2 * AAC Spectral Band Replication decoding functions
3 * Copyright (c) 2008-2009 Robert Swain ( rob opendot cl )
4 * Copyright (c) 2009-2010 Alex Converse <alex.converse@gmail.com>
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 * Note: Rounding-to-nearest used unless otherwise stated
30 #include "libavutil/attributes.h"
31 #include "libavutil/intfloat.h"
34 static SoftFloat sbr_sum_square_c(int (*x)[2], int n)
38 uint64_t accu0 = 0, accu1 = 0, accu2 = 0, accu3 = 0;
42 for (i = 0; i < n; i += 2) {
43 // Larger values are inavlid and could cause overflows of accu.
44 av_assert2(FFABS(x[i + 0][0]) >> 30 == 0);
45 accu0 += (int64_t)x[i + 0][0] * x[i + 0][0];
46 av_assert2(FFABS(x[i + 0][1]) >> 30 == 0);
47 accu1 += (int64_t)x[i + 0][1] * x[i + 0][1];
48 av_assert2(FFABS(x[i + 1][0]) >> 30 == 0);
49 accu2 += (int64_t)x[i + 1][0] * x[i + 1][0];
50 av_assert2(FFABS(x[i + 1][1]) >> 30 == 0);
51 accu3 += (int64_t)x[i + 1][1] * x[i + 1][1];
55 while ((accu0|accu1|accu2|accu3) >> 62) {
62 accu = accu0 + accu1 + accu2 + accu3;
67 while (u < 0x80000000U) {
74 round = 1ULL << (nz-1);
75 u = ((accu + round) >> nz);
77 ret = av_int2sf(u, nz0 - nz);
82 static void sbr_neg_odd_64_c(int *x)
85 for (i = 1; i < 64; i += 2)
89 static void sbr_qmf_pre_shuffle_c(int *z)
94 for (k = 1; k < 32; k++) {
95 z[64+2*k ] = -z[64 - k];
96 z[64+2*k+1] = z[ k + 1];
100 static void sbr_qmf_post_shuffle_c(int W[32][2], const int *z)
103 for (k = 0; k < 32; k++) {
109 static void sbr_qmf_deint_neg_c(int *v, const int *src)
112 for (i = 0; i < 32; i++) {
113 v[ i] = ( src[63 - 2*i ] + 0x10) >> 5;
114 v[63 - i] = (-src[63 - 2*i - 1] + 0x10) >> 5;
118 static av_always_inline SoftFloat autocorr_calc(int64_t accu)
122 int i = (int)(accu >> 32);
127 while (FFABS(i) < 0x40000000) {
134 round = 1U << (nz-1);
135 mant = (int)((accu + round) >> nz);
136 mant = (mant + 0x40)>>7;
139 return av_int2sf(mant, 30 - expo);
142 static av_always_inline void autocorrelate(const int x[40][2], SoftFloat phi[3][2][2], int lag)
145 int64_t real_sum, imag_sum;
146 int64_t accu_re = 0, accu_im = 0;
149 for (i = 1; i < 38; i++) {
150 accu_re += (int64_t)x[i][0] * x[i+lag][0];
151 accu_re += (int64_t)x[i][1] * x[i+lag][1];
152 accu_im += (int64_t)x[i][0] * x[i+lag][1];
153 accu_im -= (int64_t)x[i][1] * x[i+lag][0];
159 accu_re += (int64_t)x[ 0][0] * x[lag][0];
160 accu_re += (int64_t)x[ 0][1] * x[lag][1];
161 accu_im += (int64_t)x[ 0][0] * x[lag][1];
162 accu_im -= (int64_t)x[ 0][1] * x[lag][0];
164 phi[2-lag][1][0] = autocorr_calc(accu_re);
165 phi[2-lag][1][1] = autocorr_calc(accu_im);
170 accu_re += (int64_t)x[38][0] * x[39][0];
171 accu_re += (int64_t)x[38][1] * x[39][1];
172 accu_im += (int64_t)x[38][0] * x[39][1];
173 accu_im -= (int64_t)x[38][1] * x[39][0];
175 phi[0][0][0] = autocorr_calc(accu_re);
176 phi[0][0][1] = autocorr_calc(accu_im);
179 for (i = 1; i < 38; i++) {
180 accu_re += (int64_t)x[i][0] * x[i][0];
181 accu_re += (int64_t)x[i][1] * x[i][1];
184 accu_re += (int64_t)x[ 0][0] * x[ 0][0];
185 accu_re += (int64_t)x[ 0][1] * x[ 0][1];
