1 /* 2 dct64_i386.c: DCT64, a C variant for i386 3 4 copyright ?-2006 by the mpg123 project - free software under the terms of the LGPL 2.1 5 see COPYING and AUTHORS files in distribution or http://mpg123.org 6 initially written by Michael Hipp 7 */ 8 9 /* 10 * Discrete Cosine Tansform (DCT) for subband synthesis 11 * optimized for machines with no auto-increment. 12 * The performance is highly compiler dependend. Maybe 13 * the dct64.c version for 'normal' processor may be faster 14 * even for Intel processors. 15 */ 16 17 #include "mpg123lib_intern.h" 18 19 static void dct64_1(real *out0,real *out1,real *b1,real *b2,real *samples) 20 { 21 { 22 register real *costab = pnts[0]; 23 24 b1[0x00] = samples[0x00] + samples[0x1F]; 25 b1[0x01] = samples[0x01] + samples[0x1E]; 26 b1[0x1F] = REAL_MUL(samples[0x00] - samples[0x1F], costab[0x0]); 27 b1[0x1E] = REAL_MUL(samples[0x01] - samples[0x1E], costab[0x1]); 28 29 b1[0x02] = samples[0x02] + samples[0x1D]; 30 b1[0x03] = samples[0x03] + samples[0x1C]; 31 b1[0x1D] = REAL_MUL(samples[0x02] - samples[0x1D], costab[0x2]); 32 b1[0x1C] = REAL_MUL(samples[0x03] - samples[0x1C], costab[0x3]); 33 34 b1[0x04] = samples[0x04] + samples[0x1B]; 35 b1[0x05] = samples[0x05] + samples[0x1A]; 36 b1[0x1B] = REAL_MUL(samples[0x04] - samples[0x1B], costab[0x4]); 37 b1[0x1A] = REAL_MUL(samples[0x05] - samples[0x1A], costab[0x5]); 38 39 b1[0x06] = samples[0x06] + samples[0x19]; 40 b1[0x07] = samples[0x07] + samples[0x18]; 41 b1[0x19] = REAL_MUL(samples[0x06] - samples[0x19], costab[0x6]); 42 b1[0x18] = REAL_MUL(samples[0x07] - samples[0x18], costab[0x7]); 43 44 b1[0x08] = samples[0x08] + samples[0x17]; 45 b1[0x09] = samples[0x09] + samples[0x16]; 46 b1[0x17] = REAL_MUL(samples[0x08] - samples[0x17], costab[0x8]); 47 b1[0x16] = REAL_MUL(samples[0x09] - samples[0x16], costab[0x9]); 48 49 b1[0x0A] = samples[0x0A] + samples[0x15]; 50 b1[0x0B] = samples[0x0B] + samples[0x14]; 51 b1[0x15] = REAL_MUL(samples[0x0A] - samples[0x15], costab[0xA]); 52 b1[0x14] = REAL_MUL(samples[0x0B] - samples[0x14], costab[0xB]); 53 54 b1[0x0C] = samples[0x0C] + samples[0x13]; 55 b1[0x0D] = samples[0x0D] + samples[0x12]; 56 b1[0x13] = REAL_MUL(samples[0x0C] - samples[0x13], costab[0xC]); 57 b1[0x12] = REAL_MUL(samples[0x0D] - samples[0x12], costab[0xD]); 58 59 b1[0x0E] = samples[0x0E] + samples[0x11]; 60 b1[0x0F] = samples[0x0F] + samples[0x10]; 61 