Mercurial > libavcodec.hg
annotate dsputil.c @ 19:82d4c9be9873 libavcodec
MMX/MMXEXT iDCT support, using external functions currently defined in libmpeg2
Gives average 13-20% mpeg decoding speedup on x86 systems.
| author | arpi_esp |
|---|---|
| date | Fri, 03 Aug 2001 18:33:03 +0000 |
| parents | ec4642daa6fe |
| children | 2733a4c1c693 |
| rev | line source |
|---|---|
| 0 | 1 /* |
| 2 * DSP utils | |
| 3 * Copyright (c) 2000, 2001 Gerard Lantau. | |
| 4 * | |
| 5 * This program is free software; you can redistribute it and/or modify | |
| 6 * it under the terms of the GNU General Public License as published by | |
| 7 * the Free Software Foundation; either version 2 of the License, or | |
| 8 * (at your option) any later version. | |
| 9 * | |
| 10 * This program is distributed in the hope that it will be useful, | |
| 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
| 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
| 13 * GNU General Public License for more details. | |
| 14 * | |
| 15 * You should have received a copy of the GNU General Public License | |
| 16 * along with this program; if not, write to the Free Software | |
| 17 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
| 18 */ | |
| 19 #include <stdlib.h> | |
| 20 #include <stdio.h> | |
| 21 #include "avcodec.h" | |
| 22 #include "dsputil.h" | |
| 23 | |
|
19
82d4c9be9873
MMX/MMXEXT iDCT support, using external functions currently defined in libmpeg2
arpi_esp
parents:
6
diff
changeset
|
24 void (*ff_idct)(DCTELEM *block); |
| 0 | 25 void (*get_pixels)(DCTELEM *block, const UINT8 *pixels, int line_size); |
| 26 void (*put_pixels_clamped)(const DCTELEM *block, UINT8 *pixels, int line_size); | |
| 27 void (*add_pixels_clamped)(const DCTELEM *block, UINT8 *pixels, int line_size); | |
| 28 | |
| 29 op_pixels_abs_func pix_abs16x16; | |
| 30 op_pixels_abs_func pix_abs16x16_x2; | |
| 31 op_pixels_abs_func pix_abs16x16_y2; | |
| 32 op_pixels_abs_func pix_abs16x16_xy2; | |
| 33 | |
| 34 static UINT8 cropTbl[256 + 2 * MAX_NEG_CROP]; | |
| 35 UINT32 squareTbl[512]; | |
| 36 | |
| 37 void get_pixels_c(DCTELEM *block, const UINT8 *pixels, int line_size) | |
| 38 { | |
| 39 DCTELEM *p; | |
| 40 const UINT8 *pix; | |
| 41 int i; | |
| 42 | |
| 43 /* read the pixels */ | |
| 44 p = block; | |
| 45 pix = pixels; | |
| 46 for(i=0;i<8;i++) { | |
| 47 p[0] = pix[0]; | |
| 48 p[1] = pix[1]; | |
| 49 p[2] = pix[2]; | |
| 50 p[3] = pix[3]; | |
| 51 p[4] = pix[4]; | |
| 52 p[5] = pix[5]; | |
| 53 p[6] = pix[6]; | |
| 54 p[7] = pix[7]; | |
| 55 pix += line_size; | |
| 56 p += 8; | |
| 57 } | |
| 58 } | |
| 59 | |
| 60 void put_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size) | |
| 61 { | |
| 62 const DCTELEM *p; | |
| 63 UINT8 *pix; | |
| 64 int i; | |
| 65 UINT8 *cm = cropTbl + MAX_NEG_CROP; | |
