FFmpeg coverage


Directory: ../../../ffmpeg/
File: src/libavcodec/dcadsp.c
Date: 2022-12-09 07:38:14
Exec Total Coverage
Lines: 171 231 74.0%
Functions: 19 24 79.2%
Branches: 68 94 72.3%

Line Branch Exec Source
1 /*
2 * Copyright (C) 2016 foo86
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21 #include "libavutil/mem_internal.h"
22
23 #include "dcadsp.h"
24 #include "dcamath.h"
25
26 9771 static void decode_hf_c(int32_t **dst,
27 const int32_t *vq_index,
28 const int8_t hf_vq[1024][32],
29 int32_t scale_factors[32][2],
30 ptrdiff_t sb_start, ptrdiff_t sb_end,
31 ptrdiff_t ofs, ptrdiff_t len)
32 {
33 int i, j;
34
35
2/2
✓ Branch 0 taken 43474 times.
✓ Branch 1 taken 9771 times.
53245 for (i = sb_start; i < sb_end; i++) {
36 43474 const int8_t *coeff = hf_vq[vq_index[i]];
37 43474 int32_t scale = scale_factors[i][0];
38
2/2
✓ Branch 0 taken 695584 times.
✓ Branch 1 taken 43474 times.
739058 for (j = 0; j < len; j++)
39 695584 dst[i][j + ofs] = clip23(coeff[j] * scale + (1 << 3) >> 4);
40 }
41 9771 }
42
43 91 static void decode_joint_c(int32_t **dst, int32_t **src,
44 const int32_t *scale_factors,
45 ptrdiff_t sb_start, ptrdiff_t sb_end,
46 ptrdiff_t ofs, ptrdiff_t len)
47 {
48 int i, j;
49
50
2/2
✓ Branch 0 taken 2541 times.
✓ Branch 1 taken 91 times.
2632 for (i = sb_start; i < sb_end; i++) {
51 2541 int32_t scale = scale_factors[i];
52
2/2
✓ Branch 0 taken 40656 times.
✓ Branch 1 taken 2541 times.
43197 for (j = 0; j < len; j++)
53 40656 dst[i][j + ofs] = clip23(mul17(src[i][j + ofs], scale));
54 }
55 91 }
56
57 606 static void lfe_fir_float_c(float *pcm_samples, int32_t *lfe_samples,
58 const float *filter_coeff, ptrdiff_t npcmblocks,
59 int dec_select)
60 {
61 // Select decimation factor
62 606 int factor = 64 << dec_select;
63 606 int ncoeffs = 8 >> dec_select;
64 606 int nlfesamples = npcmblocks >> (dec_select + 1);
65 int i, j, k;
66
67
2/2
✓ Branch 0 taken 4848 times.
✓ Branch 1 taken 606 times.
5454 for (i = 0; i < nlfesamples; i++) {
68 // One decimated sample generates 64 or 128 interpolated ones
69
2/2
✓ Branch 0 taken 155136 times.
✓ Branch 1 taken 4848 times.
159984 for (j = 0; j < factor / 2; j++) {
70 155136 float a = 0;
71 155136 float b = 0;
72
73
2/2
✓ Branch 0 taken 1241088 times.
✓ Branch 1 taken 155136 times.
1396224 for (k = 0; k < ncoeffs; k++) {
74 1241088 a += filter_coeff[ j * ncoeffs + k] * lfe_samples[-k];
75 1241088 b += filter_coeff[255 - j * ncoeffs - k] * lfe_samples[-k];
76 }
77
78 155136 pcm_samples[ j] = a;
79 155136 pcm_samples[factor / 2 + j] = b;
80 }
81
82 4848 lfe_samples++;
83 4848 pcm_samples += factor;
84 }
85 606 }
86
87 606 static void lfe_fir0_float_c(float *pcm_samples, int32_t *lfe_samples,
88 const float *filter_coeff, ptrdiff_t npcmblocks)
89 {
90 606 lfe_fir_float_c(pcm_samples, lfe_samples, filter_coeff, npcmblocks, 0);
91 606 }
92
93 static void lfe_fir1_float_c(float *pcm_samples, int32_t *lfe_samples,
94 const float *filter_coeff, ptrdiff_t npcmblocks)
95 {
96 lfe_fir_float_c(pcm_samples, lfe_samples, filter_coeff, npcmblocks, 1);
97 }
98
99 35 static void lfe_x96_float_c(float *dst, const float *src,
100 float *hist, ptrdiff_t len)
101 {
102 35 float prev = *hist;
103 int i;
104
105
2/2
✓ Branch 0 taken 17920 times.
