Parent Directory
| 
Revision Log
Revision 744 - (view) (download)
| 1 : | Isibaar | 616 | // 30.10.2002 corrected qpel chroma rounding |
| 2 : | Isibaar | 581 | // 04.10.2002 added qpel support to MBMotionCompensation |
| 3 : | edgomez | 195 | // 01.05.2002 updated MBMotionCompensationBVOP |
| 4 : | // 14.04.2002 bframe compensation | ||
| 5 : | suxen_drol | 118 | |
| 6 : | Isibaar | 3 | #include "../encoder.h" |
| 7 : | #include "../utils/mbfunctions.h" | ||
| 8 : | #include "../image/interpolate8x8.h" | ||
| 9 : | suxen_drol | 701 | #include "../image/reduced.h" |
| 10 : | Isibaar | 3 | #include "../utils/timer.h" |
| 11 : | suxen_drol | 118 | #include "motion.h" |
| 12 : | Isibaar | 3 | |
| 13 : | #define ABS(X) (((X)>0)?(X):-(X)) | ||
| 14 : | #define SIGN(X) (((X)>0)?1:-1) | ||
| 15 : | |||
| 16 : | suxen_drol | 701 | |
| 17 : | edgomez | 195 | static __inline void |
| 18 : | Isibaar | 588 | compensate16x16_interpolate(int16_t * const dct_codes, |
| 19 : | uint8_t * const cur, | ||
| 20 : | const uint8_t * const ref, | ||
| 21 : | const uint8_t * const refh, | ||
| 22 : | suxen_drol | 701 | uint8_t * const refv, |
| 23 : | Isibaar | 588 | const uint8_t * const refhv, |
| 24 : | suxen_drol | 701 | uint32_t x, |
| 25 : | uint32_t y, | ||
| 26 : | Isibaar | 588 | const int32_t dx, |
| 27 : | const int32_t dy, | ||
| 28 : | const uint32_t stride, | ||
| 29 : | const uint32_t quarterpel, | ||
| 30 : | suxen_drol | 701 | const int reduced_resolution, |
| 31 : | const uint32_t rounding) | ||
| 32 : | Isibaar | 588 | { |
| 33 : | suxen_drol | 701 | |
| 34 : | syskin | 744 | if (reduced_resolution) { |
| 35 : | suxen_drol | 702 | const uint8_t * reference; |
| 36 : | x*=2; y*=2; | ||
| 37 : | Isibaar | 588 | |
| 38 : | suxen_drol | 702 | reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
| 39 : | |||
| 40 : | filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride); | ||
| 41 : | filter_diff_18x18_to_8x8(dct_codes, reference, stride); | ||
| 42 : | Isibaar | 588 | |
| 43 : | suxen_drol | 702 | filter_18x18_to_8x8(dct_codes+64, cur+y*stride + x + 16, stride); |
| 44 : | filter_diff_18x18_to_8x8(dct_codes+64, reference + 16, stride); | ||
| 45 : | Isibaar | 588 | |
| 46 : | suxen_drol | 702 | filter_18x18_to_8x8(dct_codes+128, cur+(y+16)*stride + x, stride); |
| 47 : | filter_diff_18x18_to_8x8(dct_codes+128, reference + 16*stride, stride); | ||
| 48 : | suxen_drol | 701 | |
| 49 : | suxen_drol | 702 | filter_18x18_to_8x8(dct_codes+192, cur+(y+16)*stride + x + 16, stride); |
| 50 : | filter_diff_18x18_to_8x8(dct_codes+192, reference + 16*stride + 16, stride); | ||
| 51 : | suxen_drol | 701 | |
| 52 : | suxen_drol | 702 | transfer32x32_copy(cur + y*stride + x, reference, stride); |
| 53 : | suxen_drol | 701 | |
| 54 : | syskin | 744 | } else { |
| 55 : | suxen_drol | 701 | if(quarterpel) { |
| 56 : | syskin | 744 | const uint8_t * ptr; |
| 57 : | if (dx&3 | dy&3) { | ||
| 58 : | interpolate16x16_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refv + 32, | ||
| 59 : | (uint8_t *) refv + 64, (uint8_t *) refv + 96, x, y, dx, dy, stride, rounding); | ||
| 60 : | ptr = refv + y*stride + x; | ||
