d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
b2019e1789
freebsd-skq/sys/dev/sound/pcm/feeder_rate.c
Ariff Abdullah df41a638f7 - Do aggresive saturation on various polynomial interpolators. 
This dramatically pushing 99.9% interpolations and quantizations
  error _below_ -180dB on 32bit dynamic range, resulting extremely
  high quality conversion.
- Use BSPLINE interpolator for filter oversampling factor greater or
  equal than 64 (log2 6).

Approved by:	re (kib)
2009-07-14 18:53:34 +00:00

1744 lines
48 KiB
C
Raw Blame History

/*-
* Copyright (c) 2005-2009 Ariff Abdullah <ariff@FreeBSD.org>
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
/*
* feeder_rate: (Codename: Z Resampler), which means any effort to create
* future replacement for this resampler are simply absurd unless
* the world decide to add new alphabet after Z.
*
* FreeBSD bandlimited sinc interpolator, technically based on
* "Digital Audio Resampling" by Julius O. Smith III
* - http://ccrma.stanford.edu/~jos/resample/
*
* The Good:
* + all out fixed point integer operations, no soft-float or anything like
* that.
* + classic polyphase converters with high quality coefficient's polynomial
* interpolators.
* + fast, faster, or the fastest of its kind.
* + compile time configurable.
* + etc etc..
*
* The Bad:
* - The z, z_, and Z_ . Due to mental block (or maybe just 0x7a69), I
* couldn't think of anything simpler than that (feeder_rate_xxx is just
* too long). Expect possible clashes with other zitizens (any?).
*/
#ifdef _KERNEL
#ifdef HAVE_KERNEL_OPTION_HEADERS
#include "opt_snd.h"
#endif
#include <dev/sound/pcm/sound.h>
#include <dev/sound/pcm/pcm.h>
#include "feeder_if.h"
#define SND_USE_FXDIV
#include "snd_fxdiv_gen.h"
SND_DECLARE_FILE("$FreeBSD$");
#endif
#include "feeder_rate_gen.h"
#if !defined(_KERNEL) && defined(SND_DIAGNOSTIC)
#undef Z_DIAGNOSTIC
#define Z_DIAGNOSTIC 1
#elif defined(_KERNEL)
#undef Z_DIAGNOSTIC
#endif
#ifndef Z_QUALITY_DEFAULT
#define Z_QUALITY_DEFAULT Z_QUALITY_LINEAR
#endif
#define Z_RESERVOIR 2048
#define Z_RESERVOIR_MAX 131072
#define Z_SINC_MAX 0x3fffff
#define Z_SINC_DOWNMAX 48 /* 384000 / 8000 */
#ifdef _KERNEL
#define Z_POLYPHASE_MAX 183040 /* 286 taps, 640 phases */
#else
#define Z_POLYPHASE_MAX 1464320 /* 286 taps, 5120 phases */
#endif
#define Z_RATE_DEFAULT 48000
#define Z_RATE_MIN FEEDRATE_RATEMIN
#define Z_RATE_MAX FEEDRATE_RATEMAX
#define Z_ROUNDHZ FEEDRATE_ROUNDHZ
#define Z_ROUNDHZ_MIN FEEDRATE_ROUNDHZ_MIN
#define Z_ROUNDHZ_MAX FEEDRATE_ROUNDHZ_MAX
#define Z_RATE_SRC FEEDRATE_SRC
#define Z_RATE_DST FEEDRATE_DST
#define Z_RATE_QUALITY FEEDRATE_QUALITY
#define Z_RATE_CHANNELS FEEDRATE_CHANNELS
#define Z_PARANOID 1
#define Z_MULTIFORMAT 1
#ifdef _KERNEL
#undef Z_USE_ALPHADRIFT
#define Z_USE_ALPHADRIFT 1
#endif
#define Z_FACTOR_MIN 1
#define Z_FACTOR_MAX Z_MASK
#define Z_FACTOR_SAFE(v) (!((v) < Z_FACTOR_MIN || (v) > Z_FACTOR_MAX))
struct z_info;
typedef void (*z_resampler_t)(struct z_info *, uint8_t *);
struct z_info {
int32_t rsrc, rdst; /* original source / destination rates */
int32_t src, dst; /* rounded source / destination rates */
int32_t channels; /* total channels */
int32_t bps; /* bytes-per-sample */
int32_t quality; /* resampling quality */
int32_t z_gx, z_gy; /* interpolation / decimation ratio */
int32_t z_alpha; /* output sample time phase / drift */
uint8_t *z_delay; /* FIR delay line / linear buffer */
int32_t *z_coeff; /* FIR coefficients */
int32_t *z_dcoeff; /* FIR coefficients differences */
int32_t *z_pcoeff; /* FIR polyphase coefficients */
int32_t z_scale; /* output scaling */
int32_t z_dx; /* input sample drift increment */
int32_t z_dy; /* output sample drift increment */
#ifdef Z_USE_ALPHADRIFT
int32_t z_alphadrift; /* alpha drift rate */
int32_t z_startdrift; /* buffer start position drift rate */
#endif
int32_t z_mask; /* delay line full length mask */
int32_t z_size; /* half width of FIR taps */
int32_t z_full; /* full size of delay line */
int32_t z_alloc; /* largest allocated full size of delay line */
int32_t z_start; /* buffer processing start position */
int32_t z_pos; /* current position for the next feed */
#ifdef Z_DIAGNOSTIC
uint32_t z_cycle; /* output cycle, purely for statistical */
#endif
int32_t z_maxfeed; /* maximum feed to avoid 32bit overflow */
z_resampler_t z_resample;
};
int feeder_rate_min = Z_RATE_MIN;
int feeder_rate_max = Z_RATE_MAX;
int feeder_rate_round = Z_ROUNDHZ;
int feeder_rate_quality = Z_QUALITY_DEFAULT;
static int feeder_rate_polyphase_max = Z_POLYPHASE_MAX;
#ifdef _KERNEL
static const char feeder_rate_presets[] = FEEDER_RATE_PRESETS;
SYSCTL_STRING(_hw_snd, OID_AUTO, feeder_rate_presets, CTLFLAG_RD,
&feeder_rate_presets, 0, "compile-time rate presets");
TUNABLE_INT("hw.snd.feeder_rate_min", &feeder_rate_min);
TUNABLE_INT("hw.snd.feeder_rate_max", &feeder_rate_max);
TUNABLE_INT("hw.snd.feeder_rate_round", &feeder_rate_round);
TUNABLE_INT("hw.snd.feeder_rate_quality", &feeder_rate_quality);
TUNABLE_INT("hw.snd.feeder_rate_polyphase_max", &feeder_rate_polyphase_max);
SYSCTL_INT(_hw_snd, OID_AUTO, feeder_rate_polyphase_max, CTLFLAG_RW,
&feeder_rate_polyphase_max, 0, "maximum allowable polyphase entries");
static int
sysctl_hw_snd_feeder_rate_min(SYSCTL_HANDLER_ARGS)
{
int err, val;
val = feeder_rate_min;
err = sysctl_handle_int(oidp, &val, 0, req);
if (err != 0 || req->newptr == NULL || val == feeder_rate_min)
return (err);
if (!(Z_FACTOR_SAFE(val) && val < feeder_rate_max))
return (EINVAL);
feeder_rate_min = val;
return (0);
}
SYSCTL_PROC(_hw_snd, OID_AUTO, feeder_rate_min, CTLTYPE_INT | CTLFLAG_RW,
0, sizeof(int), sysctl_hw_snd_feeder_rate_min, "I",
"minimum allowable rate");
static int
sysctl_hw_snd_feeder_rate_max(SYSCTL_HANDLER_ARGS)
{
int err, val;
val = feeder_rate_max;
err = sysctl_handle_int(oidp, &val, 0, req);
if (err != 0 || req->newptr == NULL || val == feeder_rate_max)
return (err);
if (!(Z_FACTOR_SAFE(val) && val > feeder_rate_min))
return (EINVAL);
feeder_rate_max = val;
return (0);
}
SYSCTL_PROC(_hw_snd, OID_AUTO, feeder_rate_max, CTLTYPE_INT | CTLFLAG_RW,
0, sizeof(int), sysctl_hw_snd_feeder_rate_max, "I",
"maximum allowable rate");
static int
sysctl_hw_snd_feeder_rate_round(SYSCTL_HANDLER_ARGS)
{
int err, val;
val = feeder_rate_round;
err = sysctl_handle_int(oidp, &val, 0, req);
if (err != 0 || req->newptr == NULL || val == feeder_rate_round)
return (err);
if (val < Z_ROUNDHZ_MIN || val > Z_ROUNDHZ_MAX)
return (EINVAL);
feeder_rate_round = val - (val % Z_ROUNDHZ);
return (0);
}
SYSCTL_PROC(_hw_snd, OID_AUTO, feeder_rate_round, CTLTYPE_INT | CTLFLAG_RW,
0, sizeof(int), sysctl_hw_snd_feeder_rate_round, "I",
"sample rate converter rounding threshold");
static int
sysctl_hw_snd_feeder_rate_quality(SYSCTL_HANDLER_ARGS)
{
struct snddev_info *d;
struct pcm_channel *c;
struct pcm_feeder *f;
int i, err, val;
val = feeder_rate_quality;
err = sysctl_handle_int(oidp, &val, 0, req);
if (err != 0 || req->newptr == NULL || val == feeder_rate_quality)
return (err);
if (val < Z_QUALITY_MIN || val > Z_QUALITY_MAX)
return (EINVAL);
feeder_rate_quality = val;
/*
* Traverse all available channels on each device and try to
* set resampler quality if and only if it is exist as
* part of feeder chains and the channel is idle.
