1 /*
2 * AC-3 DSP functions
3 * Copyright (c) 2011 Justin Ruggles
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 #include "avcodec.h"
23 #include "ac3.h"
24 #include "ac3dsp.h"
25 #include "mathops.h"
26
ac3_exponent_min_c(uint8_t * exp,int num_reuse_blocks,int nb_coefs)27 static void ac3_exponent_min_c(uint8_t *exp, int num_reuse_blocks, int nb_coefs)
28 {
29 int blk, i;
30
31 if (!num_reuse_blocks)
32 return;
33
34 for (i = 0; i < nb_coefs; i++) {
35 uint8_t min_exp = *exp;
36 uint8_t *exp1 = exp + 256;
37 for (blk = 0; blk < num_reuse_blocks; blk++) {
38 uint8_t next_exp = *exp1;
39 if (next_exp < min_exp)
40 min_exp = next_exp;
41 exp1 += 256;
42 }
43 *exp++ = min_exp;
44 }
45 }
46
ac3_max_msb_abs_int16_c(const int16_t * src,int len)47 static int ac3_max_msb_abs_int16_c(const int16_t *src, int len)
48 {
49 int i, v = 0;
50 for (i = 0; i < len; i++)
51 v |= abs(src[i]);
52 return v;
53 }
54
ac3_lshift_int16_c(int16_t * src,unsigned int len,unsigned int shift)55 static void ac3_lshift_int16_c(int16_t *src, unsigned int len,
56 unsigned int shift)
57 {
58 uint32_t *src32 = (uint32_t *)src;
59 const uint32_t mask = ~(((1 << shift) - 1) << 16);
60 int i;
61 len >>= 1;
62 for (i = 0; i < len; i += 8) {
63 src32[i ] = (src32[i ] << shift) & mask;
64 src32[i+1] = (src32[i+1] << shift) & mask;
65 src32[i+2] = (src32[i+2] << shift) & mask;
66 src32[i+3] = (src32[i+3] << shift) & mask;
67 src32[i+4] = (src32[i+4] << shift) & mask;
68 src32[i+5] = (src32[i+5] << shift) & mask;
69 src32[i+6] = (src32[i+6] << shift) & mask;
70 src32[i+7] = (src32[i+7] << shift) & mask;
71 }
72 }
73
ac3_rshift_int32_c(int32_t * src,unsigned int len,unsigned int shift)74 static void ac3_rshift_int32_c(int32_t *src, unsigned int len,
75 unsigned int shift)
76 {
77 do {
78 *src++ >>= shift;
79 *src++ >>= shift;
80 *src++ >>= shift;
81 *src++ >>= shift;
82 *src++ >>= shift;
83 *src++ >>= shift;
84 *src++ >>= shift;
85 *src++ >>= shift;
86 len -= 8;
87 } while (len > 0);
88 }
89
float_to_fixed24_c(int32_t * dst,const float * src,unsigned int len)90 static void float_to_fixed24_c(int32_t *dst, const float *src, unsigned int len)
91 {
92 const float scale = 1 << 24;
93 do {
94 *dst++ = lrintf(*src++ * scale);
95 *dst++ = lrintf(*src++ * scale);
96 *dst++ = lrintf(*src++ * scale);
97 *dst++ = lrintf(*src++ * scale);
98 *dst++ = lrintf(*src++ * scale);
99 *dst++ = lrintf(*src++ * scale);
100 *dst++ = lrintf(*src++ * scale);
101 *dst++ = lrintf(*src++ * scale);
102 len -= 8;
103 } while (len > 0);
104 }
105
ac3_bit_alloc_calc_bap_c(int16_t * mask,int16_t * psd,int start,int end,int snr_offset,int floor,const uint8_t * bap_tab,uint8_t * bap)106 static void ac3_bit_alloc_calc_bap_c(int16_t *mask, int16_t *psd,
107 int start, int end,
108 int snr_offset, int floor,
109 const uint8_t *bap_tab, uint8_t *bap)
110 {
111 int bin, band, band_end;
112
113 /* special case, if snr offset is -960, set all bap's to zero */
114 if (snr_offset == -960) {
115 memset(bap, 0, AC3_MAX_COEFS);
116 return;
117 }
118
119 bin = start;
120 band = ff_ac3_bin_to_band_tab[start];
121 do {
122 int m = (FFMAX(mask[band] - snr_offset - floor, 0) & 0x1FE0) + floor;
123 band_end = ff_ac3_band_start_tab[++band];
124 band_end = FFMIN(band_end, end);
125
126 for (; bin < band_end; bin++) {
127 int address = av_clip_uintp2((psd[bin] - m) >> 5, 6);
