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 Libav.
8 * Libav 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 * Libav 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 Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 #include "libavutil/attributes.h"
25 #include "libavutil/intfloat.h"
28 static void sbr_sum64x5_c(float *z)
31 for (k = 0; k < 64; k++) {
32 float f = z[k] + z[k + 64] + z[k + 128] + z[k + 192] + z[k + 256];
37 static float sbr_sum_square_c(float (*x)[2], int n)
39 float sum0 = 0.0f, sum1 = 0.0f;
42 for (i = 0; i < n; i += 2)
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];
53 static void sbr_neg_odd_64_c(float *x)
55 union av_intfloat32 *xi = (union av_intfloat32*) x;
57 for (i = 1; i < 64; i += 4) {
58 xi[i + 0].i ^= 1U << 31;
59 xi[i + 2].i ^= 1U << 31;
63 static void sbr_qmf_pre_shuffle_c(float *z)
65 union av_intfloat32 *zi = (union av_intfloat32*) z;
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;
75 zi[64 + 2 * 31 + 0].i = zi[64 - 31].i ^ (1U << 31);
76 zi[64 + 2 * 31 + 1].i = zi[31 + 1].i;
79 static void sbr_qmf_post_shuffle_c(float W[32][2], const float *z)
81 const union av_intfloat32 *zi = (const union av_intfloat32*) z;
82 union av_intfloat32 *Wi = (union av_intfloat32*) W;
84 for (k = 0; k < 32; k += 2) {
85 Wi[2 * k + 0].i = zi[63 - k].i ^ (1U << 31);
86 Wi[2 * k + 1].i = zi[ k + 0].i;
87 Wi[2 * k + 2].i = zi[62 - k].i ^ (1U << 31);
88 Wi[2 * k + 3].i = zi[ k + 1].i;
92 static void sbr_qmf_deint_neg_c(float *v, const float *src)
94 const union av_intfloat32 *si = (const union av_intfloat32*)src;
95 union av_intfloat32 *vi = (union av_intfloat32*)v;
97 for (i = 0; i < 32; i++) {
98 vi[ i].i = si[63 - 2 * i ].i;
99 vi[63 - i].i = si[63 - 2 * i - 1].i ^ (1U << 31);
103 static void sbr_qmf_deint_bfly_c(float *v, const float *src0, const float *src1)
106 for (i = 0; i < 64; i++) {
107 v[ i] = src0[i] - src1[63 - i];
108 v[127 - i] = src0[i] + src1[63 - i];
112 static av_always_inline void autocorrelate(const float x[40][2],
113 float phi[3][2][2], int lag)
116 float real_sum = 0.0f;
117 float imag_sum = 0.0f;
119 for (i = 1; i < 38; i++) {
120 real_sum += x[i][0] * x[i+lag][0] + x[i][1] * x[i+lag][1];
121 imag_sum += x[i][0] * x[i+lag][1] - x[i][1] * x[i+lag][0];
123 phi[2-lag][1][0] = real_sum + x[ 0][0] * x[lag][0] + x[ 0][1] * x[lag][1];
124 phi[2-lag][1][1] = imag_sum + x[ 0][0] * x[lag][1] - x[ 0][1] * x[lag][0];
126 phi[0][0][0] = real_sum + x[38][0] * x[39][0] + x[38][1] * x[39][1];
127 phi[0][0][1] = imag_sum + x[38][0] * x[39][1] - x[38][1] * x[39][0];
130 for (i = 1; i < 38; i++) {
131 real_sum += x[i][0] * x[i][0] + x[i][1] * x[i][1];
133 phi[2][1][0] = real_sum + x[ 0][0] * x[ 0][0] + x[ 0][1] * x[ 0][1];
134 phi[1][0][0] = real_sum + x[38][0] * x[38][0] + x[38][1] * x[38][1];
138 static void sbr_autocorrelate_c(const float x[40][2], float phi[3][2][2])
141 /* This code is slower because it multiplies memory accesses.