*/
for (i = 0; pcm_devclass != NULL &&
i < devclass_get_maxunit(pcm_devclass); i++) {
d = devclass_get_softc(pcm_devclass, i);
if (!PCM_REGISTERED(d))
continue;
PCM_LOCK(d);
PCM_WAIT(d);
PCM_ACQUIRE(d);
CHN_FOREACH(c, d, channels.pcm) {
CHN_LOCK(c);
f = chn_findfeeder(c, FEEDER_RATE);
if (f == NULL || f->data == NULL || CHN_STARTED(c)) {
CHN_UNLOCK(c);
continue;
}
(void)FEEDER_SET(f, FEEDRATE_QUALITY, val);
CHN_UNLOCK(c);
}
PCM_RELEASE(d);
PCM_UNLOCK(d);
}
return (0);
}
SYSCTL_PROC(_hw_snd, OID_AUTO, feeder_rate_quality, CTLTYPE_INT | CTLFLAG_RW,
0, sizeof(int), sysctl_hw_snd_feeder_rate_quality, "I",
"sample rate converter quality ("__XSTRING(Z_QUALITY_MIN)"=low .. "
__XSTRING(Z_QUALITY_MAX)"=high)");
#endif /* _KERNEL */
/*
* Resampler type.
*/
#define Z_IS_ZOH(i) ((i)->quality == Z_QUALITY_ZOH)
#define Z_IS_LINEAR(i) ((i)->quality == Z_QUALITY_LINEAR)
#define Z_IS_SINC(i) ((i)->quality > Z_QUALITY_LINEAR)
/*
* Macroses for accurate sample time drift calculations.
*
* gy2gx : given the amount of output, return the _exact_ required amount of
* input.
* gx2gy : given the amount of input, return the _maximum_ amount of output
* that will be generated.
* drift : given the amount of input and output, return the elapsed
* sample-time.
*/
#define _Z_GCAST(x) ((uint64_t)(x))
#if defined(__GNUCLIKE_ASM) && defined(__i386__)
/*
* This is where i386 being beaten to a pulp. Fortunately this function is
* rarely being called and if it is, it will decide the best (hopefully)
* fastest way to do the division. If we can ensure that everything is dword
* aligned, letting the compiler to call udivdi3 to do the division can be
* faster compared to this.
*
* amd64 is the clear winner here, no question about it.
*/
static __inline uint32_t
Z_DIV(uint64_t v, uint32_t d)
{
uint32_t hi, lo, quo, rem;
hi = v >> 32;
lo = v & 0xffffffff;
/*
* As much as we can, try to avoid long division like a plague.
*/
if (hi == 0)
quo = lo / d;
else
__asm("divl %2"
: "=a" (quo), "=d" (rem)
: "r" (d), "0" (lo), "1" (hi));
return (quo);
}
#else
#define Z_DIV(x, y) ((x) / (y))
#endif
#define _Z_GY2GX(i, a, v) \
Z_DIV(((_Z_GCAST((i)->z_gx) * (v)) + ((i)->z_gy - (a) - 1)), \
(i)->z_gy)
#define _Z_GX2GY(i, a, v) \
Z_DIV(((_Z_GCAST((i)->z_gy) * (v)) + (a)), (i)->z_gx)
#define _Z_DRIFT(i, x, y) \
((_Z_GCAST((i)->z_gy) * (x)) - (_Z_GCAST((i)->z_gx) * (y)))
#define z_gy2gx(i, v) _Z_GY2GX(i, (i)->z_alpha, v)
#define z_gx2gy(i, v) _Z_GX2GY(i, (i)->z_alpha, v)
#define z_drift(i, x, y) _Z_DRIFT(i, x, y)
/*
* Macroses for SINC coefficients table manipulations.. whatever.
*/
#define Z_SINC_COEFF_IDX(i) ((i)->quality - Z_QUALITY_LINEAR - 1)
#define Z_SINC_LEN(i) \
((int32_t)(((uint64_t)z_coeff_tab[Z_SINC_COEFF_IDX(i)].len << \
Z_SHIFT) / (i)->z_dy))
#define Z_SINC_BASE_LEN(i) \
((z_coeff_tab[Z_SINC_COEFF_IDX(i)].len - 1) >> (Z_DRIFT_SHIFT - 1))
/*
* Macroses for linear delay buffer operations. Alignment is not
* really necessary since we're not using true circular buffer, but it
* will help us guard against possible trespasser. To be honest,
* the linear block operations does not need guarding at all due to
* accurate drifting!
*/
#define z_align(i, v) ((v) & (i)->z_mask)
#define z_next(i, o, v) z_align(i, (o) + (v))
#define z_prev(i, o, v) z_align(i, (o) - (v))
#define z_fetched(i) (z_align(i, (i)->z_pos - (i)->z_start) - 1)
#define z_free(i) ((i)->z_full - (i)->z_pos)
/*
* Macroses for Bla Bla .. :)
*/
#define z_copy(src, dst, sz) (void)memcpy(dst, src, sz)
#define z_feed(...) FEEDER_FEED(__VA_ARGS__)
static __inline uint32_t
z_min(uint32_t x, uint32_t y)
{
return ((x < y) ? x : y);
}
static int32_t
z_gcd(int32_t x, int32_t y)
{
int32_t w;
while (y != 0) {
w = x % y;
x = y;
y = w;
}
return (x);
}
static int32_t
z_roundpow2(int32_t v)
{
int32_t i;
i = 1;
/*
* Let it overflow at will..
*/
while (i > 0 && i < v)
i <<= 1;
return (i);
}
/*
* Zero Order Hold, the worst of the worst, an insult against quality,
* but super fast.
*/
static void
z_feed_zoh(struct z_info *info, uint8_t *dst)
{
#if 0
z_copy(info->z_delay +
(info->z_start * info->channels * info->bps), dst,
info->channels * info->bps);
#else
uint32_t cnt;
uint8_t *src;
cnt = info->channels * info->bps;
src = info->z_delay + (info->z_start * cnt);
/*
* This is a bit faster than doing bcopy() since we're dealing
* with possible unaligned samples.
*/
do {
*dst++ = *src++;
} while (--cnt != 0);
#endif
}
/*
* Linear Interpolation. This at least sounds better (perceptually) and fast,
* but without any proper filtering which means aliasing still exist and
* could become worst with a right sample. Interpolation centered within
* Z_LINEAR_ONE between the present and previous sample and everything is
* done with simple 32bit scaling arithmetic.