| 61 : | } else ptr = ref + (y + dy/4)*stride + x + dx/4; // fullpixel position | ||
| 62 : | suxen_drol | 701 | |
| 63 : | transfer_8to16sub(dct_codes, cur + y*stride + x, | ||
| 64 : | syskin | 744 | ptr, stride); |
| 65 : | suxen_drol | 701 | transfer_8to16sub(dct_codes+64, cur + y*stride + x + 8, |
| 66 : | syskin | 744 | ptr + 8, stride); |
| 67 : | suxen_drol | 701 | transfer_8to16sub(dct_codes+128, cur + y*stride + x + 8*stride, |
| 68 : | syskin | 744 | ptr + 8*stride, stride); |
| 69 : | suxen_drol | 701 | transfer_8to16sub(dct_codes+192, cur + y*stride + x + 8*stride + 8, |
| 70 : | syskin | 744 | ptr + 8*stride+8, stride); |
| 71 : | suxen_drol | 701 | |
| 72 : | syskin | 744 | } else { |
| 73 : | suxen_drol | 702 | const uint8_t * reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
| 74 : | suxen_drol | 701 | |
| 75 : | transfer_8to16sub(dct_codes, cur + y * stride + x, | ||
| 76 : | reference, stride); | ||
| 77 : | transfer_8to16sub(dct_codes+64, cur + y * stride + x + 8, | ||
| 78 : | reference + 8, stride); | ||
| 79 : | transfer_8to16sub(dct_codes+128, cur + y * stride + x + 8*stride, | ||
| 80 : | reference + 8*stride, stride); | ||
| 81 : | transfer_8to16sub(dct_codes+192, cur + y * stride + x + 8*stride+8, | ||
| 82 : | reference + 8*stride + 8, stride); | ||
| 83 : | } | ||
| 84 : | Isibaar | 588 | } |
| 85 : | } | ||
| 86 : | |||
| 87 : | static __inline void | ||
| 88 : | Isibaar | 581 | compensate8x8_interpolate(int16_t * const dct_codes, |
| 89 : | uint8_t * const cur, | ||
| 90 : | const uint8_t * const ref, | ||
| 91 : | const uint8_t * const refh, | ||
| 92 : | const uint8_t * const refv, | ||
| 93 : | const uint8_t * const refhv, | ||
| 94 : | suxen_drol | 702 | uint32_t x, |
| 95 : | uint32_t y, | ||
| 96 : | Isibaar | 581 | const int32_t dx, |
| 97 : | const int32_t dy, | ||
| 98 : | const uint32_t stride, | ||
| 99 : | const uint32_t quarterpel, | ||
| 100 : | suxen_drol | 701 | const int reduced_resolution, |
| 101 : | Isibaar | 581 | const uint32_t rounding) |
| 102 : | Isibaar | 3 | { |
| 103 : | syskin | 744 | if (reduced_resolution) { |
| 104 : | suxen_drol | 702 | const uint8_t * reference; |
| 105 : | x*=2; y*=2; | ||
| 106 : | |||
| 107 : | reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); | ||
| 108 : | |||
| 109 : | filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride); | ||
| 110 : | filter_diff_18x18_to_8x8(dct_codes, reference, stride); | ||
| 111 : | |||
| 112 : | transfer16x16_copy(cur + y*stride + x, reference, stride); | ||
| 113 : | |||
| 114 : | suxen_drol | 701 | } else { |
| 115 : | |||
| 116 : | if(quarterpel) { | ||
| 117 : | syskin | 744 | const uint8_t * ptr; |
| 118 : | if (dx&3 | dy&3) { | ||
| 119 : | interpolate8x8_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refv+32, | ||
| 120 : | (uint8_t *) refv + 64, (uint8_t *) refv+96, x, y, dx, dy, stride, rounding); | ||
| 121 : | ptr = refv + y*stride + x; | ||
| 122 : | } else ptr = ref + (y + dy/4)*stride + x + dx/4; // fullpixel position | ||
| 123 : | Isibaar | 581 | |
| 124 : | syskin | 744 | |
| 125 : | suxen_drol | 701 | transfer_8to16sub(dct_codes, cur + y*stride + x, |
| 126 : | syskin | 744 | ptr, stride); |
| 127 : | } else { | ||