128 bap[bin] = bap_tab[address];
129 }
130 } while (end > band_end);
131 }
132
ac3_update_bap_counts_c(uint16_t mant_cnt[16],uint8_t * bap,int len)133 static void ac3_update_bap_counts_c(uint16_t mant_cnt[16], uint8_t *bap,
134 int len)
135 {
136 while (len-- > 0)
137 mant_cnt[bap[len]]++;
138 }
139
140 DECLARE_ALIGNED(16, const uint16_t, ff_ac3_bap_bits)[16] = {
141 0, 0, 0, 3, 0, 4, 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
142 };
143
ac3_compute_mantissa_size_c(uint16_t mant_cnt[6][16])144 static int ac3_compute_mantissa_size_c(uint16_t mant_cnt[6][16])
145 {
146 int blk, bap;
147 int bits = 0;
148
149 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
150 // bap=1 : 3 mantissas in 5 bits
151 bits += (mant_cnt[blk][1] / 3) * 5;
152 // bap=2 : 3 mantissas in 7 bits
153 // bap=4 : 2 mantissas in 7 bits
154 bits += ((mant_cnt[blk][2] / 3) + (mant_cnt[blk][4] >> 1)) * 7;
155 // bap=3 : 1 mantissa in 3 bits
156 bits += mant_cnt[blk][3] * 3;
157 // bap=5 to 15 : get bits per mantissa from table
158 for (bap = 5; bap < 16; bap++)
159 bits += mant_cnt[blk][bap] * ff_ac3_bap_bits[bap];
160 }
161 return bits;
162 }
163
ac3_extract_exponents_c(uint8_t * exp,int32_t * coef,int nb_coefs)164 static void ac3_extract_exponents_c(uint8_t *exp, int32_t *coef, int nb_coefs)
165 {
166 int i;
167
168 for (i = 0; i < nb_coefs; i++) {
169 int v = abs(coef[i]);
170 exp[i] = v ? 23 - av_log2(v) : 24;
171 }
172 }
173
ac3_sum_square_butterfly_int32_c(int64_t sum[4],const int32_t * coef0,const int32_t * coef1,int len)174 static void ac3_sum_square_butterfly_int32_c(int64_t sum[4],
175 const int32_t *coef0,
176 const int32_t *coef1,
177 int len)
178 {
179 int i;
180
181 sum[0] = sum[1] = sum[2] = sum[3] = 0;
182
183 for (i = 0; i < len; i++) {
184 int lt = coef0[i];
185 int rt = coef1[i];
186 int md = lt + rt;
187 int sd = lt - rt;
188 MAC64(sum[0], lt, lt);
189 MAC64(sum[1], rt, rt);
190 MAC64(sum[2], md, md);
191 MAC64(sum[3], sd, sd);
192 }
193 }
194
ac3_sum_square_butterfly_float_c(float sum[4],const float * coef0,const float * coef1,int len)195 static void ac3_sum_square_butterfly_float_c(float sum[4],
196 const float *coef0,
197 const float *coef1,
198 int len)
199 {
200 int i;
201
202 sum[0] = sum[1] = sum[2] = sum[3] = 0;
203
204 for (i = 0; i < len; i++) {
205 float lt = coef0[i];
206 float rt = coef1[i];
207 float md = lt + rt;
208 float sd = lt - rt;
209 sum[0] += lt * lt;
210 sum[1] += rt * rt;
211 sum[2] += md * md;
212 sum[3] += sd * sd;
213 }
214 }
215
ac3_downmix_5_to_2_symmetric_c(float ** samples,float ** matrix,int len)216 static void ac3_downmix_5_to_2_symmetric_c(float **samples, float **matrix,
217 int len)
218 {
219 int i;
220 float v0, v1;
221 float front_mix = matrix[0][0];
222 float center_mix = matrix[0][1];
223 float surround_mix = matrix[0][3];
224
225 for (i = 0; i < len; i++) {
226 v0 = samples[0][i] * front_mix +
227 samples[1][i] * center_mix +
228 samples[3][i] * surround_mix;
229
230 v1 = samples[1][i] * center_mix +
231 samples[2][i] * front_mix +
232 samples[4][i] * surround_mix;
233
234 samples[0][i] = v0;
235 samples[1][i] = v1;
236 }
237 }
238
ac3_downmix_5_to_1_symmetric_c(float ** samples,float ** matrix,int len)239 static void ac3_downmix_5_to_1_symmetric_c(float **samples, float **matrix,
240 int len)
241 {
242 int i;