b1[0x11] = REAL_MUL(samples[0x0E] - samples[0x11], costab[0xE]); 62 b1[0x10] = REAL_MUL(samples[0x0F] - samples[0x10], costab[0xF]); 63 64 } 65 66 67 { 68 register real *costab = pnts[1]; 69 70 b2[0x00] = b1[0x00] + b1[0x0F]; 71 b2[0x01] = b1[0x01] + b1[0x0E]; 72 b2[0x0F] = REAL_MUL(b1[0x00] - b1[0x0F], costab[0]); 73 b2[0x0E] = REAL_MUL(b1[0x01] - b1[0x0E], costab[1]); 74 75 b2[0x02] = b1[0x02] + b1[0x0D]; 76 b2[0x03] = b1[0x03] + b1[0x0C]; 77 b2[0x0D] = REAL_MUL(b1[0x02] - b1[0x0D], costab[2]); 78 b2[0x0C] = REAL_MUL(b1[0x03] - b1[0x0C], costab[3]); 79 80 b2[0x04] = b1[0x04] + b1[0x0B]; 81 b2[0x05] = b1[0x05] + b1[0x0A]; 82 b2[0x0B] = REAL_MUL(b1[0x04] - b1[0x0B], costab[4]); 83 b2[0x0A] = REAL_MUL(b1[0x05] - b1[0x0A], costab[5]); 84 85 b2[0x06] = b1[0x06] + b1[0x09]; 86 b2[0x07] = b1[0x07] + b1[0x08]; 87 b2[0x09] = REAL_MUL(b1[0x06] - b1[0x09], costab[6]); 88 b2[0x08] = REAL_MUL(b1[0x07] - b1[0x08], costab[7]); 89 90 /* */ 91 92 b2[0x10] = b1[0x10] + b1[0x1F]; 93 b2[0x11] = b1[0x11] + b1[0x1E]; 94 b2[0x1F] = REAL_MUL(b1[0x1F] - b1[0x10], costab[0]); 95 b2[0x1E] = REAL_MUL(b1[0x1E] - b1[0x11], costab[1]); 96 97 b2[0x12] = b1[0x12] + b1[0x1D]; 98 b2[0x13] = b1[0x13] + b1[0x1C]; 99 b2[0x1D] = REAL_MUL(b1[0x1D] - b1[0x12], costab[2]); 100 b2[0x1C] = REAL_MUL(b1[0x1C] - b1[0x13], costab[3]); 101 102 b2[0x14] = b1[0x14] + b1[0x1B]; 103 b2[0x15] = b1[0x15] + b1[0x1A]; 104 b2[0x1B] = REAL_MUL(b1[0x1B] - b1[0x14], costab[4]); 105 b2[0x1A] = REAL_MUL(b1[0x1A] - b1[0x15], costab[5]); 106 107 b2[0x16] = b1[0x16] + b1[0x19]; 108 b2[0x17] = b1[0x17] + b1[0x18]; 109 b2[0x19] = REAL_MUL(b1[0x19] - b1[0x16], costab[6]); 110 b2[0x18] = REAL_MUL(b1[0x18] - b1[0x17], costab[7]); 111 } 112 113 { 114 register real *costab = pnts[2]; 115 116 b1[0x00] = b2[0x00] + b2[0x07]; 117 b1[0x07] = REAL_MUL(b2[0x00] - b2[0x07], costab[0]); 118 b1[0x01] = b2[0x01] + b2[0x06]; 119 b1[0x06] = REAL_MUL(b2[0x01] - b2[0x06], costab[1]); 120 b1[0x02] = b2[0x02] + b2[0x05]; 121 b1[0x05] = REAL_MUL(b2[0x02] - b2[0x05], costab[2]); 122 b1[0x03] = b2[0x03] + b2[0x04]; 123 b1[0x04] = REAL_MUL(b2[0x03] - b2[0x04], costab[3]); 124 125 b1[0x08] = b2[0x08] + b2[0x0F]; 126 b1[0x0F] = REAL_MUL(b2[0x0F] - b2[0x08], costab[0]); 127 b1[0x09] = b2[0x09] + b2[0x0E]; 128 b1[0x0E] = REAL_MUL(b2[0x0E] - b2[0x09], costab[1]); 129 b1[0x0A] = b2[0x0A] + b2[0x0D]; 130 b1[0x0D] = REAL_MUL(b2[0x0D] - b2[0x0A], costab[2]); 131 b1[0x0B] = b2[0x0B] + b2[0x0C]; 132 