*/
#define Z_DECLARE_LINEAR(SIGN, BIT, ENDIAN) \
static void \
z_feed_linear_##SIGN##BIT##ENDIAN(struct z_info *info, uint8_t *dst) \
{ \
int32_t z; \
intpcm_t x, y; \
uint32_t ch; \
uint8_t *sx, *sy; \
\
z = ((uint32_t)info->z_alpha * info->z_dx) >> Z_LINEAR_UNSHIFT; \
\
sx = info->z_delay + (info->z_start * info->channels * \
PCM_##BIT##_BPS); \
sy = sx - (info->channels * PCM_##BIT##_BPS); \
\
ch = info->channels; \
\
do { \
x = _PCM_READ_##SIGN##BIT##_##ENDIAN(sx); \
y = _PCM_READ_##SIGN##BIT##_##ENDIAN(sy); \
x = Z_LINEAR_INTERPOLATE_##BIT(z, x, y); \
_PCM_WRITE_##SIGN##BIT##_##ENDIAN(dst, x); \
sx += PCM_##BIT##_BPS; \
sy += PCM_##BIT##_BPS; \
dst += PCM_##BIT##_BPS; \
} while (--ch != 0); \
}
/*
* Userland clipping diagnostic check, not enabled in kernel compilation.
* While doing sinc interpolation, unrealistic samples like full scale sine
* wav will clip, but for other things this will not make any noise at all.
* Everybody should learn how to normalized perceived loudness of their own
* music/sounds/samples (hint: ReplayGain).
*/
#ifdef Z_DIAGNOSTIC
#define Z_CLIP_CHECK(v, BIT) do { \
if ((v) > PCM_S##BIT##_MAX) { \
fprintf(stderr, "Overflow: v=%jd, max=%jd\n", \
(intmax_t)(v), (intmax_t)PCM_S##BIT##_MAX); \
} else if ((v) < PCM_S##BIT##_MIN) { \
fprintf(stderr, "Underflow: v=%jd, min=%jd\n", \
(intmax_t)(v), (intmax_t)PCM_S##BIT##_MIN); \
} \
} while (0)
#else
#define Z_CLIP_CHECK(...)
#endif
#define Z_CLAMP(v, BIT) \
(((v) > PCM_S##BIT##_MAX) ? PCM_S##BIT##_MAX : \
(((v) < PCM_S##BIT##_MIN) ? PCM_S##BIT##_MIN : (v)))
/*
* Sine Cardinal (SINC) Interpolation. Scaling is done in 64 bit, so
* there's no point to hold the plate any longer. All samples will be
* shifted to a full 32 bit, scaled and restored during write for
* maximum dynamic range (only for downsampling).
*/
#define _Z_SINC_ACCUMULATE(SIGN, BIT, ENDIAN, adv) \
c += z >> Z_SHIFT; \
z &= Z_MASK; \
coeff = Z_COEFF_INTERPOLATE(z, z_coeff[c], z_dcoeff[c]); \
x = _PCM_READ_##SIGN##BIT##_##ENDIAN(p); \
v += Z_NORM_##BIT((intpcm64_t)x * coeff); \
z += info->z_dy; \
p adv##= info->channels * PCM_##BIT##_BPS
/*
* XXX GCC4 optimization is such a !@#$%, need manual unrolling.
*/
#if defined(__GNUC__) && __GNUC__ >= 4
#define Z_SINC_ACCUMULATE(...) do { \
_Z_SINC_ACCUMULATE(__VA_ARGS__); \
_Z_SINC_ACCUMULATE(__VA_ARGS__); \
} while (0)
#define Z_SINC_ACCUMULATE_DECR 2
#else
#define Z_SINC_ACCUMULATE(...) do { \
_Z_SINC_ACCUMULATE(__VA_ARGS__); \
} while (0)
#define Z_SINC_ACCUMULATE_DECR 1
#endif
#define Z_DECLARE_SINC(SIGN, BIT, ENDIAN) \
static void \
z_feed_sinc_##SIGN##BIT##ENDIAN(struct z_info *info, uint8_t *dst) \
{ \
intpcm64_t v; \
intpcm_t x; \
uint8_t *p; \
int32_t coeff, z, *z_coeff, *z_dcoeff; \
uint32_t c, center, ch, i; \
\
z_coeff = info->z_coeff; \
z_dcoeff = info->z_dcoeff; \
center = z_prev(info, info->z_start, info->z_size); \
ch = info->channels * PCM_##BIT##_BPS; \
dst += ch; \
\
do { \
dst -= PCM_##BIT##_BPS; \
ch -= PCM_##BIT##_BPS; \
v = 0; \
z = info->z_alpha * info->z_dx; \
c = 0; \
p = info->z_delay + (z_next(info, center, 1) * \
info->channels * PCM_##BIT##_BPS) + ch; \
for (i = info->z_size; i != 0; i -= Z_SINC_ACCUMULATE_DECR) \
Z_SINC_ACCUMULATE(SIGN, BIT, ENDIAN, +); \
z = info->z_dy - (info->z_alpha * info->z_dx); \
c = 0; \
p = info->z_delay + (center * info->channels * \
PCM_##BIT##_BPS) + ch; \
for (i = info->z_size; i != 0; i -= Z_SINC_ACCUMULATE_DECR) \
Z_SINC_ACCUMULATE(SIGN, BIT, ENDIAN, -); \
if (info->z_scale != Z_ONE) \
v = Z_SCALE_##BIT(v, info->z_scale); \
else \
v >>= Z_COEFF_SHIFT - Z_GUARD_BIT_##BIT; \
Z_CLIP_CHECK(v, BIT); \
_PCM_WRITE_##SIGN##BIT##_##ENDIAN(dst, Z_CLAMP(v, BIT)); \
} while (ch != 0); \
}
#define Z_DECLARE_SINC_POLYPHASE(SIGN, BIT, ENDIAN) \
static void \
z_feed_sinc_polyphase_##SIGN##BIT##ENDIAN(struct z_info *info, uint8_t *dst) \
{ \
intpcm64_t v; \
intpcm_t x; \
uint8_t *p; \
int32_t ch, i, start, *z_pcoeff; \
\
ch = info->channels * PCM_##BIT##_BPS; \
dst += ch; \
start = z_prev(info, info->z_start, (info->z_size << 1) - 1) * ch; \
\
do { \
dst -= PCM_##BIT##_BPS; \
ch -= PCM_##BIT##_BPS; \
v = 0; \
p = info->z_delay + start + ch; \
z_pcoeff = info->z_pcoeff + \
((info->z_alpha * info->z_size) << 1); \
for (i = info->z_size; i != 0; i--) { \
x = _PCM_READ_##SIGN##BIT##_##ENDIAN(p); \
v += Z_NORM_##BIT((intpcm64_t)x * *z_pcoeff); \
z_pcoeff++; \
p += info->channels * PCM_##BIT##_BPS; \
x = _PCM_READ_##SIGN##BIT##_##ENDIAN(p); \
v += Z_NORM_##BIT((intpcm64_t)x * *z_pcoeff); \
z_pcoeff++; \
p += info->channels * PCM_##BIT##_BPS; \
} \
if (info->z_scale != Z_ONE) \
v = Z_SCALE_##BIT(v, info->z_scale); \
else \
v >>= Z_COEFF_SHIFT - Z_GUARD_BIT_##BIT; \
Z_CLIP_CHECK(v, BIT); \
_PCM_WRITE_##SIGN##BIT##_##ENDIAN(dst, Z_CLAMP(v, BIT)); \
