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
27 #include "libavutil/attributes.h"
28 #include "libavutil/intfloat.h"
31 static float sbr_sum_square_c(float (*x)[2], int n)
33 float sum0 = 0.0f, sum1 = 0.0f;
36 for (i = 0; i < n; i += 2)
38 sum0 += x[i + 0][0] * x[i + 0][0];
39 sum1 += x[i + 0][1] * x[i + 0][1];
40 sum0 += x[i + 1][0] * x[i + 1][0];
41 sum1 += x[i + 1][1] * x[i + 1][1];
47 static void sbr_neg_odd_64_c(float *x)
49 union av_intfloat32 *xi = (union av_intfloat32*) x;
51 for (i = 1; i < 64; i += 4) {
52 xi[i + 0].i ^= 1U << 31;
53 xi[i + 2].i ^= 1U << 31;
57 static void sbr_qmf_pre_shuffle_c(float *z)
59 union av_intfloat32 *zi = (union av_intfloat32*) z;
63 for (k = 1; k < 31; k += 2) {
64 zi[64 + 2 * k + 0].i = zi[64 - k].i ^ (1U << 31);
65 zi[64 + 2 * k + 1].i = zi[ k + 1].i;
66 zi[64 + 2 * k + 2].i = zi[63 - k].i ^ (1U << 31);
67 zi[64 + 2 * k + 3].i = zi[ k + 2].i;
70 zi[64 + 2 * 31 + 0].i = zi[64 - 31].i ^ (1U << 31);
71 zi[64 + 2 * 31 + 1].i = zi[31 + 1].i;
74 static void sbr_qmf_post_shuffle_c(float W[32][2], const float *z)
76 const union av_intfloat32 *zi = (const union av_intfloat32*) z;
77 union av_intfloat32 *Wi = (union av_intfloat32*) W;
79 for (k = 0; k < 32; k += 2) {
80 Wi[2 * k + 0].i = zi[63 - k].i ^ (1U << 31);
81 Wi[2 * k + 1].i = zi[ k + 0].i;
82 Wi[2 * k + 2].i = zi[62 - k].i ^ (1U << 31);
83 Wi[2 * k + 3].i = zi[ k + 1].i;
87 static void sbr_qmf_deint_neg_c(float *v, const float *src)
89 const union av_intfloat32 *si = (const union av_intfloat32*)src;
90 union av_intfloat32 *vi = (union av_intfloat32*)v;
92 for (i = 0; i < 32; i++) {
93 vi[ i].i = si[63 - 2 * i ].i;
94 vi[63 - i].i = si[63 - 2 * i - 1].i ^ (1U << 31);
99 /* This code is slower because it multiplies memory accesses.
100 * It is left for educational purposes and because it may offer
101 * a better reference for writing arch-specific DSP functions. */
102 static av_always_inline void autocorrelate(const float x[40][2],
103 float phi[3][2][2], int lag)
106 float real_sum = 0.0f;
107 float imag_sum = 0.0f;
109 for (i = 1; i < 38; i++) {
110 real_sum += x[i][0] * x[i+lag][0] + x[i][1] * x[i+lag][1];
111 imag_sum += x[i][0] * x[i+lag][1] - x[i][1] * x[i+lag][0];
113 phi[2-lag][1][0] = real_sum + x[ 0][0] * x[lag][0] + x[ 0][1] * x[lag][1];
114 phi[2-lag][1][1] = imag_sum + x[ 0][0] * x[lag][1] - x[ 0][1] * x[lag][0];
116 phi[0][0][0] = real_sum + x[38][0] * x[39][0] + x[38][1] * x[39][1];
117 phi[0][0][1] = imag_sum + x[38][0] * x[39][1] - x[38][1] * x[39][0];
120 for (i = 1; i < 38; i++) {
121 real_sum += x[i][0] * x[i][0] + x[i][1] * x[i][1];
