2 * The simplest AC-3 encoder
3 * Copyright (c) 2000 Fabrice Bellard
4 * Copyright (c) 2006-2010 Justin Ruggles <justin.ruggles@gmail.com>
5 * Copyright (c) 2006-2010 Prakash Punnoor <prakash@punnoor.de>
7 * This file is part of FFmpeg.
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 * fixed-point AC-3 encoder.
32 /** Scale a float value by 2^15, convert to an integer, and clip to range -32767..32767. */
33 #define FIX15(a) av_clip(SCALE_FLOAT(a, 15), -32767, 32767)
37 * Finalize MDCT and free allocated memory.
39 static av_cold void mdct_end(AC3MDCTContext *mdct)
42 av_freep(&mdct->costab);
43 av_freep(&mdct->sintab);
44 av_freep(&mdct->xcos1);
45 av_freep(&mdct->xsin1);
46 av_freep(&mdct->rot_tmp);
47 av_freep(&mdct->cplx_tmp);
52 * Initialize FFT tables.
53 * @param ln log2(FFT size)
55 static av_cold int fft_init(AVCodecContext *avctx, AC3MDCTContext *mdct, int ln)
63 FF_ALLOC_OR_GOTO(avctx, mdct->costab, n2 * sizeof(*mdct->costab), fft_alloc_fail);
64 FF_ALLOC_OR_GOTO(avctx, mdct->sintab, n2 * sizeof(*mdct->sintab), fft_alloc_fail);
66 for (i = 0; i < n2; i++) {
67 alpha = 2.0 * M_PI * i / n;
68 mdct->costab[i] = FIX15(cos(alpha));
69 mdct->sintab[i] = FIX15(sin(alpha));
75 return AVERROR(ENOMEM);
80 * Initialize MDCT tables.
81 * @param nbits log2(MDCT size)
83 static av_cold int mdct_init(AVCodecContext *avctx, AC3MDCTContext *mdct,
93 ret = fft_init(avctx, mdct, nbits - 2);
97 mdct->window = ff_ac3_window;
99 FF_ALLOC_OR_GOTO(avctx, mdct->xcos1, n4 * sizeof(*mdct->xcos1), mdct_alloc_fail);
100 FF_ALLOC_OR_GOTO(avctx, mdct->xsin1, n4 * sizeof(*mdct->xsin1), mdct_alloc_fail);
101 FF_ALLOC_OR_GOTO(avctx, mdct->rot_tmp, n * sizeof(*mdct->rot_tmp), mdct_alloc_fail);
102 FF_ALLOC_OR_GOTO(avctx, mdct->cplx_tmp, n4 * sizeof(*mdct->cplx_tmp), mdct_alloc_fail);
104 for (i = 0; i < n4; i++) {
105 float alpha = 2.0 * M_PI * (i + 1.0 / 8.0) / n;
106 mdct->xcos1[i] = FIX15(-cos(alpha));
107 mdct->xsin1[i] = FIX15(-sin(alpha));
113 return AVERROR(ENOMEM);
118 #define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \
120 int ax, ay, bx, by; \
125 pre = (bx + ax) >> 1; \
126 pim = (by + ay) >> 1; \
127 qre = (bx - ax) >> 1; \
128 qim = (by - ay) >> 1; \
132 /** Complex multiply */
133 #define CMUL(pre, pim, are, aim, bre, bim) \
135 pre = (MUL16(are, bre) - MUL16(aim, bim)) >> 15; \
136 pim = (MUL16(are, bim) + MUL16(bre, aim)) >> 15; \
141 * Calculate a 2^n point complex FFT on 2^ln points.
142 * @param z complex input/output samples
143 * @param ln log2(FFT size)
145 static void fft(AC3MDCTContext *mdct, IComplex *z, int ln)
149 register IComplex *p,*q;
155 for (j = 0; j < np; j++) {
156 int k = av_reverse[j] >> (8 - ln);
158 FFSWAP(IComplex, z[k], z[j]);
166 BF(p[0].re, p[0].im, p[1].re, p[1].im,
167 p[0].re, p[0].im, p[1].re, p[1].im);
176 BF(p[0].re, p[0].im, p[2].re, p[2].im,
177 p[0].re, p[0].im, p[2].re, p[2].im);
178 BF(p[1].re, p[1].im, p[3].re, p[3].im,
179 p[1].re, p[1].im, p[3].im, -p[3].re);
191 for (j = 0; j < nblocks; j++) {
192 BF(p->re, p->im, q->re, q->im,
193 p->re, p->im, q->re, q->im);
196 for(l = nblocks; l < np2; l += nblocks) {
197 CMUL(tmp_re, tmp_im, mdct->costab[l], -mdct->sintab[l], q->re, q->im);
198 BF(p->re, p->im, q->re, q->im,
199 p->re, p->im, tmp_re, tmp_im);
206 nblocks = nblocks >> 1;
207 nloops = nloops << 1;
213 * Calculate a 512-point MDCT
214 * @param out 256 output frequency coefficients
215 * @param in 512 windowed input audio samples
217 static void mdct512(AC3MDCTContext *mdct, int32_t *out, int16_t *in)
219 int i, re, im, n, n2, n4;
220 int16_t *rot = mdct->rot_tmp;
221 IComplex *x = mdct->cplx_tmp;
223 n = 1 << mdct->nbits;
227 /* shift to simplify computations */
228 for (i = 0; i <n4; i++)
229 rot[i] = -in[i + 3*n4];
230 memcpy(&rot[n4], &in[0], 3*n4*sizeof(*in));
233 for (i = 0; i < n4; i++) {
234 re = ((int)rot[ 2*i] - (int)rot[ n-1-2*i]) >> 1;
235 im = -((int)rot[n2+2*i] - (int)rot[n2-1-2*i]) >> 1;
236 CMUL(x[i].re, x[i].im, re, im, -mdct->xcos1[i], mdct->xsin1[i]);
239 fft(mdct, x, mdct->nbits - 2);
242 for (i = 0; i < n4; i++) {
245 CMUL(out[n2-1-2*i], out[2*i], re, im, mdct->xsin1[i], mdct->xcos1[i]);
251 * Apply KBD window to input samples prior to MDCT.
