FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
sbrdsp.c
Go to the documentation of this file.
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 
23 #include "config.h"
24 #include "libavutil/attributes.h"
25 #include "libavutil/intfloat.h"
26 #include "sbrdsp.h"
27 
28 static void sbr_sum64x5_c(float *z)
29 {
30  int k;
31  for (k = 0; k < 64; k++) {
32  float f = z[k] + z[k + 64] + z[k + 128] + z[k + 192] + z[k + 256];
33  z[k] = f;
34  }
35 }
36 
37 static float sbr_sum_square_c(float (*x)[2], int n)
38 {
39  float sum0 = 0.0f, sum1 = 0.0f;
40  int i;
41 
42  for (i = 0; i < n; i += 2)
43  {
44  sum0 += x[i + 0][0] * x[i + 0][0];
45  sum1 += x[i + 0][1] * x[i + 0][1];
46  sum0 += x[i + 1][0] * x[i + 1][0];
47  sum1 += x[i + 1][1] * x[i + 1][1];
48  }
49 
50  return sum0 + sum1;
51 }
52 
53 static void sbr_neg_odd_64_c(float *x)
54 {
55  union av_intfloat32 *xi = (union av_intfloat32*) x;
56  int i;
57  for (i = 1; i < 64; i += 4) {
58  xi[i + 0].i ^= 1U << 31;
59  xi[i + 2].i ^= 1U << 31;
60  }
61 }
62 
63 static void sbr_qmf_pre_shuffle_c(float *z)
64 {
65  union av_intfloat32 *zi = (union av_intfloat32*) z;
66  int k;
67  zi[64].i = zi[0].i;
68  zi[65].i = zi[1].i;
69  for (k = 1; k < 31; k += 2) {
70  zi[64 + 2 * k + 0].i = zi[64 - k].i ^ (1U << 31);
71  zi[64 + 2 * k + 1].i = zi[ k + 1].i;
72  zi[64 + 2 * k + 2].i = zi[63 - k].i ^ (1U << 31);
73  zi[64 + 2 * k + 3].i = zi[ k + 2].i;
74  }
75 
76  zi[64 + 2 * 31 + 0].i = zi[64 - 31].i ^ (1U << 31);
77  zi[64 + 2 * 31 + 1].i = zi[31 + 1].i;
78 }
79 
80 static void sbr_qmf_post_shuffle_c(float W[32][2], const float *z)
81 {
82  const union av_intfloat32 *zi = (const union av_intfloat32*) z;
83  union av_intfloat32 *Wi = (union av_intfloat32*) W;
84  int k;
85  for (k = 0; k < 32; k += 2) {
86  Wi[2 * k + 0].i = zi[63 - k].i ^ (1U << 31);
87  Wi[2 * k + 1].i = zi[ k + 0].i;
88  Wi[2 * k + 2].i = zi[62 - k].i ^ (1U << 31);
89  Wi[2 * k + 3].i = zi[ k + 1].i;
90  }
91 }
92 
93 static void sbr_qmf_deint_neg_c(float *v, const float *src)
94 {
95  const union av_intfloat32 *si = (const union av_intfloat32*)src;
96  union av_intfloat32 *vi = (union av_intfloat32*)v;
97  int i;
98  for (i = 0; i < 32; i++) {
99  vi[ i].i = si[63 - 2 * i ].i;
100  vi[63 - i].i = si[63 - 2 * i - 1].i ^ (1U << 31);
101  }
102 }
103 
104 static void sbr_qmf_deint_bfly_c(float *v, const float *src0, const float *src1)
105 {
106  int i;
107  for (i = 0; i < 64; i++) {
108  v[ i] = src0[i] - src1[63 - i];
109  v[127 - i] = src0[i] + src1[63 - i];
110  }
111 }
112 
113 static av_always_inline void autocorrelate(const float x[40][2],
114  float phi[3][2][2], int lag)
115 {
116  int i;
117  float real_sum = 0.0f;
118  float imag_sum = 0.0f;
119  if (lag) {
120  for (i = 1; i < 38; i++) {
121  real_sum += x[i][0] * x[i+lag][0] + x[i][1] * x[i+lag][1];
122  imag_sum += x[i][0] * x[i+lag][1] - x[i][1] * x[i+lag][0];
123  }
124  phi[2-lag][1][0] = real_sum + x[ 0][0] * x[lag][0] + x[ 0][1] * x[lag][1];
125  phi[2-lag][1][1] = imag_sum + x[ 0][0] * x[lag][1] - x[ 0][1] * x[lag][0];
126  if (lag == 1) {
127  phi[0][0][0] = real_sum + x[38][0] * x[39][0] + x[38][1] * x[39][1];
128  phi[0][0][1] = imag_sum + x[38][0] * x[39][1] - x[38][1] * x[39][0];
129  }
130  } else {
131  for (i = 1; i < 38; i++) {
132  real_sum += x[i][0] * x[i][0] + x[i][1] * x[i][1];
133  }
134  phi[2][1][0] = real_sum + x[ 0][0] * x[ 0][0] + x[ 0][1] * x[ 0][1];
135  phi[1][0][0] = real_sum + x[38][0] * x[38][0] + x[38][1] * x[38][1];
136  }
137 }
138 
139 static void sbr_autocorrelate_c(const float x[40][2], float phi[3][2][2])
140 {
141 #if 0
142  /* This code is slower because it multiplies memory accesses.
