#include "audioconvert.h"
-#define MDCT_NBITS 9
-#define MDCT_SAMPLES (1 << MDCT_NBITS)
+#ifndef CONFIG_AC3ENC_FLOAT
+#define CONFIG_AC3ENC_FLOAT 0
+#endif
+
/** Maximum number of exponent groups. +1 for separate DC exponent. */
#define AC3_MAX_EXP_GROUPS 85
+/* stereo rematrixing algorithms */
+#define AC3_REMATRIXING_IS_STATIC 0x1
+#define AC3_REMATRIXING_SUMS 0
+#define AC3_REMATRIXING_NONE 1
+#define AC3_REMATRIXING_ALWAYS 3
+
/** Scale a float value by 2^bits and convert to an integer. */
#define SCALE_FLOAT(a, bits) lrintf((a) * (float)(1 << (bits)))
-/** Scale a float value by 2^15, convert to an integer, and clip to int16_t range. */
-#define FIX15(a) av_clip_int16(SCALE_FLOAT(a, 15))
-
-/**
- * Compex number.
- * Used in fixed-point MDCT calculation.
- */
-typedef struct IComplex {
- int16_t re,im;
-} IComplex;
+#if CONFIG_AC3ENC_FLOAT
+#include "ac3enc_float.h"
+#else
+#include "ac3enc_fixed.h"
+#endif
-typedef struct AC3MDCTContext {
- AVCodecContext *avctx; ///< parent context for av_log()
- int16_t *rot_tmp; ///< temp buffer for pre-rotated samples
- IComplex *cplx_tmp; ///< temp buffer for complex pre-rotated samples
-} AC3MDCTContext;
/**
* Data for a single audio block.
*/
typedef struct AC3Block {
uint8_t **bap; ///< bit allocation pointers (bap)
- int32_t **mdct_coef; ///< MDCT coefficients
+ CoefType **mdct_coef; ///< MDCT coefficients
+ int32_t **fixed_coef; ///< fixed-point MDCT coefficients
uint8_t **exp; ///< original exponents
uint8_t **grouped_exp; ///< grouped exponents
int16_t **psd; ///< psd per frequency bin
int16_t **band_psd; ///< psd per critical band
int16_t **mask; ///< masking curve
uint16_t **qmant; ///< quantized mantissas
- uint8_t num_exp_groups[AC3_MAX_CHANNELS]; ///< number of exponent groups
- uint8_t exp_strategy[AC3_MAX_CHANNELS]; ///< exponent strategies
int8_t exp_shift[AC3_MAX_CHANNELS]; ///< exponent shift values
+ uint8_t new_rematrixing_strategy; ///< send new rematrixing flags in this block
+ uint8_t rematrixing_flags[4]; ///< rematrixing flags
} AC3Block;
/**
int frame_size_min; ///< minimum frame size in case rounding is necessary
int frame_size; ///< current frame size in bytes
int frame_size_code; ///< frame size code (frmsizecod)
+ uint16_t crc_inv[2];
int bits_written; ///< bit count (used to avg. bitrate)
int samples_written; ///< sample count (used to avg. bitrate)
int channel_mode; ///< channel mode (acmod)
const uint8_t *channel_map; ///< channel map used to reorder channels
+ int cutoff; ///< user-specified cutoff frequency, in Hz
int bandwidth_code[AC3_MAX_CHANNELS]; ///< bandwidth code (0 to 60) (chbwcod)
int nb_coefs[AC3_MAX_CHANNELS];
+ int rematrixing; ///< determines how rematrixing strategy is calculated
+
/* bitrate allocation control */
int slow_gain_code; ///< slow gain code (sgaincod)
int slow_decay_code; ///< slow decay code (sdcycod)
int coarse_snr_offset; ///< coarse SNR offsets (csnroffst)
int fast_gain_code[AC3_MAX_CHANNELS]; ///< fast gain codes (signal-to-mask ratio) (fgaincod)
int fine_snr_offset[AC3_MAX_CHANNELS]; ///< fine SNR offsets (fsnroffst)
+ int frame_bits_fixed; ///< number of non-coefficient bits for fixed parameters
int frame_bits; ///< all frame bits except exponents and mantissas
int exponent_bits; ///< number of bits used for exponents
int mant1_cnt, mant2_cnt, mant4_cnt; ///< mantissa counts for bap=1,2,4
uint16_t *qmant1_ptr, *qmant2_ptr, *qmant4_ptr; ///< mantissa pointers for bap=1,2,4
- int16_t **planar_samples;
+ SampleType **planar_samples;
uint8_t *bap_buffer;
uint8_t *bap1_buffer;
- int32_t *mdct_coef_buffer;
+ CoefType *mdct_coef_buffer;
+ int32_t *fixed_coef_buffer;
uint8_t *exp_buffer;
uint8_t *grouped_exp_buffer;
int16_t *psd_buffer;
int16_t *mask_buffer;
uint16_t *qmant_buffer;
- DECLARE_ALIGNED(16, int16_t, windowed_samples)[AC3_WINDOW_SIZE];
+ uint8_t exp_strategy[AC3_MAX_CHANNELS][AC3_MAX_BLOCKS]; ///< exponent strategies
+
+ DECLARE_ALIGNED(16, SampleType, windowed_samples)[AC3_WINDOW_SIZE];
} AC3EncodeContext;
-/** MDCT and FFT tables */
-static int16_t costab[64];
-static int16_t sintab[64];
-static int16_t xcos1[128];
-static int16_t xsin1[128];
+/* prototypes for functions in ac3enc_fixed.c and ac3enc_float.c */
+
+static av_cold void mdct_end(AC3MDCTContext *mdct);
+
+static av_cold int mdct_init(AVCodecContext *avctx, AC3MDCTContext *mdct,
+ int nbits);
+
+static void mdct512(AC3MDCTContext *mdct, CoefType *out, SampleType *in);
+
+static void apply_window(DSPContext *dsp, SampleType *output, const SampleType *input,
+ const SampleType *window, int n);
+
+static int normalize_samples(AC3EncodeContext *s);
+
+static void scale_coefficients(AC3EncodeContext *s);
+
+
+/**
+ * LUT for number of exponent groups.
+ * exponent_group_tab[exponent strategy-1][number of coefficients]
+ */
+static uint8_t exponent_group_tab[3][256];
+
+
+/**
+ * List of supported channel layouts.
+ */
+static const int64_t ac3_channel_layouts[] = {
+ AV_CH_LAYOUT_MONO,
+ AV_CH_LAYOUT_STEREO,
+ AV_CH_LAYOUT_2_1,
+ AV_CH_LAYOUT_SURROUND,
+ AV_CH_LAYOUT_2_2,
+ AV_CH_LAYOUT_QUAD,
+ AV_CH_LAYOUT_4POINT0,
+ AV_CH_LAYOUT_5POINT0,
+ AV_CH_LAYOUT_5POINT0_BACK,
+ (AV_CH_LAYOUT_MONO | AV_CH_LOW_FREQUENCY),
+ (AV_CH_LAYOUT_STEREO | AV_CH_LOW_FREQUENCY),
+ (AV_CH_LAYOUT_2_1 | AV_CH_LOW_FREQUENCY),
+ (AV_CH_LAYOUT_SURROUND | AV_CH_LOW_FREQUENCY),
+ (AV_CH_LAYOUT_2_2 | AV_CH_LOW_FREQUENCY),
+ (AV_CH_LAYOUT_QUAD | AV_CH_LOW_FREQUENCY),
+ (AV_CH_LAYOUT_4POINT0 | AV_CH_LOW_FREQUENCY),
+ AV_CH_LAYOUT_5POINT1,
+ AV_CH_LAYOUT_5POINT1_BACK,
+ 0
+};
/**
* Channels are reordered from FFmpeg's default order to AC-3 order.
