1 /* 2 dct64_i486.c: DCT64, a plain C variant for i486 3 4 copyright 1998-2006 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 Fabrice Bellard 7 */ 8 9 /* Discrete Cosine Tansform (DCT) for subband synthesis. 10 * 11 * This code is optimized for 80486. It should be compiled with gcc 12 * 2.7.2 or higher. 13 * 14 * Note: This code does not give the necessary accuracy. Moreover, no 15 * overflow test are done. 16 * 17 * (c) 1998 Fabrice Bellard. 18 */ 19 20 #include "mpg123lib_intern.h" 21 22 #define COS_0_0 16403 23 #define COS_0_1 16563 24 #define COS_0_2 16890 25 #define COS_0_3 17401 26 #define COS_0_4 18124 27 #define COS_0_5 19101 28 #define COS_0_6 20398 29 #define COS_0_7 22112 30 #define COS_0_8 24396 31 #define COS_0_9 27503 32 #define COS_0_10 31869 33 #define COS_0_11 38320 34 #define COS_0_12 48633 35 #define COS_0_13 67429 36 #define COS_0_14 111660 37 #define COS_0_15 333906 38 #define COS_1_0 16463 39 #define COS_1_1 17121 40 #define COS_1_2 18577 41 #define COS_1_3 21195 42 #define COS_1_4 25826 43 #define COS_1_5 34756 44 #define COS_1_6 56441 45 #define COS_1_7 167154 46 #define COS_2_0 16704 47 #define COS_2_1 19704 48 #define COS_2_2 29490 49 #define COS_2_3 83981 50 #define COS_3_0 17733 51 #define COS_3_1 42813 52 #define COS_4_0 23170 53 54 #define SETOUT(out,n,expr) out[FIR_BUFFER_SIZE*(n)]=(expr) 55 #define MULL(a,b) (((long long)(a)*(long long)(b)) >> 15) 56 #define MUL(a,b) \ 57 (\ 58 ((!(b & 0x3F)) ? (((a)*(b >> 6)) >> 9) :\ 59 ((!(b & 0x1F)) ? (((a)*(b >> 5)) >> 10) :\ 60 ((!(b & 0x0F)) ? (((a)*(b >> 4)) >> 11) :\ 61 ((!(b & 0x07)) ? (((a)*(b >> 3)) >> 12) :\ 62 ((!(b & 0x03)) ? (((a)*(b >> 2)) >> 13) :\ 63 ((!(b & 0x01)) ? (((a)*(b >> 1)) >> 14) :\ 64 (((a)*(b )) >> 15)))))))) 65 66 67 void dct64_1_486(int *out0,int *out1,int *b1,int *b2) 68 { 69 b1[0x00] = b2[0x00] + b2[0x1F]; 70 b1[0x1F] = MUL((b2[0x00] - b2[0x1F]),COS_0_0); 71 72 b1[0x01] = b2[0x01] + b2[0x1E]; 73 b1[0x1E] = MUL((b2[0x01] - b2[0x1E]),COS_0_1); 74 75 b1[0x02] = b2[0x02] + b2[0x1D]; 76 b1[0x1D] = MUL((b2[0x02] - b2[0x1D]),COS_0_2); 77 78 b1[0x03] = b2[0x03] + b2[0x1C]; 79 b1[0x1C] = MUL((b2[0x03] - b2[0x1C]),COS_0_3); 80 81 b1[0x04] = b2[0x04] + b2[0x1B]; 82 b1[0x1B] = MUL((b2[0x04] - b2[0x1B]),COS_0_4); 83 84 b1[0x05] = b2[0x05] + b2[0x1A]; 85 b1[0x1A] = MUL((b2[0x05] - b2[0x1A]),COS_0_5); 86 87 b1[0x06] = b2[0x06] + b2[0x19]; 88 b1[0x19] = MUL((b2[0x06] - b2[0x19]),COS_0_6); 89 90 b1[0x07] = b2[0x07] + b2[0x18]; 91 b1[0x18] = MUL((b2[0x07] - b2[0x18]),COS_0_7); 92 93 b1[0x08] = b2[0x08] + b2[0x17]; 94 b1[0x17] = MUL((b2[0x08] - b2[0x17]),COS_0_8); 95 96 