243 float front_mix = matrix[0][0];
244 float center_mix = matrix[0][1];
245 float surround_mix = matrix[0][3];
246
247 for (i = 0; i < len; i++) {
248 samples[0][i] = samples[0][i] * front_mix +
249 samples[1][i] * center_mix +
250 samples[2][i] * front_mix +
251 samples[3][i] * surround_mix +
252 samples[4][i] * surround_mix;
253 }
254 }
255
ac3_downmix_c(float ** samples,float ** matrix,int out_ch,int in_ch,int len)256 static void ac3_downmix_c(float **samples, float **matrix,
257 int out_ch, int in_ch, int len)
258 {
259 int i, j;
260 float v0, v1;
261
262 if (out_ch == 2) {
263 for (i = 0; i < len; i++) {
264 v0 = v1 = 0.0f;
265 for (j = 0; j < in_ch; j++) {
266 v0 += samples[j][i] * matrix[0][j];
267 v1 += samples[j][i] * matrix[1][j];
268 }
269 samples[0][i] = v0;
270 samples[1][i] = v1;
271 }
272 } else if (out_ch == 1) {
273 for (i = 0; i < len; i++) {
274 v0 = 0.0f;
275 for (j = 0; j < in_ch; j++)
276 v0 += samples[j][i] * matrix[0][j];
277 samples[0][i] = v0;
278 }
279 }
280 }
281
ac3_downmix_5_to_2_symmetric_c_fixed(int32_t ** samples,int16_t ** matrix,int len)282 static void ac3_downmix_5_to_2_symmetric_c_fixed(int32_t **samples, int16_t **matrix,
283 int len)
284 {
285 int i;
286 int64_t v0, v1;
287 int16_t front_mix = matrix[0][0];
288 int16_t center_mix = matrix[0][1];
289 int16_t surround_mix = matrix[0][3];
290
291 for (i = 0; i < len; i++) {
292 v0 = (int64_t)samples[0][i] * front_mix +
293 (int64_t)samples[1][i] * center_mix +
294 (int64_t)samples[3][i] * surround_mix;
295
296 v1 = (int64_t)samples[1][i] * center_mix +
297 (int64_t)samples[2][i] * front_mix +
298 (int64_t)samples[4][i] * surround_mix;
299
300 samples[0][i] = (v0+2048)>>12;
301 samples[1][i] = (v1+2048)>>12;
302 }
303 }
304
ac3_downmix_5_to_1_symmetric_c_fixed(int32_t ** samples,int16_t ** matrix,int len)305 static void ac3_downmix_5_to_1_symmetric_c_fixed(int32_t **samples, int16_t **matrix,
306 int len)
307 {
308 int i;
309 int64_t v0;
310 int16_t front_mix = matrix[0][0];
311 int16_t center_mix = matrix[0][1];
312 int16_t surround_mix = matrix[0][3];
313
314 for (i = 0; i < len; i++) {
315 v0 = (int64_t)samples[0][i] * front_mix +
316 (int64_t)samples[1][i] * center_mix +
317 (int64_t)samples[2][i] * front_mix +
318 (int64_t)samples[3][i] * surround_mix +
319 (int64_t)samples[4][i] * surround_mix;
320
321 samples[0][i] = (v0+2048)>>12;
322 }
323 }
324
ac3_downmix_c_fixed(int32_t ** samples,int16_t ** matrix,int out_ch,int in_ch,int len)325 static void ac3_downmix_c_fixed(int32_t **samples, int16_t **matrix,
326 int out_ch, int in_ch, int len)
327 {
328 int i, j;
329 int64_t v0, v1;
330 if (out_ch == 2) {
331 for (i = 0; i < len; i++) {
332 v0 = v1 = 0;
333 for (j = 0; j < in_ch; j++) {
334 v0 += (int64_t)samples[j][i] * matrix[0][j];
335 v1 += (int64_t)samples[j][i] * matrix[1][j];
336 }
337 samples[0][i] = (v0+2048)>>12;
338 samples[1][i] = (v1+2048)>>12;
339 }
340 } else if (out_ch == 1) {
341 for (i = 0; i < len; i++) {
342 v0 = 0;
343 for (j = 0; j < in_ch; j++)
344 v0 += (int64_t)samples[j][i] * matrix[0][j];
345 samples[0][i] = (v0+2048)>>12;
346 }
347 }
348 }
349
ff_ac3dsp_downmix_fixed(AC3DSPContext * c,int32_t ** samples,int16_t ** matrix,int out_ch,int in_ch,int len)350 void ff_ac3dsp_downmix_fixed(AC3DSPContext *c, int32_t **samples, int16_t **matrix,
351 int out_ch, int in_ch, int len)