✓ Branch 1 taken 35 times.
17955 for (i = 0; i < len; i++) {
106 17920 float a = 0.25f * src[i] + 0.75f * prev;
107 17920 float b = 0.75f * src[i] + 0.25f * prev;
108 17920 prev = src[i];
109 17920 *dst++ = a;
110 17920 *dst++ = b;
111 }
112
113 35 *hist = prev;
114 35 }
115
116 3150 static void sub_qmf32_float_c(SynthFilterContext *synth,
117 AVTXContext *imdct,
118 av_tx_fn imdct_fn,
119 float *pcm_samples,
120 int32_t **subband_samples_lo,
121 int32_t **subband_samples_hi,
122 float *hist1, int *offset, float *hist2,
123 const float *filter_coeff, ptrdiff_t npcmblocks,
124 float scale)
125 {
126 3150 LOCAL_ALIGNED_32(float, input, [32]);
127 int i, j;
128
129
2/2
✓ Branch 0 taken 50400 times.
✓ Branch 1 taken 3150 times.
53550 for (j = 0; j < npcmblocks; j++) {
130 // Load in one sample from each subband
131
2/2
✓ Branch 0 taken 1612800 times.
✓ Branch 1 taken 50400 times.
1663200 for (i = 0; i < 32; i++) {
132
2/2
✓ Branch 0 taken 806400 times.
✓ Branch 1 taken 806400 times.
1612800 if ((i - 1) & 2)
133 806400 input[i] = -subband_samples_lo[i][j];
134 else
135 806400 input[i] = subband_samples_lo[i][j];
136 }
137
138 // One subband sample generates 32 interpolated ones
139 50400 synth->synth_filter_float(imdct, hist1, offset,
140 hist2, filter_coeff,
141 pcm_samples, input, scale, imdct_fn);
142 50400 pcm_samples += 32;
143 }
144 3150 }
145
146 196 static void sub_qmf64_float_c(SynthFilterContext *synth,
147 AVTXContext *imdct,
148 av_tx_fn imdct_fn,
149 float *pcm_samples,
150 int32_t **subband_samples_lo,
151 int32_t **subband_samples_hi,
152 float *hist1, int *offset, float *hist2,
153 const float *filter_coeff, ptrdiff_t npcmblocks,
154 float scale)
155 {
156 196 LOCAL_ALIGNED_32(float, input, [64]);
157 int i, j;
158
159
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 196 times.
196 if (!subband_samples_hi)
160 memset(&input[32], 0, sizeof(input[0]) * 32);
161
162
2/2
✓ Branch 0 taken 3136 times.
✓ Branch 1 taken 196 times.
3332 for (j = 0; j < npcmblocks; j++) {
163 // Load in one sample from each subband
164
1/2
✓ Branch 0 taken 3136 times.
✗ Branch 1 not taken.
3136 if (subband_samples_hi) {
165 // Full 64 subbands, first 32 are residual coded
166
2/2
✓ Branch 0 taken 100352 times.
✓ Branch 1 taken 3136 times.
103488 for (i = 0; i < 32; i++) {
167
2/2
✓ Branch 0 taken 50176 times.
✓ Branch 1 taken 50176 times.
100352 if ((i - 1) & 2)
168 50176 input[i] = -subband_samples_lo[i][j] - subband_samples_hi[i][j];
169 else
170 50176 input[i] = subband_samples_lo[i][j] + subband_samples_hi[i][j];
171 }
172
2/2
✓ Branch 0 taken 100352 times.
✓ Branch 1 taken 3136 times.
103488 for (i = 32; i < 64; i++) {
173
2/2
✓ Branch 0 taken 50176 times.
✓ Branch 1 taken 50176 times.
100352 if ((i - 1) & 2)
174 50176 input[i] = -subband_samples_hi[i][j];
175 else
176 50176 input[i] = subband_samples_hi[i][j];
177 }
178 } else {
179 // Only first 32 subbands
180 for (i = 0; i < 32; i++) {
181 if ((i - 1) & 2)
182 input[i] = -subband_samples_lo[i][j];
183 else
184 input[i] = subband_samples_lo[i][j];
185 }
186 }
187
188 // One subband sample generates 64 interpolated ones
189 3136 synth->synth_filter_float_64(imdct, hist1, offset,
190 hist2, filter_coeff,
191 pcm_samples, input, scale, imdct_fn);
192 3136 pcm_samples += 64;
193 }
194 196 }
195
196 1347 static void lfe_fir_fixed_c(int32_t *pcm_samples, int32_t *lfe_samples,
197 const int32_t *filter_coeff, ptrdiff_t npcmblocks)
198 {
199 // Select decimation factor
200 1347 int nlfesamples = npcmblocks >> 1;
201 int i, j, k;
202
203
2/2
✓ Branch 0 taken 10776 times.
