| 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 | |
| 114 | #if 0 |
| 115 | /* This code is slower because it multiplies memory accesses. |
| 116 | * It is left for educational purposes and because it may offer |
| 117 | * a better reference for writing arch-specific DSP functions. */ |
| 118 | static av_always_inline void autocorrelate(const float x[40][2], |
| 119 | float phi[3][2][2], int lag) |
| 120 | { |
| 121 | int i; |
| 122 | float real_sum = 0.0f; |
| 123 | float imag_sum = 0.0f; |
| 124 | if (lag) { |
| 125 | for (i = 1; i < 38; i++) { |
| 126 | real_sum += x[i][0] * x[i+lag][0] + x[i][1] * x[i+lag][1]; |
| 127 | imag_sum += x[i][0] * x[i+lag][1] - x[i][1] * x[i+lag][0]; |
| 128 | } |
| 129 | phi[2-lag][1][0] = real_sum + x[ 0][0] * x[lag][0] + x[ 0][1] * x[lag][1]; |
| 130 | phi[2-lag][1][1] = imag_sum + x[ 0][0] * x[lag][1] - x[ 0][1] * x[lag][0]; |
| 131 | if (lag == 1) { |
| 132 | phi[0][0][0] = real_sum + x[38][0] * x[39][0] + x[38][1] * x[39][1]; |
| 133 | phi[0][0][1] = imag_sum + x[38][0] * x[39][1] - x[38][1] * x[39][0]; |
| 134 | } |
| 135 | } else { |
| 136 | for (i = 1; i < 38; i++) { |
| 137 | real_sum += x[i][0] * x[i][0] + x[i][1] * x[i][1]; |
| 138 | } |
| 139 | phi[2][1][0] = real_sum + x[ 0][0] * x[ 0][0] + x[ 0][1] * x[ 0][1]; |
| 140 | phi[1][0][0] = real_sum + x[38][0] * x[38][0] + x[38][1] * x[38][1]; |
| 141 | } |
| 142 | } |
| 143 | |
| 144 | static void sbr_autocorrelate_c(const float x[40][2], float phi[3][2][2]) |
| 145 | { |
| 146 | autocorrelate(x, phi, 0); |
| 147 | autocorrelate(x, phi, 1); |
| 148 | autocorrelate(x, phi, 2); |
| 149 | } |
| 150 | #else |
| 151 | static void sbr_autocorrelate_c(const float x[40][2], float phi[3][2][2]) |
| 152 | { |
| 153 | float real_sum2 = x[0][0] * x[2][0] + x[0][1] * x[2][1]; |
| 154 | float imag_sum2 = x[0][0] * x[2][1] - x[0][1] * x[2][0]; |
| 155 | float real_sum1 = 0.0f, imag_sum1 = 0.0f, real_sum0 = 0.0f; |
| 156 | int i; |
| 157 | for (i = 1; i < 38; i++) { |
| 158 | real_sum0 += x[i][0] * x[i ][0] + x[i][1] * x[i ][1]; |
| 159 | real_sum1 += x[i][0] * x[i + 1][0] + x[i][1] * x[i + 1][1]; |
| 160 | imag_sum1 += x[i][0] * x[i + 1][1] - x[i][1] * x[i + 1][0]; |
| 161 | real_sum2 += x[i][0] * x[i + 2][0] + x[i][1] * x[i + 2][1]; |
| 162 | imag_sum2 += x[i][0] * x[i + 2][1] - x[i][1] * x[i + 2][0]; |
| 163 | } |
| 164 | phi[2 - 2][1][0] = real_sum2; |
| 165 | phi[2 - 2][1][1] = imag_sum2; |
| 166 | phi[2 ][1][0] = real_sum0 + x[ 0][0] * x[ 0][0] + x[ 0][1] * x[ 0][1]; |
| 167 | phi[1 ][0][0] = real_sum0 + x[38][0] * x[38][0] + x[38][1] * x[38][1]; |
| 168 | phi[2 - 1][1][0] = real_sum1 + x[ 0][0] * x[ 1][0] + x[ 0][1] * x[ 1][1]; |
| 169 | phi[2 - 1][1][1] = imag_sum1 + x[ 0][0] * x[ 1][1] - x[ 0][1] * x[ 1][0]; |
| 170 | phi[0 ][0][0] = real_sum1 + x[38][0] * x[39][0] + x[38][1] * x[39][1]; |
| 171 | phi[0 ][0][1] = imag_sum1 + x[38][0] * x[39][1] - x[38][1] * x[39][0]; |
| 172 | #endif |
| 173 | } |
| 174 | |
| 175 | static void sbr_hf_gen_c(float (*X_high)[2], const float (*X_low)[2], |
| 176 | const float alpha0[2], const float alpha1[2], |
| 177 | float bw, int start, int end) |
| 178 | { |
| 179 | float alpha[4]; |
| 180 | int i; |
| 181 | |
| 182 | alpha[0] = alpha1[0] * bw * bw; |
| 183 | alpha[1] = alpha1[1] * bw * bw; |
| 184 | alpha[2] = alpha0[0] * bw; |
| 185 | alpha[3] = alpha0[1] * bw; |
| 186 | |
| 187 | for (i = start; i < end; i++) { |
| 188 | X_high[i][0] = |
| 189 | X_low[i - 2][0] * alpha[0] - |
| 190 | X_low[i - 2][1] * alpha[1] + |