} while (ch != 0); \
}
#define Z_DECLARE(SIGN, BIT, ENDIAN) \
Z_DECLARE_LINEAR(SIGN, BIT, ENDIAN) \
Z_DECLARE_SINC(SIGN, BIT, ENDIAN) \
Z_DECLARE_SINC_POLYPHASE(SIGN, BIT, ENDIAN)
#if BYTE_ORDER == LITTLE_ENDIAN || defined(SND_FEEDER_MULTIFORMAT)
Z_DECLARE(S, 16, LE)
Z_DECLARE(S, 32, LE)
#endif
#if BYTE_ORDER == BIG_ENDIAN || defined(SND_FEEDER_MULTIFORMAT)
Z_DECLARE(S, 16, BE)
Z_DECLARE(S, 32, BE)
#endif
#ifdef SND_FEEDER_MULTIFORMAT
Z_DECLARE(S, 8, NE)
Z_DECLARE(S, 24, LE)
Z_DECLARE(S, 24, BE)
Z_DECLARE(U, 8, NE)
Z_DECLARE(U, 16, LE)
Z_DECLARE(U, 24, LE)
Z_DECLARE(U, 32, LE)
Z_DECLARE(U, 16, BE)
Z_DECLARE(U, 24, BE)
Z_DECLARE(U, 32, BE)
#endif
enum {
Z_RESAMPLER_ZOH,
Z_RESAMPLER_LINEAR,
Z_RESAMPLER_SINC,
Z_RESAMPLER_SINC_POLYPHASE,
Z_RESAMPLER_LAST
};
#define Z_RESAMPLER_IDX(i) \
(Z_IS_SINC(i) ? Z_RESAMPLER_SINC : (i)->quality)
#define Z_RESAMPLER_ENTRY(SIGN, BIT, ENDIAN) \
{ \
AFMT_##SIGN##BIT##_##ENDIAN, \
{ \
[Z_RESAMPLER_ZOH] = z_feed_zoh, \
[Z_RESAMPLER_LINEAR] = z_feed_linear_##SIGN##BIT##ENDIAN, \
[Z_RESAMPLER_SINC] = z_feed_sinc_##SIGN##BIT##ENDIAN, \
[Z_RESAMPLER_SINC_POLYPHASE] = \
z_feed_sinc_polyphase_##SIGN##BIT##ENDIAN \
} \
}
static const struct {
uint32_t format;
z_resampler_t resampler[Z_RESAMPLER_LAST];
} z_resampler_tab[] = {
#if BYTE_ORDER == LITTLE_ENDIAN || defined(SND_FEEDER_MULTIFORMAT)
Z_RESAMPLER_ENTRY(S, 16, LE),
Z_RESAMPLER_ENTRY(S, 32, LE),
#endif
#if BYTE_ORDER == BIG_ENDIAN || defined(SND_FEEDER_MULTIFORMAT)
Z_RESAMPLER_ENTRY(S, 16, BE),
Z_RESAMPLER_ENTRY(S, 32, BE),
#endif
#ifdef SND_FEEDER_MULTIFORMAT
Z_RESAMPLER_ENTRY(S, 8, NE),
Z_RESAMPLER_ENTRY(S, 24, LE),
Z_RESAMPLER_ENTRY(S, 24, BE),
Z_RESAMPLER_ENTRY(U, 8, NE),
Z_RESAMPLER_ENTRY(U, 16, LE),
Z_RESAMPLER_ENTRY(U, 24, LE),
Z_RESAMPLER_ENTRY(U, 32, LE),
Z_RESAMPLER_ENTRY(U, 16, BE),
Z_RESAMPLER_ENTRY(U, 24, BE),
Z_RESAMPLER_ENTRY(U, 32, BE),
#endif
};
#define Z_RESAMPLER_TAB_SIZE \
((int32_t)(sizeof(z_resampler_tab) / sizeof(z_resampler_tab[0])))
static void
z_resampler_reset(struct z_info *info)
{
info->src = info->rsrc - (info->rsrc % ((feeder_rate_round > 0 &&
info->rsrc > feeder_rate_round) ? feeder_rate_round : 1));
info->dst = info->rdst - (info->rdst % ((feeder_rate_round > 0 &&
info->rdst > feeder_rate_round) ? feeder_rate_round : 1));
info->z_gx = 1;
info->z_gy = 1;
info->z_alpha = 0;
info->z_resample = NULL;
info->z_size = 1;
info->z_coeff = NULL;
info->z_dcoeff = NULL;
if (info->z_pcoeff != NULL) {
free(info->z_pcoeff, M_DEVBUF);
info->z_pcoeff = NULL;
}
info->z_scale = Z_ONE;
info->z_dx = Z_FULL_ONE;
info->z_dy = Z_FULL_ONE;
#ifdef Z_DIAGNOSTIC
info->z_cycle = 0;
#endif
if (info->quality < Z_QUALITY_MIN)
info->quality = Z_QUALITY_MIN;
else if (info->quality > Z_QUALITY_MAX)
info->quality = Z_QUALITY_MAX;
}
#ifdef Z_PARANOID
static int32_t
z_resampler_sinc_len(struct z_info *info)
{
int32_t c, z, len, lmax;
if (!Z_IS_SINC(info))
return (1);
/*
* A rather careful (or useless) way to calculate filter length.
* Z_SINC_LEN() itself is accurate enough to do its job. Extra
* sanity checking is not going to hurt though..
*/
c = 0;
z = info->z_dy;
len = 0;
lmax = z_coeff_tab[Z_SINC_COEFF_IDX(info)].len;
do {
c += z >> Z_SHIFT;
z &= Z_MASK;
z += info->z_dy;
} while (c < lmax && ++len > 0);
if (len != Z_SINC_LEN(info)) {
#ifdef _KERNEL
printf("%s(): sinc l=%d != Z_SINC_LEN=%d\n",
__func__, len, Z_SINC_LEN(info));
#else
fprintf(stderr, "%s(): sinc l=%d != Z_SINC_LEN=%d\n",
__func__, len, Z_SINC_LEN(info));
return (-1);
#endif
}
return (len);
}
#else
#define z_resampler_sinc_len(i) (Z_IS_SINC(i) ? Z_SINC_LEN(i) : 1)
#endif
#define Z_POLYPHASE_COEFF_SHIFT 0
/*
* Pick suitable polynomial interpolators based on filter oversampled ratio
* (2 ^ Z_DRIFT_SHIFT).
*/
#if !(defined(Z_COEFF_INTERP_ZOH) || defined(Z_COEFF_INTERP_LINEAR) || \
defined(Z_COEFF_INTERP_QUADRATIC) || defined(Z_COEFF_INTERP_HERMITE) || \
defined(Z_COEFF_INTER_BSPLINE) || defined(Z_COEFF_INTERP_OPT32X) || \
defined(Z_COEFF_INTERP_OPT16X) || defined(Z_COEFF_INTERP_OPT8X) || \
defined(Z_COEFF_INTERP_OPT4X) || defined(Z_COEFF_INTERP_OPT2X))
#if Z_DRIFT_SHIFT >= 6
#define Z_COEFF_INTERP_BSPLINE 1
#elif Z_DRIFT_SHIFT >= 5
#define Z_COEFF_INTERP_OPT32X 1
#elif Z_DRIFT_SHIFT == 4
#define Z_COEFF_INTERP_OPT16X 1
#elif Z_DRIFT_SHIFT == 3
#define Z_COEFF_INTERP_OPT8X 1
#elif Z_DRIFT_SHIFT == 2
#define Z_COEFF_INTERP_OPT4X 1
#elif Z_DRIFT_SHIFT == 1
#define Z_COEFF_INTERP_OPT2X 1
#else
#error "Z_DRIFT_SHIFT screwed!"
#endif
#endif
/*
* In classic polyphase mode, the actual coefficients for each phases need to
* be calculated based on default prototype filters. For highly oversampled
* filter, linear or quadradatic interpolator should be enough. Anything less
* than that require 'special' interpolators to reduce interpolation errors.