b1[0x0C] = REAL_MUL(b2[0x0C] - b2[0x0B], costab[3]); 133 134 b1[0x10] = b2[0x10] + b2[0x17]; 135 b1[0x17] = REAL_MUL(b2[0x10] - b2[0x17], costab[0]); 136 b1[0x11] = b2[0x11] + b2[0x16]; 137 b1[0x16] = REAL_MUL(b2[0x11] - b2[0x16], costab[1]); 138 b1[0x12] = b2[0x12] + b2[0x15]; 139 b1[0x15] = REAL_MUL(b2[0x12] - b2[0x15], costab[2]); 140 b1[0x13] = b2[0x13] + b2[0x14]; 141 b1[0x14] = REAL_MUL(b2[0x13] - b2[0x14], costab[3]); 142 143 b1[0x18] = b2[0x18] + b2[0x1F]; 144 b1[0x1F] = REAL_MUL(b2[0x1F] - b2[0x18], costab[0]); 145 b1[0x19] = b2[0x19] + b2[0x1E]; 146 b1[0x1E] = REAL_MUL(b2[0x1E] - b2[0x19], costab[1]); 147 b1[0x1A] = b2[0x1A] + b2[0x1D]; 148 b1[0x1D] = REAL_MUL(b2[0x1D] - b2[0x1A], costab[2]); 149 b1[0x1B] = b2[0x1B] + b2[0x1C]; 150 b1[0x1C] = REAL_MUL(b2[0x1C] - b2[0x1B], costab[3]); 151 } 152 153 { 154 register real const cos0 = pnts[3][0]; 155 register real const cos1 = pnts[3][1]; 156 157 b2[0x00] = b1[0x00] + b1[0x03]; 158 b2[0x03] = REAL_MUL(b1[0x00] - b1[0x03], cos0); 159 b2[0x01] = b1[0x01] + b1[0x02]; 160 b2[0x02] = REAL_MUL(b1[0x01] - b1[0x02], cos1); 161 162 b2[0x04] = b1[0x04] + b1[0x07]; 163 b2[0x07] = REAL_MUL(b1[0x07] - b1[0x04], cos0); 164 b2[0x05] = b1[0x05] + b1[0x06]; 165 b2[0x06] = REAL_MUL(b1[0x06] - b1[0x05], cos1); 166 167 b2[0x08] = b1[0x08] + b1[0x0B]; 168 b2[0x0B] = REAL_MUL(b1[0x08] - b1[0x0B], cos0); 169 b2[0x09] = b1[0x09] + b1[0x0A]; 170 b2[0x0A] = REAL_MUL(b1[0x09] - b1[0x0A], cos1); 171 172 b2[0x0C] = b1[0x0C] + b1[0x0F]; 173 b2[0x0F] = REAL_MUL(b1[0x0F] - b1[0x0C], cos0); 174 b2[0x0D] = b1[0x0D] + b1[0x0E]; 175 b2[0x0E] = REAL_MUL(b1[0x0E] - b1[0x0D], cos1); 176 177 b2[0x10] = b1[0x10] + b1[0x13]; 178 b2[0x13] = REAL_MUL(b1[0x10] - b1[0x13], cos0); 179 b2[0x11] = b1[0x11] + b1[0x12]; 180 b2[0x12] = REAL_MUL(b1[0x11] - b1[0x12], cos1); 181 182 b2[0x14] = b1[0x14] + b1[0x17]; 183 b2[0x17] = REAL_MUL(b1[0x17] - b1[0x14], cos0); 184 b2[0x15] = b1[0x15] + b1[0x16]; 185 b2[0x16] = REAL_MUL(b1[0x16] - b1[0x15], cos1); 186 187 b2[0x18] = b1[0x18] + b1[0x1B]; 188 b2[0x1B] = REAL_MUL(b1[0x18] - b1[0x1B], cos0); 189 b2[0x19] = b1[0x19] + b1[0x1A]; 190 b2[0x1A] = REAL_MUL(b1[0x19] - b1[0x1A], cos1); 191 192 b2[0x1C] = b1[0x1C] + b1[0x1F]; 193 b2[0x1F] = REAL_MUL(b1[0x1F] - b1[0x1C], cos0); 194 b2[0x1D] = b1[0x1D] + b1[0x1E]; 195 b2[0x1E] = REAL_MUL(b1[0x1E] - b1[0x1D], cos1); 196 } 197 198 { 199 register real const cos0 = pnts[4][0]; 200 201 