| 66 | |
| 67 /* read the pixels */ | |
| 68 p = block; | |
| 69 pix = pixels; | |
| 70 for(i=0;i<8;i++) { | |
| 71 pix[0] = cm[p[0]]; | |
| 72 pix[1] = cm[p[1]]; | |
| 73 pix[2] = cm[p[2]]; | |
| 74 pix[3] = cm[p[3]]; | |
| 75 pix[4] = cm[p[4]]; | |
| 76 pix[5] = cm[p[5]]; | |
| 77 pix[6] = cm[p[6]]; | |
| 78 pix[7] = cm[p[7]]; | |
| 79 pix += line_size; | |
| 80 p += 8; | |
| 81 } | |
| 82 } | |
| 83 | |
| 84 void add_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size) | |
| 85 { | |
| 86 const DCTELEM *p; | |
| 87 UINT8 *pix; | |
| 88 int i; | |
| 89 UINT8 *cm = cropTbl + MAX_NEG_CROP; | |
| 90 | |
| 91 /* read the pixels */ | |
| 92 p = block; | |
| 93 pix = pixels; | |
| 94 for(i=0;i<8;i++) { | |
| 95 pix[0] = cm[pix[0] + p[0]]; | |
| 96 pix[1] = cm[pix[1] + p[1]]; | |
| 97 pix[2] = cm[pix[2] + p[2]]; | |
| 98 pix[3] = cm[pix[3] + p[3]]; | |
| 99 pix[4] = cm[pix[4] + p[4]]; | |
| 100 pix[5] = cm[pix[5] + p[5]]; | |
| 101 pix[6] = cm[pix[6] + p[6]]; | |
| 102 pix[7] = cm[pix[7] + p[7]]; | |
| 103 pix += line_size; | |
| 104 p += 8; | |
| 105 } | |
| 106 } | |
| 107 | |
| 108 #define PIXOP(BTYPE, OPNAME, OP, INCR) \ | |
| 109 \ | |
| 110 static void OPNAME ## _pixels(BTYPE *block, const UINT8 *pixels, int line_size, int h) \ | |
| 111 { \ | |
| 112 BTYPE *p; \ | |
| 113 const UINT8 *pix; \ | |
| 114 \ | |
| 115 p = block; \ | |
| 116 pix = pixels; \ | |
| 117 do { \ | |
| 118 OP(p[0], pix[0]); \ | |
| 119 OP(p[1], pix[1]); \ | |
| 120 OP(p[2], pix[2]); \ | |
| 121 OP(p[3], pix[3]); \ | |
| 122 OP(p[4], pix[4]); \ | |
| 123 OP(p[5], pix[5]); \ | |
| 124 OP(p[6], pix[6]); \ | |
| 125 OP(p[7], pix[7]); \ | |
| 126 pix += line_size; \ | |
| 127 p += INCR; \ | |
| 128 } while (--h);; \ | |
| 129 } \ | |
| 130 \ | |
| 131 static void OPNAME ## _pixels_x2(BTYPE *block, const UINT8 *pixels, int line_size, int h) \ | |
| 132 { \ | |
| 133 BTYPE *p; \ | |
| 134 const UINT8 *pix; \ | |
| 135 \ | |
| 136 p = block; \ | |
| 137 pix = pixels; \ | |
| 138 do { \ | |
| 139 OP(p[0], avg2(pix[0], pix[1])); \ | |
| 140 OP(p[1], avg2(pix[1], pix[2])); \ | |
| 141 OP(p[2], avg2(pix[2], pix[3])); \ | |
| 142 OP(p[3], avg2(pix[3], pix[4])); \ | |
| 143 OP(p[4], avg2(pix[4], pix[5])); \ | |
| 144 OP(p[5], avg2(pix[5], pix[6])); \ | |
| 145 OP(p[6], avg2(pix[6], pix[7])); \ | |
| 146 OP(p[7], avg2(pix[7], pix[8])); \ | |
| 147 pix += line_size; \ | |
| 148 p += INCR; \ | |
| 149 } while (--h); \ | |
| 150 } \ | |
| 151 \ | |
| 152 static void OPNAME ## _pixels_y2(BTYPE *block, const UINT8 *pixels, int line_size, int h) \ | |
| 153 { \ | |
| 154 BTYPE *p; \ | |
| 155 const UINT8 *pix; \ | |
| 156 const UINT8 *pix1; \ | |
| 157 \ | |
| 158 p = block; \ | |
| 159 pix = pixels; \ | |
| 160 pix1 = pixels + line_size; \ | |
| 161 do { \ | |
| 162 OP(p[0], avg2(pix[0], pix1[0])); \ | |