123 phi[2][1][0] = real_sum + x[ 0][0] * x[ 0][0] + x[ 0][1] * x[ 0][1];
124 phi[1][0][0] = real_sum + x[38][0] * x[38][0] + x[38][1] * x[38][1];
128 static void sbr_autocorrelate_c(const float x[40][2], float phi[3][2][2])
130 autocorrelate(x, phi, 0);
131 autocorrelate(x, phi, 1);
132 autocorrelate(x, phi, 2);
135 static void sbr_autocorrelate_c(const float x[40][2], float phi[3][2][2])
137 float real_sum2 = x[0][0] * x[2][0] + x[0][1] * x[2][1];
138 float imag_sum2 = x[0][0] * x[2][1] - x[0][1] * x[2][0];
139 float real_sum1 = 0.0f, imag_sum1 = 0.0f, real_sum0 = 0.0f;
141 for (i = 1; i < 38; i++) {
142 real_sum0 += x[i][0] * x[i ][0] + x[i][1] * x[i ][1];
143 real_sum1 += x[i][0] * x[i + 1][0] + x[i][1] * x[i + 1][1];
144 imag_sum1 += x[i][0] * x[i + 1][1] - x[i][1] * x[i + 1][0];
145 real_sum2 += x[i][0] * x[i + 2][0] + x[i][1] * x[i + 2][1];
146 imag_sum2 += x[i][0] * x[i + 2][1] - x[i][1] * x[i + 2][0];
148 phi[2 - 2][1][0] = real_sum2;
149 phi[2 - 2][1][1] = imag_sum2;
150 phi[2 ][1][0] = real_sum0 + x[ 0][0] * x[ 0][0] + x[ 0][1] * x[ 0][1];
151 phi[1 ][0][0] = real_sum0 + x[38][0] * x[38][0] + x[38][1] * x[38][1];
152 phi[2 - 1][1][0] = real_sum1 + x[ 0][0] * x[ 1][0] + x[ 0][1] * x[ 1][1];
153 phi[2 - 1][1][1] = imag_sum1 + x[ 0][0] * x[ 1][1] - x[ 0][1] * x[ 1][0];
154 phi[0 ][0][0] = real_sum1 + x[38][0] * x[39][0] + x[38][1] * x[39][1];
155 phi[0 ][0][1] = imag_sum1 + x[38][0] * x[39][1] - x[38][1] * x[39][0];
159 static void sbr_hf_gen_c(float (*X_high)[2], const float (*X_low)[2],
160 const float alpha0[2], const float alpha1[2],
161 float bw, int start, int end)
166 alpha[0] = alpha1[0] * bw * bw;
167 alpha[1] = alpha1[1] * bw * bw;
168 alpha[2] = alpha0[0] * bw;
169 alpha[3] = alpha0[1] * bw;
171 for (i = start; i < end; i++) {
173 X_low[i - 2][0] * alpha[0] -
174 X_low[i - 2][1] * alpha[1] +
175 X_low[i - 1][0] * alpha[2] -
176 X_low[i - 1][1] * alpha[3] +
179 X_low[i - 2][1] * alpha[0] +
180 X_low[i - 2][0] * alpha[1] +
181 X_low[i - 1][1] * alpha[2] +
182 X_low[i - 1][0] * alpha[3] +
187 static void sbr_hf_g_filt_c(float (*Y)[2], const float (*X_high)[40][2],
188 const float *g_filt, int m_max, intptr_t ixh)
192 for (m = 0; m < m_max; m++) {
193 Y[m][0] = X_high[m][ixh][0] * g_filt[m];
194 Y[m][1] = X_high[m][ixh][1] * g_filt[m];
198 static av_always_inline void sbr_hf_apply_noise(float (*Y)[2],
208 for (m = 0; m < m_max; m++) {
211 noise = (noise + 1) & 0x1ff;
213 y0 += s_m[m] * phi_sign0;
214 y1 += s_m[m] * phi_sign1;
216 y0 += q_filt[m] * ff_sbr_noise_table[noise][0];
217 y1 += q_filt[m] * ff_sbr_noise_table[noise][1];
221 phi_sign1 = -phi_sign1;
225 #include "sbrdsp_template.c"