187 phi[2][1][0] = autocorr_calc(accu_re);
190 accu_re += (int64_t)x[38][0] * x[38][0];
191 accu_re += (int64_t)x[38][1] * x[38][1];
193 phi[1][0][0] = autocorr_calc(accu_re);
197 static void sbr_autocorrelate_c(const int x[40][2], SoftFloat phi[3][2][2])
199 autocorrelate(x, phi, 0);
200 autocorrelate(x, phi, 1);
201 autocorrelate(x, phi, 2);
204 static void sbr_hf_gen_c(int (*X_high)[2], const int (*X_low)[2],
205 const int alpha0[2], const int alpha1[2],
206 int bw, int start, int end)
212 accu = (int64_t)alpha0[0] * bw;
213 alpha[2] = (int)((accu + 0x40000000) >> 31);
214 accu = (int64_t)alpha0[1] * bw;
215 alpha[3] = (int)((accu + 0x40000000) >> 31);
216 accu = (int64_t)bw * bw;
217 bw = (int)((accu + 0x40000000) >> 31);
218 accu = (int64_t)alpha1[0] * bw;
219 alpha[0] = (int)((accu + 0x40000000) >> 31);
220 accu = (int64_t)alpha1[1] * bw;
221 alpha[1] = (int)((accu + 0x40000000) >> 31);
223 for (i = start; i < end; i++) {
224 accu = (int64_t)X_low[i][0] * 0x20000000;
225 accu += (int64_t)X_low[i - 2][0] * alpha[0];
226 accu -= (int64_t)X_low[i - 2][1] * alpha[1];
227 accu += (int64_t)X_low[i - 1][0] * alpha[2];
228 accu -= (int64_t)X_low[i - 1][1] * alpha[3];
229 X_high[i][0] = (int)((accu + 0x10000000) >> 29);
231 accu = (int64_t)X_low[i][1] * 0x20000000;
232 accu += (int64_t)X_low[i - 2][1] * alpha[0];
233 accu += (int64_t)X_low[i - 2][0] * alpha[1];
234 accu += (int64_t)X_low[i - 1][1] * alpha[2];
235 accu += (int64_t)X_low[i - 1][0] * alpha[3];
236 X_high[i][1] = (int)((accu + 0x10000000) >> 29);
240 static void sbr_hf_g_filt_c(int (*Y)[2], const int (*X_high)[40][2],
241 const SoftFloat *g_filt, int m_max, intptr_t ixh)
246 for (m = 0; m < m_max; m++) {
247 int64_t r = 1LL << (22-g_filt[m].exp);
248 accu = (int64_t)X_high[m][ixh][0] * ((g_filt[m].mant + 0x40)>>7);
249 Y[m][0] = (int)((accu + r) >> (23-g_filt[m].exp));
251 accu = (int64_t)X_high[m][ixh][1] * ((g_filt[m].mant + 0x40)>>7);
252 Y[m][1] = (int)((accu + r) >> (23-g_filt[m].exp));
256 static av_always_inline int sbr_hf_apply_noise(int (*Y)[2],
257 const SoftFloat *s_m,
258 const SoftFloat *q_filt,
266 for (m = 0; m < m_max; m++) {
269 noise = (noise + 1) & 0x1ff;
273 shift = 22 - s_m[m].exp;
275 av_log(NULL, AV_LOG_ERROR, "Overflow in sbr_hf_apply_noise, shift=%d\n", shift);
276 return AVERROR(ERANGE);
277 } else if (shift < 30) {
278 round = 1 << (shift-1);
279 y0 += (s_m[m].mant * phi_sign0 + round) >> shift;
280 y1 += (s_m[m].mant * phi_sign1 + round) >> shift;
283 int shift, round, tmp;
286 shift = 22 - q_filt[m].exp;
288 av_log(NULL, AV_LOG_ERROR, "Overflow in sbr_hf_apply_noise, shift=%d\n", shift);
289 return AVERROR(ERANGE);
290 } else if (shift < 30) {
291 round = 1 << (shift-1);
293 accu = (int64_t)q_filt[m].mant * ff_sbr_noise_table_fixed[noise][0];
294 tmp = (int)((accu + 0x40000000) >> 31);
295 y0 += (tmp + round) >> shift;
297 accu = (int64_t)q_filt[m].mant * ff_sbr_noise_table_fixed[noise][1];
298 tmp = (int)((accu + 0x40000000) >> 31);
299 y1 += (tmp + round) >> shift;
304 phi_sign1 = -phi_sign1;
309 #include "sbrdsp_template.c"