| 128 : | suxen_drol | 702 | const uint8_t * reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
| 129 : | suxen_drol | 701 | |
| 130 : | transfer_8to16sub(dct_codes, cur + y * stride + x, | ||
| 131 : | reference, stride); | ||
| 132 : | Isibaar | 581 | } |
| 133 : | edgomez | 195 | } |
| 134 : | Isibaar | 3 | } |
| 135 : | |||
| 136 : | suxen_drol | 703 | |
| 137 : | |||
| 138 : | /* XXX: slow, inelegant... */ | ||
| 139 : | static void | ||
| 140 : | interpolate18x18_switch(uint8_t * const cur, | ||
| 141 : | const uint8_t * const refn, | ||
| 142 : | const uint32_t x, | ||
| 143 : | const uint32_t y, | ||
| 144 : | const int32_t dx, | ||
| 145 : | const int dy, | ||
| 146 : | const uint32_t stride, | ||
| 147 : | const uint32_t rounding) | ||
| 148 : | { | ||
| 149 : | interpolate8x8_switch(cur, refn, x-1, y-1, dx, dy, stride, rounding); | ||
| 150 : | interpolate8x8_switch(cur, refn, x+7, y-1, dx, dy, stride, rounding); | ||
| 151 : | interpolate8x8_switch(cur, refn, x+9, y-1, dx, dy, stride, rounding); | ||
| 152 : | |||
| 153 : | interpolate8x8_switch(cur, refn, x-1, y+7, dx, dy, stride, rounding); | ||
| 154 : | interpolate8x8_switch(cur, refn, x+7, y+7, dx, dy, stride, rounding); | ||
| 155 : | interpolate8x8_switch(cur, refn, x+9, y+7, dx, dy, stride, rounding); | ||
| 156 : | |||
| 157 : | interpolate8x8_switch(cur, refn, x-1, y+9, dx, dy, stride, rounding); | ||
| 158 : | interpolate8x8_switch(cur, refn, x+7, y+9, dx, dy, stride, rounding); | ||
| 159 : | interpolate8x8_switch(cur, refn, x+9, y+9, dx, dy, stride, rounding); | ||
| 160 : | } | ||
| 161 : | |||
| 162 : | |||
| 163 : | edgomez | 195 | void |
| 164 : | MBMotionCompensation(MACROBLOCK * const mb, | ||
| 165 : | const uint32_t i, | ||
| 166 : | const uint32_t j, | ||
| 167 : | const IMAGE * const ref, | ||
| 168 : | const IMAGE * const refh, | ||
| 169 : | const IMAGE * const refv, | ||
| 170 : | const IMAGE * const refhv, | ||
| 171 : | IMAGE * const cur, | ||
| 172 : | int16_t * dct_codes, | ||
| 173 : | const uint32_t width, | ||
| 174 : | const uint32_t height, | ||
| 175 : | const uint32_t edged_width, | ||
| 176 : | Isibaar | 581 | const uint32_t quarterpel, |
| 177 : | suxen_drol | 701 | const int reduced_resolution, |
| 178 : | edgomez | 195 | const uint32_t rounding) |
| 179 : | Isibaar | 3 | { |
| 180 : | syskin | 744 | int32_t dx = (quarterpel ? mb->qmvs[0].x : mb->mvs[0].x); |
| 181 : | int32_t dy = (quarterpel ? mb->qmvs[0].y : mb->mvs[0].y); | ||
| 182 : | suxen_drol | 701 | |
| 183 : | syskin | 744 | if ( (!reduced_resolution) && (mb->mode == MODE_NOT_CODED) && (dx==0) && (dy==0) ) { /* quick copy */ |
| 184 : | transfer16x16_copy(cur->y + 16 * (i + j * edged_width), | ||
| 185 : | ref->y + 16 * (i + j * edged_width), | ||
| 186 : | edged_width); | ||
| 187 : | |||
| 188 : | transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), | ||
| 189 : | ref->u + 8 * (i + j * edged_width/2), | ||
| 190 : | edged_width / 2); | ||
| 191 : | transfer8x8_copy(cur->v + 8 * (i + j * edged_width/2), | ||
| 192 : | ref->v + 8 * (i + j * edged_width/2), | ||
| 193 : | edged_width / 2); | ||
| 194 : | return; | ||
| 195 : | } | ||
| 196 : | Isibaar | 3 | |