*/
static void deinterleave_input_samples(AC3EncodeContext *s,
- const int16_t *samples)
+ const SampleType *samples)
{
int ch, i;
/* deinterleave and remap input samples */
for (ch = 0; ch < s->channels; ch++) {
- const int16_t *sptr;
+ const SampleType *sptr;
int sinc;
/* copy last 256 samples of previous frame to the start of the current frame */
/**
- * Finalize MDCT and free allocated memory.
- */
-static av_cold void mdct_end(AC3MDCTContext *mdct)
-{
- av_freep(&mdct->rot_tmp);
- av_freep(&mdct->cplx_tmp);
-}
-
-
-
-/**
- * Initialize FFT tables.
- * @param ln log2(FFT size)
- */
-static av_cold void fft_init(int ln)
-{
- int i, n, n2;
- float alpha;
-
- n = 1 << ln;
- n2 = n >> 1;
-
- for (i = 0; i < n2; i++) {
- alpha = 2.0 * M_PI * i / n;
- costab[i] = FIX15(cos(alpha));
- sintab[i] = FIX15(sin(alpha));
- }
-}
-
-
-/**
- * Initialize MDCT tables.
- * @param nbits log2(MDCT size)
+ * Apply the MDCT to input samples to generate frequency coefficients.
+ * This applies the KBD window and normalizes the input to reduce precision
+ * loss due to fixed-point calculations.
*/
-static av_cold int mdct_init(AC3MDCTContext *mdct, int nbits)
+static void apply_mdct(AC3EncodeContext *s)
{
- int i, n, n4;
+ int blk, ch;
- n = 1 << nbits;
- n4 = n >> 2;
+ for (ch = 0; ch < s->channels; ch++) {
+ for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
+ AC3Block *block = &s->blocks[blk];
+ const SampleType *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE];
- fft_init(nbits - 2);
+ apply_window(&s->dsp, s->windowed_samples, input_samples, s->mdct.window, AC3_WINDOW_SIZE);
- FF_ALLOC_OR_GOTO(mdct->avctx, mdct->rot_tmp, n * sizeof(*mdct->rot_tmp),
- mdct_alloc_fail);
- FF_ALLOC_OR_GOTO(mdct->avctx, mdct->cplx_tmp, n4 * sizeof(*mdct->cplx_tmp),
- mdct_alloc_fail);
+ block->exp_shift[ch] = normalize_samples(s);
- for (i = 0; i < n4; i++) {
- float alpha = 2.0 * M_PI * (i + 1.0 / 8.0) / n;
- xcos1[i] = FIX15(-cos(alpha));
- xsin1[i] = FIX15(-sin(alpha));
+ mdct512(&s->mdct, block->mdct_coef[ch], s->windowed_samples);
+ }
}
-
- return 0;
-mdct_alloc_fail:
- return AVERROR(ENOMEM);
-}
-
-
-/** Butterfly op */
-#define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \
-{ \
- int ax, ay, bx, by; \
- bx = pre1; \
- by = pim1; \
- ax = qre1; \
- ay = qim1; \
- pre = (bx + ax) >> 1; \
- pim = (by + ay) >> 1; \
- qre = (bx - ax) >> 1; \
- qim = (by - ay) >> 1; \
-}
-
-
-/** Complex multiply */
-#define CMUL(pre, pim, are, aim, bre, bim) \
-{ \
- pre = (MUL16(are, bre) - MUL16(aim, bim)) >> 15; \
- pim = (MUL16(are, bim) + MUL16(bre, aim)) >> 15; \
}
/**
- * Calculate a 2^n point complex FFT on 2^ln points.
- * @param z complex input/output samples
- * @param ln log2(FFT size)
+ * Initialize stereo rematrixing.
+ * If the strategy does not change for each frame, set the rematrixing flags.
*/
-static void fft(IComplex *z, int ln)
+static void rematrixing_init(AC3EncodeContext *s)
{
- int j, l, np, np2;
- int nblocks, nloops;
- register IComplex *p,*q;
- int tmp_re, tmp_im;
-
- np = 1 << ln;
-
- /* reverse */
- for (j = 0; j < np; j++) {
- int k = av_reverse[j] >> (8 - ln);
- if (k < j)
- FFSWAP(IComplex, z[k], z[j]);
+ if (s->channel_mode == AC3_CHMODE_STEREO)
+ s->rematrixing = AC3_REMATRIXING_SUMS;
+ else
+ s->rematrixing = AC3_REMATRIXING_NONE;
+ /* NOTE: AC3_REMATRIXING_ALWAYS might be used in
+ the future in conjunction with channel coupling. */
+
+ if (s->rematrixing & AC3_REMATRIXING_IS_STATIC) {
+ int flag = (s->rematrixing == AC3_REMATRIXING_ALWAYS);
+ s->blocks[0].new_rematrixing_strategy = 1;
+ memset(s->blocks[0].rematrixing_flags, flag,
+ sizeof(s->blocks[0].rematrixing_flags));
}
-
- /* pass 0 */
-
- p = &z[0];
- j = np >> 1;
- do {
- BF(p[0].re, p[0].im, p[1].re, p[1].im,
- p[0].re, p[0].im, p[1].re, p[1].im);
- p += 2;
- } while (--j);
-
- /* pass 1 */
-
- p = &z[0];
- j = np >> 2;
- do {
- BF(p[0].re, p[0].im, p[2].re, p[2].im,
- p[0].re, p[0].im, p[2].re, p[2].im);
- BF(p[1].re, p[1].im, p[3].re, p[3].im,
- p[1].re, p[1].im, p[3].im, -p[3].re);
- p+=4;
- } while (--j);
-
- /* pass 2 .. ln-1 */
-
- nblocks = np >> 3;
- nloops = 1 << 2;
- np2 = np >> 1;
- do {
- p = z;
- q = z + nloops;
- for (j = 0; j < nblocks; j++) {
- BF(p->re, p->im, q->re, q->im,
- p->re, p->im, q->re, q->im);
- p++;
- q++;
- for(l = nblocks; l < np2; l += nblocks) {
- CMUL(tmp_re, tmp_im, costab[l], -sintab[l], q->re, q->im);
- BF(p->re, p->im, q->re, q->im,
- p->re, p->im, tmp_re, tmp_im);
- p++;
- q++;
- }
- p += nloops;
- q += nloops;
- }
- nblocks = nblocks >> 1;
- nloops = nloops << 1;
- } while (nblocks);
}
/**
- * Calculate a 512-point MDCT
- * @param out 256 output frequency coefficients
- * @param in 512 windowed input audio samples
+ * Determine rematrixing flags for each block and band.
*/
-static void mdct512(AC3MDCTContext *mdct, int32_t *out, int16_t *in)
+static void compute_rematrixing_strategy(AC3EncodeContext *s)
{
- int i, re, im;
- int16_t *rot = mdct->rot_tmp;
- IComplex *x = mdct->cplx_tmp;
-
- /* shift to simplify computations */
- for (i = 0; i < MDCT_SAMPLES/4; i++)
- rot[i] = -in[i + 3*MDCT_SAMPLES/4];
- memcpy(&rot[MDCT_SAMPLES/4], &in[0], 3*MDCT_SAMPLES/4*sizeof(*in));
-
- /* pre rotation */
- for (i = 0; i < MDCT_SAMPLES/4; i++) {
- re = ((int)rot[ 2*i] - (int)rot[MDCT_SAMPLES -1-2*i]) >> 1;
- im = -((int)rot[MDCT_SAMPLES/2+2*i] - (int)rot[MDCT_SAMPLES/2-1-2*i]) >> 1;
- CMUL(x[i].re, x[i].im, re, im, -xcos1[i], xsin1[i]);
- }
+ int nb_coefs;
+ int blk, bnd, i;
+ AC3Block *block, *block0;
- fft(x, MDCT_NBITS - 2);
-
- /* post rotation */
- for (i = 0; i < MDCT_SAMPLES/4; i++) {
- re = x[i].re;
- im = x[i].im;
- CMUL(out[MDCT_SAMPLES/2-1-2*i], out[2*i], re, im, xsin1[i], xcos1[i]);
- }
-}
+ if (s->rematrixing & AC3_REMATRIXING_IS_STATIC)
+ return;
+ nb_coefs = FFMIN(s->nb_coefs[0], s->nb_coefs[1]);
-/**
- * Apply KBD window to input samples prior to MDCT.
