ac3enc: Reshuffle some float/fixed-mode ifdefs to avoid a dummy function
[libav.git] / libavcodec / ac3enc_template.c
1 /*
2 * AC-3 encoder float/fixed template
3 * Copyright (c) 2000 Fabrice Bellard
4 * Copyright (c) 2006-2011 Justin Ruggles <justin.ruggles@gmail.com>
5 * Copyright (c) 2006-2010 Prakash Punnoor <prakash@punnoor.de>
6 *
7 * This file is part of Libav.
8 *
9 * Libav 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.
13 *
14 * Libav 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.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with Libav; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 */
23
24 /**
25 * @file
26 * AC-3 encoder float/fixed template
27 */
28
29 #include <stdint.h>
30
31 #include "libavutil/attributes.h"
32 #include "libavutil/internal.h"
33
34 #include "audiodsp.h"
35 #include "internal.h"
36 #include "ac3enc.h"
37 #include "eac3enc.h"
38
39 /* prototypes for static functions in ac3enc_fixed.c and ac3enc_float.c */
40
41 static void scale_coefficients(AC3EncodeContext *s);
42
43 static void clip_coefficients(AudioDSPContext *adsp, CoefType *coef,
44 unsigned int len);
45
46 static CoefType calc_cpl_coord(CoefSumType energy_ch, CoefSumType energy_cpl);
47
48
49 int AC3_NAME(allocate_sample_buffers)(AC3EncodeContext *s)
50 {
51 int ch;
52
53 FF_ALLOC_OR_GOTO(s->avctx, s->windowed_samples, AC3_WINDOW_SIZE *
54 sizeof(*s->windowed_samples), alloc_fail);
55 FF_ALLOC_OR_GOTO(s->avctx, s->planar_samples, s->channels * sizeof(*s->planar_samples),
56 alloc_fail);
57 for (ch = 0; ch < s->channels; ch++) {
58 FF_ALLOCZ_OR_GOTO(s->avctx, s->planar_samples[ch],
59 (AC3_FRAME_SIZE+AC3_BLOCK_SIZE) * sizeof(**s->planar_samples),
60 alloc_fail);
61 }
62
63 return 0;
64 alloc_fail:
65 return AVERROR(ENOMEM);
66 }
67
68
69 /*
70 * Copy input samples.
71 * Channels are reordered from Libav's default order to AC-3 order.
72 */
73 static void copy_input_samples(AC3EncodeContext *s, SampleType **samples)
74 {
75 int ch;
76
77 /* copy and remap input samples */
78 for (ch = 0; ch < s->channels; ch++) {
79 /* copy last 256 samples of previous frame to the start of the current frame */
80 memcpy(&s->planar_samples[ch][0], &s->planar_samples[ch][AC3_BLOCK_SIZE * s->num_blocks],
81 AC3_BLOCK_SIZE * sizeof(s->planar_samples[0][0]));
82
83 /* copy new samples for current frame */
84 memcpy(&s->planar_samples[ch][AC3_BLOCK_SIZE],
85 samples[s->channel_map[ch]],
86 AC3_BLOCK_SIZE * s->num_blocks * sizeof(s->planar_samples[0][0]));
87 }
88 }
89
90
91 /*
92 * Apply the MDCT to input samples to generate frequency coefficients.
93 * This applies the KBD window and normalizes the input to reduce precision
94 * loss due to fixed-point calculations.
95 */
96 static void apply_mdct(AC3EncodeContext *s)
97 {
98 int blk, ch;
99
100 for (ch = 0; ch < s->channels; ch++) {
101 for (blk = 0; blk < s->num_blocks; blk++) {
102 AC3Block *block = &s->blocks[blk];
103 const SampleType *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE];
104
105 #if CONFIG_AC3ENC_FLOAT
106 s->fdsp.vector_fmul(s->windowed_samples, input_samples,
107 s->mdct_window, AC3_WINDOW_SIZE);
108 #else
109 s->ac3dsp.apply_window_int16(s->windowed_samples, input_samples,
110 s->mdct_window, AC3_WINDOW_SIZE);
111
112 if (s->fixed_point)
113 block->coeff_shift[ch+1] = normalize_samples(s);
114 #endif
115
116 s->mdct.mdct_calcw(&s->mdct, block->mdct_coef[ch+1],
117 s->windowed_samples);
118 }
119 }
120 }
121
122
123 /*
124 * Calculate coupling channel and coupling coordinates.