| 197 : | syskin | 744 | if ((mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) /*&& !quarterpel*/) { |
| 198 : | chl | 530 | |
| 199 : | chl | 619 | /* quick MODE_NOT_CODED for GMC with MV!=(0,0) is still needed */ |
| 200 : | Isibaar | 581 | |
| 201 : | syskin | 744 | if (reduced_resolution) { |
| 202 : | suxen_drol | 702 | dx = RRV_MV_SCALEUP(dx); |
| 203 : | dy = RRV_MV_SCALEUP(dy); | ||
| 204 : | } | ||
| 205 : | |||
| 206 : | Isibaar | 588 | compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
| 207 : | suxen_drol | 701 | refv->y, refhv->y, 16 * i, 16 * j, dx, dy, |
| 208 : | edged_width, quarterpel, reduced_resolution, rounding); | ||
| 209 : | Isibaar | 3 | |
| 210 : | syskin | 744 | if (quarterpel) { |
| 211 : | Isibaar | 617 | dx /= 2; |
| 212 : | dy /= 2; | ||
| 213 : | Isibaar | 581 | } |
| 214 : | |||
| 215 : | Isibaar | 617 | dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
| 216 : | dy = (dy >> 1) + roundtab_79[dy & 0x3]; | ||
| 217 : | |||
| 218 : | suxen_drol | 706 | /* uv-block-based compensation */ |
| 219 : | syskin | 744 | if (reduced_resolution) { |
| 220 : | suxen_drol | 721 | const int stride = edged_width/2; |
| 221 : | suxen_drol | 702 | uint8_t * current, * reference; |
| 222 : | |||
| 223 : | current = cur->u + 16*j*stride + 16*i; | ||
| 224 : | reference = refv->u + 16*j*stride + 16*i; | ||
| 225 : | suxen_drol | 703 | interpolate18x18_switch(refv->u, ref->u, 16*i, 16*j, dx, dy, stride, rounding); |
| 226 : | suxen_drol | 702 | filter_18x18_to_8x8(dct_codes + 4*64, current, stride); |
| 227 : | filter_diff_18x18_to_8x8(dct_codes + 4*64, reference, stride); | ||
| 228 : | transfer16x16_copy(current, reference, stride); | ||
| 229 : | |||
| 230 : | current = cur->v + 16*j*stride + 16*i; | ||
| 231 : | reference = refv->v + 16*j*stride + 16*i; | ||
| 232 : | suxen_drol | 703 | interpolate18x18_switch(refv->v, ref->v, 16*i, 16*j, dx, dy, stride, rounding); |
| 233 : | suxen_drol | 702 | filter_18x18_to_8x8(dct_codes + 5*64, current, stride); |
| 234 : | filter_diff_18x18_to_8x8(dct_codes + 5*64, reference, stride); | ||
| 235 : | transfer16x16_copy(current, reference, stride); | ||
| 236 : | syskin | 744 | } else { |
| 237 : | suxen_drol | 701 | transfer_8to16sub(&dct_codes[4 * 64], |
| 238 : | cur->u + 8 * j * edged_width / 2 + 8 * i, | ||
| 239 : | interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, | ||
| 240 : | dx, dy, edged_width / 2, rounding), | ||
| 241 : | syskin | 744 | edged_width / 2); |
| 242 : | Isibaar | 3 | |
| 243 : | suxen_drol | 701 | transfer_8to16sub(&dct_codes[5 * 64], |
| 244 : | cur->v + 8 * j * edged_width / 2 + 8 * i, | ||
| 245 : | interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, | ||
| 246 : | dx, dy, edged_width / 2, rounding), | ||
| 247 : | syskin | 744 | edged_width / 2); |
| 248 : | suxen_drol | 701 | } |
| 249 : | suxen_drol | 449 | |
| 250 : | chl | 530 | } else { // mode == MODE_INTER4V |
| 251 : | suxen_drol | 706 | int k; |
| 252 : | suxen_drol | 702 | VECTOR mvs[4]; |
| 253 : | Isibaar | 3 | |
| 254 : | Isibaar | 581 | if(quarterpel) |
| 255 : | suxen_drol | 706 | for (k = 0; k < 4; k++) mvs[k] = mb->qmvs[k]; |
| 256 : | Isibaar | 581 | else |
