0)
1454 return (0);
1455 info->quality = oquality;
1456 break;
1457 case Z_RATE_CHANNELS:
1458 if (value < SND_CHN_MIN || value > SND_CHN_MAX)
1459 return (EINVAL);
1460 if (value == info->channels)
1461 return (0);
1462 info->channels = value;
1463 break;
1464 default:
1465 return (EINVAL);
1466 break;
1467 }
1468
1469 return (z_resampler_setup(f));
1470 }
1471
1472 static int
z_resampler_get(struct pcm_feeder * f,int what)1473 z_resampler_get(struct pcm_feeder *f, int what)
1474 {
1475 struct z_info *info;
1476
1477 info = f->data;
1478
1479 switch (what) {
1480 case Z_RATE_SRC:
1481 return (info->rsrc);
1482 break;
1483 case Z_RATE_DST:
1484 return (info->rdst);
1485 break;
1486 case Z_RATE_QUALITY:
1487 return (info->quality);
1488 break;
1489 case Z_RATE_CHANNELS:
1490 return (info->channels);
1491 break;
1492 default:
1493 break;
1494 }
1495
1496 return (-1);
1497 }
1498
1499 static int
z_resampler_init(struct pcm_feeder * f)1500 z_resampler_init(struct pcm_feeder *f)
1501 {
1502 struct z_info *info;
1503 int ret;
1504
1505 if (f->desc->in != f->desc->out)
1506 return (EINVAL);
1507
1508 info = malloc(sizeof(*info), M_DEVBUF, M_NOWAIT | M_ZERO);
1509 if (info == NULL)
1510 return (ENOMEM);
1511
1512 info->rsrc = Z_RATE_DEFAULT;
1513 info->rdst = Z_RATE_DEFAULT;
1514 info->quality = feeder_rate_quality;
1515 info->channels = AFMT_CHANNEL(f->desc->in);
1516
1517 f->data = info;
1518
1519 ret = z_resampler_setup(f);
1520 if (ret != 0) {
1521 if (info->z_pcoeff != NULL)
1522 free(info->z_pcoeff, M_DEVBUF);
1523 if (info->z_delay != NULL)
1524 free(info->z_delay, M_DEVBUF);
1525 free(info, M_DEVBUF);
1526 f->data = NULL;
1527 }
1528
1529 return (ret);
1530 }
1531
1532 static int
z_resampler_free(struct pcm_feeder * f)1533 z_resampler_free(struct pcm_feeder *f)
1534 {
1535 struct z_info *info;
1536
1537 info = f->data;
1538 if (info != NULL) {
1539 if (info->z_pcoeff != NULL)
1540 free(info->z_pcoeff, M_DEVBUF);
1541 if (info->z_delay != NULL)
1542 free(info->z_delay, M_DEVBUF);
1543 free(info, M_DEVBUF);
1544 }
1545
1546 f->data = NULL;
1547
1548 return (0);
1549 }
1550
1551 static uint32_t
z_resampler_feed_internal(struct pcm_feeder * f,struct pcm_channel * c,uint8_t * b,uint32_t count,void * source)1552 z_resampler_feed_internal(struct pcm_feeder *f, struct pcm_channel *c,
1553 uint8_t *b, uint32_t count, void *source)
1554 {
1555 struct z_info *info;
1556 int32_t alphadrift, startdrift, reqout, ocount, reqin, align;
1557 int32_t fetch, fetched, start, cp;
1558 uint8_t *dst;
1559
1560 info = f->data;
1561 if (info->z_resample == NULL)
1562 return (z_feed(f->source, c, b, count, source));
1563
1564 /*
1565 * Calculate sample size alignment and amount of sample output.
1566 * We will do everything in sample domain, but at the end we
1567 * will jump back to byte domain.
1568 */
1569 align = info->channels * info->bps;
1570 ocount = SND_FXDIV(count, align);
1571 if (ocount == 0)
1572 return (0);
1573
1574 /*
1575 * Calculate amount of input samples that is needed to generate
1576 * exact amount of output.
1577 */
1578 reqin = z_gy2gx(info, ocount) - z_fetched(info);
1579
1580 #ifdef Z_USE_ALPHADRIFT
1581 startdrift = info->z_startdrift;
1582 alphadrift = info->z_alphadrift;
1583 #else
1584 startdrift = _Z_GY2GX(info, 0, 1);
1585 alphadrift = z_drift(info, startdrift, 1);
1586 #endif
1587
1588 dst = b;
1589
1590 do {
1591 if (reqin != 0) {
1592 fetch = z_min(z_free(info), reqin);
1593 if (fetch == 0) {
1594 /*
1595 * No more free spaces, so wind enough
1596 * samples back to the head of delay line
1597 * in byte domain.