253 static void apply_window(int16_t *output, const int16_t *input,
254 const int16_t *window, int n)
259 for (i = 0; i < n2; i++) {
260 output[i] = MUL16(input[i], window[i]) >> 15;
261 output[n-i-1] = MUL16(input[n-i-1], window[i]) >> 15;
267 * Calculate the log2() of the maximum absolute value in an array.
268 * @param tab input array
269 * @param n number of values in the array
270 * @return log2(max(abs(tab[])))
272 static int log2_tab(int16_t *tab, int n)
277 for (i = 0; i < n; i++)
285 * Left-shift each value in an array by a specified amount.
286 * @param tab input array
287 * @param n number of values in the array
288 * @param lshift left shift amount. a negative value means right shift.
290 static void lshift_tab(int16_t *tab, int n, int lshift)
295 for (i = 0; i < n; i++)
297 } else if (lshift < 0) {
299 for (i = 0; i < n; i++)
306 * Normalize the input samples to use the maximum available precision.
307 * This assumes signed 16-bit input samples. Exponents are reduced by 9 to
308 * match the 24-bit internal precision for MDCT coefficients.
310 * @return exponent shift
312 static int normalize_samples(AC3EncodeContext *s)
314 int v = 14 - log2_tab(s->windowed_samples, AC3_WINDOW_SIZE);
316 lshift_tab(s->windowed_samples, AC3_WINDOW_SIZE, v);
322 /*************************************************************************/
325 #include "libavutil/lfg.h"
328 #define MDCT_SAMPLES (1 << MDCT_NBITS)
329 #define FN (MDCT_SAMPLES/4)
332 static void fft_test(AC3MDCTContext *mdct, AVLFG *lfg)
334 IComplex in[FN], in1[FN];
336 float sum_re, sum_im, a;
338 for (i = 0; i < FN; i++) {
339 in[i].re = av_lfg_get(lfg) % 65535 - 32767;
340 in[i].im = av_lfg_get(lfg) % 65535 - 32767;
346 for (k = 0; k < FN; k++) {
349 for (n = 0; n < FN; n++) {
350 a = -2 * M_PI * (n * k) / FN;
351 sum_re += in1[n].re * cos(a) - in1[n].im * sin(a);
352 sum_im += in1[n].re * sin(a) + in1[n].im * cos(a);
354 av_log(NULL, AV_LOG_DEBUG, "%3d: %6d,%6d %6.0f,%6.0f\n",
355 k, in[k].re, in[k].im, sum_re / FN, sum_im / FN);
360 static void mdct_test(AC3MDCTContext *mdct, AVLFG *lfg)
362 int16_t input[MDCT_SAMPLES];
363 int32_t output[AC3_MAX_COEFS];
364 float input1[MDCT_SAMPLES];
365 float output1[AC3_MAX_COEFS];
366 float s, a, err, e, emax;
369 for (i = 0; i < MDCT_SAMPLES; i++) {
370 input[i] = (av_lfg_get(lfg) % 65535 - 32767) * 9 / 10;
371 input1[i] = input[i];
374 mdct512(mdct, output, input);
377 for (k = 0; k < AC3_MAX_COEFS; k++) {
379 for (n = 0; n < MDCT_SAMPLES; n++) {
380 a = (2*M_PI*(2*n+1+MDCT_SAMPLES/2)*(2*k+1) / (4 * MDCT_SAMPLES));
381 s += input1[n] * cos(a);
383 output1[k] = -2 * s / MDCT_SAMPLES;
388 for (i = 0; i < AC3_MAX_COEFS; i++) {
389 av_log(NULL, AV_LOG_DEBUG, "%3d: %7d %7.0f\n", i, output[i], output1[i]);
390 e = output[i] - output1[i];
395 av_log(NULL, AV_LOG_DEBUG, "err2=%f emax=%f\n", err / AC3_MAX_COEFS, emax);
405 av_log_set_level(AV_LOG_DEBUG);
408 fft_test(&mdct, &lfg);
409 mdct_test(&mdct, &lfg);
416 AVCodec ac3_encoder = {
420 sizeof(AC3EncodeContext),
425 .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
426 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52A (AC-3)"),
427 .channel_layouts = ac3_channel_layouts,