| 1 |
/***************************************************************************** |
| 2 |
* |
| 3 |
* XVID MPEG-4 VIDEO CODEC |
| 4 |
* - QPel interpolation - |
| 5 |
* |
| 6 |
* Copyright(C) 2003 Pascal Massimino <skal@planet-d.net> |
| 7 |
* |
| 8 |
* This program is free software ; you can redistribute it and/or modify |
| 9 |
* it under the terms of the GNU General Public License as published by |
| 10 |
* the Free Software Foundation ; either version 2 of the License, or |
| 11 |
* (at your option) any later version. |
| 12 |
* |
| 13 |
* This program 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 |
| 16 |
* GNU General Public License for more details. |
| 17 |
* |
| 18 |
* You should have received a copy of the GNU General Public License |
| 19 |
* along with this program ; if not, write to the Free Software |
| 20 |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
| 21 |
* |
| 22 |
* $Id: qpel.h,v 1.2 2004-03-22 22:36:23 edgomez Exp $ |
| 23 |
* |
| 24 |
****************************************************************************/ |
| 25 |
|
| 26 |
#ifndef _XVID_QPEL_H_ |
| 27 |
#define _XVID_QPEL_H_ |
| 28 |
|
| 29 |
#include "../utils/mem_transfer.h" |
| 30 |
|
| 31 |
/***************************************************************************** |
| 32 |
* Signatures |
| 33 |
****************************************************************************/ |
| 34 |
|
| 35 |
#define XVID_QP_PASS_SIGNATURE(NAME) \ |
| 36 |
void (NAME)(uint8_t *dst, const uint8_t *src, int32_t length, int32_t BpS, int32_t rounding) |
| 37 |
|
| 38 |
typedef XVID_QP_PASS_SIGNATURE(XVID_QP_PASS); |
| 39 |
|
| 40 |
/* We put everything in a single struct so it can easily be passed |
| 41 |
* to prediction functions as a whole... */ |
| 42 |
|
| 43 |
typedef struct _XVID_QP_FUNCS { |
| 44 |
|
| 45 |
/* filter for QPel 16x? prediction */ |
| 46 |
|
| 47 |
XVID_QP_PASS *H_Pass; |
| 48 |
XVID_QP_PASS *H_Pass_Avrg; |
| 49 |
XVID_QP_PASS *H_Pass_Avrg_Up; |
| 50 |
XVID_QP_PASS *V_Pass; |
| 51 |
XVID_QP_PASS *V_Pass_Avrg; |
| 52 |
XVID_QP_PASS *V_Pass_Avrg_Up; |
| 53 |
|
| 54 |
/* filter for QPel 8x? prediction */ |
| 55 |
|
| 56 |
XVID_QP_PASS *H_Pass_8; |
| 57 |
XVID_QP_PASS *H_Pass_Avrg_8; |
| 58 |
XVID_QP_PASS *H_Pass_Avrg_Up_8; |
| 59 |
XVID_QP_PASS *V_Pass_8; |
| 60 |
XVID_QP_PASS *V_Pass_Avrg_8; |
| 61 |
XVID_QP_PASS *V_Pass_Avrg_Up_8; |
| 62 |
} XVID_QP_FUNCS; |
| 63 |
|
| 64 |
/***************************************************************************** |
| 65 |
* fwd dcl |
| 66 |
****************************************************************************/ |
| 67 |
extern void xvid_Init_QP(); |
| 68 |
|
| 69 |
extern XVID_QP_FUNCS xvid_QP_Funcs_C; /* for P-frames */ |
| 70 |
extern XVID_QP_FUNCS xvid_QP_Add_Funcs_C; /* for B-frames */ |