125 */
126 static void apply_channel_coupling(AC3EncodeContext *s)
127 {
128 LOCAL_ALIGNED_16(CoefType, cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
129 #if CONFIG_AC3ENC_FLOAT
130 LOCAL_ALIGNED_16(int32_t, fixed_cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
131 #else
132 int32_t (*fixed_cpl_coords)[AC3_MAX_CHANNELS][16] = cpl_coords;
133 #endif
134 int blk, ch, bnd, i, j;
135 CoefSumType energy[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][16] = {{{0}}};
136 int cpl_start, num_cpl_coefs;
137
138 memset(cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*cpl_coords));
139 #if CONFIG_AC3ENC_FLOAT
140 memset(fixed_cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*cpl_coords));
141 #endif
142
143 /* align start to 16-byte boundary. align length to multiple of 32.
144 note: coupling start bin % 4 will always be 1 */
145 cpl_start = s->start_freq[CPL_CH] - 1;
146 num_cpl_coefs = FFALIGN(s->num_cpl_subbands * 12 + 1, 32);
147 cpl_start = FFMIN(256, cpl_start + num_cpl_coefs) - num_cpl_coefs;
148
149 /* calculate coupling channel from fbw channels */
150 for (blk = 0; blk < s->num_blocks; blk++) {
151 AC3Block *block = &s->blocks[blk];
152 CoefType *cpl_coef = &block->mdct_coef[CPL_CH][cpl_start];
153 if (!block->cpl_in_use)
154 continue;
155 memset(cpl_coef, 0, num_cpl_coefs * sizeof(*cpl_coef));
156 for (ch = 1; ch <= s->fbw_channels; ch++) {
157 CoefType *ch_coef = &block->mdct_coef[ch][cpl_start];
158 if (!block->channel_in_cpl[ch])
159 continue;
160 for (i = 0; i < num_cpl_coefs; i++)
161 cpl_coef[i] += ch_coef[i];
162 }
163
164 /* coefficients must be clipped in order to be encoded */
165 clip_coefficients(&s->adsp, cpl_coef, num_cpl_coefs);
166 }
167
168 /* calculate energy in each band in coupling channel and each fbw channel */
169 /* TODO: possibly use SIMD to speed up energy calculation */
170 bnd = 0;
171 i = s->start_freq[CPL_CH];
172 while (i < s->cpl_end_freq) {
173 int band_size = s->cpl_band_sizes[bnd];
174 for (ch = CPL_CH; ch <= s->fbw_channels; ch++) {
175 for (blk = 0; blk < s->num_blocks; blk++) {
176 AC3Block *block = &s->blocks[blk];
177 if (!block->cpl_in_use || (ch > CPL_CH && !block->channel_in_cpl[ch]))
178 continue;
179 for (j = 0; j < band_size; j++) {
180 CoefType v = block->mdct_coef[ch][i+j];
181 MAC_COEF(energy[blk][ch][bnd], v, v);
182 }
183 }
184 }
185 i += band_size;
186 bnd++;
187 }
188
189 /* calculate coupling coordinates for all blocks for all channels */
190 for (blk = 0; blk < s->num_blocks; blk++) {
191 AC3Block *block = &s->blocks[blk];
192 if (!block->cpl_in_use)
193 continue;
194 for (ch = 1; ch <= s->fbw_channels; ch++) {
195 if (!block->channel_in_cpl[ch])
196 continue;
197 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
198 cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy[blk][ch][bnd],
199 energy[blk][CPL_CH][bnd]);
200 }
201 }
202 }
203
204 /* determine which blocks to send new coupling coordinates for */
205 for (blk = 0; blk < s->num_blocks; blk++) {
206 AC3Block *block = &s->blocks[blk];
207 AC3Block *block0 = blk ? &s->blocks[blk-1] : NULL;
208
209 memset(block->new_cpl_coords, 0, sizeof(block->new_cpl_coords));
210
211 if (block->cpl_in_use) {
212 /* send new coordinates if this is the first block, if previous
213 * block did not use coupling but this block does, the channels
214 * using coupling has changed from the previous block, or the