| 257 : | suxen_drol | 706 | for (k = 0; k < 4; k++) mvs[k] = mb->mvs[k]; |
| 258 : | Isibaar | 581 | |
| 259 : | suxen_drol | 701 | if (reduced_resolution) |
| 260 : | syskin | 744 | for (k = 0; k < 4; k++) { |
| 261 : | suxen_drol | 706 | mvs[k].x = RRV_MV_SCALEUP(mvs[k].x); |
| 262 : | mvs[k].y = RRV_MV_SCALEUP(mvs[k].y); | ||
| 263 : | suxen_drol | 702 | } |
| 264 : | Isibaar | 3 | |
| 265 : | suxen_drol | 702 | compensate8x8_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
| 266 : | refv->y, refhv->y, 16 * i, 16 * j, mvs[0].x, | ||
| 267 : | mvs[0].y, edged_width, quarterpel, reduced_resolution, rounding); | ||
| 268 : | compensate8x8_interpolate(&dct_codes[1 * 64], cur->y, ref->y, refh->y, | ||
| 269 : | refv->y, refhv->y, 16 * i + 8, 16 * j, | ||
| 270 : | mvs[1].x, mvs[1].y, edged_width, quarterpel, reduced_resolution, rounding); | ||
| 271 : | compensate8x8_interpolate(&dct_codes[2 * 64], cur->y, ref->y, refh->y, | ||
| 272 : | refv->y, refhv->y, 16 * i, 16 * j + 8, | ||
| 273 : | mvs[2].x, mvs[2].y, edged_width, quarterpel, reduced_resolution, rounding); | ||
| 274 : | compensate8x8_interpolate(&dct_codes[3 * 64], cur->y, ref->y, refh->y, | ||
| 275 : | refv->y, refhv->y, 16 * i + 8, 16 * j + 8, | ||
| 276 : | mvs[3].x, mvs[3].y, edged_width, quarterpel, reduced_resolution, rounding); | ||
| 277 : | |||
| 278 : | syskin | 744 | if (quarterpel) { |
| 279 : | dx = (mvs[0].x / 2) + (mvs[1].x / 2) + (mvs[2].x / 2) + (mvs[3].x / 2); | ||
| 280 : | dy = (mvs[0].y / 2) + (mvs[1].y / 2) + (mvs[2].y / 2) + (mvs[3].y / 2); | ||
| 281 : | } else { | ||
| 282 : | dx = mvs[0].x + mvs[1].x + mvs[2].x + mvs[3].x; | ||
| 283 : | dy = mvs[0].y + mvs[1].y + mvs[2].y + mvs[3].y; | ||
| 284 : | } | ||
| 285 : | Isibaar | 581 | |
| 286 : | syskin | 744 | dx = (dx >> 3) + roundtab_76[dx & 0xf]; |
| 287 : | dy = (dy >> 3) + roundtab_76[dy & 0xf]; | ||
| 288 : | Isibaar | 3 | |
| 289 : | Isibaar | 581 | |
| 290 : | suxen_drol | 706 | /* uv-block-based compensation */ |
| 291 : | syskin | 744 | if (reduced_resolution) { |
| 292 : | suxen_drol | 721 | const int stride = edged_width/2; |
| 293 : | suxen_drol | 702 | uint8_t * current, * reference; |
| 294 : | |||
| 295 : | current = cur->u + 16*j*stride + 16*i; | ||
| 296 : | reference = refv->u + 16*j*stride + 16*i; | ||
| 297 : | suxen_drol | 706 | interpolate18x18_switch(refv->u, ref->u, 16*i, 16*j, dx, dy, stride, rounding); |
| 298 : | suxen_drol | 702 | filter_18x18_to_8x8(dct_codes + 4*64, current, stride); |
| 299 : | filter_diff_18x18_to_8x8(dct_codes + 4*64, reference, stride); | ||
| 300 : | transfer16x16_copy(current, reference, stride); | ||
| 301 : | |||
| 302 : | current = cur->v + 16*j*stride + 16*i; | ||
| 303 : | reference = refv->v + 16*j*stride + 16*i; | ||
| 304 : | suxen_drol | 706 | interpolate18x18_switch(refv->v, ref->v, 16*i, 16*j, dx, dy, stride, rounding); |
| 305 : | suxen_drol | 702 | filter_18x18_to_8x8(dct_codes + 5*64, current, stride); |
| 306 : | filter_diff_18x18_to_8x8(dct_codes + 5*64, reference, stride); | ||
| 307 : | transfer16x16_copy(current, reference, stride); | ||
| 308 : | |||
| 309 : | syskin | 744 | } else { |