| 163 OP(p[1], avg2(pix[1], pix1[1])); \ | |
| 164 OP(p[2], avg2(pix[2], pix1[2])); \ | |
| 165 OP(p[3], avg2(pix[3], pix1[3])); \ | |
| 166 OP(p[4], avg2(pix[4], pix1[4])); \ | |
| 167 OP(p[5], avg2(pix[5], pix1[5])); \ | |
| 168 OP(p[6], avg2(pix[6], pix1[6])); \ | |
| 169 OP(p[7], avg2(pix[7], pix1[7])); \ | |
| 170 pix += line_size; \ | |
| 171 pix1 += line_size; \ | |
| 172 p += INCR; \ | |
| 173 } while(--h); \ | |
| 174 } \ | |
| 175 \ | |
| 176 static void OPNAME ## _pixels_xy2(BTYPE *block, const UINT8 *pixels, int line_size, int h) \ | |
| 177 { \ | |
| 178 BTYPE *p; \ | |
| 179 const UINT8 *pix; \ | |
| 180 const UINT8 *pix1; \ | |
| 181 \ | |
| 182 p = block; \ | |
| 183 pix = pixels; \ | |
| 184 pix1 = pixels + line_size; \ | |
| 185 do { \ | |
| 186 OP(p[0], avg4(pix[0], pix[1], pix1[0], pix1[1])); \ | |
| 187 OP(p[1], avg4(pix[1], pix[2], pix1[1], pix1[2])); \ | |
| 188 OP(p[2], avg4(pix[2], pix[3], pix1[2], pix1[3])); \ | |
| 189 OP(p[3], avg4(pix[3], pix[4], pix1[3], pix1[4])); \ | |
| 190 OP(p[4], avg4(pix[4], pix[5], pix1[4], pix1[5])); \ | |
| 191 OP(p[5], avg4(pix[5], pix[6], pix1[5], pix1[6])); \ | |
| 192 OP(p[6], avg4(pix[6], pix[7], pix1[6], pix1[7])); \ | |
| 193 OP(p[7], avg4(pix[7], pix[8], pix1[7], pix1[8])); \ | |
| 194 pix += line_size; \ | |
| 195 pix1 += line_size; \ | |
| 196 p += INCR; \ | |
| 197 } while(--h); \ | |
| 198 } \ | |
| 199 \ | |
| 200 void (*OPNAME ## _pixels_tab[4])(BTYPE *block, const UINT8 *pixels, int line_size, int h) = { \ | |
| 201 OPNAME ## _pixels, \ | |
| 202 OPNAME ## _pixels_x2, \ | |
| 203 OPNAME ## _pixels_y2, \ | |
| 204 OPNAME ## _pixels_xy2, \ | |
| 205 }; | |
| 206 | |
| 207 | |
| 208 /* rounding primitives */ | |
| 209 #define avg2(a,b) ((a+b+1)>>1) | |
| 210 #define avg4(a,b,c,d) ((a+b+c+d+2)>>2) | |
| 211 | |
| 212 #define op_put(a, b) a = b | |
| 213 #define op_avg(a, b) a = avg2(a, b) | |
| 214 #define op_sub(a, b) a -= b | |
| 215 | |
| 216 PIXOP(UINT8, put, op_put, line_size) | |
| 217 PIXOP(UINT8, avg, op_avg, line_size) | |
| 218 | |
| 219 PIXOP(DCTELEM, sub, op_sub, 8) | |
| 220 | |
| 221 /* not rounding primitives */ | |
| 222 #undef avg2 | |
| 223 #undef avg4 | |
| 224 #define avg2(a,b) ((a+b)>>1) | |
| 225 #define avg4(a,b,c,d) ((a+b+c+d+1)>>2) | |
| 226 | |
| 227 PIXOP(UINT8, put_no_rnd, op_put, line_size) | |
| 228 PIXOP(UINT8, avg_no_rnd, op_avg, line_size) | |
| 229 | |
| 230 /* motion estimation */ | |
| 231 | |
| 232 #undef avg2 | |
| 233 #undef avg4 | |
| 234 #define avg2(a,b) ((a+b+1)>>1) | |
| 235 #define avg4(a,b,c,d) ((a+b+c+d+2)>>2) | |
| 236 | |
| 237 int pix_abs16x16_c(UINT8 *pix1, UINT8 *pix2, int line_size, int h) | |
| 238 { | |
| 239 int s, i; | |
| 240 | |
| 241 s = 0; | |
| 242 for(i=0;i<h;i++) { | |
| 243 s += abs(pix1[0] - pix2[0]); | |