- */
-static void apply_window(int16_t *output, const int16_t *input,
- const int16_t *window, int n)
-{
- int i;
- int n2 = n >> 1;
+ s->blocks[0].new_rematrixing_strategy = 1;
+ for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
+ block = &s->blocks[blk];
+ for (bnd = 0; bnd < 4; bnd++) {
+ /* calculate calculate sum of squared coeffs for one band in one block */
+ int start = ff_ac3_rematrix_band_tab[bnd];
+ int end = FFMIN(nb_coefs, ff_ac3_rematrix_band_tab[bnd+1]);
+ CoefSumType sum[4] = {0,};
+ for (i = start; i < end; i++) {
+ CoefType lt = block->mdct_coef[0][i];
+ CoefType rt = block->mdct_coef[1][i];
+ CoefType md = lt + rt;
+ CoefType sd = lt - rt;
+ sum[0] += lt * lt;
+ sum[1] += rt * rt;
+ sum[2] += md * md;
+ sum[3] += sd * sd;
+ }
- for (i = 0; i < n2; i++) {
- output[i] = MUL16(input[i], window[i]) >> 15;
- output[n-i-1] = MUL16(input[n-i-1], window[i]) >> 15;
+ /* compare sums to determine if rematrixing will be used for this band */
+ if (FFMIN(sum[2], sum[3]) < FFMIN(sum[0], sum[1]))
+ block->rematrixing_flags[bnd] = 1;
+ else
+ block->rematrixing_flags[bnd] = 0;
+
+ /* determine if new rematrixing flags will be sent */
+ if (blk &&
+ !block->new_rematrixing_strategy &&
+ block->rematrixing_flags[bnd] != block0->rematrixing_flags[bnd]) {
+ block->new_rematrixing_strategy = 1;
+ }
+ }
+ block0 = block;
}
}
/**
- * Calculate the log2() of the maximum absolute value in an array.
- * @param tab input array
- * @param n number of values in the array
- * @return log2(max(abs(tab[])))
+ * Apply stereo rematrixing to coefficients based on rematrixing flags.
*/
-static int log2_tab(int16_t *tab, int n)
+static void apply_rematrixing(AC3EncodeContext *s)
{
- int i, v;
-
- v = 0;
- for (i = 0; i < n; i++)
- v |= abs(tab[i]);
-
- return av_log2(v);
-}
+ int nb_coefs;
+ int blk, bnd, i;
+ int start, end;
+ uint8_t *flags;
+ if (s->rematrixing == AC3_REMATRIXING_NONE)
+ return;
-/**
- * Left-shift each value in an array by a specified amount.
- * @param tab input array
- * @param n number of values in the array
- * @param lshift left shift amount. a negative value means right shift.
- */
-static void lshift_tab(int16_t *tab, int n, int lshift)
-{
- int i;
+ nb_coefs = FFMIN(s->nb_coefs[0], s->nb_coefs[1]);
- if (lshift > 0) {
- for (i = 0; i < n; i++)
- tab[i] <<= lshift;
- } else if (lshift < 0) {
- lshift = -lshift;
- for (i = 0; i < n; i++)
- tab[i] >>= lshift;
+ for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
+ AC3Block *block = &s->blocks[blk];
+ if (block->new_rematrixing_strategy)
+ flags = block->rematrixing_flags;
+ for (bnd = 0; bnd < 4; bnd++) {
+ if (flags[bnd]) {
+ start = ff_ac3_rematrix_band_tab[bnd];
+ end = FFMIN(nb_coefs, ff_ac3_rematrix_band_tab[bnd+1]);
+ for (i = start; i < end; i++) {
+ int32_t lt = block->fixed_coef[0][i];
+ int32_t rt = block->fixed_coef[1][i];
+ block->fixed_coef[0][i] = (lt + rt) >> 1;
+ block->fixed_coef[1][i] = (lt - rt) >> 1;
+ }
+ }
+ }
}
}
/**
- * Normalize the input samples to use the maximum available precision.
- * This assumes signed 16-bit input samples. Exponents are reduced by 9 to
- * match the 24-bit internal precision for MDCT coefficients.
- *
- * @return exponent shift
- */
-static int normalize_samples(AC3EncodeContext *s)
-{
- int v = 14 - log2_tab(s->windowed_samples, AC3_WINDOW_SIZE);
- v = FFMAX(0, v);
- lshift_tab(s->windowed_samples, AC3_WINDOW_SIZE, v);
- return v - 9;
-}
-
-
-/**
- * Apply the MDCT to input samples to generate frequency coefficients.
- * This applies the KBD window and normalizes the input to reduce precision
- * loss due to fixed-point calculations.
+ * Initialize exponent tables.
*/
-static void apply_mdct(AC3EncodeContext *s)
+static av_cold void exponent_init(AC3EncodeContext *s)
{
- int blk, ch;
-
- for (ch = 0; ch < s->channels; ch++) {
- for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
- AC3Block *block = &s->blocks[blk];
- const int16_t *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE];
-
- apply_window(s->windowed_samples, input_samples, ff_ac3_window, AC3_WINDOW_SIZE);
-
- block->exp_shift[ch] = normalize_samples(s);
-
- mdct512(&s->mdct, block->mdct_coef[ch], s->windowed_samples);
- }
+ int i;
+ for (i = 73; i < 256; i++) {
+ exponent_group_tab[0][i] = (i - 1) / 3;
+ exponent_group_tab[1][i] = (i + 2) / 6;
+ exponent_group_tab[2][i] = (i + 8) / 12;
}
+ /* LFE */
+ exponent_group_tab[0][7] = 2;
}
for (ch = 0; ch < s->channels; ch++) {
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
AC3Block *block = &s->blocks[blk];
+ uint8_t *exp = block->exp[ch];
+ int32_t *coef = block->fixed_coef[ch];
+ int exp_shift = block->exp_shift[ch];
for (i = 0; i < AC3_MAX_COEFS; i++) {
int e;
- int v = abs(block->mdct_coef[ch][i]);
+ int v = abs(coef[i]);
if (v == 0)
e = 24;
else {
- e = 23 - av_log2(v) + block->exp_shift[ch];
+ e = 23 - av_log2(v) + exp_shift;
if (e >= 24) {
e = 24;
- block->mdct_coef[ch][i] = 0;
+ coef[i] = 0;
}
}
- block->exp[ch][i] = e;
+ exp[i] = e;
}
}
}
/**
* Calculate exponent strategies for all blocks in a single channel.