b1[0x00] = b2[0x00] + b2[0x01]; 202 b1[0x01] = REAL_MUL(b2[0x00] - b2[0x01], cos0); 203 b1[0x02] = b2[0x02] + b2[0x03]; 204 b1[0x03] = REAL_MUL(b2[0x03] - b2[0x02], cos0); 205 b1[0x02] += b1[0x03]; 206 207 b1[0x04] = b2[0x04] + b2[0x05]; 208 b1[0x05] = REAL_MUL(b2[0x04] - b2[0x05], cos0); 209 b1[0x06] = b2[0x06] + b2[0x07]; 210 b1[0x07] = REAL_MUL(b2[0x07] - b2[0x06], cos0); 211 b1[0x06] += b1[0x07]; 212 b1[0x04] += b1[0x06]; 213 b1[0x06] += b1[0x05]; 214 b1[0x05] += b1[0x07]; 215 216 b1[0x08] = b2[0x08] + b2[0x09]; 217 b1[0x09] = REAL_MUL(b2[0x08] - b2[0x09], cos0); 218 b1[0x0A] = b2[0x0A] + b2[0x0B]; 219 b1[0x0B] = REAL_MUL(b2[0x0B] - b2[0x0A], cos0); 220 b1[0x0A] += b1[0x0B]; 221 222 b1[0x0C] = b2[0x0C] + b2[0x0D]; 223 b1[0x0D] = REAL_MUL(b2[0x0C] - b2[0x0D], cos0); 224 b1[0x0E] = b2[0x0E] + b2[0x0F]; 225 b1[0x0F] = REAL_MUL(b2[0x0F] - b2[0x0E], cos0); 226 b1[0x0E] += b1[0x0F]; 227 b1[0x0C] += b1[0x0E]; 228 b1[0x0E] += b1[0x0D]; 229 b1[0x0D] += b1[0x0F]; 230 231 b1[0x10] = b2[0x10] + b2[0x11]; 232 b1[0x11] = REAL_MUL(b2[0x10] - b2[0x11], cos0); 233 b1[0x12] = b2[0x12] + b2[0x13]; 234 b1[0x13] = REAL_MUL(b2[0x13] - b2[0x12], cos0); 235 b1[0x12] += b1[0x13]; 236 237 b1[0x14] = b2[0x14] + b2[0x15]; 238 b1[0x15] = REAL_MUL(b2[0x14] - b2[0x15], cos0); 239 b1[0x16] = b2[0x16] + b2[0x17]; 240 b1[0x17] = REAL_MUL(b2[0x17] - b2[0x16], cos0); 241 b1[0x16] += b1[0x17]; 242 b1[0x14] += b1[0x16]; 243 b1[0x16] += b1[0x15]; 244 b1[0x15] += b1[0x17]; 245 246 b1[0x18] = b2[0x18] + b2[0x19]; 247 b1[0x19] = REAL_MUL(b2[0x18] - b2[0x19], cos0); 248 b1[0x1A] = b2[0x1A] + b2[0x1B]; 249 b1[0x1B] = REAL_MUL(b2[0x1B] - b2[0x1A], cos0); 250 b1[0x1A] += b1[0x1B]; 251 252 b1[0x1C] = b2[0x1C] + b2[0x1D]; 253 b1[0x1D] = REAL_MUL(b2[0x1C] - b2[0x1D], cos0); 254 b1[0x1E] = b2[0x1E] + b2[0x1F]; 255 b1[0x1F] = REAL_MUL(b2[0x1F] - b2[0x1E], cos0); 256 b1[0x1E] += b1[0x1F]; 257 b1[0x1C] += b1[0x1E]; 258 b1[0x1E] += b1[0x1D]; 259 b1[0x1D] += b1[0x1F]; 260 } 261 262 out0[0x10*16] = REAL_SCALE_DCT64(b1[0x00]); 263 out0[0x10*12] = REAL_SCALE_DCT64(b1[0x04]); 264 out0[0x10* 8] = REAL_SCALE_DCT64(b1[0x02]); 265 out0[0x10* 4] = REAL_SCALE_DCT64(b1[0x06]); 266 out0[0x10* 0] = REAL_SCALE_DCT64(b1[0x01]); 267 out1[0x10* 0] = REAL_SCALE_DCT64(b1[0x01]); 268 out1[0x10* 4] = REAL_SCALE_DCT64(b1[0x05]); 269 out1[0x10* 8] = REAL_SCALE_DCT64(b1[0x03]); 270 out1[0x10*12] = REAL_SCALE_DCT64(b1[0x07]); 271 272 #if 1 273 