215 * coordinate difference from the last block for any channel is
216 * greater than a threshold value. */
217 if (blk == 0 || !block0->cpl_in_use) {
218 for (ch = 1; ch <= s->fbw_channels; ch++)
219 block->new_cpl_coords[ch] = 1;
220 } else {
221 for (ch = 1; ch <= s->fbw_channels; ch++) {
222 if (!block->channel_in_cpl[ch])
223 continue;
224 if (!block0->channel_in_cpl[ch]) {
225 block->new_cpl_coords[ch] = 1;
226 } else {
227 CoefSumType coord_diff = 0;
228 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
229 coord_diff += FFABS(cpl_coords[blk-1][ch][bnd] -
230 cpl_coords[blk ][ch][bnd]);
231 }
232 coord_diff /= s->num_cpl_bands;
233 if (coord_diff > NEW_CPL_COORD_THRESHOLD)
234 block->new_cpl_coords[ch] = 1;
235 }
236 }
237 }
238 }
239 }
240
241 /* calculate final coupling coordinates, taking into account reusing of
242 coordinates in successive blocks */
243 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
244 blk = 0;
245 while (blk < s->num_blocks) {
246 int av_uninit(blk1);
247 AC3Block *block = &s->blocks[blk];
248
249 if (!block->cpl_in_use) {
250 blk++;
251 continue;
252 }
253
254 for (ch = 1; ch <= s->fbw_channels; ch++) {
255 CoefSumType energy_ch, energy_cpl;
256 if (!block->channel_in_cpl[ch])
257 continue;
258 energy_cpl = energy[blk][CPL_CH][bnd];
259 energy_ch = energy[blk][ch][bnd];
260 blk1 = blk+1;
261 while (blk1 < s->num_blocks && !s->blocks[blk1].new_cpl_coords[ch]) {
262 if (s->blocks[blk1].cpl_in_use) {
263 energy_cpl += energy[blk1][CPL_CH][bnd];
264 energy_ch += energy[blk1][ch][bnd];
265 }
266 blk1++;
267 }
268 cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy_ch, energy_cpl);
269 }
270 blk = blk1;
271 }
272 }
273
274 /* calculate exponents/mantissas for coupling coordinates */
275 for (blk = 0; blk < s->num_blocks; blk++) {
276 AC3Block *block = &s->blocks[blk];
277 if (!block->cpl_in_use)
278 continue;
279
280 #if CONFIG_AC3ENC_FLOAT
281 s->ac3dsp.float_to_fixed24(fixed_cpl_coords[blk][1],
282 cpl_coords[blk][1],
283 s->fbw_channels * 16);
284 #endif
285 s->ac3dsp.extract_exponents(block->cpl_coord_exp[1],
286 fixed_cpl_coords[blk][1],
287 s->fbw_channels * 16);
288
289 for (ch = 1; ch <= s->fbw_channels; ch++) {
290 int bnd, min_exp, max_exp, master_exp;
291
292 if (!block->new_cpl_coords[ch])
293 continue;
294
295 /* determine master exponent */
296 min_exp = max_exp = block->cpl_coord_exp[ch][0];
297 for (bnd = 1; bnd < s->num_cpl_bands; bnd++) {
298 int exp = block->cpl_coord_exp[ch][bnd];
299 min_exp = FFMIN(exp, min_exp);
300 max_exp = FFMAX(exp, max_exp);
301 }
302 master_exp = ((max_exp - 15) + 2) / 3;
303 master_exp = FFMAX(master_exp, 0);
304 while (min_exp < master_exp * 3)
305 master_exp--;
306 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
307 block->cpl_coord_exp[ch][bnd] = av_clip(block->cpl_coord_exp[ch][bnd] -
308 master_exp * 3, 0, 15);
309 }
310 block->cpl_master_exp[ch] = master_exp;
311
312 /* quantize mantissas */
313 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
314 int cpl_exp = block->cpl_coord_exp[ch][bnd];
315 int cpl_mant = (fixed_cpl_coords[blk][ch][bnd] << (5 + cpl_exp + master_exp * 3)) >> 24;
316 if (cpl_exp == 15)
317 cpl_mant >>= 1;
318 else
319 cpl_mant -= 16;
320
321 block->cpl_coord_mant[ch][bnd] = cpl_mant;
322 }
323 }
324 }
325
326 if (CONFIG_EAC3_ENCODER && s->eac3)
327 ff_eac3_set_cpl_states(s);
328 }
329
330
331 /*
332 * Determine rematrixing flags for each block and band.