| 310 : | suxen_drol | 701 | transfer_8to16sub(&dct_codes[4 * 64], |
| 311 : | cur->u + 8 * j * edged_width / 2 + 8 * i, | ||
| 312 : | interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, | ||
| 313 : | dx, dy, edged_width / 2, rounding), | ||
| 314 : | syskin | 744 | edged_width / 2); |
| 315 : | Isibaar | 3 | |
| 316 : | suxen_drol | 701 | transfer_8to16sub(&dct_codes[5 * 64], |
| 317 : | cur->v + 8 * j * edged_width / 2 + 8 * i, | ||
| 318 : | interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, | ||
| 319 : | dx, dy, edged_width / 2, rounding), | ||
| 320 : | syskin | 744 | edged_width / 2); |
| 321 : | suxen_drol | 701 | } |
| 322 : | Isibaar | 3 | } |
| 323 : | } | ||
| 324 : | suxen_drol | 118 | |
| 325 : | |||
| 326 : | edgomez | 195 | void |
| 327 : | MBMotionCompensationBVOP(MBParam * pParam, | ||
| 328 : | MACROBLOCK * const mb, | ||
| 329 : | const uint32_t i, | ||
| 330 : | const uint32_t j, | ||
| 331 : | IMAGE * const cur, | ||
| 332 : | const IMAGE * const f_ref, | ||
| 333 : | const IMAGE * const f_refh, | ||
| 334 : | const IMAGE * const f_refv, | ||
| 335 : | const IMAGE * const f_refhv, | ||
| 336 : | const IMAGE * const b_ref, | ||
| 337 : | const IMAGE * const b_refh, | ||
| 338 : | const IMAGE * const b_refv, | ||
| 339 : | const IMAGE * const b_refhv, | ||
| 340 : | int16_t * dct_codes) | ||
| 341 : | suxen_drol | 118 | { |
| 342 : | static const uint32_t roundtab[16] = | ||
| 343 : | { 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2 }; | ||
| 344 : | |||
| 345 : | const int32_t edged_width = pParam->edged_width; | ||
| 346 : | int32_t dx, dy; | ||
| 347 : | int32_t b_dx, b_dy; | ||
| 348 : | syskin | 744 | int k, sum; |
| 349 : | suxen_drol | 118 | int x = i; |
| 350 : | int y = j; | ||
| 351 : | syskin | 744 | const uint32_t quarterpel = pParam->m_quarterpel; |
| 352 : | const uint8_t * ptr1, * ptr2; | ||
| 353 : | suxen_drol | 118 | |
| 354 : | edgomez | 195 | switch (mb->mode) { |
| 355 : | case MODE_FORWARD: | ||
| 356 : | suxen_drol | 118 | |
| 357 : | syskin | 658 | if (quarterpel) { |
| 358 : | dx = mb->qmvs[0].x; | ||
| 359 : | dy = mb->qmvs[0].y; | ||
| 360 : | } else { | ||
| 361 : | dx = mb->mvs[0].x; | ||
| 362 : | dy = mb->mvs[0].y; | ||
| 363 : | } | ||
| 364 : | suxen_drol | 118 | |
| 365 : | syskin | 658 | compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, f_ref->y, f_refh->y, |
| 366 : | f_refv->y, f_refhv->y, 16 * i, 16 * j, dx, | ||
| 367 : | syskin | 744 | dy, edged_width, quarterpel, 0, 0); |
| 368 : | suxen_drol | 118 | |
| 369 : | syskin | 658 | if (quarterpel) { |
| 370 : | dx /= 2; | ||
| 371 : | dy /= 2; | ||
| 372 : | } | ||
| 373 : | suxen_drol | 118 | |
| 374 : | syskin | 658 | dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
| 375 : | dy = (dy >> 1) + roundtab_79[dy & 0x3]; | ||
| 376 : | suxen_drol | 118 | |
| 377 : | chl | 530 | /* uv-block-based compensation */ |
| 378 : | suxen_drol | 449 | transfer_8to16sub(&dct_codes[4 * 64], |
| 379 : | chl | 530 | cur->u + 8 * j * edged_width / 2 + 8 * i, |
| 380 : | interpolate8x8_switch2(f_refv->u, f_ref->u, 8 * i, 8 * j, | ||
| 381 : | dx, dy, edged_width / 2, 0), | ||
| 382 : | suxen_drol | 449 | edged_width / 2); |
| 383 : | |||