✓ Branch 1 taken 1347 times.
12123 for (i = 0; i < nlfesamples; i++) {
204 // One decimated sample generates 64 interpolated ones
205
2/2
✓ Branch 0 taken 344832 times.
✓ Branch 1 taken 10776 times.
355608 for (j = 0; j < 32; j++) {
206 344832 int64_t a = 0;
207 344832 int64_t b = 0;
208
209
2/2
✓ Branch 0 taken 2758656 times.
✓ Branch 1 taken 344832 times.
3103488 for (k = 0; k < 8; k++) {
210 2758656 a += (int64_t)filter_coeff[ j * 8 + k] * lfe_samples[-k];
211 2758656 b += (int64_t)filter_coeff[255 - j * 8 - k] * lfe_samples[-k];
212 }
213
214 344832 pcm_samples[ j] = clip23(norm23(a));
215 344832 pcm_samples[32 + j] = clip23(norm23(b));
216 }
217
218 10776 lfe_samples++;
219 10776 pcm_samples += 64;
220 }
221 1347 }
222
223 84 static void lfe_x96_fixed_c(int32_t *dst, const int32_t *src,
224 int32_t *hist, ptrdiff_t len)
225 {
226 84 int32_t prev = *hist;
227 int i;
228
229
2/2
✓ Branch 0 taken 43008 times.
✓ Branch 1 taken 84 times.
43092 for (i = 0; i < len; i++) {
230 43008 int64_t a = INT64_C(2097471) * src[i] + INT64_C(6291137) * prev;
231 43008 int64_t b = INT64_C(6291137) * src[i] + INT64_C(2097471) * prev;
232 43008 prev = src[i];
233 43008 *dst++ = clip23(norm23(a));
234 43008 *dst++ = clip23(norm23(b));
235 }
236
237 84 *hist = prev;
238 84 }
239
240 7356 static void sub_qmf32_fixed_c(SynthFilterContext *synth,
241 DCADCTContext *imdct,
242 int32_t *pcm_samples,
243 int32_t **subband_samples_lo,
244 int32_t **subband_samples_hi,
245 int32_t *hist1, int *offset, int32_t *hist2,
246 const int32_t *filter_coeff, ptrdiff_t npcmblocks)
247 {
248 7356 LOCAL_ALIGNED_32(int32_t, input, [32]);
249 int i, j;
250
251
2/2
✓ Branch 0 taken 117696 times.
✓ Branch 1 taken 7356 times.
125052 for (j = 0; j < npcmblocks; j++) {
252 // Load in one sample from each subband
253
2/2
✓ Branch 0 taken 3766272 times.
✓ Branch 1 taken 117696 times.
3883968 for (i = 0; i < 32; i++)
254 3766272 input[i] = subband_samples_lo[i][j];
255
256 // One subband sample generates 32 interpolated ones
257 117696 synth->synth_filter_fixed(imdct, hist1, offset,
258 hist2, filter_coeff,
259 pcm_samples, input);
260 117696 pcm_samples += 32;
261 }
262 7356 }
263
264 420 static void sub_qmf64_fixed_c(SynthFilterContext *synth,
265 DCADCTContext *imdct,
266 int32_t *pcm_samples,
267 int32_t **subband_samples_lo,
268 int32_t **subband_samples_hi,
269 int32_t *hist1, int *offset, int32_t *hist2,
270 const int32_t *filter_coeff, ptrdiff_t npcmblocks)
271 {
272 420 LOCAL_ALIGNED_32(int32_t, input, [64]);
273 int i, j;
274
275
1/2
✓ Branch 0 taken 420 times.
✗ Branch 1 not taken.
420 if (!subband_samples_hi)
276 420 memset(&input[32], 0, sizeof(input[0]) * 32);
277
278
2/2
✓ Branch 0 taken 6720 times.
✓ Branch 1 taken 420 times.
7140 for (j = 0; j < npcmblocks; j++) {
279 // Load in one sample from each subband
280
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 6720 times.