1598 */
1599 fetched = z_fetched(info);
1600 start = z_prev(info, info->z_start,
1601 (info->z_size << 1) - 1);
1602 cp = (info->z_size << 1) + fetched;
1603 z_copy(info->z_delay + (start * align),
1604 info->z_delay, cp * align);
1605 info->z_start =
1606 z_prev(info, info->z_size << 1, 1);
1607 info->z_pos =
1608 z_next(info, info->z_start, fetched + 1);
1609 fetch = z_min(z_free(info), reqin);
1610 #ifdef Z_DIAGNOSTIC
1611 if (1) {
1612 static uint32_t kk = 0;
1613 fprintf(stderr,
1614 "Buffer Move: "
1615 "start=%d fetched=%d cp=%d "
1616 "cycle=%u [%u]\r",
1617 start, fetched, cp, info->z_cycle,
1618 ++kk);
1619 }
1620 info->z_cycle = 0;
1621 #endif
1622 }
1623 if (fetch != 0) {
1624 /*
1625 * Fetch in byte domain and jump back
1626 * to sample domain.
1627 */
1628 fetched = SND_FXDIV(z_feed(f->source, c,
1629 info->z_delay + (info->z_pos * align),
1630 fetch * align, source), align);
1631 /*
1632 * Prepare to convert fetched buffer,
1633 * or mark us done if we cannot fulfill
1634 * the request.
1635 */
1636 reqin -= fetched;
1637 info->z_pos += fetched;
1638 if (fetched != fetch)
1639 reqin = 0;
1640 }
1641 }
1642
1643 reqout = z_min(z_gx2gy(info, z_fetched(info)), ocount);
1644 if (reqout != 0) {
1645 ocount -= reqout;
1646
1647 /*
1648 * Drift.. drift.. drift..
1649 *
1650 * Notice that there are 2 methods of doing the drift
1651 * operations: The former is much cleaner (in a sense
1652 * of mathematical readings of my eyes), but slower
1653 * due to integer division in z_gy2gx(). Nevertheless,
1654 * both should give the same exact accurate drifting
1655 * results, so the later is favourable.
1656 */
1657 do {
1658 info->z_resample(info, dst);
1659 info->z_alpha += alphadrift;
1660 if (info->z_alpha < info->z_gy)
1661 info->z_start += startdrift;
1662 else {
1663 info->z_start += startdrift - 1;
1664 info->z_alpha -= info->z_gy;
1665 }
1666 dst += align;
1667 #ifdef Z_DIAGNOSTIC
1668 info->z_cycle++;
1669 #endif
1670 } while (--reqout != 0);
1671 }
1672 } while (reqin != 0 && ocount != 0);
1673
1674 /*
1675 * Back to byte domain..
1676 */
1677 return (dst - b);
1678 }
1679
1680 static int
z_resampler_feed(struct pcm_feeder * f,struct pcm_channel * c,uint8_t * b,uint32_t count,void * source)1681 z_resampler_feed(struct pcm_feeder *f, struct pcm_channel *c, uint8_t *b,
1682 uint32_t count, void *source)
1683 {
1684 uint32_t feed, maxfeed, left;
1685
1686 /*
1687 * Split count to smaller chunks to avoid possible 32bit overflow.
1688 */
1689 maxfeed = ((struct z_info *)(f->data))->z_maxfeed;
1690 left = count;
1691
1692 do {
1693 feed = z_resampler_feed_internal(f, c, b,
1694 z_min(maxfeed, left), source);
1695 b += feed;
1696 left -= feed;
1697 } while (left != 0 && feed != 0);
1698
1699 return (count - left);
1700 }
1701
1702 static struct pcm_feederdesc feeder_rate_desc[] = {
1703 { FEEDER_RATE, 0, 0, 0, 0 },
1704 { 0, 0, 0, 0, 0 },
1705 };
1706
1707 static kobj_method_t feeder_rate_methods[] = {
1708 KOBJMETHOD(feeder_init, z_resampler_init),
1709 KOBJMETHOD(feeder_free, z_resampler_free),
1710 KOBJMETHOD(feeder_set, z_resampler_set),
1711 KOBJMETHOD(feeder_get, z_resampler_get),
1712 KOBJMETHOD(feeder_feed, z_resampler_feed),
1713 KOBJMETHOD_END
1714 };
1715
1716 FEEDER_DECLARE(feeder_rate, NULL);
1717