| 244 s += abs(pix1[1] - pix2[1]); | |
| 245 s += abs(pix1[2] - pix2[2]); | |
| 246 s += abs(pix1[3] - pix2[3]); | |
| 247 s += abs(pix1[4] - pix2[4]); | |
| 248 s += abs(pix1[5] - pix2[5]); | |
| 249 s += abs(pix1[6] - pix2[6]); | |
| 250 s += abs(pix1[7] - pix2[7]); | |
| 251 s += abs(pix1[8] - pix2[8]); | |
| 252 s += abs(pix1[9] - pix2[9]); | |
| 253 s += abs(pix1[10] - pix2[10]); | |
| 254 s += abs(pix1[11] - pix2[11]); | |
| 255 s += abs(pix1[12] - pix2[12]); | |
| 256 s += abs(pix1[13] - pix2[13]); | |
| 257 s += abs(pix1[14] - pix2[14]); | |
| 258 s += abs(pix1[15] - pix2[15]); | |
| 259 pix1 += line_size; | |
| 260 pix2 += line_size; | |
| 261 } | |
| 262 return s; | |
| 263 } | |
| 264 | |
| 265 int pix_abs16x16_x2_c(UINT8 *pix1, UINT8 *pix2, int line_size, int h) | |
| 266 { | |
| 267 int s, i; | |
| 268 | |
| 269 s = 0; | |
| 270 for(i=0;i<h;i++) { | |
| 271 s += abs(pix1[0] - avg2(pix2[0], pix2[1])); | |
| 272 s += abs(pix1[1] - avg2(pix2[1], pix2[2])); | |
| 273 s += abs(pix1[2] - avg2(pix2[2], pix2[3])); | |
| 274 s += abs(pix1[3] - avg2(pix2[3], pix2[4])); | |
| 275 s += abs(pix1[4] - avg2(pix2[4], pix2[5])); | |
| 276 s += abs(pix1[5] - avg2(pix2[5], pix2[6])); | |
| 277 s += abs(pix1[6] - avg2(pix2[6], pix2[7])); | |
| 278 s += abs(pix1[7] - avg2(pix2[7], pix2[8])); | |
| 279 s += abs(pix1[8] - avg2(pix2[8], pix2[9])); | |
| 280 s += abs(pix1[9] - avg2(pix2[9], pix2[10])); | |
| 281 s += abs(pix1[10] - avg2(pix2[10], pix2[11])); | |
| 282 s += abs(pix1[11] - avg2(pix2[11], pix2[12])); | |
| 283 s += abs(pix1[12] - avg2(pix2[12], pix2[13])); | |
| 284 s += abs(pix1[13] - avg2(pix2[13], pix2[14])); | |
| 285 s += abs(pix1[14] - avg2(pix2[14], pix2[15])); | |
| 286 s += abs(pix1[15] - avg2(pix2[15], pix2[16])); | |
| 287 pix1 += line_size; | |
| 288 pix2 += line_size; | |
| 289 } | |
| 290 return s; | |
| 291 } | |
| 292 | |
| 293 int pix_abs16x16_y2_c(UINT8 *pix1, UINT8 *pix2, int line_size, int h) | |
| 294 { | |
| 295 int s, i; | |
| 296 UINT8 *pix3 = pix2 + line_size; | |
| 297 | |
| 298 s = 0; | |
| 299 for(i=0;i<h;i++) { | |
| 300 s += abs(pix1[0] - avg2(pix2[0], pix3[0])); | |
| 301 s += abs(pix1[1] - avg2(pix2[1], pix3[1])); | |
| 302 s += abs(pix1[2] - avg2(pix2[2], pix3[2])); | |
| 303 s += abs(pix1[3] - avg2(pix2[3], pix3[3])); | |
| 304 s += abs(pix1[4] - avg2(pix2[4], pix3[4])); | |
| 305 s += abs(pix1[5] - avg2(pix2[5], pix3[5])); | |
| 306 s += abs(pix1[6] - avg2(pix2[6], pix3[6])); | |
| 307 s += abs(pix1[7] - avg2(pix2[7], pix3[7])); | |
| 308 s += abs(pix1[8] - avg2(pix2[8], pix3[8])); | |
| 309 s += abs(pix1[9] - avg2(pix2[9], pix3[9])); | |
| 310 s += abs(pix1[10] - avg2(pix2[10], pix3[10])); | |
| 311 s += abs(pix1[11] - avg2(pix2[11], pix3[11])); | |
| 312 s += abs(pix1[12] - avg2(pix2[12], pix3[12])); | |