*
* "Polynomial Interpolators for High-Quality Resampling of Oversampled Audio"
* by Olli Niemitalo
* - http://www.student.oulu.fi/~oniemita/dsp/deip.pdf
*
*/
static int32_t
z_coeff_interpolate(int32_t z, int32_t *z_coeff)
{
int32_t coeff;
#if defined(Z_COEFF_INTERP_ZOH)
/* 1-point, 0th-order (Zero Order Hold) */
z = z;
coeff = z_coeff[0];
#elif defined(Z_COEFF_INTERP_LINEAR)
int32_t zl0, zl1;
/* 2-point, 1st-order Linear */
zl0 = z_coeff[0];
zl1 = z_coeff[1] - z_coeff[0];
coeff = Z_RSHIFT((int64_t)zl1 * z, Z_SHIFT) + zl0;
#elif defined(Z_COEFF_INTERP_QUADRATIC)
int32_t zq0, zq1, zq2;
/* 3-point, 2nd-order Quadratic */
zq0 = z_coeff[0];
zq1 = z_coeff[1] - z_coeff[-1];
zq2 = z_coeff[1] + z_coeff[-1] - (z_coeff[0] << 1);
coeff = Z_RSHIFT((Z_RSHIFT((int64_t)zq2 * z, Z_SHIFT) +
zq1) * z, Z_SHIFT + 1) + zq0;
#elif defined(Z_COEFF_INTERP_HERMITE)
int32_t zh0, zh1, zh2, zh3;
/* 4-point, 3rd-order Hermite */
zh0 = z_coeff[0];
zh1 = z_coeff[1] - z_coeff[-1];
zh2 = (z_coeff[-1] << 1) - (z_coeff[0] * 5) + (z_coeff[1] << 2) -
z_coeff[2];
zh3 = z_coeff[2] - z_coeff[-1] + ((z_coeff[0] - z_coeff[1]) * 3);
coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((int64_t)zh3 * z, Z_SHIFT) +
zh2) * z, Z_SHIFT) + zh1) * z, Z_SHIFT + 1) + zh0;
#elif defined(Z_COEFF_INTERP_BSPLINE)
int32_t zb0, zb1, zb2, zb3;
/* 4-point, 3rd-order B-Spline */
zb0 = Z_RSHIFT(0x15555555LL * (((int64_t)z_coeff[0] << 2) +
z_coeff[-1] + z_coeff[1]), 30);
zb1 = z_coeff[1] - z_coeff[-1];
zb2 = z_coeff[-1] + z_coeff[1] - (z_coeff[0] << 1);
zb3 = Z_RSHIFT(0x15555555LL * (((z_coeff[0] - z_coeff[1]) * 3) +
z_coeff[2] - z_coeff[-1]), 30);
coeff = (Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((int64_t)zb3 * z, Z_SHIFT) +
zb2) * z, Z_SHIFT) + zb1) * z, Z_SHIFT) + zb0 + 1) >> 1;
#elif defined(Z_COEFF_INTERP_OPT32X)
int32_t zoz, zoe1, zoe2, zoe3, zoo1, zoo2, zoo3;
int32_t zoc0, zoc1, zoc2, zoc3, zoc4, zoc5;
/* 6-point, 5th-order Optimal 32x */
zoz = z - (Z_ONE >> 1);
zoe1 = z_coeff[1] + z_coeff[0];
zoe2 = z_coeff[2] + z_coeff[-1];
zoe3 = z_coeff[3] + z_coeff[-2];
zoo1 = z_coeff[1] - z_coeff[0];
zoo2 = z_coeff[2] - z_coeff[-1];
zoo3 = z_coeff[3] - z_coeff[-2];
zoc0 = Z_RSHIFT((0x1ac2260dLL * zoe1) + (0x0526cdcaLL * zoe2) +
(0x00170c29LL * zoe3), 30);
zoc1 = Z_RSHIFT((0x14f8a49aLL * zoo1) + (0x0d6d1109LL * zoo2) +
(0x008cd4dcLL * zoo3), 30);
zoc2 = Z_RSHIFT((-0x0d3e94a4LL * zoe1) + (0x0bddded4LL * zoe2) +
(0x0160b5d0LL * zoe3), 30);
zoc3 = Z_RSHIFT((-0x0de10cc4LL * zoo1) + (0x019b2a7dLL * zoo2) +
(0x01cfe914LL * zoo3), 30);
zoc4 = Z_RSHIFT((0x02aa12d7LL * zoe1) + (-0x03ff1bb3LL * zoe2) +
(0x015508ddLL * zoe3), 30);
zoc5 = Z_RSHIFT((0x051d29e5LL * zoo1) + (-0x028e7647LL * zoo2) +
(0x0082d81aLL * zoo3), 30);
coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT(
(int64_t)zoc5 * zoz, Z_SHIFT) +
zoc4) * zoz, Z_SHIFT) + zoc3) * zoz, Z_SHIFT) +
zoc2) * zoz, Z_SHIFT) + zoc1) * zoz, Z_SHIFT) + zoc0;
#elif defined(Z_COEFF_INTERP_OPT16X)
int32_t zoz, zoe1, zoe2, zoe3, zoo1, zoo2, zoo3;
int32_t zoc0, zoc1, zoc2, zoc3, zoc4, zoc5;
/* 6-point, 5th-order Optimal 16x */
zoz = z - (Z_ONE >> 1);
zoe1 = z_coeff[1] + z_coeff[0];
zoe2 = z_coeff[2] + z_coeff[-1];
zoe3 = z_coeff[3] + z_coeff[-2];
zoo1 = z_coeff[1] - z_coeff[0];
zoo2 = z_coeff[2] - z_coeff[-1];
zoo3 = z_coeff[3] - z_coeff[-2];
zoc0 = Z_RSHIFT((0x1ac2260dLL * zoe1) + (0x0526cdcaLL * zoe2) +
(0x00170c29LL * zoe3), 30);
zoc1 = Z_RSHIFT((0x14f8a49aLL * zoo1) + (0x0d6d1109LL * zoo2) +
(0x008cd4dcLL * zoo3), 30);
zoc2 = Z_RSHIFT((-0x0d3e94a4LL * zoe1) + (0x0bddded4LL * zoe2) +
(0x0160b5d0LL * zoe3), 30);
zoc3 = Z_RSHIFT((-0x0de10cc4LL * zoo1) + (0x019b2a7dLL * zoo2) +
(0x01cfe914LL * zoo3), 30);
zoc4 = Z_RSHIFT((0x02aa12d7LL * zoe1) + (-0x03ff1bb3LL * zoe2) +
(0x015508ddLL * zoe3), 30);
zoc5 = Z_RSHIFT((0x051d29e5LL * zoo1) + (-0x028e7647LL * zoo2) +
(0x0082d81aLL * zoo3), 30);
coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT(
(int64_t)zoc5 * zoz, Z_SHIFT) +
zoc4) * zoz, Z_SHIFT) + zoc3) * zoz, Z_SHIFT) +
zoc2) * zoz, Z_SHIFT) + zoc1) * zoz, Z_SHIFT) + zoc0;
#elif defined(Z_COEFF_INTERP_OPT8X)
int32_t zoz, zoe1, zoe2, zoe3, zoo1, zoo2, zoo3;
int32_t zoc0, zoc1, zoc2, zoc3, zoc4, zoc5;
/* 6-point, 5th-order Optimal 8x */
zoz = z - (Z_ONE >> 1);
zoe1 = z_coeff[1] + z_coeff[0];
zoe2 = z_coeff[2] + z_coeff[-1];
zoe3 = z_coeff[3] + z_coeff[-2];
zoo1 = z_coeff[1] - z_coeff[0];
zoo2 = z_coeff[2] - z_coeff[-1];
zoo3 = z_coeff[3] - z_coeff[-2];
zoc0 = Z_RSHIFT((0x1aa9b47dLL * zoe1) + (0x053d9944LL * zoe2) +
(0x0018b23fLL * zoe3), 30);
zoc1 = Z_RSHIFT((0x14a104d1LL * zoo1) + (0x0d7d2504LL * zoo2) +
(0x0094b599LL * zoo3), 30);
zoc2 = Z_RSHIFT((-0x0d22530bLL * zoe1) + (0x0bb37a2cLL * zoe2) +
(0x016ed8e0LL * zoe3), 30);
zoc3 = Z_RSHIFT((-0x0d744b1cLL * zoo1) + (0x01649591LL * zoo2) +
(0x01dae93aLL * zoo3), 30);
zoc4 = Z_RSHIFT((0x02a7ee1bLL * zoe1) + (-0x03fbdb24LL * zoe2) +
(0x0153ed07LL * zoe3), 30);
zoc5 = Z_RSHIFT((0x04cf9b6cLL * zoo1) + (-0x0266b378LL * zoo2) +
(0x007a7c26LL * zoo3), 30);
coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT(
(int64_t)zoc5 * zoz, Z_SHIFT) +
zoc4) * zoz, Z_SHIFT) + zoc3) * zoz, Z_SHIFT) +
zoc2) * zoz, Z_SHIFT) + zoc1) * zoz, Z_SHIFT) + zoc0;
#elif defined(Z_COEFF_INTERP_OPT4X)
int32_t zoz, zoe1, zoe2, zoe3, zoo1, zoo2, zoo3;
int32_t zoc0, zoc1, zoc2, zoc3, zoc4, zoc5;
/* 6-point, 5th-order Optimal 4x */
zoz = z - (Z_ONE >> 1);
zoe1 = z_coeff[1] + z_coeff[0];
zoe2 = z_coeff[2] + z_coeff[-1];
zoe3 = z_coeff[3] + z_coeff[-2];
zoo1 = z_coeff[1] - z_coeff[0];
zoo2 = z_coeff[2] - z_coeff[-1];
zoo3 = z_coeff[3] - z_coeff[-2];
zoc0 = Z_RSHIFT((0x1a8eda43LL * zoe1) + (0x0556ee38LL * zoe2) +
(0x001a3784LL * zoe3), 30);