| 71 |
|
| 72 |
#ifdef ARCH_IS_IA32 |
| 73 |
extern XVID_QP_FUNCS xvid_QP_Funcs_mmx; |
| 74 |
extern XVID_QP_FUNCS xvid_QP_Add_Funcs_mmx; |
| 75 |
#endif |
| 76 |
|
| 77 |
extern XVID_QP_FUNCS *xvid_QP_Funcs; /* <- main pointer for enc/dec structure */ |
| 78 |
extern XVID_QP_FUNCS *xvid_QP_Add_Funcs; /* <- main pointer for enc/dec structure */ |
| 79 |
|
| 80 |
/***************************************************************************** |
| 81 |
* macros |
| 82 |
****************************************************************************/ |
| 83 |
|
| 84 |
/***************************************************************************** |
| 85 |
|
| 86 |
Passes to be performed |
| 87 |
|
| 88 |
case 0: copy |
| 89 |
case 2: h-pass |
| 90 |
case 1/3: h-pass + h-avrg |
| 91 |
case 8: v-pass |
| 92 |
case 10: h-pass + v-pass |
| 93 |
case 9/11: h-pass + h-avrg + v-pass |
| 94 |
case 4/12: v-pass + v-avrg |
| 95 |
case 6/14: h-pass + v-pass + v-avrg |
| 96 |
case 5/13/7/15: h-pass + h-avrg + v-pass + v-avrg |
| 97 |
|
| 98 |
****************************************************************************/ |
| 99 |
|
| 100 |
static void __inline |
| 101 |
new_interpolate16x16_quarterpel(uint8_t * const cur, |
| 102 |
uint8_t * const refn, |
| 103 |
uint8_t * const refh, |
| 104 |
uint8_t * const refv, |
| 105 |
uint8_t * const refhv, |
| 106 |
const uint32_t x, const uint32_t y, |
| 107 |
const int32_t dx, const int dy, |
| 108 |
const uint32_t stride, |
| 109 |
const uint32_t rounding) |
| 110 |
{ |
| 111 |
const uint8_t *src; |
| 112 |
uint8_t *dst; |
| 113 |
uint8_t *tmp; |
| 114 |
int32_t quads; |
| 115 |
const XVID_QP_FUNCS *Ops; |
| 116 |
|
| 117 |
int32_t x_int, y_int; |
| 118 |
|
| 119 |
const int32_t xRef = x*4 + dx; |
| 120 |
const int32_t yRef = y*4 + dy; |
| 121 |
|
| 122 |
Ops = xvid_QP_Funcs; /* TODO: pass as argument */ |
| 123 |
quads = (dx&3) | ((dy&3)<<2); |
| 124 |
|
| 125 |
x_int = xRef/4; |
| 126 |
if (xRef < 0 && xRef % 4) |
| 127 |
x_int--; |
| 128 |
|
| 129 |
y_int = yRef/4; |
| 130 |
if (yRef < 0 && yRef % 4) |
| 131 |
y_int--; |
| 132 |
|
| 133 |
dst = cur + y * stride + x; |
| 134 |
src = refn + y_int * stride + x_int; |
| 135 |
|
| 136 |
tmp = refh; /* we need at least a 16 x stride scratch block */ |
| 137 |
|
| 138 |
switch(quads) { |
| 139 |
case 0: |
| 140 |
transfer8x8_copy( dst, src, stride); |
| 141 |
transfer8x8_copy( dst+8, src+8, stride); |
| 142 |
transfer8x8_copy( dst+8*stride, src+8*stride, stride); |
| 143 |
transfer8x8_copy( dst+8*stride+8, src+8*stride+8, stride); |
| 144 |
break; |
| 145 |
case 1: |
| 146 |
Ops->H_Pass_Avrg(dst, src, 16, stride, rounding); |
| 147 |
break; |
| 148 |
case 2: |
| 149 |
Ops->H_Pass(dst, src, 16, stride, rounding); |
| 150 |
break; |
| 151 |
case 3: |
| 152 |