b1[0x09] = b2[0x09] + b2[0x16]; 97 b1[0x16] = MUL((b2[0x09] - b2[0x16]),COS_0_9); 98 99 b1[0x0A] = b2[0x0A] + b2[0x15]; 100 b1[0x15] = MUL((b2[0x0A] - b2[0x15]),COS_0_10); 101 102 b1[0x0B] = b2[0x0B] + b2[0x14]; 103 b1[0x14] = MUL((b2[0x0B] - b2[0x14]),COS_0_11); 104 105 b1[0x0C] = b2[0x0C] + b2[0x13]; 106 b1[0x13] = MUL((b2[0x0C] - b2[0x13]),COS_0_12); 107 108 b1[0x0D] = b2[0x0D] + b2[0x12]; 109 b1[0x12] = MULL((b2[0x0D] - b2[0x12]),COS_0_13); 110 111 b1[0x0E] = b2[0x0E] + b2[0x11]; 112 b1[0x11] = MULL((b2[0x0E] - b2[0x11]),COS_0_14); 113 114 b1[0x0F] = b2[0x0F] + b2[0x10]; 115 b1[0x10] = MULL((b2[0x0F] - b2[0x10]),COS_0_15); 116 117 118 b2[0x00] = b1[0x00] + b1[0x0F]; 119 b2[0x0F] = MUL((b1[0x00] - b1[0x0F]),COS_1_0); 120 b2[0x01] = b1[0x01] + b1[0x0E]; 121 b2[0x0E] = MUL((b1[0x01] - b1[0x0E]),COS_1_1); 122 b2[0x02] = b1[0x02] + b1[0x0D]; 123 b2[0x0D] = MUL((b1[0x02] - b1[0x0D]),COS_1_2); 124 b2[0x03] = b1[0x03] + b1[0x0C]; 125 b2[0x0C] = MUL((b1[0x03] - b1[0x0C]),COS_1_3); 126 b2[0x04] = b1[0x04] + b1[0x0B]; 127 b2[0x0B] = MUL((b1[0x04] - b1[0x0B]),COS_1_4); 128 b2[0x05] = b1[0x05] + b1[0x0A]; 129 b2[0x0A] = MUL((b1[0x05] - b1[0x0A]),COS_1_5); 130 b2[0x06] = b1[0x06] + b1[0x09]; 131 b2[0x09] = MUL((b1[0x06] - b1[0x09]),COS_1_6); 132 b2[0x07] = b1[0x07] + b1[0x08]; 133 b2[0x08] = MULL((b1[0x07] - b1[0x08]),COS_1_7); 134 135 b2[0x10] = b1[0x10] + b1[0x1F]; 136 b2[0x1F] = MUL((b1[0x1F] - b1[0x10]),COS_1_0); 137 b2[0x11] = b1[0x11] + b1[0x1E]; 138 b2[0x1E] = MUL((b1[0x1E] - b1[0x11]),COS_1_1); 139 b2[0x12] = b1[0x12] + b1[0x1D]; 140 b2[0x1D] = MUL((b1[0x1D] - b1[0x12]),COS_1_2); 141 b2[0x13] = b1[0x13] + b1[0x1C]; 142 b2[0x1C] = MUL((b1[0x1C] - b1[0x13]),COS_1_3); 143 b2[0x14] = b1[0x14] + b1[0x1B]; 144 b2[0x1B] = MUL((b1[0x1B] - b1[0x14]),COS_1_4); 145 b2[0x15] = b1[0x15] + b1[0x1A]; 146 b2[0x1A] = MUL((b1[0x1A] - b1[0x15]),COS_1_5); 147 b2[0x16] = b1[0x16] + b1[0x19]; 148 b2[0x19] = MUL((b1[0x19] - b1[0x16]),COS_1_6); 149 b2[0x17] = b1[0x17] + b1[0x18]; 150 b2[0x18] = MULL((b1[0x18] - b1[0x17]),COS_1_7); 151 152 153 b1[0x00] = b2[0x00] + b2[0x07]; 154 b1[0x07] = MUL((b2[0x00] - b2[0x07]),COS_2_0); 155 b1[0x01] = b2[0x01] + b2[0x06]; 156 b1[0x06] = MUL((b2[0x01] - b2[0x06]),COS_2_1); 157 b1[0x02] = b2[0x02] + b2[0x05]; 158 b1[0x05] = MUL((b2[0x02] - b2[0x05]),COS_2_2); 159 b1[0x03] = b2[0x03] + b2[0x04]; 160 b1[0x04] = MULL((b2[0x03] - b2[0x04]),COS_2_3); 161 162 b1[0x08] = b2[0x08] + b2[0x0F]; 163 b1[0x0F] = MUL((b2[0x0F] - b2[0x08]),COS_2_0); 164 b1[0x09] = b2[0x09] + b2[0x0E]; 165 b1[0x0E] = MUL((b2[0x0E] - b2[0x09]),COS_2_1); 166 b1[0x0A] = b2[0x0A] + b2[0x0D]; 167 