zoc1 = Z_RSHIFT((0x143d863eLL * zoo1) + (0x0d910e36LL * zoo2) +
(0x009ca889LL * zoo3), 30);
zoc2 = Z_RSHIFT((-0x0d026821LL * zoe1) + (0x0b837773LL * zoe2) +
(0x017ef0c6LL * zoe3), 30);
zoc3 = Z_RSHIFT((-0x0cef1502LL * zoo1) + (0x01207a8eLL * zoo2) +
(0x01e936dbLL * zoo3), 30);
zoc4 = Z_RSHIFT((0x029fe643LL * zoe1) + (-0x03ef3fc8LL * zoe2) +
(0x014f5923LL * zoe3), 30);
zoc5 = Z_RSHIFT((0x043a9d08LL * zoo1) + (-0x02154febLL * zoo2) +
(0x00670dbdLL * zoo3), 30);
coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT(
(int64_t)zoc5 * zoz, Z_SHIFT) +
zoc4) * zoz, Z_SHIFT) + zoc3) * zoz, Z_SHIFT) +
zoc2) * zoz, Z_SHIFT) + zoc1) * zoz, Z_SHIFT) + zoc0;
#elif defined(Z_COEFF_INTERP_OPT2X)
int32_t zoz, zoe1, zoe2, zoe3, zoo1, zoo2, zoo3;
int32_t zoc0, zoc1, zoc2, zoc3, zoc4, zoc5;
/* 6-point, 5th-order Optimal 2x */
zoz = z - (Z_ONE >> 1);
zoe1 = z_coeff[1] + z_coeff[0];
zoe2 = z_coeff[2] + z_coeff[-1];
zoe3 = z_coeff[3] + z_coeff[-2];
zoo1 = z_coeff[1] - z_coeff[0];
zoo2 = z_coeff[2] - z_coeff[-1];
zoo3 = z_coeff[3] - z_coeff[-2];
zoc0 = Z_RSHIFT((0x19edb6fdLL * zoe1) + (0x05ebd062LL * zoe2) +
(0x00267881LL * zoe3), 30);
zoc1 = Z_RSHIFT((0x1223af76LL * zoo1) + (0x0de3dd6bLL * zoo2) +
(0x00d683cdLL * zoo3), 30);
zoc2 = Z_RSHIFT((-0x0c3ee068LL * zoe1) + (0x0a5c3769LL * zoe2) +
(0x01e2aceaLL * zoe3), 30);
zoc3 = Z_RSHIFT((-0x0a8ab614LL * zoo1) + (-0x0019522eLL * zoo2) +
(0x022cefc7LL * zoo3), 30);
zoc4 = Z_RSHIFT((0x0276187dLL * zoe1) + (-0x03a801e8LL * zoe2) +
(0x0131d935LL * zoe3), 30);
zoc5 = Z_RSHIFT((0x02c373f5LL * zoo1) + (-0x01275f83LL * zoo2) +
(0x0018ee79LL * zoo3), 30);
coeff = Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT((Z_RSHIFT(
(int64_t)zoc5 * zoz, Z_SHIFT) +
zoc4) * zoz, Z_SHIFT) + zoc3) * zoz, Z_SHIFT) +
zoc2) * zoz, Z_SHIFT) + zoc1) * zoz, Z_SHIFT) + zoc0;
#else
#error "Interpolation type screwed!"
#endif
#if Z_POLYPHASE_COEFF_SHIFT > 0
coeff = Z_RSHIFT(coeff, Z_POLYPHASE_COEFF_SHIFT);
#endif
return (coeff);
}
static int
z_resampler_build_polyphase(struct z_info *info)
{
int32_t alpha, c, i, z, idx;
/* Let this be here first. */
if (info->z_pcoeff != NULL) {
free(info->z_pcoeff, M_DEVBUF);
info->z_pcoeff = NULL;
}
if (feeder_rate_polyphase_max < 1)
return (ENOTSUP);
if (((int64_t)info->z_size * info->z_gy * 2) >
feeder_rate_polyphase_max) {
#ifndef _KERNEL
fprintf(stderr, "Polyphase entries exceed: [%d/%d] %jd > %d\n",
info->z_gx, info->z_gy,
(intmax_t)info->z_size * info->z_gy * 2,
feeder_rate_polyphase_max);
#endif
return (E2BIG);
}
info->z_pcoeff = malloc(sizeof(int32_t) *
info->z_size * info->z_gy * 2, M_DEVBUF, M_NOWAIT | M_ZERO);
if (info->z_pcoeff == NULL)
return (ENOMEM);
for (alpha = 0; alpha < info->z_gy; alpha++) {
z = alpha * info->z_dx;
c = 0;
for (i = info->z_size; i != 0; i--) {
c += z >> Z_SHIFT;
z &= Z_MASK;
idx = (alpha * info->z_size * 2) +
(info->z_size * 2) - i;
info->z_pcoeff[idx] =
z_coeff_interpolate(z, info->z_coeff + c);
z += info->z_dy;
}
z = info->z_dy - (alpha * info->z_dx);
c = 0;
for (i = info->z_size; i != 0; i--) {
c += z >> Z_SHIFT;
z &= Z_MASK;
idx = (alpha * info->z_size * 2) + i - 1;
info->z_pcoeff[idx] =
z_coeff_interpolate(z, info->z_coeff + c);
z += info->z_dy;
}
}
#ifndef _KERNEL
fprintf(stderr, "Polyphase: [%d/%d] %d entries\n",
info->z_gx, info->z_gy, info->z_size * info->z_gy * 2);
#endif
return (0);
}
static int
z_resampler_setup(struct pcm_feeder *f)
{
struct z_info *info;
int64_t gy2gx_max, gx2gy_max;
uint32_t format;
int32_t align, i, z_scale;
int adaptive;
info = f->data;
z_resampler_reset(info);
if (info->src == info->dst)
return (0);
/* Shrink by greatest common divisor. */
i = z_gcd(info->src, info->dst);
info->z_gx = info->src / i;
info->z_gy = info->dst / i;
/* Too big, or too small. Bail out. */
if (!(Z_FACTOR_SAFE(info->z_gx) && Z_FACTOR_SAFE(info->z_gy)))
return (EINVAL);
format = f->desc->in;
adaptive = 0;
z_scale = 0;
/*
* Setup everything: filter length, conversion factor, etc.
*/
if (Z_IS_SINC(info)) {
/*
* Downsampling, or upsampling scaling factor. As long as the
* factor can be represented by a fraction of 1 << Z_SHIFT,
* we're pretty much in business. Scaling is not needed for
* upsampling, so we just slap Z_ONE there.
*/
if (info->z_gx > info->z_gy)
/*
* If the downsampling ratio is beyond sanity,
* enable semi-adaptive mode. Although handling
* extreme ratio is possible, the result of the
* conversion is just pointless, unworthy,
* nonsensical noises, etc.
*/
if ((info->z_gx / info->z_gy) > Z_SINC_DOWNMAX)
z_scale = Z_ONE / Z_SINC_DOWNMAX;
else
z_scale = ((uint64_t)info->z_gy << Z_SHIFT) /
info->z_gx;
else
z_scale = Z_ONE;
/*
* This is actually impossible, unless anything above
* overflow.
*/
if (z_scale < 1)
return (E2BIG);
/*
* Calculate sample time/coefficients index drift. It is
* a constant for upsampling, but downsampling require
* heavy duty filtering with possible too long filters.
* If anything goes wrong, revisit again and enable
* adaptive mode.
*/
z_setup_adaptive_sinc:
if (info->z_pcoeff != NULL) {
free(info->z_pcoeff, M_DEVBUF);
info->z_pcoeff = NULL;
}
if (adaptive == 0) {
info->z_dy = z_scale << Z_DRIFT_SHIFT;
if (info->z_dy < 1)
return (E2BIG);
info->z_scale = z_scale;
} else {
info->z_dy = Z_FULL_ONE;
info->z_scale = Z_ONE;
}
#if 0
#define Z_SCALE_DIV 10000
#define Z_SCALE_LIMIT(s, v) \
((((uint64_t)(s) * (v)) + (Z_SCALE_DIV >> 1)) / Z_SCALE_DIV)
info->z_scale = Z_SCALE_LIMIT(info->z_scale, 9780);
#endif
/* Smallest drift increment. */
info->z_dx = info->z_dy / info->z_gy;
/*
* Overflow or underflow. Try adaptive, let it continue and
* retry.
*/
if (info->z_dx < 1) {
if (adaptive == 0) {
adaptive = 1;
goto z_setup_adaptive_sinc;
}
return (E2BIG);
}
/*
* Round back output drift.