Ops->H_Pass_Avrg_Up(dst, src, 16, stride, rounding); |
| 153 |
break; |
| 154 |
case 4: |
| 155 |
Ops->V_Pass_Avrg(dst, src, 16, stride, rounding); |
| 156 |
break; |
| 157 |
case 5: |
| 158 |
Ops->H_Pass_Avrg(tmp, src, 17, stride, rounding); |
| 159 |
Ops->V_Pass_Avrg(dst, tmp, 16, stride, rounding); |
| 160 |
break; |
| 161 |
case 6: |
| 162 |
Ops->H_Pass(tmp, src, 17, stride, rounding); |
| 163 |
Ops->V_Pass_Avrg(dst, tmp, 16, stride, rounding); |
| 164 |
break; |
| 165 |
case 7: |
| 166 |
Ops->H_Pass_Avrg_Up(tmp, src, 17, stride, rounding); |
| 167 |
Ops->V_Pass_Avrg(dst, tmp, 16, stride, rounding); |
| 168 |
break; |
| 169 |
case 8: |
| 170 |
Ops->V_Pass(dst, src, 16, stride, rounding); |
| 171 |
break; |
| 172 |
case 9: |
| 173 |
Ops->H_Pass_Avrg(tmp, src, 17, stride, rounding); |
| 174 |
Ops->V_Pass(dst, tmp, 16, stride, rounding); |
| 175 |
break; |
| 176 |
case 10: |
| 177 |
Ops->H_Pass(tmp, src, 17, stride, rounding); |
| 178 |
Ops->V_Pass(dst, tmp, 16, stride, rounding); |
| 179 |
break; |
| 180 |
case 11: |
| 181 |
Ops->H_Pass_Avrg_Up(tmp, src, 17, stride, rounding); |
| 182 |
Ops->V_Pass(dst, tmp, 16, stride, rounding); |
| 183 |
break; |
| 184 |
case 12: |
| 185 |
Ops->V_Pass_Avrg_Up(dst, src, 16, stride, rounding); |
| 186 |
break; |
| 187 |
case 13: |
| 188 |
Ops->H_Pass_Avrg(tmp, src, 17, stride, rounding); |
| 189 |
Ops->V_Pass_Avrg_Up(dst, tmp, 16, stride, rounding); |
| 190 |
break; |
| 191 |
case 14: |
| 192 |
Ops->H_Pass(tmp, src, 17, stride, rounding); |
| 193 |
Ops->V_Pass_Avrg_Up( dst, tmp, 16, stride, rounding); |
| 194 |
break; |
| 195 |
case 15: |
| 196 |
Ops->H_Pass_Avrg_Up(tmp, src, 17, stride, rounding); |
| 197 |
Ops->V_Pass_Avrg_Up(dst, tmp, 16, stride, rounding); |
| 198 |
break; |
| 199 |
} |
| 200 |
} |
| 201 |
|
| 202 |
static void __inline |
| 203 |
new_interpolate16x8_quarterpel(uint8_t * const cur, |
| 204 |
uint8_t * const refn, |
| 205 |
uint8_t * const refh, |
| 206 |
uint8_t * const refv, |
| 207 |
uint8_t * const refhv, |
| 208 |
const uint32_t x, const uint32_t y, |
| 209 |
const int32_t dx, const int dy, |
| 210 |
const uint32_t stride, |
| 211 |
const uint32_t rounding) |
| 212 |
{ |
| 213 |
const uint8_t *src; |
| 214 |
uint8_t *dst; |
| 215 |
uint8_t *tmp; |
| 216 |
int32_t quads; |
| 217 |
const XVID_QP_FUNCS *Ops; |
| 218 |
|
| 219 |
int32_t x_int, y_int; |
| 220 |
|
| 221 |
const int32_t xRef = x*4 + dx; |
| 222 |
const int32_t yRef = y*4 + dy; |
| 223 |
|
| 224 |
Ops = xvid_QP_Funcs; /* TODO: pass as argument */ |
| 225 |
quads = (dx&3) | ((dy&3)<<2); |
| 226 |
|
| 227 |
x_int = xRef/4; |
| 228 |
if (xRef < 0 && xRef % 4) |
| 229 |
x_int--; |
| 230 |
|
| 231 |
y_int = yRef/4; |
| 232 |
if (yRef < 0 && yRef % 4) |
| 233 |
y_int--; |
| 234 |
|
| 235 |
dst = cur + y * stride + x; |
| 236 |
