2 * Copyright (C) 2016 foo86
4 * This file is part of FFmpeg.
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.
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.
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
21 #include "libavutil/mem_internal.h"
26 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)
35 for (i = sb_start; i < sb_end; i++) {
36 const int8_t *coeff = hf_vq[vq_index[i]];
37 int32_t scale = scale_factors[i][0];
38 for (j = 0; j < len; j++)
39 dst[i][j + ofs] = clip23(coeff[j] * scale + (1 << 3) >> 4);
43 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)
50 for (i = sb_start; i < sb_end; i++) {
51 int32_t scale = scale_factors[i];
52 for (j = 0; j < len; j++)
53 dst[i][j + ofs] = clip23(mul17(src[i][j + ofs], scale));
57 static void lfe_fir_float_c(float *pcm_samples, int32_t *lfe_samples,
58 const float *filter_coeff, ptrdiff_t npcmblocks,
61 // Select decimation factor
62 int factor = 64 << dec_select;
63 int ncoeffs = 8 >> dec_select;
64 int nlfesamples = npcmblocks >> (dec_select + 1);
67 for (i = 0; i < nlfesamples; i++) {
68 // One decimated sample generates 64 or 128 interpolated ones
69 for (j = 0; j < factor / 2; j++) {
73 for (k = 0; k < ncoeffs; k++) {
74 a += filter_coeff[ j * ncoeffs + k] * lfe_samples[-k];
75 b += filter_coeff[255 - j * ncoeffs - k] * lfe_samples[-k];
79 pcm_samples[factor / 2 + j] = b;
83 pcm_samples += factor;
87 static void lfe_fir0_float_c(float *pcm_samples, int32_t *lfe_samples,
88 const float *filter_coeff, ptrdiff_t npcmblocks)
90 lfe_fir_float_c(pcm_samples, lfe_samples, filter_coeff, npcmblocks, 0);
93 static void lfe_fir1_float_c(float *pcm_samples, int32_t *lfe_samples,
94 const float *filter_coeff, ptrdiff_t npcmblocks)
96 lfe_fir_float_c(pcm_samples, lfe_samples, filter_coeff, npcmblocks, 1);
99 static void lfe_x96_float_c(float *dst, const float *src,
100 float *hist, ptrdiff_t len)
105 for (i = 0; i < len; i++) {
106 float a = 0.25f * src[i] + 0.75f * prev;
107 float b = 0.75f * src[i] + 0.25f * prev;
116 static void sub_qmf32_float_c(SynthFilterContext *synth,
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,
126 LOCAL_ALIGNED_32(float, input, [32]);
129 for (j = 0; j < npcmblocks; j++) {
130 // Load in one sample from each subband
131 for (i = 0; i < 32; i++) {
133 input[i] = -subband_samples_lo[i][j];
135 input[i] = subband_samples_lo[i][j];
138 // One subband sample generates 32 interpolated ones
139 synth->synth_filter_float(imdct, hist1, offset,
141 pcm_samples, input, scale, imdct_fn);
146 static void sub_qmf64_float_c(SynthFilterContext *synth,
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,
156 LOCAL_ALIGNED_32(float, input, [64]);
159 if (!subband_samples_hi)
160 memset(&input[32], 0, sizeof(input[0]) * 32);
162 for (j = 0; j < npcmblocks; j++) {
163 // Load in one sample from each subband
164 if (subband_samples_hi) {
165 // Full 64 subbands, first 32 are residual coded
166 for (i = 0; i < 32; i++) {
168 input[i] = -subband_samples_lo[i][j] - subband_samples_hi[i][j];
170 input[i] = subband_samples_lo[i][j] + subband_samples_hi[i][j];
172 for (i = 32; i < 64; i++) {
174 input[i] = -subband_samples_hi[i][j];
176 input[i] = subband_samples_hi[i][j];
179 // Only first 32 subbands
180 for (i = 0; i < 32; i++) {
182 input[i] = -subband_samples_lo[i][j];
184 input[i] = subband_samples_lo[i][j];
188 // One subband sample generates 64 interpolated ones
189 synth->synth_filter_float_64(imdct, hist1, offset,
191 pcm_samples, input, scale, imdct_fn);
196 static void lfe_fir_fixed_c(int32_t *pcm_samples, int32_t *lfe_samples,
197 const int32_t *filter_coeff, ptrdiff_t npcmblocks)
199 // Select decimation factor
200 int nlfesamples = npcmblocks >> 1;
203 for (i = 0; i < nlfesamples; i++) {
204 // One decimated sample generates 64 interpolated ones
205 for (j = 0; j < 32; j++) {
209 for (k = 0; k < 8; k++) {
210 a += (int64_t)filter_coeff[ j * 8 + k] * lfe_samples[-k];
211 b += (int64_t)filter_coeff[255 - j * 8 - k] * lfe_samples[-k];
214 pcm_samples[ j] = clip23(norm23(a));
215 pcm_samples[32 + j] = clip23(norm23(b));
223 static void lfe_x96_fixed_c(int32_t *dst, const int32_t *src,
224 int32_t *hist, ptrdiff_t len)
226 int32_t prev = *hist;
229 for (i = 0; i < len; i++) {
230 int64_t a = INT64_C(2097471) * src[i] + INT64_C(6291137) * prev;
231 int64_t b = INT64_C(6291137) * src[i] + INT64_C(2097471) * prev;