*/
-static void compute_exp_strategy_ch(AC3EncodeContext *s, uint8_t *exp_strategy, uint8_t **exp)
+static void compute_exp_strategy_ch(AC3EncodeContext *s, uint8_t *exp_strategy,
+ uint8_t *exp)
{
int blk, blk1;
int exp_diff;
/* estimate if the exponent variation & decide if they should be
reused in the next frame */
exp_strategy[0] = EXP_NEW;
+ exp += AC3_MAX_COEFS;
for (blk = 1; blk < AC3_MAX_BLOCKS; blk++) {
- exp_diff = s->dsp.sad[0](NULL, exp[blk], exp[blk-1], 16, 16);
+ exp_diff = s->dsp.sad[0](NULL, exp, exp - AC3_MAX_COEFS, 16, 16);
if (exp_diff > EXP_DIFF_THRESHOLD)
exp_strategy[blk] = EXP_NEW;
else
exp_strategy[blk] = EXP_REUSE;
+ exp += AC3_MAX_COEFS;
}
+ emms_c();
/* now select the encoding strategy type : if exponents are often
recoded, we use a coarse encoding */
*/
static void compute_exp_strategy(AC3EncodeContext *s)
{
- uint8_t *exp1[AC3_MAX_CHANNELS][AC3_MAX_BLOCKS];
- uint8_t exp_str1[AC3_MAX_CHANNELS][AC3_MAX_BLOCKS];
int ch, blk;
for (ch = 0; ch < s->fbw_channels; ch++) {
- for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
- exp1[ch][blk] = s->blocks[blk].exp[ch];
- exp_str1[ch][blk] = s->blocks[blk].exp_strategy[ch];
- }
-
- compute_exp_strategy_ch(s, exp_str1[ch], exp1[ch]);
-
- for (blk = 0; blk < AC3_MAX_BLOCKS; blk++)
- s->blocks[blk].exp_strategy[ch] = exp_str1[ch][blk];
+ compute_exp_strategy_ch(s, s->exp_strategy[ch], s->blocks[0].exp[ch]);
}
if (s->lfe_on) {
ch = s->lfe_channel;
- s->blocks[0].exp_strategy[ch] = EXP_D15;
+ s->exp_strategy[ch][0] = EXP_D15;
for (blk = 1; blk < AC3_MAX_BLOCKS; blk++)
- s->blocks[blk].exp_strategy[ch] = EXP_REUSE;
+ s->exp_strategy[ch][blk] = EXP_REUSE;
}
}
/**
* Set each encoded exponent in a block to the minimum of itself and the
- * exponent in the same frequency bin of a following block.
- * exp[i] = min(exp[i], exp1[i]
+ * exponents in the same frequency bin of up to 5 following blocks.
*/
-static void exponent_min(uint8_t *exp, uint8_t *exp1, int n)
+static void exponent_min(uint8_t *exp, int num_reuse_blocks, int nb_coefs)
{
- int i;
- for (i = 0; i < n; i++) {
- if (exp1[i] < exp[i])
- exp[i] = exp1[i];
+ int blk, i;
+
+ if (!num_reuse_blocks)
+ return;
+
+ for (i = 0; i < nb_coefs; i++) {
+ uint8_t min_exp = *exp;
+ uint8_t *exp1 = exp + AC3_MAX_COEFS;
+ for (blk = 0; blk < num_reuse_blocks; blk++) {
+ uint8_t next_exp = *exp1;
+ if (next_exp < min_exp)
+ min_exp = next_exp;
+ exp1 += AC3_MAX_COEFS;
+ }
+ *exp++ = min_exp;
}
}
/**
* Update the exponents so that they are the ones the decoder will decode.
*/
-static void encode_exponents_blk_ch(uint8_t *exp,
- int nb_exps, int exp_strategy,
- uint8_t *num_exp_groups)
+static void encode_exponents_blk_ch(uint8_t *exp, int nb_exps, int exp_strategy)
{
- int group_size, nb_groups, i, k;
+ int nb_groups, i, k;
- group_size = exp_strategy + (exp_strategy == EXP_D45);
- *num_exp_groups = (nb_exps + (group_size * 3) - 4) / (3 * group_size);
- nb_groups = *num_exp_groups * 3;
+ nb_groups = exponent_group_tab[exp_strategy-1][nb_exps] * 3;
/* for each group, compute the minimum exponent */
switch(exp_strategy) {
*/
static void encode_exponents(AC3EncodeContext *s)
{
- int blk, blk1, blk2, ch;
- AC3Block *block, *block1, *block2;
+ int blk, blk1, ch;
+ uint8_t *exp, *exp1, *exp_strategy;
+ int nb_coefs, num_reuse_blocks;
for (ch = 0; ch < s->channels; ch++) {
+ exp = s->blocks[0].exp[ch];
+ exp_strategy = s->exp_strategy[ch];
+ nb_coefs = s->nb_coefs[ch];
+
blk = 0;
- block = &s->blocks[0];
while (blk < AC3_MAX_BLOCKS) {
blk1 = blk + 1;
- block1 = block + 1;
- /* for the EXP_REUSE case we select the min of the exponents */
- while (blk1 < AC3_MAX_BLOCKS && block1->exp_strategy[ch] == EXP_REUSE) {
- exponent_min(block->exp[ch], block1->exp[ch], s->nb_coefs[ch]);
+
+ /* count the number of EXP_REUSE blocks after the current block */
+ while (blk1 < AC3_MAX_BLOCKS && exp_strategy[blk1] == EXP_REUSE)
blk1++;
- block1++;
- }
- encode_exponents_blk_ch(block->exp[ch], s->nb_coefs[ch],
- block->exp_strategy[ch],
- &block->num_exp_groups[ch]);
+ num_reuse_blocks = blk1 - blk - 1;
+
+ /* for the EXP_REUSE case we select the min of the exponents */
+ exponent_min(exp, num_reuse_blocks, nb_coefs);
+
+ encode_exponents_blk_ch(exp, nb_coefs, exp_strategy[blk]);
+
/* copy encoded exponents for reuse case */
- block2 = block + 1;
- for (blk2 = blk+1; blk2 < blk1; blk2++, block2++) {
- memcpy(block2->exp[ch], block->exp[ch],
- s->nb_coefs[ch] * sizeof(uint8_t));
+ exp1 = exp + AC3_MAX_COEFS;
+ while (blk < blk1-1) {
+ memcpy(exp1, exp, nb_coefs * sizeof(*exp));
+ exp1 += AC3_MAX_COEFS;
+ blk++;
}
blk = blk1;
- block = block1;
+ exp = exp1;
}
}
}
static void group_exponents(AC3EncodeContext *s)
{
int blk, ch, i;
- int group_size, bit_count;
+ int group_size, nb_groups, bit_count;
uint8_t *p;
int delta0, delta1, delta2;
int exp0, exp1;
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
AC3Block *block = &s->blocks[blk];
for (ch = 0; ch < s->channels; ch++) {
- if (block->exp_strategy[ch] == EXP_REUSE) {
- block->num_exp_groups[ch] = 0;
+ int exp_strategy = s->exp_strategy[ch][blk];
+ if (exp_strategy == EXP_REUSE)
continue;
- }
- group_size = block->exp_strategy[ch] + (block->exp_strategy[ch] == EXP_D45);
- bit_count += 4 + (block->num_exp_groups[ch] * 7);
+ group_size = exp_strategy + (exp_strategy == EXP_D45);
+ nb_groups = exponent_group_tab[exp_strategy-1][s->nb_coefs[ch]];
+ bit_count += 4 + (nb_groups * 7);
p = block->exp[ch];
/* DC exponent */
block->grouped_exp[ch][0] = exp1;
/* remaining exponents are delta encoded */
- for (i = 1; i <= block->num_exp_groups[ch]; i++) {
+ for (i = 1; i <= nb_groups; i++) {
/* merge three delta in one code */
exp0 = exp1;
exp1 = p[0];
}
+/**
+ * Count frame bits that are based solely on fixed parameters.
+ * This only has to be run once when the encoder is initialized.
+ */
+static void count_frame_bits_fixed(AC3EncodeContext *s)
+{
+ static const int frame_bits_inc[8] = { 0, 0, 2, 2, 2, 4, 2, 4 };
+ int blk;
+ int frame_bits;
+
+ /* assumptions:
+ * no dynamic range codes
+ * no channel coupling
+ * bit allocation parameters do not change between blocks
+ * SNR offsets do not change between blocks
+ * no delta bit allocation
+ * no skipped data
+ * no auxilliary data
+ */
+
+ /* header size */
+ frame_bits = 65;
+ frame_bits += frame_bits_inc[s->channel_mode];
+
+ /* audio blocks */
+ for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
+ frame_bits += s->fbw_channels * 2 + 2; /* blksw * c, dithflag * c, dynrnge, cplstre */
+ if (s->channel_mode == AC3_CHMODE_STEREO) {
+ frame_bits++; /* rematstr */
+ }
+ frame_bits += 2 * s->fbw_channels; /* chexpstr[2] * c */
+ if (s->lfe_on)
+ frame_bits++; /* lfeexpstr */
+ frame_bits++; /* baie */
+ frame_bits++; /* snr */
+ frame_bits += 2; /* delta / skip */
+ }
+ frame_bits++; /* cplinu for block 0 */
+ /* bit alloc info */
+ /* sdcycod[2], fdcycod[2], sgaincod[2], dbpbcod[2], floorcod[3] */
+ /* csnroffset[6] */
+ /* (fsnoffset[4] + fgaincod[4]) * c */
+ frame_bits += 2*4 + 3 + 6 + s->channels * (4 + 3);
+
+ /* auxdatae, crcrsv */
+ frame_bits += 2;
+
+ /* CRC */
+ frame_bits += 16;
+
+ s->frame_bits_fixed = frame_bits;
+}
+
+
/**
* Initialize bit allocation.