*/
info->z_dy = info->z_dx * info->z_gy;
for (i = 0; i < Z_COEFF_TAB_SIZE; i++) {
if (Z_SINC_COEFF_IDX(info) != i)
continue;
/*
* Calculate required filter length and guard
* against possible abusive result. Note that
* this represents only 1/2 of the entire filter
* length.
*/
info->z_size = z_resampler_sinc_len(info);
/*
* Multiple of 2 rounding, for better accumulator
* performance.
*/
info->z_size &= ~1;
if (info->z_size < 2 || info->z_size > Z_SINC_MAX) {
if (adaptive == 0) {
adaptive = 1;
goto z_setup_adaptive_sinc;
}
return (E2BIG);
}
info->z_coeff = z_coeff_tab[i].coeff + Z_COEFF_OFFSET;
info->z_dcoeff = z_coeff_tab[i].dcoeff;
break;
}
if (info->z_coeff == NULL || info->z_dcoeff == NULL)
return (EINVAL);
} else if (Z_IS_LINEAR(info)) {
/*
* Don't put much effort if we're doing linear interpolation.
* Just center the interpolation distance within Z_LINEAR_ONE,
* and be happy about it.
*/
info->z_dx = Z_LINEAR_FULL_ONE / info->z_gy;
}
/*
* We're safe for now, lets continue.. Look for our resampler
* depending on configured format and quality.
*/
for (i = 0; i < Z_RESAMPLER_TAB_SIZE; i++) {
int ridx;
if (AFMT_ENCODING(format) != z_resampler_tab[i].format)
continue;
if (Z_IS_SINC(info) && adaptive == 0 &&
z_resampler_build_polyphase(info) == 0)
ridx = Z_RESAMPLER_SINC_POLYPHASE;
else
ridx = Z_RESAMPLER_IDX(info);
info->z_resample = z_resampler_tab[i].resampler[ridx];
break;
}
if (info->z_resample == NULL)
return (EINVAL);
info->bps = AFMT_BPS(format);
align = info->channels * info->bps;
/*
* Calculate largest value that can be fed into z_gy2gx() and
* z_gx2gy() without causing (signed) 32bit overflow. z_gy2gx() will
* be called early during feeding process to determine how much input
* samples that is required to generate requested output, while
* z_gx2gy() will be called just before samples filtering /
* accumulation process based on available samples that has been
* calculated using z_gx2gy().
*
* Now that is damn confusing, I guess ;-) .
*/
gy2gx_max = (((uint64_t)info->z_gy * INT32_MAX) - info->z_gy + 1) /
info->z_gx;
if ((gy2gx_max * align) > SND_FXDIV_MAX)
gy2gx_max = SND_FXDIV_MAX / align;
if (gy2gx_max < 1)
return (E2BIG);
gx2gy_max = (((uint64_t)info->z_gx * INT32_MAX) - info->z_gy) /
info->z_gy;
if (gx2gy_max > INT32_MAX)
gx2gy_max = INT32_MAX;
if (gx2gy_max < 1)
return (E2BIG);
/*
* Ensure that z_gy2gx() at its largest possible calculated value
* (alpha = 0) will not cause overflow further late during z_gx2gy()
* stage.
*/
if (z_gy2gx(info, gy2gx_max) > _Z_GCAST(gx2gy_max))
return (E2BIG);
info->z_maxfeed = gy2gx_max * align;
#ifdef Z_USE_ALPHADRIFT
info->z_startdrift = z_gy2gx(info, 1);
info->z_alphadrift = z_drift(info, info->z_startdrift, 1);
#endif
i = z_gy2gx(info, 1);
info->z_full = z_roundpow2((info->z_size << 1) + i);
/*
* Too big to be true, and overflowing left and right like mad ..
*/
if ((info->z_full * align) < 1) {
if (adaptive == 0 && Z_IS_SINC(info)) {
adaptive = 1;
goto z_setup_adaptive_sinc;
}
return (E2BIG);
}
/*
* Increase full buffer size if its too small to reduce cyclic
* buffer shifting in main conversion/feeder loop.
*/
while (info->z_full < Z_RESERVOIR_MAX &&
(info->z_full - (info->z_size << 1)) < Z_RESERVOIR)
info->z_full <<= 1;
/* Initialize buffer position. */
info->z_mask = info->z_full - 1;
info->z_start = z_prev(info, info->z_size << 1, 1);
info->z_pos = z_next(info, info->z_start, 1);
/*
* Allocate or reuse delay line buffer, whichever makes sense.
*/
i = info->z_full * align;
if (i < 1)
return (E2BIG);
if (info->z_delay == NULL || info->z_alloc < i ||
i <= (info->z_alloc >> 1)) {
if (info->z_delay != NULL)
free(info->z_delay, M_DEVBUF);
info->z_delay = malloc(i, M_DEVBUF, M_NOWAIT | M_ZERO);
if (info->z_delay == NULL)
return (ENOMEM);
info->z_alloc = i;
}
/*
* Zero out head of buffer to avoid pops and clicks.
*/
memset(info->z_delay, sndbuf_zerodata(f->desc->out),
info->z_pos * align);
#ifdef Z_DIAGNOSTIC
/*
* XXX Debuging mess !@#$%^
*/
#define dumpz(x) fprintf(stderr, "\t%12s = %10u : %-11d\n", \
"z_"__STRING(x), (uint32_t)info->z_##x, \
(int32_t)info->z_##x)
fprintf(stderr, "\n%s():\n", __func__);
fprintf(stderr, "\tchannels=%d, bps=%d, format=0x%08x, quality=%d\n",
info->channels, info->bps, format, info->quality);
fprintf(stderr, "\t%d (%d) -> %d (%d), ",
info->src, info->rsrc, info->dst, info->rdst);
fprintf(stderr, "[%d/%d]\n", info->z_gx, info->z_gy);
fprintf(stderr, "\tminreq=%d, ", z_gy2gx(info, 1));
if (adaptive != 0)
z_scale = Z_ONE;
fprintf(stderr, "factor=0x%08x/0x%08x (%f)\n",
z_scale, Z_ONE, (double)z_scale / Z_ONE);
fprintf(stderr, "\tbase_length=%d, ", Z_SINC_BASE_LEN(info));
fprintf(stderr, "adaptive=%s\n", (adaptive != 0) ? "YES" : "NO");
dumpz(size);
dumpz(alloc);
if (info->z_alloc < 1024)
fprintf(stderr, "\t%15s%10d Bytes\n",
"", info->z_alloc);
else if (info->z_alloc < (1024 << 10))
fprintf(stderr, "\t%15s%10d KBytes\n",
"", info->z_alloc >> 10);
else if (info->z_alloc < (1024 << 20))
fprintf(stderr, "\t%15s%10d MBytes\n",
"", info->z_alloc >> 20);
else
fprintf(stderr, "\t%15s%10d GBytes\n",
"", info->z_alloc >> 30);
fprintf(stderr, "\t%12s %10d (min output samples)\n",
"",
(int32_t)z_gx2gy(info, info->z_full - (info->z_size << 1)));
fprintf(stderr, "\t%12s %10d (min allocated output samples)\n",
"",
(int32_t)z_gx2gy(info, (info->z_alloc / align) -
(info->z_size << 1)));
fprintf(stderr, "\t%12s = %10d\n",
"z_gy2gx()", (int32_t)z_gy2gx(info, 1));
fprintf(stderr, "\t%12s = %10d -> z_gy2gx() -> %d\n",
"Max", (int32_t)gy2gx_max, (int32_t)z_gy2gx(info, gy2gx_max));
fprintf(stderr, "\t%12s = %10d\n",
"z_gx2gy()", (int32_t)z_gx2gy(info, 1));
fprintf(stderr, "\t%12s = %10d -> z_gx2gy() -> %d\n",
"Max", (int32_t)gx2gy_max, (int32_t)z_gx2gy(info, gx2gy_max));
dumpz(maxfeed);
dumpz(full);
dumpz(start);
dumpz(pos);
dumpz(scale);
fprintf(stderr, "\t%12s %10f\n", "",
(double)info->z_scale / Z_ONE);
dumpz(dx);
fprintf(stderr, "\t%12s %10f\n", "",
(double)info->z_dx / info->z_dy);
dumpz(dy);
fprintf(stderr, "\t%12s %10d (drift step)\n", "",
info->z_dy >> Z_SHIFT);
fprintf(stderr, "\t%12s %10d (scaling differences)\n", "",
(z_scale << Z_DRIFT_SHIFT) - info->z_dy);
fprintf(stderr, "\t%12s = %u bytes\n",
"intpcm32_t", sizeof(intpcm32_t));
fprintf(stderr, "\t%12s = 0x%08x, smallest=%.16lf\n",
"Z_ONE", Z_ONE, (double)1.0 / (double)Z_ONE);
#endif
return (0);
}
static int
z_resampler_set(struct pcm_feeder *f, int what, int32_t value)
{
struct z_info *info;
int32_t oquality;
info = f->data;
switch (what) {
case Z_RATE_SRC:
if (value < feeder_rate_min || value > feeder_rate_max)
return (E2BIG);
if (value == info->rsrc)
return (0);
info->rsrc = value;
break;
case Z_RATE_DST:
if (value < feeder_rate_min || value > feeder_rate_max)
return (E2BIG);
if (value == info->rdst)
return (0);
info->rdst = value;
break;
case Z_RATE_QUALITY:
if (value < Z_QUALITY_MIN || value > Z_QUALITY_MAX)
return (EINVAL);
if (value == info->quality)
return (0);
/*
* If we failed to set the requested quality, restore
* the old one. We cannot afford leaving it broken since
* passive feeder chains like vchans never reinitialize
* itself.