6720 if (subband_samples_hi) {
281 // Full 64 subbands, first 32 are residual coded
282 for (i = 0; i < 32; i++)
283 input[i] = subband_samples_lo[i][j] + subband_samples_hi[i][j];
284 for (i = 32; i < 64; i++)
285 input[i] = subband_samples_hi[i][j];
286 } else {
287 // Only first 32 subbands
288
2/2
✓ Branch 0 taken 215040 times.
✓ Branch 1 taken 6720 times.
221760 for (i = 0; i < 32; i++)
289 215040 input[i] = subband_samples_lo[i][j];
290 }
291
292 // One subband sample generates 64 interpolated ones
293 6720 synth->synth_filter_fixed_64(imdct, hist1, offset,
294 hist2, filter_coeff,
295 pcm_samples, input);
296 6720 pcm_samples += 64;
297 }
298 420 }
299
300 1815 static void decor_c(int32_t *dst, const int32_t *src, int coeff, ptrdiff_t len)
301 {
302 int i;
303
304
2/2
✓ Branch 0 taken 1030144 times.
✓ Branch 1 taken 1815 times.
1031959 for (i = 0; i < len; i++)
305 1030144 dst[i] += (SUINT)((int)(src[i] * (SUINT)coeff + (1 << 2)) >> 3);
306 1815 }
307
308 static void dmix_sub_xch_c(int32_t *dst1, int32_t *dst2,
309 const int32_t *src, ptrdiff_t len)
310 {
311 int i;
312
313 for (i = 0; i < len; i++) {
314 int32_t cs = mul23(src[i], 5931520 /* M_SQRT1_2 * (1 << 23) */);
315 dst1[i] -= cs;
316 dst2[i] -= cs;
317 }
318 }
319
320 120 static void dmix_sub_c(int32_t *dst, const int32_t *src, int coeff, ptrdiff_t len)
321 {
322 int i;
323
324
2/2
✓ Branch 0 taken 76960 times.
✓ Branch 1 taken 120 times.
77080 for (i = 0; i < len; i++)
325 76960 dst[i] -= (unsigned)mul15(src[i], coeff);
326 120 }
327
328 14 static void dmix_add_c(int32_t *dst, const int32_t *src, int coeff, ptrdiff_t len)
329 {
330 int i;
331
332
2/2
✓ Branch 0 taken 7168 times.
✓ Branch 1 taken 14 times.
7182 for (i = 0; i < len; i++)
333 7168 dst[i] += (unsigned)mul15(src[i], coeff);
334 14 }
335
336 204 static void dmix_scale_c(int32_t *dst, int scale, ptrdiff_t len)
337 {
338 int i;
339
340
2/2
✓ Branch 0 taken 140368 times.
✓ Branch 1 taken 204 times.
140572 for (i = 0; i < len; i++)
341 140368 dst[i] = mul15(dst[i], scale);
342 204 }
343
344 static void dmix_scale_inv_c(int32_t *dst, int scale_inv, ptrdiff_t len)
345 {
346 int i;
347
348 for (i = 0; i < len; i++)
349 dst[i] = mul16(dst[i], scale_inv);
350 }
351
352 896 static void filter0(SUINT32 *dst, const int32_t *src, int32_t coeff, ptrdiff_t len)
353 {
354 int i;
355
356
2/2
✓ Branch 0 taken 917504 times.
✓ Branch 1 taken 896 times.
918400 for (i = 0; i < len; i++)
357 917504 dst[i] -= mul22(src[i], coeff);
358 896 }
359
360 5376 static void filter1(SUINT32 *dst, const int32_t *src, int32_t coeff, ptrdiff_t len)
361 {
362 int i;
363
364
2/2
✓ Branch 0 taken 5505024 times.
✓ Branch 1 taken 5376 times.
5510400 for (i = 0; i < len; i++)
365 5505024 dst[i] -= mul23(src[i], coeff);
366 5376 }
367
368 224 static void assemble_freq_bands_c(int32_t *dst, int32_t *src0, int32_t *src1,
369 const int32_t *coeff, ptrdiff_t len)
370 {
371 int i;
372
373 224 filter0(src0, src1, coeff[0], len);
374 224 filter0(src1, src0, coeff[1], len);
375 224 filter0(src0, src1, coeff[2], len);
376 224 filter0(src1, src0, coeff[3], len);
377
378
2/2
✓ Branch 0 taken 1792 times.
✓ Branch 1 taken 224 times.
2016 for (i = 0; i < 8; i++, src0--) {
379 1792 filter1(src0, src1, coeff[i + 4], len);
380 1792 filter1(src1, src0, coeff[i + 12], len);
381 1792 filter1(src0, src1, coeff[i + 4], len);
382 }
383
384
2/2
✓ Branch 0 taken 229376 times.