b1[0x0D] = MUL((b2[0x0D] - b2[0x0A]),COS_2_2); 168 b1[0x0B] = b2[0x0B] + b2[0x0C]; 169 b1[0x0C] = MULL((b2[0x0C] - b2[0x0B]),COS_2_3); 170 171 b1[0x10] = b2[0x10] + b2[0x17]; 172 b1[0x17] = MUL((b2[0x10] - b2[0x17]),COS_2_0); 173 b1[0x11] = b2[0x11] + b2[0x16]; 174 b1[0x16] = MUL((b2[0x11] - b2[0x16]),COS_2_1); 175 b1[0x12] = b2[0x12] + b2[0x15]; 176 b1[0x15] = MUL((b2[0x12] - b2[0x15]),COS_2_2); 177 b1[0x13] = b2[0x13] + b2[0x14]; 178 b1[0x14] = MULL((b2[0x13] - b2[0x14]),COS_2_3); 179 180 b1[0x18] = b2[0x18] + b2[0x1F]; 181 b1[0x1F] = MUL((b2[0x1F] - b2[0x18]),COS_2_0); 182 b1[0x19] = b2[0x19] + b2[0x1E]; 183 b1[0x1E] = MUL((b2[0x1E] - b2[0x19]),COS_2_1); 184 b1[0x1A] = b2[0x1A] + b2[0x1D]; 185 b1[0x1D] = MUL((b2[0x1D] - b2[0x1A]),COS_2_2); 186 b1[0x1B] = b2[0x1B] + b2[0x1C]; 187 b1[0x1C] = MULL((b2[0x1C] - b2[0x1B]),COS_2_3); 188 189 190 b2[0x00] = b1[0x00] + b1[0x03]; 191 b2[0x03] = MUL((b1[0x00] - b1[0x03]),COS_3_0); 192 b2[0x01] = b1[0x01] + b1[0x02]; 193 b2[0x02] = MUL((b1[0x01] - b1[0x02]),COS_3_1); 194 195 b2[0x04] = b1[0x04] + b1[0x07]; 196 b2[0x07] = MUL((b1[0x07] - b1[0x04]),COS_3_0); 197 b2[0x05] = b1[0x05] + b1[0x06]; 198 b2[0x06] = MUL((b1[0x06] - b1[0x05]),COS_3_1); 199 200 b2[0x08] = b1[0x08] + b1[0x0B]; 201 b2[0x0B] = MUL((b1[0x08] - b1[0x0B]),COS_3_0); 202 b2[0x09] = b1[0x09] + b1[0x0A]; 203 b2[0x0A] = MUL((b1[0x09] - b1[0x0A]),COS_3_1); 204 205 b2[0x0C] = b1[0x0C] + b1[0x0F]; 206 b2[0x0F] = MUL((b1[0x0F] - b1[0x0C]),COS_3_0); 207 b2[0x0D] = b1[0x0D] + b1[0x0E]; 208 b2[0x0E] = MUL((b1[0x0E] - b1[0x0D]),COS_3_1); 209 210 b2[0x10] = b1[0x10] + b1[0x13]; 211 b2[0x13] = MUL((b1[0x10] - b1[0x13]),COS_3_0); 212 b2[0x11] = b1[0x11] + b1[0x12]; 213 b2[0x12] = MUL((b1[0x11] - b1[0x12]),COS_3_1); 214 215 b2[0x14] = b1[0x14] + b1[0x17]; 216 b2[0x17] = MUL((b1[0x17] - b1[0x14]),COS_3_0); 217 b2[0x15] = b1[0x15] + b1[0x16]; 218 b2[0x16] = MUL((b1[0x16] - b1[0x15]),COS_3_1); 219 220 b2[0x18] = b1[0x18] + b1[0x1B]; 221 b2[0x1B] = MUL((b1[0x18] - b1[0x1B]),COS_3_0); 222 b2[0x19] = b1[0x19] + b1[0x1A]; 223 b2[0x1A] = MUL((b1[0x19] - b1[0x1A]),COS_3_1); 224 225 b2[0x1C] = b1[0x1C] + b1[0x1F]; 226 b2[0x1F] = MUL((b1[0x1F] - b1[0x1C]),COS_3_0); 227 b2[0x1D] = b1[0x1D] + b1[0x1E]; 228 b2[0x1E] = MUL((b1[0x1E] - b1[0x1D]),COS_3_1); 229 230 { 231 int i; 232 for(i=0;i<32;i+=4) { 233 b1[i+0x00] = b2[i+0x00] + b2[i+0x01]; 234 b1[i+0x01] = MUL((b2[i+0x00] - b2[i+0x01]),COS_4_0); 235 b1[i+0x02] = b2[i+0x02] + b2[i+0x03]; 236 b1[i+0x03] = MUL((b2[i+0x03] - b2[i+0x02]),COS_4_0); 237 } 238 } 239 240 b1[0x02] += b1[0x03]; 241 b1[0x06] += b1[0x07]; 242 b1[0x04] += b1[0x06]; 243 b1[0x06] += b1[0x05]; 244 