333 */
334 static void compute_rematrixing_strategy(AC3EncodeContext *s)
335 {
336 int nb_coefs;
337 int blk, bnd, i;
338 AC3Block *block, *block0;
339
340 if (s->channel_mode != AC3_CHMODE_STEREO)
341 return;
342
343 for (blk = 0; blk < s->num_blocks; blk++) {
344 block = &s->blocks[blk];
345 block->new_rematrixing_strategy = !blk;
346
347 block->num_rematrixing_bands = 4;
348 if (block->cpl_in_use) {
349 block->num_rematrixing_bands -= (s->start_freq[CPL_CH] <= 61);
350 block->num_rematrixing_bands -= (s->start_freq[CPL_CH] == 37);
351 if (blk && block->num_rematrixing_bands != block0->num_rematrixing_bands)
352 block->new_rematrixing_strategy = 1;
353 }
354 nb_coefs = FFMIN(block->end_freq[1], block->end_freq[2]);
355
356 if (!s->rematrixing_enabled) {
357 block0 = block;
358 continue;
359 }
360
361 for (bnd = 0; bnd < block->num_rematrixing_bands; bnd++) {
362 /* calculate calculate sum of squared coeffs for one band in one block */
363 int start = ff_ac3_rematrix_band_tab[bnd];
364 int end = FFMIN(nb_coefs, ff_ac3_rematrix_band_tab[bnd+1]);
365 CoefSumType sum[4] = {0,};
366 for (i = start; i < end; i++) {
367 CoefType lt = block->mdct_coef[1][i];
368 CoefType rt = block->mdct_coef[2][i];
369 CoefType md = lt + rt;
370 CoefType sd = lt - rt;
371 MAC_COEF(sum[0], lt, lt);
372 MAC_COEF(sum[1], rt, rt);
373 MAC_COEF(sum[2], md, md);
374 MAC_COEF(sum[3], sd, sd);
375 }
376
377 /* compare sums to determine if rematrixing will be used for this band */
378 if (FFMIN(sum[2], sum[3]) < FFMIN(sum[0], sum[1]))
379 block->rematrixing_flags[bnd] = 1;
380 else
381 block->rematrixing_flags[bnd] = 0;
382
383 /* determine if new rematrixing flags will be sent */
384 if (blk &&
385 block->rematrixing_flags[bnd] != block0->rematrixing_flags[bnd]) {
386 block->new_rematrixing_strategy = 1;
387 }
388 }
389 block0 = block;
390 }
391 }
392
393
394 int AC3_NAME(encode_frame)(AVCodecContext *avctx, AVPacket *avpkt,
395 const AVFrame *frame, int *got_packet_ptr)
396 {
397 AC3EncodeContext *s = avctx->priv_data;
398 int ret;
399
400 if (s->options.allow_per_frame_metadata) {
401 ret = ff_ac3_validate_metadata(s);
402 if (ret)
403 return ret;
404 }
405
406 if (s->bit_alloc.sr_code == 1 || s->eac3)
407 ff_ac3_adjust_frame_size(s);
408
409 copy_input_samples(s, (SampleType **)frame->extended_data);
410
411 apply_mdct(s);
412
413 if (s->fixed_point)
414 scale_coefficients(s);
415
416 clip_coefficients(&s->adsp, s->blocks[0].mdct_coef[1],
417 AC3_MAX_COEFS * s->num_blocks * s->channels);
418
419 s->cpl_on = s->cpl_enabled;
420 ff_ac3_compute_coupling_strategy(s);
421
422 if (s->cpl_on)
423 apply_channel_coupling(s);
424
425 compute_rematrixing_strategy(s);
426
427 if (!s->fixed_point)
428 scale_coefficients(s);
429
430 ff_ac3_apply_rematrixing(s);
431
432 ff_ac3_process_exponents(s);
433
434 ret = ff_ac3_compute_bit_allocation(s);
435 if (ret) {
436 av_log(avctx, AV_LOG_ERROR, "Bit allocation failed. Try increasing the bitrate.\n");
437 return ret;
438 }
439
440 ff_ac3_group_exponents(s);
441
442 ff_ac3_quantize_mantissas(s);
443
444 if ((ret = ff_alloc_packet(avpkt, s->frame_size))) {
445 av_log(avctx, AV_LOG_ERROR, "Error getting output packet\n");
446 return ret;
447 }
448 ff_ac3_output_frame(s, avpkt->data);
449
450 if (frame->pts != AV_NOPTS_VALUE)
451 avpkt->pts = frame->pts - ff_samples_to_time_base(avctx, avctx->initial_padding);
452
453 *got_packet_ptr = 1;
454 return 0;
455 }