| 191 | X_low[i - 1][0] * alpha[2] - |
| 192 | X_low[i - 1][1] * alpha[3] + |
| 193 | X_low[i][0]; |
| 194 | X_high[i][1] = |
| 195 | X_low[i - 2][1] * alpha[0] + |
| 196 | X_low[i - 2][0] * alpha[1] + |
| 197 | X_low[i - 1][1] * alpha[2] + |
| 198 | X_low[i - 1][0] * alpha[3] + |
| 199 | X_low[i][1]; |
| 200 | } |
| 201 | } |
| 202 | |
| 203 | static void sbr_hf_g_filt_c(float (*Y)[2], const float (*X_high)[40][2], |
| 204 | const float *g_filt, int m_max, intptr_t ixh) |
| 205 | { |
| 206 | int m; |
| 207 | |
| 208 | for (m = 0; m < m_max; m++) { |
| 209 | Y[m][0] = X_high[m][ixh][0] * g_filt[m]; |
| 210 | Y[m][1] = X_high[m][ixh][1] * g_filt[m]; |
| 211 | } |
| 212 | } |
| 213 | |
| 214 | static av_always_inline void sbr_hf_apply_noise(float (*Y)[2], |
| 215 | const float *s_m, |
| 216 | const float *q_filt, |
| 217 | int noise, |
| 218 | float phi_sign0, |
| 219 | float phi_sign1, |
| 220 | int m_max) |
| 221 | { |
| 222 | int m; |
| 223 | |
| 224 | for (m = 0; m < m_max; m++) { |
| 225 | float y0 = Y[m][0]; |
| 226 | float y1 = Y[m][1]; |
| 227 | noise = (noise + 1) & 0x1ff; |
| 228 | if (s_m[m]) { |
| 229 | y0 += s_m[m] * phi_sign0; |
| 230 | y1 += s_m[m] * phi_sign1; |
| 231 | } else { |
| 232 | y0 += q_filt[m] * ff_sbr_noise_table[noise][0]; |
| 233 | y1 += q_filt[m] * ff_sbr_noise_table[noise][1]; |
| 234 | } |
| 235 | Y[m][0] = y0; |
| 236 | Y[m][1] = y1; |
| 237 | phi_sign1 = -phi_sign1; |
| 238 | } |
| 239 | } |
| 240 | |
| 241 | static void sbr_hf_apply_noise_0(float (*Y)[2], const float *s_m, |
| 242 | const float *q_filt, int noise, |
| 243 | int kx, int m_max) |
| 244 | { |
| 245 | sbr_hf_apply_noise(Y, s_m, q_filt, noise, 1.0, 0.0, m_max); |
| 246 | } |
| 247 | |
| 248 | static void sbr_hf_apply_noise_1(float (*Y)[2], const float *s_m, |
| 249 | const float *q_filt, int noise, |
| 250 | int kx, int m_max) |
| 251 | { |
| 252 | float phi_sign = 1 - 2 * (kx & 1); |
| 253 | sbr_hf_apply_noise(Y, s_m, q_filt, noise, 0.0, phi_sign, m_max); |
| 254 | } |
| 255 | |
| 256 | static void sbr_hf_apply_noise_2(float (*Y)[2], const float *s_m, |
| 257 | const float *q_filt, int noise, |
| 258 | int kx, int m_max) |
| 259 | { |
| 260 | sbr_hf_apply_noise(Y, s_m, q_filt, noise, -1.0, 0.0, m_max); |
| 261 | } |
| 262 | |
| 263 | static void sbr_hf_apply_noise_3(float (*Y)[2], const float *s_m, |
| 264 | const float *q_filt, int noise, |
| 265 | int kx, int m_max) |
| 266 | { |
| 267 | float phi_sign = 1 - 2 * (kx & 1); |
| 268 | sbr_hf_apply_noise(Y, s_m, q_filt, noise, 0.0, -phi_sign, m_max); |
| 269 | } |
| 270 | |
| 271 | av_cold void ff_sbrdsp_init(SBRDSPContext *s) |
| 272 | { |
| 273 | s->sum64x5 = sbr_sum64x5_c; |
| 274 | s->sum_square = sbr_sum_square_c; |
| 275 | s->neg_odd_64 = sbr_neg_odd_64_c; |
| 276 | s->qmf_pre_shuffle = sbr_qmf_pre_shuffle_c; |
| 277 | s->qmf_post_shuffle = sbr_qmf_post_shuffle_c; |
| 278 | s->qmf_deint_neg = sbr_qmf_deint_neg_c; |
| 279 | s->qmf_deint_bfly = sbr_qmf_deint_bfly_c; |
| 280 | s->autocorrelate = sbr_autocorrelate_c; |
| 281 | s->hf_gen = sbr_hf_gen_c; |
| 282 | s->hf_g_filt = sbr_hf_g_filt_c; |
| 283 | |
| 284 | s->hf_apply_noise[0] = sbr_hf_apply_noise_0; |
| 285 | s->hf_apply_noise[1] = sbr_hf_apply_noise_1; |
| 286 | s->hf_apply_noise[2] = sbr_hf_apply_noise_2; |
| 287 | s->hf_apply_noise[3] = sbr_hf_apply_noise_3; |
| 288 | |
| 289 | if (ARCH_ARM) |
| 290 | ff_sbrdsp_init_arm(s); |
| 291 | if (ARCH_X86) |
| 292 | ff_sbrdsp_init_x86(s); |
| 293 | if (ARCH_MIPS) |
| 294 | ff_sbrdsp_init_mips(s); |
| 295 | } |