* Set default parameter codes and calculate parameter values.
s->slow_decay_code = 2;
s->fast_decay_code = 1;
s->slow_gain_code = 1;
- s->db_per_bit_code = 2;
+ s->db_per_bit_code = 3;
s->floor_code = 4;
for (ch = 0; ch < s->channels; ch++)
s->fast_gain_code[ch] = 4;
s->bit_alloc.slow_gain = ff_ac3_slow_gain_tab[s->slow_gain_code];
s->bit_alloc.db_per_bit = ff_ac3_db_per_bit_tab[s->db_per_bit_code];
s->bit_alloc.floor = ff_ac3_floor_tab[s->floor_code];
+
+ count_frame_bits_fixed(s);
}
/**
* Count the bits used to encode the frame, minus exponents and mantissas.
+ * Bits based on fixed parameters have already been counted, so now we just
+ * have to add the bits based on parameters that change during encoding.
*/
static void count_frame_bits(AC3EncodeContext *s)
{
- static const int frame_bits_inc[8] = { 0, 0, 2, 2, 2, 4, 2, 4 };
int blk, ch;
- int frame_bits;
+ int frame_bits = 0;
- /* header size */
- frame_bits = 65;
- frame_bits += frame_bits_inc[s->channel_mode];
-
- /* audio blocks */
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
- frame_bits += s->fbw_channels * 2 + 2; /* blksw * c, dithflag * c, dynrnge, cplstre */
- if (s->channel_mode == AC3_CHMODE_STEREO) {
- frame_bits++; /* rematstr */
- if (!blk)
- frame_bits += 4;
+ /* stereo rematrixing */
+ if (s->channel_mode == AC3_CHMODE_STEREO &&
+ s->blocks[blk].new_rematrixing_strategy) {
+ frame_bits += 4;
}
- frame_bits += 2 * s->fbw_channels; /* chexpstr[2] * c */
- if (s->lfe_on)
- frame_bits++; /* lfeexpstr */
+
for (ch = 0; ch < s->fbw_channels; ch++) {
- if (s->blocks[blk].exp_strategy[ch] != EXP_REUSE)
+ if (s->exp_strategy[ch][blk] != EXP_REUSE)
frame_bits += 6 + 2; /* chbwcod[6], gainrng[2] */
}
- frame_bits++; /* baie */
- frame_bits++; /* snr */
- frame_bits += 2; /* delta / skip */
}
- frame_bits++; /* cplinu for block 0 */
- /* bit alloc info */
- /* sdcycod[2], fdcycod[2], sgaincod[2], dbpbcod[2], floorcod[3] */
- /* csnroffset[6] */
- /* (fsnoffset[4] + fgaincod[4]) * c */
- frame_bits += 2*4 + 3 + 6 + s->channels * (4 + 3);
-
- /* auxdatae, crcrsv */
- frame_bits += 2;
-
- /* CRC */
- frame_bits += 16;
-
- s->frame_bits = frame_bits;
+ s->frame_bits = s->frame_bits_fixed + frame_bits;
}
/**
* Calculate the number of bits needed to encode a set of mantissas.
*/
-static int compute_mantissa_size(AC3EncodeContext *s, uint8_t *bap, int nb_coefs)
+static int compute_mantissa_size(int mant_cnt[5], uint8_t *bap, int nb_coefs)
{
int bits, b, i;
bits = 0;
for (i = 0; i < nb_coefs; i++) {
b = bap[i];
- switch (b) {
- case 0:
- /* bap=0 mantissas are not encoded */
- break;
- case 1:
- /* 3 mantissas in 5 bits */
- if (s->mant1_cnt == 0)
- bits += 5;
- if (++s->mant1_cnt == 3)
- s->mant1_cnt = 0;
- break;
- case 2:
- /* 3 mantissas in 7 bits */
- if (s->mant2_cnt == 0)
- bits += 7;
- if (++s->mant2_cnt == 3)
- s->mant2_cnt = 0;
- break;
- case 3:
- bits += 3;
- break;
- case 4:
- /* 2 mantissas in 7 bits */
- if (s->mant4_cnt == 0)
- bits += 7;
- if (++s->mant4_cnt == 2)
- s->mant4_cnt = 0;
- break;
- case 14:
- bits += 14;
- break;
- case 15:
- bits += 16;
- break;
- default:
+ if (b <= 4) {
+ // bap=1 to bap=4 will be counted in compute_mantissa_size_final
+ mant_cnt[b]++;
+ } else if (b <= 13) {
+ // bap=5 to bap=13 use (bap-1) bits
bits += b - 1;
- break;
+ } else {
+ // bap=14 uses 14 bits and bap=15 uses 16 bits
+ bits += (b == 14) ? 14 : 16;
}
}
return bits;
}
+/**
+ * Finalize the mantissa bit count by adding in the grouped mantissas.
+ */
+static int compute_mantissa_size_final(int mant_cnt[5])
+{
+ // bap=1 : 3 mantissas in 5 bits
+ int bits = (mant_cnt[1] / 3) * 5;
+ // bap=2 : 3 mantissas in 7 bits
+ // bap=4 : 2 mantissas in 7 bits
+ bits += ((mant_cnt[2] / 3) + (mant_cnt[4] >> 1)) * 7;
+ // bap=3 : each mantissa is 3 bits
+ bits += mant_cnt[3] * 3;
+ return bits;
+}
+
+
/**
* Calculate masking curve based on the final exponents.
* Also calculate the power spectral densities to use in future calculations.
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
AC3Block *block = &s->blocks[blk];
for (ch = 0; ch < s->channels; ch++) {
- if (block->exp_strategy[ch] == EXP_REUSE) {
- AC3Block *block1 = &s->blocks[blk-1];
- memcpy(block->psd[ch], block1->psd[ch], AC3_MAX_COEFS*sizeof(block->psd[0][0]));
- memcpy(block->mask[ch], block1->mask[ch], AC3_CRITICAL_BANDS*sizeof(block->mask[0][0]));
- } else {
+ /* We only need psd and mask for calculating bap.
+ Since we currently do not calculate bap when exponent
+ strategy is EXP_REUSE we do not need to calculate psd or mask. */
+ if (s->exp_strategy[ch][blk] != EXP_REUSE) {
ff_ac3_bit_alloc_calc_psd(block->exp[ch], 0,
s->nb_coefs[ch],
block->psd[ch], block->band_psd[ch]);
* @return the number of bits needed for mantissas if the given SNR offset is
* is used.