src = refn + y_int * stride + x_int; |
| 237 |
|
| 238 |
tmp = refh; /* we need at least a 16 x stride scratch block */ |
| 239 |
|
| 240 |
switch(quads) { |
| 241 |
case 0: |
| 242 |
transfer8x8_copy( dst, src, stride); |
| 243 |
transfer8x8_copy( dst+8, src+8, stride); |
| 244 |
break; |
| 245 |
case 1: |
| 246 |
Ops->H_Pass_Avrg(dst, src, 8, stride, rounding); |
| 247 |
break; |
| 248 |
case 2: |
| 249 |
Ops->H_Pass(dst, src, 8, stride, rounding); |
| 250 |
break; |
| 251 |
case 3: |
| 252 |
Ops->H_Pass_Avrg_Up(dst, src, 8, stride, rounding); |
| 253 |
break; |
| 254 |
case 4: |
| 255 |
Ops->V_Pass_Avrg_8(dst, src, 16, stride, rounding); |
| 256 |
break; |
| 257 |
case 5: |
| 258 |
Ops->H_Pass_Avrg(tmp, src, 9, stride, rounding); |
| 259 |
Ops->V_Pass_Avrg_8(dst, tmp, 16, stride, rounding); |
| 260 |
break; |
| 261 |
case 6: |
| 262 |
Ops->H_Pass(tmp, src, 9, stride, rounding); |
| 263 |
Ops->V_Pass_Avrg_8(dst, tmp, 16, stride, rounding); |
| 264 |
break; |
| 265 |
case 7: |
| 266 |
Ops->H_Pass_Avrg_Up(tmp, src, 9, stride, rounding); |
| 267 |
Ops->V_Pass_Avrg_8(dst, tmp, 16, stride, rounding); |
| 268 |
break; |
| 269 |
case 8: |
| 270 |
Ops->V_Pass_8(dst, src, 16, stride, rounding); |
| 271 |
break; |
| 272 |
case 9: |
| 273 |
Ops->H_Pass_Avrg(tmp, src, 9, stride, rounding); |
| 274 |
Ops->V_Pass_8(dst, tmp, 16, stride, rounding); |
| 275 |
break; |
| 276 |
case 10: |
| 277 |
Ops->H_Pass(tmp, src, 9, stride, rounding); |
| 278 |
Ops->V_Pass_8(dst, tmp, 16, stride, rounding); |
| 279 |
break; |
| 280 |
case 11: |
| 281 |
Ops->H_Pass_Avrg_Up(tmp, src, 9, stride, rounding); |
| 282 |
Ops->V_Pass_8(dst, tmp, 16, stride, rounding); |
| 283 |
break; |
| 284 |
case 12: |
| 285 |
Ops->V_Pass_Avrg_Up_8(dst, src, 16, stride, rounding); |
| 286 |
break; |
| 287 |
case 13: |
| 288 |
Ops->H_Pass_Avrg(tmp, src, 9, stride, rounding); |
| 289 |
Ops->V_Pass_Avrg_Up_8(dst, tmp, 16, stride, rounding); |
| 290 |
break; |
| 291 |
case 14: |
| 292 |
Ops->H_Pass(tmp, src, 9, stride, rounding); |
| 293 |
Ops->V_Pass_Avrg_Up_8( dst, tmp, 16, stride, rounding); |
| 294 |
break; |
| 295 |
case 15: |
| 296 |
Ops->H_Pass_Avrg_Up(tmp, src, 9, stride, rounding); |
| 297 |
Ops->V_Pass_Avrg_Up_8(dst, tmp, 16, stride, rounding); |
| 298 |
break; |
| 299 |
} |
| 300 |
} |
| 301 |
|
| 302 |
static void __inline |
| 303 |
new_interpolate8x8_quarterpel(uint8_t * const cur, |
| 304 |
uint8_t * const refn, |
| 305 |
uint8_t * const refh, |
| 306 |
uint8_t * const refv, |
| 307 |
uint8_t * const refhv, |
| 308 |
const uint32_t x, const uint32_t y, |
| 309 |
const int32_t dx, const int dy, |
| 310 |
const uint32_t stride, |
| 311 |
const uint32_t rounding) |
| 312 |
{ |
| 313 |
const uint8_t *src; |
| 314 |
uint8_t *dst; |
| 315 |
uint8_t *tmp; |
| 316 |
int32_t quads; |
| 317 |