352 {
353 if (c->in_channels != in_ch || c->out_channels != out_ch) {
354 c->in_channels = in_ch;
355 c->out_channels = out_ch;
356 c->downmix_fixed = NULL;
357
358 if (in_ch == 5 && out_ch == 2 &&
359 !(matrix[1][0] | matrix[0][2] |
360 matrix[1][3] | matrix[0][4] |
361 (matrix[0][1] ^ matrix[1][1]) |
362 (matrix[0][0] ^ matrix[1][2]))) {
363 c->downmix_fixed = ac3_downmix_5_to_2_symmetric_c_fixed;
364 } else if (in_ch == 5 && out_ch == 1 &&
365 matrix[0][0] == matrix[0][2] &&
366 matrix[0][3] == matrix[0][4]) {
367 c->downmix_fixed = ac3_downmix_5_to_1_symmetric_c_fixed;
368 }
369 }
370
371 if (c->downmix_fixed)
372 c->downmix_fixed(samples, matrix, len);
373 else
374 ac3_downmix_c_fixed(samples, matrix, out_ch, in_ch, len);
375 }
376
apply_window_int16_c(int16_t * output,const int16_t * input,const int16_t * window,unsigned int len)377 static void apply_window_int16_c(int16_t *output, const int16_t *input,
378 const int16_t *window, unsigned int len)
379 {
380 int i;
381 int len2 = len >> 1;
382
383 for (i = 0; i < len2; i++) {
384 int16_t w = window[i];
385 output[i] = (MUL16(input[i], w) + (1 << 14)) >> 15;
386 output[len-i-1] = (MUL16(input[len-i-1], w) + (1 << 14)) >> 15;
387 }
388 }
389
ff_ac3dsp_downmix(AC3DSPContext * c,float ** samples,float ** matrix,int out_ch,int in_ch,int len)390 void ff_ac3dsp_downmix(AC3DSPContext *c, float **samples, float **matrix,
391 int out_ch, int in_ch, int len)
392 {
393 if (c->in_channels != in_ch || c->out_channels != out_ch) {
394 int **matrix_cmp = (int **)matrix;
395
396 c->in_channels = in_ch;
397 c->out_channels = out_ch;
398 c->downmix = NULL;
399
400 if (in_ch == 5 && out_ch == 2 &&
401 !(matrix_cmp[1][0] | matrix_cmp[0][2] |
402 matrix_cmp[1][3] | matrix_cmp[0][4] |
403 (matrix_cmp[0][1] ^ matrix_cmp[1][1]) |
404 (matrix_cmp[0][0] ^ matrix_cmp[1][2]))) {
405 c->downmix = ac3_downmix_5_to_2_symmetric_c;
406 } else if (in_ch == 5 && out_ch == 1 &&
407 matrix_cmp[0][0] == matrix_cmp[0][2] &&
408 matrix_cmp[0][3] == matrix_cmp[0][4]) {
409 c->downmix = ac3_downmix_5_to_1_symmetric_c;
410 }
411
412 if (ARCH_X86)
413 ff_ac3dsp_set_downmix_x86(c);
414 }
415
416 if (c->downmix)
417 c->downmix(samples, matrix, len);
418 else
419 ac3_downmix_c(samples, matrix, out_ch, in_ch, len);
420 }
421
ff_ac3dsp_init(AC3DSPContext * c,int bit_exact)422 av_cold void ff_ac3dsp_init(AC3DSPContext *c, int bit_exact)
423 {
424 c->ac3_exponent_min = ac3_exponent_min_c;
425 c->ac3_max_msb_abs_int16 = ac3_max_msb_abs_int16_c;
426 c->ac3_lshift_int16 = ac3_lshift_int16_c;
427 c->ac3_rshift_int32 = ac3_rshift_int32_c;
428 c->float_to_fixed24 = float_to_fixed24_c;
429 c->bit_alloc_calc_bap = ac3_bit_alloc_calc_bap_c;
430 c->update_bap_counts = ac3_update_bap_counts_c;
431 c->compute_mantissa_size = ac3_compute_mantissa_size_c;
432 c->extract_exponents = ac3_extract_exponents_c;
433 c->sum_square_butterfly_int32 = ac3_sum_square_butterfly_int32_c;
434 c->sum_square_butterfly_float = ac3_sum_square_butterfly_float_c;
435 c->in_channels = 0;
436 c->out_channels = 0;
437 c->downmix = NULL;
438 c->downmix_fixed = NULL;
439 c->apply_window_int16 = apply_window_int16_c;
440
441 if (ARCH_ARM)
442 ff_ac3dsp_init_arm(c, bit_exact);
443 if (ARCH_X86)
444 ff_ac3dsp_init_x86(c, bit_exact);
445 if (ARCH_MIPS)
446 ff_ac3dsp_init_mips(c, bit_exact);
447 }
448