*/
-static int bit_alloc(AC3EncodeContext *s,
- int snr_offset)
+static int bit_alloc(AC3EncodeContext *s, int snr_offset)
{
int blk, ch;
int mantissa_bits;
+ int mant_cnt[5];
snr_offset = (snr_offset - 240) << 2;
mantissa_bits = 0;
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
AC3Block *block = &s->blocks[blk];
- s->mant1_cnt = 0;
- s->mant2_cnt = 0;
- s->mant4_cnt = 0;
+ // initialize grouped mantissa counts. these are set so that they are
+ // padded to the next whole group size when bits are counted in
+ // compute_mantissa_size_final
+ mant_cnt[0] = mant_cnt[3] = 0;
+ mant_cnt[1] = mant_cnt[2] = 2;
+ mant_cnt[4] = 1;
for (ch = 0; ch < s->channels; ch++) {
- ff_ac3_bit_alloc_calc_bap(block->mask[ch], block->psd[ch], 0,
- s->nb_coefs[ch], snr_offset,
- s->bit_alloc.floor, ff_ac3_bap_tab,
- block->bap[ch]);
- mantissa_bits += compute_mantissa_size(s, block->bap[ch], s->nb_coefs[ch]);
+ /* Currently the only bit allocation parameters which vary across
+ blocks within a frame are the exponent values. We can take
+ advantage of that by reusing the bit allocation pointers
+ whenever we reuse exponents. */
+ if (s->exp_strategy[ch][blk] == EXP_REUSE) {
+ memcpy(block->bap[ch], s->blocks[blk-1].bap[ch], AC3_MAX_COEFS);
+ } else {
+ ff_ac3_bit_alloc_calc_bap(block->mask[ch], block->psd[ch], 0,
+ s->nb_coefs[ch], snr_offset,
+ s->bit_alloc.floor, ff_ac3_bap_tab,
+ block->bap[ch]);
+ }
+ mantissa_bits += compute_mantissa_size(mant_cnt, block->bap[ch], s->nb_coefs[ch]);
}
+ mantissa_bits += compute_mantissa_size_final(mant_cnt);
}
return mantissa_bits;
}
{
int ch;
int bits_left;
- int snr_offset;
+ int snr_offset, snr_incr;
bits_left = 8 * s->frame_size - (s->frame_bits + s->exponent_bits);
snr_offset = s->coarse_snr_offset << 4;
+ /* if previous frame SNR offset was 1023, check if current frame can also
+ use SNR offset of 1023. if so, skip the search. */
+ if ((snr_offset | s->fine_snr_offset[0]) == 1023) {
+ if (bit_alloc(s, 1023) <= bits_left)
+ return 0;
+ }
+
while (snr_offset >= 0 &&
bit_alloc(s, snr_offset) > bits_left) {
snr_offset -= 64;
return AVERROR(EINVAL);
FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer);
- while (snr_offset + 64 <= 1023 &&
- bit_alloc(s, snr_offset + 64) <= bits_left) {
- snr_offset += 64;
- FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer);
- }
- while (snr_offset + 16 <= 1023 &&
- bit_alloc(s, snr_offset + 16) <= bits_left) {
- snr_offset += 16;
- FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer);
- }
- while (snr_offset + 4 <= 1023 &&
- bit_alloc(s, snr_offset + 4) <= bits_left) {
- snr_offset += 4;
- FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer);
- }
- while (snr_offset + 1 <= 1023 &&
- bit_alloc(s, snr_offset + 1) <= bits_left) {
- snr_offset++;
- FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer);
+ for (snr_incr = 64; snr_incr > 0; snr_incr >>= 2) {
+ while (snr_offset + snr_incr <= 1023 &&
+ bit_alloc(s, snr_offset + snr_incr) <= bits_left) {
+ snr_offset += snr_incr;
+ FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer);
+ }
}
FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer);
reset_block_bap(s);
}
+/**
+ * Downgrade exponent strategies to reduce the bits used by the exponents.
+ * This is a fallback for when bit allocation fails with the normal exponent
+ * strategies. Each time this function is run it only downgrades the
+ * strategy in 1 channel of 1 block.
+ * @return non-zero if downgrade was unsuccessful
+ */
+static int downgrade_exponents(AC3EncodeContext *s)
+{
+ int ch, blk;
+
+ for (ch = 0; ch < s->fbw_channels; ch++) {
+ for (blk = AC3_MAX_BLOCKS-1; blk >= 0; blk--) {
+ if (s->exp_strategy[ch][blk] == EXP_D15) {
+ s->exp_strategy[ch][blk] = EXP_D25;
+ return 0;
+ }
+ }
+ }
+ for (ch = 0; ch < s->fbw_channels; ch++) {
+ for (blk = AC3_MAX_BLOCKS-1; blk >= 0; blk--) {
+ if (s->exp_strategy[ch][blk] == EXP_D25) {
+ s->exp_strategy[ch][blk] = EXP_D45;
+ return 0;
+ }
+ }
+ }
+ for (ch = 0; ch < s->fbw_channels; ch++) {
+ /* block 0 cannot reuse exponents, so only downgrade D45 to REUSE if
+ the block number > 0 */
+ for (blk = AC3_MAX_BLOCKS-1; blk > 0; blk--) {
+ if (s->exp_strategy[ch][blk] > EXP_REUSE) {
+ s->exp_strategy[ch][blk] = EXP_REUSE;
+ return 0;
+ }
+ }
+ }
+ return -1;
+}
+
+
+/**
+ * Reduce the bandwidth to reduce the number of bits used for a given SNR offset.
+ * This is a second fallback for when bit allocation still fails after exponents
+ * have been downgraded.
+ * @return non-zero if bandwidth reduction was unsuccessful
+ */
+static int reduce_bandwidth(AC3EncodeContext *s, int min_bw_code)
+{
+ int ch;
+
+ if (s->bandwidth_code[0] > min_bw_code) {
+ for (ch = 0; ch < s->fbw_channels; ch++) {
+ s->bandwidth_code[ch]--;
+ s->nb_coefs[ch] = s->bandwidth_code[ch] * 3 + 73;
+ }
+ return 0;
+ }
+ return -1;
+}
+
+
/**
* Perform bit allocation search.
* Finds the SNR offset value that maximizes quality and fits in the specified
*/
static int compute_bit_allocation(AC3EncodeContext *s)
{
+ int ret;
+
count_frame_bits(s);
bit_alloc_masking(s);
- return cbr_bit_allocation(s);
+ ret = cbr_bit_allocation(s);
+ while (ret) {
+ /* fallback 1: downgrade exponents */
+ if (!downgrade_exponents(s)) {
+ extract_exponents(s);
+ encode_exponents(s);
+ group_exponents(s);
+ ret = compute_bit_allocation(s);
+ continue;
+ }
+
+ /* fallback 2: reduce bandwidth */
+ /* only do this if the user has not specified a specific cutoff
+ frequency */
+ if (!s->cutoff && !reduce_bandwidth(s, 0)) {
+ process_exponents(s);
+ ret = compute_bit_allocation(s);
+ continue;
+ }
+
+ /* fallbacks were not enough... */
+ break;
+ }
+
+ return ret;
}
/**
* Quantize a set of mantissas for a single channel in a single block.
*/
-static void quantize_mantissas_blk_ch(AC3EncodeContext *s,
- int32_t *mdct_coef, int8_t exp_shift,
- uint8_t *exp, uint8_t *bap,
- uint16_t *qmant, int n)
+static void quantize_mantissas_blk_ch(AC3EncodeContext *s, int32_t *fixed_coef,
+ int8_t exp_shift, uint8_t *exp,
+ uint8_t *bap, uint16_t *qmant, int n)
{
int i;
for (i = 0; i < n; i++) {
int v;
- int c = mdct_coef[i];
+ int c = fixed_coef[i];
int e = exp[i] - exp_shift;
int b = bap[i];
switch (b) {
s->qmant1_ptr = s->qmant2_ptr = s->qmant4_ptr = NULL;
for (ch = 0; ch < s->channels; ch++) {
- quantize_mantissas_blk_ch(s, block->mdct_coef[ch], block->exp_shift[ch],
+ quantize_mantissas_blk_ch(s, block->fixed_coef[ch], block->exp_shift[ch],
block->exp[ch], block->bap[ch],
block->qmant[ch], s->nb_coefs[ch]);
}
/**
* Write one audio block to the output bitstream.