| 313 s += abs(pix1[13] - avg2(pix2[13], pix3[13])); | |
| 314 s += abs(pix1[14] - avg2(pix2[14], pix3[14])); | |
| 315 s += abs(pix1[15] - avg2(pix2[15], pix3[15])); | |
| 316 pix1 += line_size; | |
| 317 pix2 += line_size; | |
| 318 pix3 += line_size; | |
| 319 } | |
| 320 return s; | |
| 321 } | |
| 322 | |
| 323 int pix_abs16x16_xy2_c(UINT8 *pix1, UINT8 *pix2, int line_size, int h) | |
| 324 { | |
| 325 int s, i; | |
| 326 UINT8 *pix3 = pix2 + line_size; | |
| 327 | |
| 328 s = 0; | |
| 329 for(i=0;i<h;i++) { | |
| 330 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1])); | |
| 331 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2])); | |
| 332 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3])); | |
| 333 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4])); | |
| 334 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5])); | |
| 335 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6])); | |
| 336 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7])); | |
| 337 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8])); | |
| 338 s += abs(pix1[8] - avg4(pix2[8], pix2[9], pix3[8], pix3[9])); | |
| 339 s += abs(pix1[9] - avg4(pix2[9], pix2[10], pix3[9], pix3[10])); | |
| 340 s += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11])); | |
| 341 s += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12])); | |
| 342 s += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13])); | |
| 343 s += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14])); | |
| 344 s += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15])); | |
| 345 s += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16])); | |
| 346 pix1 += line_size; | |
| 347 pix2 += line_size; | |
| 348 pix3 += line_size; | |
| 349 } | |
| 350 return s; | |
| 351 } | |
| 352 | |
| 353 void dsputil_init(void) | |
| 354 { | |
| 355 int i; | |
| 356 | |
| 357 for(i=0;i<256;i++) cropTbl[i + MAX_NEG_CROP] = i; | |
| 358 for(i=0;i<MAX_NEG_CROP;i++) { | |
| 359 cropTbl[i] = 0; | |
| 360 cropTbl[i + MAX_NEG_CROP + 256] = 255; | |
| 361 } | |
| 362 | |
| 363 for(i=0;i<512;i++) { | |
| 364 squareTbl[i] = (i - 256) * (i - 256); | |
| 365 } | |
| 366 | |
|
19
82d4c9be9873
MMX/MMXEXT iDCT support, using external functions currently defined in libmpeg2
arpi_esp
parents:
6
diff
changeset
|
367 ff_idct = j_rev_dct; |
| 0 | 368 get_pixels = get_pixels_c; |
| 369 put_pixels_clamped = put_pixels_clamped_c; | |
| 370 add_pixels_clamped = add_pixels_clamped_c; | |
| 371 | |
| 372 pix_abs16x16 = pix_abs16x16_c; | |
| 373 pix_abs16x16_x2 = pix_abs16x16_x2_c; | |
| 374 pix_abs16x16_y2 = pix_abs16x16_y2_c; | |
| 375 pix_abs16x16_xy2 = pix_abs16x16_xy2_c; | |
| 376 av_fdct = jpeg_fdct_ifast; | |
| 377 | |
| 2 | 378 #ifdef HAVE_MMX |
| 0 | 379 dsputil_init_mmx(); |
| 380 #endif | |
| 381 } |