| 384 : | transfer_8to16sub(&dct_codes[5 * 64], | ||
| 385 : | chl | 530 | cur->v + 8 * j * edged_width / 2 + 8 * i, |
| 386 : | interpolate8x8_switch2(f_refv->u, f_ref->v, 8 * i, 8 * j, | ||
| 387 : | dx, dy, edged_width / 2, 0), | ||
| 388 : | suxen_drol | 449 | edged_width / 2); |
| 389 : | |||
| 390 : | suxen_drol | 118 | break; |
| 391 : | |||
| 392 : | edgomez | 195 | case MODE_BACKWARD: |
| 393 : | syskin | 658 | if (quarterpel) { |
| 394 : | b_dx = mb->b_qmvs[0].x; | ||
| 395 : | b_dy = mb->b_qmvs[0].y; | ||
| 396 : | } else { | ||
| 397 : | b_dx = mb->b_mvs[0].x; | ||
| 398 : | b_dy = mb->b_mvs[0].y; | ||
| 399 : | } | ||
| 400 : | suxen_drol | 118 | |
| 401 : | syskin | 658 | compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, b_ref->y, b_refh->y, |
| 402 : | b_refv->y, b_refhv->y, 16 * i, 16 * j, b_dx, | ||
| 403 : | syskin | 744 | b_dy, edged_width, quarterpel, 0, 0); |
| 404 : | suxen_drol | 118 | |
| 405 : | syskin | 658 | if (quarterpel) { |
| 406 : | b_dx /= 2; | ||
| 407 : | b_dy /= 2; | ||
| 408 : | } | ||
| 409 : | suxen_drol | 118 | |
| 410 : | syskin | 658 | b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
| 411 : | b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3]; | ||
| 412 : | suxen_drol | 118 | |
| 413 : | |||
| 414 : | chl | 530 | /* uv-block-based compensation */ |
| 415 : | suxen_drol | 449 | transfer_8to16sub(&dct_codes[4 * 64], |
| 416 : | chl | 530 | cur->u + 8 * j * edged_width / 2 + 8 * i, |
| 417 : | interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, | ||
| 418 : | b_dx, b_dy, edged_width / 2, 0), | ||
| 419 : | suxen_drol | 449 | |
| 420 : | chl | 530 | edged_width / 2); |
| 421 : | |||
| 422 : | suxen_drol | 449 | transfer_8to16sub(&dct_codes[5 * 64], |
| 423 : | chl | 530 | cur->v + 8 * j * edged_width / 2 + 8 * i, |
| 424 : | interpolate8x8_switch2(f_refv->u, b_ref->v, 8 * i, 8 * j, | ||
| 425 : | b_dx, b_dy, edged_width / 2, 0), | ||
| 426 : | suxen_drol | 449 | |
| 427 : | chl | 530 | edged_width / 2); |
| 428 : | |||
| 429 : | suxen_drol | 118 | break; |
| 430 : | |||
| 431 : | syskin | 744 | case MODE_INTERPOLATE: /* _could_ use DIRECT, but would be overkill (no 4MV there) */ |
| 432 : | chl | 530 | case MODE_DIRECT_NO4V: |
| 433 : | chl | 374 | |
| 434 : | syskin | 658 | if (quarterpel) { |
| 435 : | dx = mb->qmvs[0].x; | ||
| 436 : | dy = mb->qmvs[0].y; | ||
| 437 : | b_dx = mb->b_qmvs[0].x; | ||
| 438 : | b_dy = mb->b_qmvs[0].y; | ||
| 439 : | syskin | 744 | |
| 440 : | if (dx&3 | dy&3) { | ||
| 441 : | interpolate16x16_quarterpel((uint8_t *) f_refv->y, (uint8_t *) f_ref->y, (uint8_t *) f_refv->y + 32, | ||
| 442 : | (uint8_t *) f_refv->y + 64, (uint8_t *) f_refv->y + 96, 16*i, 16*j, dx, dy, edged_width, 0); | ||
| 443 : | ptr1 = f_refv->y + i * 16 + j * 16 * edged_width; | ||
| 444 : | } else ptr1 = f_ref->y + (16*j + dy/4)*edged_width + 16*i + dx/4; // fullpixel position | ||
| 445 : | chl | 374 | |
| 446 : | syskin | 744 | if (b_dx&3 | b_dy&3) { |
| 447 : | interpolate16x16_quarterpel((uint8_t *) b_refv->y, (uint8_t *) b_ref->y, (uint8_t *) f_refv->y + 32, | ||
| 448 : | (uint8_t *) f_refv->y + 64, (uint8_t *) f_refv->y + 96, 16*i, 16*j, b_dx, b_dy, edged_width, 0); | ||