*/
-static void output_audio_block(AC3EncodeContext *s,
- int block_num)
+static void output_audio_block(AC3EncodeContext *s, int blk)
{
int ch, i, baie, rbnd;
- AC3Block *block = &s->blocks[block_num];
+ AC3Block *block = &s->blocks[blk];
/* block switching */
for (ch = 0; ch < s->fbw_channels; ch++)
put_bits(&s->pb, 1, 0);
/* channel coupling */
- if (!block_num) {
+ if (!blk) {
put_bits(&s->pb, 1, 1); /* coupling strategy present */
put_bits(&s->pb, 1, 0); /* no coupling strategy */
} else {
/* stereo rematrixing */
if (s->channel_mode == AC3_CHMODE_STEREO) {
- if (!block_num) {
- /* first block must define rematrixing (rematstr) */
- put_bits(&s->pb, 1, 1);
-
- /* dummy rematrixing rematflg(1:4)=0 */
+ put_bits(&s->pb, 1, block->new_rematrixing_strategy);
+ if (block->new_rematrixing_strategy) {
+ /* rematrixing flags */
for (rbnd = 0; rbnd < 4; rbnd++)
- put_bits(&s->pb, 1, 0);
- } else {
- /* no matrixing (but should be used in the future) */
- put_bits(&s->pb, 1, 0);
+ put_bits(&s->pb, 1, block->rematrixing_flags[rbnd]);
}
}
/* exponent strategy */
for (ch = 0; ch < s->fbw_channels; ch++)
- put_bits(&s->pb, 2, block->exp_strategy[ch]);
+ put_bits(&s->pb, 2, s->exp_strategy[ch][blk]);
if (s->lfe_on)
- put_bits(&s->pb, 1, block->exp_strategy[s->lfe_channel]);
+ put_bits(&s->pb, 1, s->exp_strategy[s->lfe_channel][blk]);
/* bandwidth */
for (ch = 0; ch < s->fbw_channels; ch++) {
- if (block->exp_strategy[ch] != EXP_REUSE)
+ if (s->exp_strategy[ch][blk] != EXP_REUSE)
put_bits(&s->pb, 6, s->bandwidth_code[ch]);
}
/* exponents */
for (ch = 0; ch < s->channels; ch++) {
- if (block->exp_strategy[ch] == EXP_REUSE)
+ int nb_groups;
+
+ if (s->exp_strategy[ch][blk] == EXP_REUSE)
continue;
/* DC exponent */
put_bits(&s->pb, 4, block->grouped_exp[ch][0]);
/* exponent groups */
- for (i = 1; i <= block->num_exp_groups[ch]; i++)
+ nb_groups = exponent_group_tab[s->exp_strategy[ch][blk]-1][s->nb_coefs[ch]];
+ for (i = 1; i <= nb_groups; i++)
put_bits(&s->pb, 7, block->grouped_exp[ch][i]);
/* gain range info */
}
/* bit allocation info */
- baie = (block_num == 0);
+ baie = (blk == 0);
put_bits(&s->pb, 1, baie);
if (baie) {
put_bits(&s->pb, 2, s->slow_decay_code);
*/
static void output_frame_end(AC3EncodeContext *s)
{
- int frame_size, frame_size_58, pad_bytes, crc1, crc2, crc_inv;
+ const AVCRC *crc_ctx = av_crc_get_table(AV_CRC_16_ANSI);
+ int frame_size_58, pad_bytes, crc1, crc2_partial, crc2, crc_inv;
uint8_t *frame;
- frame_size = s->frame_size;
- frame_size_58 = ((frame_size >> 2) + (frame_size >> 4)) << 1;
+ frame_size_58 = ((s->frame_size >> 2) + (s->frame_size >> 4)) << 1;
/* pad the remainder of the frame with zeros */
flush_put_bits(&s->pb);
/* compute crc1 */
/* this is not so easy because it is at the beginning of the data... */
- crc1 = av_bswap16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0,
- frame + 4, frame_size_58 - 4));
- /* XXX: could precompute crc_inv */
- crc_inv = pow_poly((CRC16_POLY >> 1), (8 * frame_size_58) - 16, CRC16_POLY);
+ crc1 = av_bswap16(av_crc(crc_ctx, 0, frame + 4, frame_size_58 - 4));
+ crc_inv = s->crc_inv[s->frame_size > s->frame_size_min];
crc1 = mul_poly(crc_inv, crc1, CRC16_POLY);
AV_WB16(frame + 2, crc1);
/* compute crc2 */
- crc2 = av_bswap16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0,
- frame + frame_size_58,
- frame_size - frame_size_58 - 2));
- AV_WB16(frame + frame_size - 2, crc2);
+ crc2_partial = av_crc(crc_ctx, 0, frame + frame_size_58,
+ s->frame_size - frame_size_58 - 3);
+ crc2 = av_crc(crc_ctx, crc2_partial, frame + s->frame_size - 3, 1);
+ /* ensure crc2 does not match sync word by flipping crcrsv bit if needed */
+ if (crc2 == 0x770B) {
+ frame[s->frame_size - 3] ^= 0x1;
+ crc2 = av_crc(crc_ctx, crc2_partial, frame + s->frame_size - 3, 1);
+ }
+ crc2 = av_bswap16(crc2);
+ AV_WB16(frame + s->frame_size - 2, crc2);
}
/**
* Write the frame to the output bitstream.
*/
-static void output_frame(AC3EncodeContext *s,
- unsigned char *frame)
+static void output_frame(AC3EncodeContext *s, unsigned char *frame)
{
int blk;
/**
* Encode a single AC-3 frame.
*/
-static int ac3_encode_frame(AVCodecContext *avctx,
- unsigned char *frame, int buf_size, void *data)
+static int ac3_encode_frame(AVCodecContext *avctx, unsigned char *frame,
+ int buf_size, void *data)
{
AC3EncodeContext *s = avctx->priv_data;
- const int16_t *samples = data;
+ const SampleType *samples = data;
int ret;
if (s->bit_alloc.sr_code == 1)
apply_mdct(s);
+ compute_rematrixing_strategy(s);
+
+ scale_coefficients(s);
+
+ apply_rematrixing(s);
+
process_exponents(s);
ret = compute_bit_allocation(s);
av_freep(&s->bap_buffer);
av_freep(&s->bap1_buffer);
av_freep(&s->mdct_coef_buffer);
+ av_freep(&s->fixed_coef_buffer);
av_freep(&s->exp_buffer);
av_freep(&s->grouped_exp_buffer);
av_freep(&s->psd_buffer);
AC3Block *block = &s->blocks[blk];
av_freep(&block->bap);
av_freep(&block->mdct_coef);
+ av_freep(&block->fixed_coef);
av_freep(&block->exp);
av_freep(&block->grouped_exp);
av_freep(&block->psd);
s->bit_rate = avctx->bit_rate;
s->frame_size_code = i << 1;
+ /* validate cutoff */
+ if (avctx->cutoff < 0) {
+ av_log(avctx, AV_LOG_ERROR, "invalid cutoff frequency\n");
+ return AVERROR(EINVAL);
+ }
+ s->cutoff = avctx->cutoff;
+ if (s->cutoff > (s->sample_rate >> 1))
+ s->cutoff = s->sample_rate >> 1;
+
return 0;
}
* The user can optionally supply a cutoff frequency. Otherwise an appropriate
* default value will be used.