*/
oquality = info->quality;
info->quality = value;
if (z_resampler_setup(f) == 0)
return (0);
info->quality = oquality;
break;
case Z_RATE_CHANNELS:
if (value < SND_CHN_MIN || value > SND_CHN_MAX)
return (EINVAL);
if (value == info->channels)
return (0);
info->channels = value;
break;
default:
return (EINVAL);
break;
}
return (z_resampler_setup(f));
}
static int
z_resampler_get(struct pcm_feeder *f, int what)
{
struct z_info *info;
info = f->data;
switch (what) {
case Z_RATE_SRC:
return (info->rsrc);
break;
case Z_RATE_DST:
return (info->rdst);
break;
case Z_RATE_QUALITY:
return (info->quality);
break;
case Z_RATE_CHANNELS:
return (info->channels);
break;
default:
break;
}
return (-1);
}
static int
z_resampler_init(struct pcm_feeder *f)
{
struct z_info *info;
int ret;
if (f->desc->in != f->desc->out)
return (EINVAL);
info = malloc(sizeof(*info), M_DEVBUF, M_NOWAIT | M_ZERO);
if (info == NULL)
return (ENOMEM);
info->rsrc = Z_RATE_DEFAULT;
info->rdst = Z_RATE_DEFAULT;
info->quality = feeder_rate_quality;
info->channels = AFMT_CHANNEL(f->desc->in);
f->data = info;
ret = z_resampler_setup(f);
if (ret != 0) {
if (info->z_pcoeff != NULL)
free(info->z_pcoeff, M_DEVBUF);
if (info->z_delay != NULL)
free(info->z_delay, M_DEVBUF);
free(info, M_DEVBUF);
f->data = NULL;
}
return (ret);
}
static int
z_resampler_free(struct pcm_feeder *f)
{
struct z_info *info;
info = f->data;
if (info != NULL) {
if (info->z_pcoeff != NULL)
free(info->z_pcoeff, M_DEVBUF);
if (info->z_delay != NULL)
free(info->z_delay, M_DEVBUF);
free(info, M_DEVBUF);
}
f->data = NULL;
return (0);
}
static uint32_t
z_resampler_feed_internal(struct pcm_feeder *f, struct pcm_channel *c,
uint8_t *b, uint32_t count, void *source)
{
struct z_info *info;
int32_t alphadrift, startdrift, reqout, ocount, reqin, align;
int32_t fetch, fetched, start, cp;
uint8_t *dst;
info = f->data;
if (info->z_resample == NULL)
return (z_feed(f->source, c, b, count, source));
/*
* Calculate sample size alignment and amount of sample output.
* We will do everything in sample domain, but at the end we
* will jump back to byte domain.
*/
align = info->channels * info->bps;
ocount = SND_FXDIV(count, align);
if (ocount == 0)
return (0);
/*
* Calculate amount of input samples that is needed to generate
* exact amount of output.
*/
reqin = z_gy2gx(info, ocount) - z_fetched(info);
#ifdef Z_USE_ALPHADRIFT
startdrift = info->z_startdrift;
alphadrift = info->z_alphadrift;
#else
startdrift = _Z_GY2GX(info, 0, 1);
alphadrift = z_drift(info, startdrift, 1);
#endif
dst = b;
do {
if (reqin != 0) {
fetch = z_min(z_free(info), reqin);
if (fetch == 0) {
/*
* No more free spaces, so wind enough
* samples back to the head of delay line
* in byte domain.
*/
fetched = z_fetched(info);
start = z_prev(info, info->z_start,
(info->z_size << 1) - 1);
cp = (info->z_size << 1) + fetched;
z_copy(info->z_delay + (start * align),
info->z_delay, cp * align);
info->z_start =
z_prev(info, info->z_size << 1, 1);
info->z_pos =
z_next(info, info->z_start, fetched + 1);
fetch = z_min(z_free(info), reqin);
#ifdef Z_DIAGNOSTIC
if (1) {
static uint32_t kk = 0;
fprintf(stderr,
"Buffer Move: "
"start=%d fetched=%d cp=%d "
"cycle=%u [%u]\r",
start, fetched, cp, info->z_cycle,
++kk);
}
info->z_cycle = 0;
#endif
}
if (fetch != 0) {
/*
* Fetch in byte domain and jump back
* to sample domain.
*/
fetched = SND_FXDIV(z_feed(f->source, c,
info->z_delay + (info->z_pos * align),
fetch * align, source), align);
/*
* Prepare to convert fetched buffer,
* or mark us done if we cannot fulfill
* the request.
*/
reqin -= fetched;
info->z_pos += fetched;
if (fetched != fetch)
reqin = 0;
}
}
reqout = z_min(z_gx2gy(info, z_fetched(info)), ocount);
if (reqout != 0) {
ocount -= reqout;
/*
* Drift.. drift.. drift..
*
* Notice that there are 2 methods of doing the drift
* operations: The former is much cleaner (in a sense
* of mathematical readings of my eyes), but slower
* due to integer division in z_gy2gx(). Nevertheless,
* both should give the same exact accurate drifting
* results, so the later is favourable.
*/
do {
info->z_resample(info, dst);
#if 0
startdrift = z_gy2gx(info, 1);
alphadrift = z_drift(info, startdrift, 1);
info->z_start += startdrift;
info->z_alpha += alphadrift;
#else
info->z_alpha += alphadrift;
if (info->z_alpha < info->z_gy)
info->z_start += startdrift;
else {
info->z_start += startdrift - 1;
info->z_alpha -= info->z_gy;
}
#endif
dst += align;
#ifdef Z_DIAGNOSTIC
info->z_cycle++;
#endif
} while (--reqout != 0);
}
} while (reqin != 0 && ocount != 0);
/*
* Back to byte domain..
*/
return (dst - b);
}
static int
z_resampler_feed(struct pcm_feeder *f, struct pcm_channel *c, uint8_t *b,
uint32_t count, void *source)
{
uint32_t feed, maxfeed, left;
/*
* Split count to smaller chunks to avoid possible 32bit overflow.
*/
maxfeed = ((struct z_info *)(f->data))->z_maxfeed;
left = count;
do {
feed = z_resampler_feed_internal(f, c, b,
z_min(maxfeed, left), source);
b += feed;
left -= feed;
} while (left != 0 && feed != 0);
return (count - left);
}
static struct pcm_feederdesc feeder_rate_desc[] = {
{ FEEDER_RATE, 0, 0, 0, 0 },
{ 0, 0, 0, 0, 0 },
};
static kobj_method_t feeder_rate_methods[] = {
KOBJMETHOD(feeder_init, z_resampler_init),
KOBJMETHOD(feeder_free, z_resampler_free),
KOBJMETHOD(feeder_set, z_resampler_set),
KOBJMETHOD(feeder_get, z_resampler_get),
KOBJMETHOD(feeder_feed, z_resampler_feed),
KOBJMETHOD_END
};
FEEDER_DECLARE(feeder_rate, NULL);
Reference in New Issue View Git Blame Copy Permalink