143  * It is left for educational purposes and because it may offer
144  * a better reference for writing arch-specific DSP functions. */
145  autocorrelate(x, phi, 0);
146  autocorrelate(x, phi, 1);
147  autocorrelate(x, phi, 2);
148 #else
149  float real_sum2 = x[0][0] * x[2][0] + x[0][1] * x[2][1];
150  float imag_sum2 = x[0][0] * x[2][1] - x[0][1] * x[2][0];
151  float real_sum1 = 0.0f, imag_sum1 = 0.0f, real_sum0 = 0.0f;
152  int i;
153  for (i = 1; i < 38; i++) {
154  real_sum0 += x[i][0] * x[i ][0] + x[i][1] * x[i ][1];
155  real_sum1 += x[i][0] * x[i + 1][0] + x[i][1] * x[i + 1][1];
156  imag_sum1 += x[i][0] * x[i + 1][1] - x[i][1] * x[i + 1][0];
157  real_sum2 += x[i][0] * x[i + 2][0] + x[i][1] * x[i + 2][1];
158  imag_sum2 += x[i][0] * x[i + 2][1] - x[i][1] * x[i + 2][0];
159  }
160  phi[2 - 2][1][0] = real_sum2;
161  phi[2 - 2][1][1] = imag_sum2;
162  phi[2 ][1][0] = real_sum0 + x[ 0][0] * x[ 0][0] + x[ 0][1] * x[ 0][1];
163  phi[1 ][0][0] = real_sum0 + x[38][0] * x[38][0] + x[38][1] * x[38][1];
164  phi[2 - 1][1][0] = real_sum1 + x[ 0][0] * x[ 1][0] + x[ 0][1] * x[ 1][1];
165  phi[2 - 1][1][1] = imag_sum1 + x[ 0][0] * x[ 1][1] - x[ 0][1] * x[ 1][0];
166  phi[0 ][0][0] = real_sum1 + x[38][0] * x[39][0] + x[38][1] * x[39][1];
167  phi[0 ][0][1] = imag_sum1 + x[38][0] * x[39][1] - x[38][1] * x[39][0];
168 #endif
169 }
170 
171 static void sbr_hf_gen_c(float (*X_high)[2], const float (*X_low)[2],
172  const float alpha0[2], const float alpha1[2],
173  float bw, int start, int end)
174 {
175  float alpha[4];
176  int i;
177 
178  alpha[0] = alpha1[0] * bw * bw;
179  alpha[1] = alpha1[1] * bw * bw;
180  alpha[2] = alpha0[0] * bw;
181  alpha[3] = alpha0[1] * bw;
182 
183  for (i = start; i < end; i++) {
184  X_high[i][0] =
185  X_low[i - 2][0] * alpha[0] -
186  X_low[i - 2][1] * alpha[1] +
187  X_low[i - 1][0] * alpha[2] -
188  X_low[i - 1][1] * alpha[3] +
189  X_low[i][0];
190  X_high[i][1] =
191  X_low[i - 2][1] * alpha[0] +
192  X_low[i - 2][0] * alpha[1] +
193  X_low[i - 1][1] * alpha[2] +
194  X_low[i - 1][0] * alpha[3] +
195  X_low[i][1];
196  }
197 }
198 
199 static void sbr_hf_g_filt_c(float (*Y)[2], const float (*X_high)[40][2],
200  const float *g_filt, int m_max, intptr_t ixh)
201 {
202  int m;
203 
204  for (m = 0; m < m_max; m++) {
205  Y[m][0] = X_high[m][ixh][0] * g_filt[m];
206  Y[m][1] = X_high[m][ixh][1] * g_filt[m];
207  }
208 }
209 
210 static av_always_inline void sbr_hf_apply_noise(float (*Y)[2],
211  const float *s_m,
212  const float *q_filt,
213  int noise,
214  float phi_sign0,
215  float phi_sign1,
216  int m_max)
217 {
218  int m;
219 
220  for (m = 0; m < m_max; m++) {
221  float y0 = Y[m][0];
222  float y1 = Y[m][1];
223  noise = (noise + 1) & 0x1ff;
224  if (s_m[m]) {
225  y0 += s_m[m] * phi_sign0;
226  y1 += s_m[m] * phi_sign1;
227  } else {
228  y0 += q_filt[m] * ff_sbr_noise_table[noise][0];
229  y1 += q_filt[m] * ff_sbr_noise_table[noise][1];
230  }
231  Y[m][0] = y0;
232  Y[m][1] = y1;
233  phi_sign1 = -phi_sign1;
234  }
235 }
236 
237 static void sbr_hf_apply_noise_0(float (*Y)[2], const float *s_m,
238  const float *q_filt, int noise,
239  int kx, int m_max)
240 {
241  sbr_hf_apply_noise(Y, s_m, q_filt, noise, 1.0, 0.0, m_max);
242 }
243 
244 static void sbr_hf_apply_noise_1(float (*Y)[2], const float *s_m,
245  const float *q_filt, int noise,
246  int kx, int m_max)
247 {
248  float phi_sign = 1 - 2 * (kx & 1);
249  sbr_hf_apply_noise(Y, s_m, q_filt, noise, 0.0, phi_sign, m_max);
250 }
251 
252 static void sbr_hf_apply_noise_2(float (*Y)[2], const float *s_m,
253  const float *q_filt, int noise,
254  int kx, int m_max)
255 {
256  sbr_hf_apply_noise(Y, s_m, q_filt, noise, -1.0, 0.0, m_max);
257 }
258 
259 static void sbr_hf_apply_noise_3(float (*Y)[2], const float *s_m,
260  const float *q_filt, int noise,
261  int kx, int m_max)
262 {
263  float phi_sign = 1 - 2 * (kx & 1);
264  sbr_hf_apply_noise(Y, s_m, q_filt, noise, 0.0, -phi_sign, m_max);
265 }
266 
268 {
269  s->sum64x5 = sbr_sum64x5_c;
277  s->hf_gen = sbr_hf_gen_c;
279 
284 
285  if (ARCH_ARM)
287  if (ARCH_X86)
289  if (ARCH_MIPS)
291 }