*/
-static av_cold void set_bandwidth(AC3EncodeContext *s, int cutoff)
+static av_cold void set_bandwidth(AC3EncodeContext *s)
{
int ch, bw_code;
- if (cutoff) {
+ if (s->cutoff) {
/* calculate bandwidth based on user-specified cutoff frequency */
int fbw_coeffs;
- cutoff = av_clip(cutoff, 1, s->sample_rate >> 1);
- fbw_coeffs = cutoff * 2 * AC3_MAX_COEFS / s->sample_rate;
+ fbw_coeffs = s->cutoff * 2 * AC3_MAX_COEFS / s->sample_rate;
bw_code = av_clip((fbw_coeffs - 73) / 3, 0, 60);
} else {
/* use default bandwidth setting */
alloc_fail);
for (ch = 0; ch < s->channels; ch++) {
+ /* arrangement: block, channel, coeff */
block->bap[ch] = &s->bap_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)];
block->mdct_coef[ch] = &s->mdct_coef_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)];
- block->exp[ch] = &s->exp_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)];
block->grouped_exp[ch] = &s->grouped_exp_buffer[128 * (blk * s->channels + ch)];
block->psd[ch] = &s->psd_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)];
block->band_psd[ch] = &s->band_psd_buffer [64 * (blk * s->channels + ch)];
block->mask[ch] = &s->mask_buffer [64 * (blk * s->channels + ch)];
block->qmant[ch] = &s->qmant_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)];
+
+ /* arrangement: channel, block, coeff */
+ block->exp[ch] = &s->exp_buffer [AC3_MAX_COEFS * (AC3_MAX_BLOCKS * ch + blk)];
+ }
+ }
+
+ if (CONFIG_AC3ENC_FLOAT) {
+ FF_ALLOC_OR_GOTO(avctx, s->fixed_coef_buffer, AC3_MAX_BLOCKS * s->channels *
+ AC3_MAX_COEFS * sizeof(*s->fixed_coef_buffer), alloc_fail);
+ for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
+ AC3Block *block = &s->blocks[blk];
+ FF_ALLOCZ_OR_GOTO(avctx, block->fixed_coef, s->channels *
+ sizeof(*block->fixed_coef), alloc_fail);
+ for (ch = 0; ch < s->channels; ch++)
+ block->fixed_coef[ch] = &s->fixed_coef_buffer[AC3_MAX_COEFS * (blk * s->channels + ch)];
+ }
+ } else {
+ for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
+ AC3Block *block = &s->blocks[blk];
+ FF_ALLOCZ_OR_GOTO(avctx, block->fixed_coef, s->channels *
+ sizeof(*block->fixed_coef), alloc_fail);
+ for (ch = 0; ch < s->channels; ch++)
+ block->fixed_coef[ch] = (int32_t *)block->mdct_coef[ch];
}
}
static av_cold int ac3_encode_init(AVCodecContext *avctx)
{
AC3EncodeContext *s = avctx->priv_data;
- int ret;
+ int ret, frame_size_58;
avctx->frame_size = AC3_FRAME_SIZE;
- ac3_common_init();
+ ff_ac3_common_init();
ret = validate_options(avctx, s);
if (ret)
s->samples_written = 0;
s->frame_size = s->frame_size_min;
- set_bandwidth(s, avctx->cutoff);
+ /* calculate crc_inv for both possible frame sizes */
+ frame_size_58 = (( s->frame_size >> 2) + ( s->frame_size >> 4)) << 1;
+ s->crc_inv[0] = pow_poly((CRC16_POLY >> 1), (8 * frame_size_58) - 16, CRC16_POLY);
+ if (s->bit_alloc.sr_code == 1) {
+ frame_size_58 = (((s->frame_size+2) >> 2) + ((s->frame_size+2) >> 4)) << 1;
+ s->crc_inv[1] = pow_poly((CRC16_POLY >> 1), (8 * frame_size_58) - 16, CRC16_POLY);
+ }
+
+ set_bandwidth(s);
+
+ rematrixing_init(s);
+
+ exponent_init(s);
bit_alloc_init(s);
- s->mdct.avctx = avctx;
- ret = mdct_init(&s->mdct, 9);
+ ret = mdct_init(avctx, &s->mdct, 9);
if (ret)
goto init_fail;
ac3_encode_close(avctx);
return ret;
}
-
-
-#ifdef TEST
-/*************************************************************************/
-/* TEST */
-
-#include "libavutil/lfg.h"
-
-#define FN (MDCT_SAMPLES/4)
-
-
-static void fft_test(AVLFG *lfg)
-{
- IComplex in[FN], in1[FN];
- int k, n, i;
- float sum_re, sum_im, a;
-
- for (i = 0; i < FN; i++) {
- in[i].re = av_lfg_get(lfg) % 65535 - 32767;
- in[i].im = av_lfg_get(lfg) % 65535 - 32767;
- in1[i] = in[i];
- }
- fft(in, 7);
-
- /* do it by hand */
- for (k = 0; k < FN; k++) {
- sum_re = 0;
- sum_im = 0;
- for (n = 0; n < FN; n++) {
- a = -2 * M_PI * (n * k) / FN;
- sum_re += in1[n].re * cos(a) - in1[n].im * sin(a);
- sum_im += in1[n].re * sin(a) + in1[n].im * cos(a);
- }
- av_log(NULL, AV_LOG_DEBUG, "%3d: %6d,%6d %6.0f,%6.0f\n",
- k, in[k].re, in[k].im, sum_re / FN, sum_im / FN);
- }
-}
-
-
-static void mdct_test(AVLFG *lfg)
-{
- int16_t input[MDCT_SAMPLES];
- int32_t output[AC3_MAX_COEFS];
- float input1[MDCT_SAMPLES];
- float output1[AC3_MAX_COEFS];
- float s, a, err, e, emax;
- int i, k, n;
-
- for (i = 0; i < MDCT_SAMPLES; i++) {
- input[i] = (av_lfg_get(lfg) % 65535 - 32767) * 9 / 10;
- input1[i] = input[i];
- }
-
- mdct512(output, input);
-
- /* do it by hand */
- for (k = 0; k < AC3_MAX_COEFS; k++) {
- s = 0;
- for (n = 0; n < MDCT_SAMPLES; n++) {
- a = (2*M_PI*(2*n+1+MDCT_SAMPLES/2)*(2*k+1) / (4 * MDCT_SAMPLES));
- s += input1[n] * cos(a);
- }
- output1[k] = -2 * s / MDCT_SAMPLES;
- }
-
- err = 0;
- emax = 0;
- for (i = 0; i < AC3_MAX_COEFS; i++) {
- av_log(NULL, AV_LOG_DEBUG, "%3d: %7d %7.0f\n", i, output[i], output1[i]);
- e = output[i] - output1[i];
- if (e > emax)
- emax = e;
- err += e * e;
- }
- av_log(NULL, AV_LOG_DEBUG, "err2=%f emax=%f\n", err / AC3_MAX_COEFS, emax);
-}
-
-
-int main(void)
-{
- AVLFG lfg;
-
- av_log_set_level(AV_LOG_DEBUG);
- mdct_init(9);
-
- fft_test(&lfg);
- mdct_test(&lfg);
-
- return 0;
-}
-#endif /* TEST */
-
-
-AVCodec ac3_encoder = {
- "ac3",
- AVMEDIA_TYPE_AUDIO,
- CODEC_ID_AC3,
- sizeof(AC3EncodeContext),
- ac3_encode_init,
- ac3_encode_frame,
- ac3_encode_close,
- NULL,
- .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
- .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52A (AC-3)"),
- .channel_layouts = (const int64_t[]){
- AV_CH_LAYOUT_MONO,
- AV_CH_LAYOUT_STEREO,
- AV_CH_LAYOUT_2_1,
- AV_CH_LAYOUT_SURROUND,
- AV_CH_LAYOUT_2_2,
- AV_CH_LAYOUT_QUAD,
- AV_CH_LAYOUT_4POINT0,
- AV_CH_LAYOUT_5POINT0,
- AV_CH_LAYOUT_5POINT0_BACK,
- (AV_CH_LAYOUT_MONO | AV_CH_LOW_FREQUENCY),
- (AV_CH_LAYOUT_STEREO | AV_CH_LOW_FREQUENCY),
- (AV_CH_LAYOUT_2_1 | AV_CH_LOW_FREQUENCY),
- (AV_CH_LAYOUT_SURROUND | AV_CH_LOW_FREQUENCY),
- (AV_CH_LAYOUT_2_2 | AV_CH_LOW_FREQUENCY),
- (AV_CH_LAYOUT_QUAD | AV_CH_LOW_FREQUENCY),
- (AV_CH_LAYOUT_4POINT0 | AV_CH_LOW_FREQUENCY),
- AV_CH_LAYOUT_5POINT1,
- AV_CH_LAYOUT_5POINT1_BACK,
- 0 },
-};