1 /*
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>
5 *
6 * This file is part of FFmpeg.
7 *
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.
12 *
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.
17 *
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
21 *
22 * Note: Rounding-to-nearest used unless otherwise stated
23 *
24 */
25
26 #define USE_FIXED 1
27
28 #include "aac.h"
29 #include "config.h"
30 #include "libavutil/attributes.h"
31 #include "libavutil/intfloat.h"
32 #include "sbrdsp.h"
33
sbr_sum_square_c(int (* x)[2],int n)34 static SoftFloat sbr_sum_square_c(int (*x)[2], int n)
35 {
36 SoftFloat ret;
37 uint64_t accu = 0, round;
38 uint64_t accu0 = 0, accu1 = 0, accu2 = 0, accu3 = 0;
39 int i, nz, nz0;
40 unsigned u;
41
42 nz = 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) {
49 accu0 >>= nz;
50 accu1 >>= nz;
51 accu2 >>= nz;
52 accu3 >>= nz;
53 while ((accu0|accu1|accu2|accu3) > (UINT64_MAX - accu) >> 2) {
54 accu0 >>= 1;
55 accu1 >>= 1;
56 accu2 >>= 1;
57 accu3 >>= 1;
58 accu >>= 1;
59 nz ++;
60 }
61 accu += accu0 + accu1 + accu2 + accu3;
62 accu0 = accu1 = accu2 = accu3 = 0;
63 }
64 }
65
66 nz0 = 15 - nz;
67
68 u = accu >> 32;
69 if (u) {
70 nz = 33;
71 while (u < 0x80000000U) {
72 u <<= 1;
73 nz--;
74 }
75 } else
76 nz = 1;
77
78 round = 1ULL << (nz-1);
79 u = ((accu + round) >> nz);
80 u >>= 1;
81 ret = av_int2sf(u, nz0 - nz);
82
83 return ret;
84 }
85
sbr_neg_odd_64_c(int * x)86 static void sbr_neg_odd_64_c(int *x)
87 {
88 int i;
89 for (i = 1; i < 64; i += 2)
90 x[i] = -(unsigned)x[i];
91 }
92
sbr_qmf_pre_shuffle_c(int * z)93 static void sbr_qmf_pre_shuffle_c(int *z)
94 {
95 int k;
96 z[64] = z[0];
97 z[65] = z[1];
98 for (k = 1; k < 32; k++) {
99 z[64+2*k ] = -z[64 - k];
100 z[64+2*k+1] = z[ k + 1];
101 }
102 }
103
sbr_qmf_post_shuffle_c(int W[32][2],const int * z)104 static void sbr_qmf_post_shuffle_c(int W[32][2], const int *z)
105 {
106 int k;
107 for (k = 0; k < 32; k++) {
108 W[k][0] = -z[63-k];
109 W[k][1] = z[k];
110 }
111 }
112
sbr_qmf_deint_neg_c(int * v,const int * src)113 static void sbr_qmf_deint_neg_c(int *v, const int *src)
114 {
115 int i;
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;
119 }
120 }
121
autocorr_calc(int64_t accu)122 static av_always_inline SoftFloat autocorr_calc(int64_t accu)
123 {
124 int nz, mant, expo;
125 unsigned round;
126 int i = (int)(accu >> 32);
127 if (i == 0) {
128 nz = 1;
129 } else {
130 nz = 0;
131 while (FFABS(i) < 0x40000000) {
132 i *= 2;
133 nz++;
134 }
135 nz = 32-nz;
136 }
137
138 round = 1U << (nz-1);
139 mant = (int)((accu + round) >> nz);
140 mant = (mant + 0x40LL)>>7;
141 mant *= 64;
142 expo = nz + 15;
143 return av_int2sf(mant, 30 - expo);
144 }
145
autocorrelate(const int x[40][2],SoftFloat phi[3][2][2],int lag)146 static av_always_inline void autocorrelate(const int x[40][2], SoftFloat phi[3][2][2], int lag)
147 {
148 int i;
149 int64_t real_sum, imag_sum;
150 int64_t accu_re = 0, accu_im = 0;
151
152 if (lag) {
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];
158 }
159
160 real_sum = accu_re;
161 imag_sum = accu_im;
162
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];
167
168 phi[2-lag][1][0] = autocorr_calc(accu_re);
169 phi[2-lag][1][1] = autocorr_calc(accu_im);
170
171 if (lag == 1) {
172 accu_re = real_sum;
173 accu_im = imag_sum;
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];
178
179 phi[0][0][0] = autocorr_calc(accu_re);
180 phi[0][0][1] = autocorr_calc(accu_im);
181 }
182 } else {
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];
186 }
187 real_sum = accu_re;
188 accu_re += (uint64_t)x[ 0][0] * x[ 0][0];
189 accu_re += (uint64_t)x[ 0][1] * x[ 0][1];
190
191 phi[2][1][0] = autocorr_calc(accu_re);
192
193 accu_re = real_sum;
194 accu_re += (uint64_t)x[38][0] * x[38][0];
195 accu_re += (uint64_t)x[38][1] * x[38][1];
196
197 phi[1][0][0] = autocorr_calc(accu_re);
198 }
199 }
200
sbr_autocorrelate_c(const int x[40][2],SoftFloat phi[3][2][2])201 static void sbr_autocorrelate_c(const int x[40][2], SoftFloat phi[3][2][2])
202 {
203 autocorrelate(x, phi, 0);
204 autocorrelate(x, phi, 1);
205 autocorrelate(x, phi, 2);
206 }
207
sbr_hf_gen_c(int (* X_high)[2],const int (* X_low)[2],const int alpha0[2],const int alpha1[2],int bw,int start,int end)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)
211 {
212 int alpha[4];
213 int i;
214 int64_t accu;
215
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);
226
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);
234
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);
241 }
242 }
243
sbr_hf_g_filt_c(int (* Y)[2],const int (* X_high)[40][2],const SoftFloat * g_filt,int m_max,intptr_t ixh)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)
246 {
247 int m;
248 int64_t accu;
249
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));
255
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));
258 }
259 }
260 }
261
sbr_hf_apply_noise(int (* Y)[2],const SoftFloat * s_m,const SoftFloat * q_filt,int noise,int phi_sign0,int phi_sign1,int m_max)262 static av_always_inline int sbr_hf_apply_noise(int (*Y)[2],
263 const SoftFloat *s_m,
264 const SoftFloat *q_filt,
265 int noise,
266 int phi_sign0,
267 int phi_sign1,
268 int m_max)
269 {
270 int m;
271
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;
276 if (s_m[m].mant) {
277 int shift, round;
278
279 shift = 22 - s_m[m].exp;
280 if (shift < 1) {
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;
287 }
288 } else {
289 int shift, round, tmp;
290 int64_t accu;
291
292 shift = 22 - q_filt[m].exp;
293 if (shift < 1) {
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);
298
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;
302
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;
306 }
307 }
308 Y[m][0] = y0;
309 Y[m][1] = y1;
310 phi_sign1 = -phi_sign1;
311 }
312 return 0;
313 }
314
315 #include "sbrdsp_template.c"
316