b1[0x05] += b1[0x07]; 245 246 b1[0x0A] += b1[0x0B]; 247 b1[0x0E] += b1[0x0F]; 248 b1[0x0C] += b1[0x0E]; 249 b1[0x0E] += b1[0x0D]; 250 b1[0x0D] += b1[0x0F]; 251 252 b1[0x12] += b1[0x13]; 253 b1[0x16] += b1[0x17]; 254 b1[0x14] += b1[0x16]; 255 b1[0x16] += b1[0x15]; 256 b1[0x15] += b1[0x17]; 257 258 b1[0x1A] += b1[0x1B]; 259 b1[0x1E] += b1[0x1F]; 260 b1[0x1C] += b1[0x1E]; 261 b1[0x1E] += b1[0x1D]; 262 b1[0x1D] += b1[0x1F]; 263 264 SETOUT(out0,16,b1[0x00]); 265 SETOUT(out0,12,b1[0x04]); 266 SETOUT(out0, 8,b1[0x02]); 267 SETOUT(out0, 4,b1[0x06]); 268 SETOUT(out0, 0,b1[0x01]); 269 SETOUT(out1, 0,b1[0x01]); 270 SETOUT(out1, 4,b1[0x05]); 271 SETOUT(out1, 8,b1[0x03]); 272 SETOUT(out1,12,b1[0x07]); 273 274 b1[0x08] += b1[0x0C]; 275 SETOUT(out0,14,b1[0x08]); 276 b1[0x0C] += b1[0x0a]; 277 SETOUT(out0,10,b1[0x0C]); 278 b1[0x0A] += b1[0x0E]; 279 SETOUT(out0, 6,b1[0x0A]); 280 b1[0x0E] += b1[0x09]; 281 SETOUT(out0, 2,b1[0x0E]); 282 b1[0x09] += b1[0x0D]; 283 SETOUT(out1, 2,b1[0x09]); 284 b1[0x0D] += b1[0x0B]; 285 SETOUT(out1, 6,b1[0x0D]); 286 b1[0x0B] += b1[0x0F]; 287 SETOUT(out1,10,b1[0x0B]); 288 SETOUT(out1,14,b1[0x0F]); 289 290 b1[0x18] += b1[0x1C]; 291 SETOUT(out0,15,b1[0x10] + b1[0x18]); 292 SETOUT(out0,13,b1[0x18] + b1[0x14]); 293 b1[0x1C] += b1[0x1a]; 294 SETOUT(out0,11,b1[0x14] + b1[0x1C]); 295 SETOUT(out0, 9,b1[0x1C] + b1[0x12]); 296 b1[0x1A] += b1[0x1E]; 297 SETOUT(out0, 7,b1[0x12] + b1[0x1A]); 298 SETOUT(out0, 5,b1[0x1A] + b1[0x16]); 299 b1[0x1E] += b1[0x19]; 300 SETOUT(out0, 3,b1[0x16] + b1[0x1E]); 301 SETOUT(out0, 1,b1[0x1E] + b1[0x11]); 302 b1[0x19] += b1[0x1D]; 303 SETOUT(out1, 1,b1[0x11] + b1[0x19]); 304 SETOUT(out1, 3,b1[0x19] + b1[0x15]); 305 b1[0x1D] += b1[0x1B]; 306 SETOUT(out1, 5,b1[0x15] + b1[0x1D]); 307 SETOUT(out1, 7,b1[0x1D] + b1[0x13]); 308 b1[0x1B] += b1[0x1F]; 309 SETOUT(out1, 9,b1[0x13] + b1[0x1B]); 310 SETOUT(out1,11,b1[0x1B] + b1[0x17]); 311 SETOUT(out1,13,b1[0x17] + b1[0x1F]); 312 SETOUT(out1,15,b1[0x1F]); 313 } 314 315 316 /* 317 * the call via dct64 is a trick to force GCC to use 318 * (new) registers for the b1,b2 pointer to the bufs[xx] field 319 */ 320 void dct64_i486(int *a,int *b,real *samples) 321 { 322 int bufs[64]; 323 int i; 324 325 #ifdef REAL_IS_FIXED 326 #define TOINT(a) ((a) * 32768 / (int)REAL_FACTOR) 327 328 for(i=0;i<32;i++) { 329 bufs[i]=TOINT(samples[i]); 330 } 331 #else 332 int *p = bufs; 333 register double const scale = ((65536.0 * 32) + 1) * 65536.0; 334 335 for(i=0;i<32;i++) { 336 *((double *) (p++)) = scale + *samples++; /* beware on bufs overrun: 8B store from x87 */ 337 } 338 #endif 339 340 dct64_1_486(a,b,bufs+32,bufs); 341 } 342 343