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)
37 uint64_t accu = 0, round;
38 uint64_t accu0 = 0, accu1 = 0, accu2 = 0, accu3 = 0;
43 for (i = 0; i < n; i += 2) {
44 accu0 += (int64_t)x[i + 0][0] * x[i + 0][0];
45 accu1 += (int64_t)x[i + 0][1] * x[i + 0][1];
46 accu2 += (int64_t)x[i + 1][0] * x[i + 1][0];
47 accu3 += (int64_t)x[i + 1][1] * x[i + 1][1];
48 if ((accu0|accu1|accu2|accu3) > UINT64_MAX - INT32_MIN*(int64_t)INT32_MIN || i+2>=n) {
53 while ((accu0|accu1|accu2|accu3) > (UINT64_MAX - accu) >> 2) {
61 accu += accu0 + accu1 + accu2 + accu3;
62 accu0 = accu1 = accu2 = accu3 = 0;
71 while (u < 0x80000000U) {
78 round = 1ULL << (nz-1);
79 u = ((accu + round) >> nz);
81 ret = av_int2sf(u, nz0 - nz);
86 static void sbr_neg_odd_64_c(int *x)
89 for (i = 1; i < 64; i += 2)
93 static void sbr_qmf_pre_shuffle_c(int *z)
98 for (k = 1; k < 32; k++) {
99 z[64+2*k ] = -z[64 - k];
100 z[64+2*k+1] = z[ k + 1];
104 static void sbr_qmf_post_shuffle_c(int W[32][2], const int *z)
107 for (k = 0; k < 32; k++) {
113 static void sbr_qmf_deint_neg_c(int *v, const int *src)
116 for (i = 0; i < 32; i++) {
117 v[ i] = ( src[63 - 2*i ] + 0x10) >> 5;
118 v[63 - i] = (-src[63 - 2*i - 1] + 0x10) >> 5;
122 static av_always_inline SoftFloat autocorr_calc(int64_t accu)
126 int i = (int)(accu >> 32);
131 while (FFABS(i) < 0x40000000) {
138 round = 1U << (nz-1);
139 mant = (int)((accu + round) >> nz);
140 mant = (mant + 0x40LL)>>7;
143 return av_int2sf(mant, 30 - expo);
146 static av_always_inline void autocorrelate(const int x[40][2], SoftFloat phi[3][2][2], int lag)
149 int64_t real_sum, imag_sum;
150 int64_t accu_re = 0, accu_im = 0;
153 for (i = 1; i < 38; i++) {
154 accu_re += (uint64_t)x[i][0] * x[i+lag][0];
155 accu_re += (uint64_t)x[i][1] * x[i+lag][1];
156 accu_im += (uint64_t)x[i][0] * x[i+lag][1];
157 accu_im -= (uint64_t)x[i][1] * x[i+lag][0];
163 accu_re += (uint64_t)x[ 0][0] * x[lag][0];
164 accu_re += (uint64_t)x[ 0][1] * x[lag][1];
165 accu_im += (uint64_t)x[ 0][0] * x[lag][1];
166 accu_im -= (uint64_t)x[ 0][1] * x[lag][0];
168 phi[2-lag][1][0] = autocorr_calc(accu_re);
169 phi[2-lag][1][1] = autocorr_calc(accu_im);
174 accu_re += (uint64_t)x[38][0] * x[39][0];
175 accu_re += (uint64_t)x[38][1] * x[39][1];
176 accu_im += (uint64_t)x[38][0] * x[39][1];
177 accu_im -= (uint64_t)x[38][1] * x[39][0];
179 phi[0][0][0] = autocorr_calc(accu_re);
180 phi[0][0][1] = autocorr_calc(accu_im);
183 for (i = 1; i < 38; i++) {
184 accu_re += (uint64_t)x[i][0] * x[i][0];
185 accu_re += (uint64_t)x[i][1] * x[i][1];
188 accu_re += (uint64_t)x[ 0][0] * x[ 0][0];
189 accu_re += (uint64_t)x[ 0][1] * x[ 0][1];
191 phi[2][1][0] = autocorr_calc(accu_re);
194 accu_re += (uint64_t)x[38][0] * x[38][0];
195 accu_re += (uint64_t)x[38][1] * x[38][1];
197 phi[1][0][0] = autocorr_calc(accu_re);
201 static void sbr_autocorrelate_c(const int x[40][2], SoftFloat phi[3][2][2])