out0[0x10*14] = REAL_SCALE_DCT64(b1[0x08] + b1[0x0C]); 274 out0[0x10*10] = REAL_SCALE_DCT64(b1[0x0C] + b1[0x0a]); 275 out0[0x10* 6] = REAL_SCALE_DCT64(b1[0x0A] + b1[0x0E]); 276 out0[0x10* 2] = REAL_SCALE_DCT64(b1[0x0E] + b1[0x09]); 277 out1[0x10* 2] = REAL_SCALE_DCT64(b1[0x09] + b1[0x0D]); 278 out1[0x10* 6] = REAL_SCALE_DCT64(b1[0x0D] + b1[0x0B]); 279 out1[0x10*10] = REAL_SCALE_DCT64(b1[0x0B] + b1[0x0F]); 280 out1[0x10*14] = REAL_SCALE_DCT64(b1[0x0F]); 281 #else 282 b1[0x08] += b1[0x0C]; 283 out0[0x10*14] = REAL_SCALE_DCT64(b1[0x08]); 284 b1[0x0C] += b1[0x0a]; 285 out0[0x10*10] = REAL_SCALE_DCT64(b1[0x0C]); 286 b1[0x0A] += b1[0x0E]; 287 out0[0x10* 6] = REAL_SCALE_DCT64(b1[0x0A]); 288 b1[0x0E] += b1[0x09]; 289 out0[0x10* 2] = REAL_SCALE_DCT64(b1[0x0E]); 290 b1[0x09] += b1[0x0D]; 291 out1[0x10* 2] = REAL_SCALE_DCT64(b1[0x09]); 292 b1[0x0D] += b1[0x0B]; 293 out1[0x10* 6] = REAL_SCALE_DCT64(b1[0x0D]); 294 b1[0x0B] += b1[0x0F]; 295 out1[0x10*10] = REAL_SCALE_DCT64(b1[0x0B]); 296 out1[0x10*14] = REAL_SCALE_DCT64(b1[0x0F]); 297 #endif 298 299 { 300 real tmp; 301 tmp = b1[0x18] + b1[0x1C]; 302 out0[0x10*15] = REAL_SCALE_DCT64(tmp + b1[0x10]); 303 out0[0x10*13] = REAL_SCALE_DCT64(tmp + b1[0x14]); 304 tmp = b1[0x1C] + b1[0x1A]; 305 out0[0x10*11] = REAL_SCALE_DCT64(tmp + b1[0x14]); 306 out0[0x10* 9] = REAL_SCALE_DCT64(tmp + b1[0x12]); 307 tmp = b1[0x1A] + b1[0x1E]; 308 out0[0x10* 7] = REAL_SCALE_DCT64(tmp + b1[0x12]); 309 out0[0x10* 5] = REAL_SCALE_DCT64(tmp + b1[0x16]); 310 tmp = b1[0x1E] + b1[0x19]; 311 out0[0x10* 3] = REAL_SCALE_DCT64(tmp + b1[0x16]); 312 out0[0x10* 1] = REAL_SCALE_DCT64(tmp + b1[0x11]); 313 tmp = b1[0x19] + b1[0x1D]; 314 out1[0x10* 1] = REAL_SCALE_DCT64(tmp + b1[0x11]); 315 out1[0x10* 3] = REAL_SCALE_DCT64(tmp + b1[0x15]); 316 tmp = b1[0x1D] + b1[0x1B]; 317 out1[0x10* 5] = REAL_SCALE_DCT64(tmp + b1[0x15]); 318 out1[0x10* 7] = REAL_SCALE_DCT64(tmp + b1[0x13]); 319 tmp = b1[0x1B] + b1[0x1F]; 320 out1[0x10* 9] = REAL_SCALE_DCT64(tmp + b1[0x13]); 321 out1[0x10*11] = REAL_SCALE_DCT64(tmp + b1[0x17]); 322 out1[0x10*13] = REAL_SCALE_DCT64(b1[0x17] + b1[0x1F]); 323 out1[0x10*15] = REAL_SCALE_DCT64(b1[0x1F]); 324 } 325 } 326 327 /* 328 * the call via dct64 is a trick to force GCC to use 329 * (new) registers for the b1,b2 pointer to the bufs[xx] field 330 */ 331 void dct64_i386(real *a,real *b,real *c) 332 { 333 real bufs[0x40]; 334 dct64_1(a,b,bufs,bufs+0x20,c); 335 } 336 337