✓ Branch 1 taken 224 times.
229600 for (i = 0; i < len; i++) {
385 229376 *dst++ = *src1++;
386 229376 *dst++ = *++src0;
387 }
388 224 }
389
390 static void lbr_bank_c(float output[32][4], float **input,
391 const float *coeff, ptrdiff_t ofs, ptrdiff_t len)
392 {
393 float SW0 = coeff[0];
394 float SW1 = coeff[1];
395 float SW2 = coeff[2];
396 float SW3 = coeff[3];
397
398 float C1 = coeff[4];
399 float C2 = coeff[5];
400 float C3 = coeff[6];
401 float C4 = coeff[7];
402
403 float AL1 = coeff[8];
404 float AL2 = coeff[9];
405
406 int i;
407
408 // Short window and 8 point forward MDCT
409 for (i = 0; i < len; i++) {
410 float *src = input[i] + ofs;
411
412 float a = src[-4] * SW0 - src[-1] * SW3;
413 float b = src[-3] * SW1 - src[-2] * SW2;
414 float c = src[ 2] * SW1 + src[ 1] * SW2;
415 float d = src[ 3] * SW0 + src[ 0] * SW3;
416
417 output[i][0] = C1 * b - C2 * c + C4 * a - C3 * d;
418 output[i][1] = C1 * d - C2 * a - C4 * b - C3 * c;
419 output[i][2] = C3 * b + C2 * d - C4 * c + C1 * a;
420 output[i][3] = C3 * a - C2 * b + C4 * d - C1 * c;
421 }
422
423 // Aliasing cancellation for high frequencies
424 for (i = 12; i < len - 1; i++) {
425 float a = output[i ][3] * AL1;
426 float b = output[i+1][0] * AL1;
427 output[i ][3] += b - a;
428 output[i+1][0] -= b + a;
429 a = output[i ][2] * AL2;
430 b = output[i+1][1] * AL2;
431 output[i ][2] += b - a;
432 output[i+1][1] -= b + a;
433 }
434 }
435
436 static void lfe_iir_c(float *output, const float *input,
437 const float iir[5][4], float hist[5][2],
438 ptrdiff_t factor)
439 {
440 float res, tmp;
441 int i, j, k;
442
443 for (i = 0; i < 64; i++) {
444 res = *input++;
445
446 for (j = 0; j < factor; j++) {
447 for (k = 0; k < 5; k++) {
448 tmp = hist[k][0] * iir[k][0] + hist[k][1] * iir[k][1] + res;
449 res = hist[k][0] * iir[k][2] + hist[k][1] * iir[k][3] + tmp;
450
451 hist[k][0] = hist[k][1];
452 hist[k][1] = tmp;
453 }
454
455 *output++ = res;
456 res = 0;
457 }
458 }
459 }
460
461 99 av_cold void ff_dcadsp_init(DCADSPContext *s)
462 {
463 99 s->decode_hf = decode_hf_c;
464 99 s->decode_joint = decode_joint_c;
465
466 99 s->lfe_fir_float[0] = lfe_fir0_float_c;
467 99 s->lfe_fir_float[1] = lfe_fir1_float_c;
468 99 s->lfe_x96_float = lfe_x96_float_c;
469 99 s->sub_qmf_float[0] = sub_qmf32_float_c;
470 99 s->sub_qmf_float[1] = sub_qmf64_float_c;
471
472 99 s->lfe_fir_fixed = lfe_fir_fixed_c;
473 99 s->lfe_x96_fixed = lfe_x96_fixed_c;
474 99 s->sub_qmf_fixed[0] = sub_qmf32_fixed_c;
475 99 s->sub_qmf_fixed[1] = sub_qmf64_fixed_c;
476
477 99 s->decor = decor_c;
478
479 99 s->dmix_sub_xch = dmix_sub_xch_c;
480 99 s->dmix_sub = dmix_sub_c;
481 99 s->dmix_add = dmix_add_c;
482 99 s->dmix_scale = dmix_scale_c;
483 99 s->dmix_scale_inv = dmix_scale_inv_c;
484
485 99 s->assemble_freq_bands = assemble_freq_bands_c;
486
487 99 s->lbr_bank = lbr_bank_c;
488 99 s->lfe_iir = lfe_iir_c;
489
490 #if ARCH_X86
491 99 ff_dcadsp_init_x86(s);
492 #endif
493 99 }
494