203 autocorrelate(x, phi, 0);
204 autocorrelate(x, phi, 1);
205 autocorrelate(x, phi, 2);
208 static void sbr_hf_gen_c(int (*X_high)[2], const int (*X_low)[2],
209 const int alpha0[2], const int alpha1[2],
210 int bw, int start, int end)
216 accu = (int64_t)alpha0[0] * bw;
217 alpha[2] = (int)((accu + 0x40000000) >> 31);
218 accu = (int64_t)alpha0[1] * bw;
219 alpha[3] = (int)((accu + 0x40000000) >> 31);
220 accu = (int64_t)bw * bw;
221 bw = (int)((accu + 0x40000000) >> 31);
222 accu = (int64_t)alpha1[0] * bw;
223 alpha[0] = (int)((accu + 0x40000000) >> 31);
224 accu = (int64_t)alpha1[1] * bw;
225 alpha[1] = (int)((accu + 0x40000000) >> 31);
227 for (i = start; i < end; i++) {
228 accu = (int64_t)X_low[i][0] * 0x20000000;
229 accu += (int64_t)X_low[i - 2][0] * alpha[0];
230 accu -= (int64_t)X_low[i - 2][1] * alpha[1];
231 accu += (int64_t)X_low[i - 1][0] * alpha[2];
232 accu -= (int64_t)X_low[i - 1][1] * alpha[3];
233 X_high[i][0] = (int)((accu + 0x10000000) >> 29);
235 accu = (int64_t)X_low[i][1] * 0x20000000;
236 accu += (int64_t)X_low[i - 2][1] * alpha[0];
237 accu += (int64_t)X_low[i - 2][0] * alpha[1];
238 accu += (int64_t)X_low[i - 1][1] * alpha[2];
239 accu += (int64_t)X_low[i - 1][0] * alpha[3];
240 X_high[i][1] = (int)((accu + 0x10000000) >> 29);
244 static void sbr_hf_g_filt_c(int (*Y)[2], const int (*X_high)[40][2],
245 const SoftFloat *g_filt, int m_max, intptr_t ixh)
250 for (m = 0; m < m_max; m++) {
251 if (22 - g_filt[m].exp < 61) {
252 int64_t r = 1LL << (22-g_filt[m].exp);
253 accu = (int64_t)X_high[m][ixh][0] * ((g_filt[m].mant + 0x40)>>7);
254 Y[m][0] = (int)((accu + r) >> (23-g_filt[m].exp));
256 accu = (int64_t)X_high[m][ixh][1] * ((g_filt[m].mant + 0x40)>>7);
257 Y[m][1] = (int)((accu + r) >> (23-g_filt[m].exp));
262 static av_always_inline int sbr_hf_apply_noise(int (*Y)[2],
263 const SoftFloat *s_m,
264 const SoftFloat *q_filt,
272 for (m = 0; m < m_max; m++) {
273 unsigned y0 = Y[m][0];
274 unsigned y1 = Y[m][1];
275 noise = (noise + 1) & 0x1ff;
279 shift = 22 - s_m[m].exp;
281 av_log(NULL, AV_LOG_ERROR, "Overflow in sbr_hf_apply_noise, shift=%d\n", shift);
282 return AVERROR(ERANGE);
283 } else if (shift < 30) {
284 round = 1 << (shift-1);
285 y0 += (s_m[m].mant * phi_sign0 + round) >> shift;
286 y1 += (s_m[m].mant * phi_sign1 + round) >> shift;
289 int shift, round, tmp;
292 shift = 22 - q_filt[m].exp;
294 av_log(NULL, AV_LOG_ERROR, "Overflow in sbr_hf_apply_noise, shift=%d\n", shift);
295 return AVERROR(ERANGE);
296 } else if (shift < 30) {
297 round = 1 << (shift-1);
299 accu = (int64_t)q_filt[m].mant * ff_sbr_noise_table_fixed[noise][0];
300 tmp = (int)((accu + 0x40000000) >> 31);
301 y0 += (tmp + round) >> shift;
303 accu = (int64_t)q_filt[m].mant * ff_sbr_noise_table_fixed[noise][1];
304 tmp = (int)((accu + 0x40000000) >> 31);
305 y1 += (tmp + round) >> shift;
310 phi_sign1 = -phi_sign1;
315 #include "sbrdsp_template.c"