const XVID_QP_FUNCS *Ops; |
| 318 |
|
| 319 |
int32_t x_int, y_int; |
| 320 |
|
| 321 |
const int32_t xRef = x*4 + dx; |
| 322 |
const int32_t yRef = y*4 + dy; |
| 323 |
|
| 324 |
Ops = xvid_QP_Funcs; /* TODO: pass as argument */ |
| 325 |
quads = (dx&3) | ((dy&3)<<2); |
| 326 |
|
| 327 |
x_int = xRef/4; |
| 328 |
if (xRef < 0 && xRef % 4) |
| 329 |
x_int--; |
| 330 |
|
| 331 |
y_int = yRef/4; |
| 332 |
if (yRef < 0 && yRef % 4) |
| 333 |
y_int--; |
| 334 |
|
| 335 |
dst = cur + y * stride + x; |
| 336 |
src = refn + y_int * stride + x_int; |
| 337 |
|
| 338 |
tmp = refh; /* we need at least a 16 x stride scratch block */ |
| 339 |
|
| 340 |
switch(quads) { |
| 341 |
case 0: |
| 342 |
transfer8x8_copy( dst, src, stride); |
| 343 |
break; |
| 344 |
case 1: |
| 345 |
Ops->H_Pass_Avrg_8(dst, src, 8, stride, rounding); |
| 346 |
break; |
| 347 |
case 2: |
| 348 |
Ops->H_Pass_8(dst, src, 8, stride, rounding); |
| 349 |
break; |
| 350 |
case 3: |
| 351 |
Ops->H_Pass_Avrg_Up_8(dst, src, 8, stride, rounding); |
| 352 |
break; |
| 353 |
case 4: |
| 354 |
Ops->V_Pass_Avrg_8(dst, src, 8, stride, rounding); |
| 355 |
break; |
| 356 |
case 5: |
| 357 |
Ops->H_Pass_Avrg_8(tmp, src, 9, stride, rounding); |
| 358 |
Ops->V_Pass_Avrg_8(dst, tmp, 8, stride, rounding); |
| 359 |
break; |
| 360 |
case 6: |
| 361 |
Ops->H_Pass_8(tmp, src, 9, stride, rounding); |
| 362 |
Ops->V_Pass_Avrg_8(dst, tmp, 8, stride, rounding); |
| 363 |
break; |
| 364 |
case 7: |
| 365 |
Ops->H_Pass_Avrg_Up_8(tmp, src, 9, stride, rounding); |
| 366 |
Ops->V_Pass_Avrg_8(dst, tmp, 8, stride, rounding); |
| 367 |
break; |
| 368 |
case 8: |
| 369 |
Ops->V_Pass_8(dst, src, 8, stride, rounding); |
| 370 |
break; |
| 371 |
case 9: |
| 372 |
Ops->H_Pass_Avrg_8(tmp, src, 9, stride, rounding); |
| 373 |
Ops->V_Pass_8(dst, tmp, 8, stride, rounding); |
| 374 |
break; |
| 375 |
case 10: |
| 376 |
Ops->H_Pass_8(tmp, src, 9, stride, rounding); |
| 377 |
Ops->V_Pass_8(dst, tmp, 8, stride, rounding); |
| 378 |
break; |
| 379 |
case 11: |
| 380 |
Ops->H_Pass_Avrg_Up_8(tmp, src, 9, stride, rounding); |
| 381 |
Ops->V_Pass_8(dst, tmp, 8, stride, rounding); |
| 382 |
break; |
| 383 |
case 12: |
| 384 |
Ops->V_Pass_Avrg_Up_8(dst, src, 8, stride, rounding); |
| 385 |
break; |
| 386 |
case 13: |
| 387 |
Ops->H_Pass_Avrg_8(tmp, src, 9, stride, rounding); |
| 388 |
Ops->V_Pass_Avrg_Up_8(dst, tmp, 8, stride, rounding); |
| 389 |
break; |
| 390 |
case 14: |
| 391 |
Ops->H_Pass_8(tmp, src, 9, stride, rounding); |
| 392 |
Ops->V_Pass_Avrg_Up_8( dst, tmp, 8, stride, rounding); |
| 393 |
break; |
| 394 |
case 15: |
| 395 |
Ops->H_Pass_Avrg_Up_8(tmp, src, 9, stride, rounding); |
| 396 |
Ops->V_Pass_Avrg_Up_8(dst, tmp, 8, stride, rounding); |
| 397 |
break; |
| 398 |
} |
| 399 |
} |
| 400 |
|
| 401 |
#endif /* _XVID_QPEL_H_ */ |