142 * It is left for educational purposes and because it may offer
143 * a better reference for writing arch-specific DSP functions. */
144 autocorrelate(x, phi, 0);
145 autocorrelate(x, phi, 1);
146 autocorrelate(x, phi, 2);
148 float real_sum2 = x[0][0] * x[2][0] + x[0][1] * x[2][1];
149 float imag_sum2 = x[0][0] * x[2][1] - x[0][1] * x[2][0];
150 float real_sum1 = 0.0f, imag_sum1 = 0.0f, real_sum0 = 0.0f;
152 for (i = 1; i < 38; i++) {
153 real_sum0 += x[i][0] * x[i ][0] + x[i][1] * x[i ][1];
154 real_sum1 += x[i][0] * x[i + 1][0] + x[i][1] * x[i + 1][1];
155 imag_sum1 += x[i][0] * x[i + 1][1] - x[i][1] * x[i + 1][0];
156 real_sum2 += x[i][0] * x[i + 2][0] + x[i][1] * x[i + 2][1];
157 imag_sum2 += x[i][0] * x[i + 2][1] - x[i][1] * x[i + 2][0];
159 phi[2 - 2][1][0] = real_sum2;
160 phi[2 - 2][1][1] = imag_sum2;
161 phi[2 ][1][0] = real_sum0 + x[ 0][0] * x[ 0][0] + x[ 0][1] * x[ 0][1];
162 phi[1 ][0][0] = real_sum0 + x[38][0] * x[38][0] + x[38][1] * x[38][1];
163 phi[2 - 1][1][0] = real_sum1 + x[ 0][0] * x[ 1][0] + x[ 0][1] * x[ 1][1];
164 phi[2 - 1][1][1] = imag_sum1 + x[ 0][0] * x[ 1][1] - x[ 0][1] * x[ 1][0];
165 phi[0 ][0][0] = real_sum1 + x[38][0] * x[39][0] + x[38][1] * x[39][1];
166 phi[0 ][0][1] = imag_sum1 + x[38][0] * x[39][1] - x[38][1] * x[39][0];
170 static void sbr_hf_gen_c(float (*X_high)[2], const float (*X_low)[2],
171 const float alpha0[2], const float alpha1[2],
172 float bw, int start, int end)
177 alpha[0] = alpha1[0] * bw * bw;
178 alpha[1] = alpha1[1] * bw * bw;
179 alpha[2] = alpha0[0] * bw;
180 alpha[3] = alpha0[1] * bw;
182 for (i = start; i < end; i++) {
184 X_low[i - 2][0] * alpha[0] -
185 X_low[i - 2][1] * alpha[1] +
186 X_low[i - 1][0] * alpha[2] -
187 X_low[i - 1][1] * alpha[3] +
190 X_low[i - 2][1] * alpha[0] +
191 X_low[i - 2][0] * alpha[1] +
192 X_low[i - 1][1] * alpha[2] +
193 X_low[i - 1][0] * alpha[3] +
198 static void sbr_hf_g_filt_c(float (*Y)[2], const float (*X_high)[40][2],
199 const float *g_filt, int m_max, intptr_t ixh)
203 for (m = 0; m < m_max; m++) {
204 Y[m][0] = X_high[m][ixh][0] * g_filt[m];
205 Y[m][1] = X_high[m][ixh][1] * g_filt[m];
209 static av_always_inline void sbr_hf_apply_noise(float (*Y)[2],
219 for (m = 0; m < m_max; m++) {
222 noise = (noise + 1) & 0x1ff;
224 y0 += s_m[m] * phi_sign0;
225 y1 += s_m[m] * phi_sign1;
227 y0 += q_filt[m] * ff_sbr_noise_table[noise][0];
228 y1 += q_filt[m] * ff_sbr_noise_table[noise][1];
232 phi_sign1 = -phi_sign1;
236 static void sbr_hf_apply_noise_0(float (*Y)[2], const float *s_m,
237 const float *q_filt, int noise,
240 sbr_hf_apply_noise(Y, s_m, q_filt, noise, 1.0, 0.0, m_max);
243 static void sbr_hf_apply_noise_1(float (*Y)[2], const float *s_m,
244 const float *q_filt, int noise,
247 float phi_sign = 1 - 2 * (kx & 1);
248 sbr_hf_apply_noise(Y, s_m, q_filt, noise, 0.0, phi_sign, m_max);
251 static void sbr_hf_apply_noise_2(float (*Y)[2], const float *s_m,
252 const float *q_filt, int noise,
255 sbr_hf_apply_noise(Y, s_m, q_filt, noise, -1.0, 0.0, m_max);
258 static void sbr_hf_apply_noise_3(float (*Y)[2], const float *s_m,
259 const float *q_filt, int noise,
262 float phi_sign = 1 - 2 * (kx & 1);
263 sbr_hf_apply_noise(Y, s_m, q_filt, noise, 0.0, -phi_sign, m_max);
266 av_cold void ff_sbrdsp_init(SBRDSPContext *s)
268 s->sum64x5 = sbr_sum64x5_c;
269 s->sum_square = sbr_sum_square_c;
270 s->neg_odd_64 = sbr_neg_odd_64_c;
271 s->qmf_pre_shuffle = sbr_qmf_pre_shuffle_c;
272 s->qmf_post_shuffle = sbr_qmf_post_shuffle_c;
273 s->qmf_deint_neg = sbr_qmf_deint_neg_c;
274 s->qmf_deint_bfly = sbr_qmf_deint_bfly_c;
275 s->autocorrelate = sbr_autocorrelate_c;
276 s->hf_gen = sbr_hf_gen_c;
277 s->hf_g_filt = sbr_hf_g_filt_c;
279 s->hf_apply_noise[0] = sbr_hf_apply_noise_0;
280 s->hf_apply_noise[1] = sbr_hf_apply_noise_1;
281 s->hf_apply_noise[2] = sbr_hf_apply_noise_2;
282 s->hf_apply_noise[3] = sbr_hf_apply_noise_3;
285 ff_sbrdsp_init_arm(s);
287 ff_sbrdsp_init_x86(s);