233 *dst++ = clip23(norm23(a));
234 *dst++ = clip23(norm23(b));
240 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)
248 LOCAL_ALIGNED_32(int32_t, input, [32]);
251 for (j = 0; j < npcmblocks; j++) {
252 // Load in one sample from each subband
253 for (i = 0; i < 32; i++)
254 input[i] = subband_samples_lo[i][j];
256 // One subband sample generates 32 interpolated ones
257 synth->synth_filter_fixed(imdct, hist1, offset,
264 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)
272 LOCAL_ALIGNED_32(int32_t, input, [64]);
275 if (!subband_samples_hi)
276 memset(&input[32], 0, sizeof(input[0]) * 32);
278 for (j = 0; j < npcmblocks; j++) {
279 // Load in one sample from each subband
280 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];
287 // Only first 32 subbands
288 for (i = 0; i < 32; i++)
289 input[i] = subband_samples_lo[i][j];
292 // One subband sample generates 64 interpolated ones
293 synth->synth_filter_fixed_64(imdct, hist1, offset,
300 static void decor_c(int32_t *dst, const int32_t *src, int coeff, ptrdiff_t len)
304 for (i = 0; i < len; i++)
305 dst[i] += (SUINT)((int)(src[i] * (SUINT)coeff + (1 << 2)) >> 3);
308 static void dmix_sub_xch_c(int32_t *dst1, int32_t *dst2,
309 const int32_t *src, ptrdiff_t len)
313 for (i = 0; i < len; i++) {
314 int32_t cs = mul23(src[i], 5931520 /* M_SQRT1_2 * (1 << 23) */);
320 static void dmix_sub_c(int32_t *dst, const int32_t *src, int coeff, ptrdiff_t len)
324 for (i = 0; i < len; i++)
325 dst[i] -= (unsigned)mul15(src[i], coeff);
328 static void dmix_add_c(int32_t *dst, const int32_t *src, int coeff, ptrdiff_t len)
332 for (i = 0; i < len; i++)
333 dst[i] += (unsigned)mul15(src[i], coeff);
336 static void dmix_scale_c(int32_t *dst, int scale, ptrdiff_t len)
340 for (i = 0; i < len; i++)
341 dst[i] = mul15(dst[i], scale);
344 static void dmix_scale_inv_c(int32_t *dst, int scale_inv, ptrdiff_t len)
348 for (i = 0; i < len; i++)
349 dst[i] = mul16(dst[i], scale_inv);
352 static void filter0(SUINT32 *dst, const int32_t *src, int32_t coeff, ptrdiff_t len)
356 for (i = 0; i < len; i++)
357 dst[i] -= mul22(src[i], coeff);
360 static void filter1(SUINT32 *dst, const int32_t *src, int32_t coeff, ptrdiff_t len)
364 for (i = 0; i < len; i++)
365 dst[i] -= mul23(src[i], coeff);
368 static void assemble_freq_bands_c(int32_t *dst, int32_t *src0, int32_t *src1,
369 const int32_t *coeff, ptrdiff_t len)
373 filter0(src0, src1, coeff[0], len);
374 filter0(src1, src0, coeff[1], len);
375 filter0(src0, src1, coeff[2], len);
376 filter0(src1, src0, coeff[3], len);
378 for (i = 0; i < 8; i++, src0--) {
379 filter1(src0, src1, coeff[i + 4], len);
380 filter1(src1, src0, coeff[i + 12], len);
381 filter1(src0, src1, coeff[i + 4], len);
384 for (i = 0; i < len; i++) {
390 static void lbr_bank_c(float output[32][4], float **input,
391 const float *coeff, ptrdiff_t ofs, ptrdiff_t len)
393 float SW0 = coeff[0];
394 float SW1 = coeff[1];
395 float SW2 = coeff[2];
396 float SW3 = coeff[3];
403 float AL1 = coeff[8];
404 float AL2 = coeff[9];
408 // Short window and 8 point forward MDCT
409 for (i = 0; i < len; i++) {
410 float *src = input[i] + ofs;
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;
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;
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;
436 static void lfe_iir_c(float *output, const float *input,
437 const float iir[5][4], float hist[5][2],
443 for (i = 0; i < 64; i++) {
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;
451 hist[k][0] = hist[k][1];
461 av_cold void ff_dcadsp_init(DCADSPContext *s)
463 s->decode_hf = decode_hf_c;
464 s->decode_joint = decode_joint_c;
466 s->lfe_fir_float[0] = lfe_fir0_float_c;
467 s->lfe_fir_float[1] = lfe_fir1_float_c;
468 s->lfe_x96_float = lfe_x96_float_c;
469 s->sub_qmf_float[0] = sub_qmf32_float_c;
470 s->sub_qmf_float[1] = sub_qmf64_float_c;
472 s->lfe_fir_fixed = lfe_fir_fixed_c;
473 s->lfe_x96_fixed = lfe_x96_fixed_c;
474 s->sub_qmf_fixed[0] = sub_qmf32_fixed_c;
475 s->sub_qmf_fixed[1] = sub_qmf64_fixed_c;
479 s->dmix_sub_xch = dmix_sub_xch_c;
480 s->dmix_sub = dmix_sub_c;
481 s->dmix_add = dmix_add_c;
482 s->dmix_scale = dmix_scale_c;
483 s->dmix_scale_inv = dmix_scale_inv_c;
485 s->assemble_freq_bands = assemble_freq_bands_c;
487 s->lbr_bank = lbr_bank_c;
488 s->lfe_iir = lfe_iir_c;
491 ff_dcadsp_init_x86(s);