| 449 : | ptr2 = b_refv->y + i * 16 + j * 16 * edged_width; | ||
| 450 : | } else ptr2 = b_ref->y + (16*j + b_dy/4)*edged_width + 16*i + b_dx/4; // fullpixel position | ||
| 451 : | suxen_drol | 118 | |
| 452 : | syskin | 658 | for (k = 0; k < 4; k++) { |
| 453 : | transfer_8to16sub2(&dct_codes[k * 64], | ||
| 454 : | cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, | ||
| 455 : | syskin | 744 | ptr1 + (k&1)*8 + (k>>1)*8*edged_width, |
| 456 : | ptr2 + (k&1)*8 + (k>>1)*8*edged_width, edged_width); | ||
| 457 : | syskin | 658 | } |
| 458 : | b_dx /= 2; | ||
| 459 : | b_dy /= 2; | ||
| 460 : | dx /= 2; | ||
| 461 : | dy /= 2; | ||
| 462 : | |||
| 463 : | } else { | ||
| 464 : | dx = mb->mvs[0].x; | ||
| 465 : | dy = mb->mvs[0].y; | ||
| 466 : | b_dx = mb->b_mvs[0].x; | ||
| 467 : | b_dy = mb->b_mvs[0].y; | ||
| 468 : | |||
| 469 : | for (k = 0; k < 4; k++) { | ||
| 470 : | transfer_8to16sub2(&dct_codes[k * 64], | ||
| 471 : | cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, | ||
| 472 : | get_ref(f_ref->y, f_refh->y, f_refv->y, | ||
| 473 : | f_refhv->y, 2*i + (k&1), 2*j + (k>>1), 8, dx, dy, | ||
| 474 : | edged_width), | ||
| 475 : | get_ref(b_ref->y, b_refh->y, b_refv->y, | ||
| 476 : | b_refhv->y, 2*i + (k&1), 2 * j+(k>>1), 8, b_dx, b_dy, | ||
| 477 : | edged_width), | ||
| 478 : | edged_width); | ||
| 479 : | } | ||
| 480 : | |||
| 481 : | chl | 374 | } |
| 482 : | suxen_drol | 118 | |
| 483 : | |||
| 484 : | syskin | 658 | dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
| 485 : | dy = (dy >> 1) + roundtab_79[dy & 0x3]; | ||
| 486 : | suxen_drol | 118 | |
| 487 : | syskin | 658 | b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
| 488 : | b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3]; | ||
| 489 : | |||
| 490 : | chl | 530 | transfer_8to16sub2(&dct_codes[4 * 64], |
| 491 : | cur->u + (y * 8) * edged_width / 2 + (x * 8), | ||
| 492 : | interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, | ||
| 493 : | b_dx, b_dy, edged_width / 2, 0), | ||
| 494 : | interpolate8x8_switch2(f_refv->u + 8, f_ref->u, 8 * i, 8 * j, | ||
| 495 : | dx, dy, edged_width / 2, 0), | ||
| 496 : | suxen_drol | 449 | edged_width / 2); |
| 497 : | |||
| 498 : | chl | 530 | transfer_8to16sub2(&dct_codes[5 * 64], |
| 499 : | cur->v + (y * 8) * edged_width / 2 + (x * 8), | ||
| 500 : | interpolate8x8_switch2(f_refv->u, b_ref->v, 8 * i, 8 * j, | ||
| 501 : | b_dx, b_dy, edged_width / 2, 0), | ||
| 502 : | interpolate8x8_switch2(f_refv->u + 8, f_ref->v, 8 * i, 8 * j, | ||
| 503 : | dx, dy, edged_width / 2, 0), | ||
| 504 : | suxen_drol | 449 | edged_width / 2); |
| 505 : | chl | 530 | |
| 506 : | suxen_drol | 118 | break; |
| 507 : | chl | 374 | |
| 508 : | case MODE_DIRECT: | ||
| 509 : | syskin | 744 | |
| 510 : | syskin | 658 | if (quarterpel) { |
| 511 : | syskin | 744 | for (k = 0; k < 4; k++) { |
| 512 : | syskin | 658 | dx = mb->qmvs[k].x; |
| 513 : | dy = mb->qmvs[k].y; | ||
| 514 : | b_dx = mb->b_qmvs[k].x; | ||
| 515 : | b_dy = mb->b_qmvs[k].y; | ||
| 516 : | syskin | 744 | |
| 517 : | if (dx&3 | dy&3) { | ||
| 518 : | interpolate8x8_quarterpel((uint8_t *) f_refv->y, | ||
| 519 : | (uint8_t *) f_ref->y, | ||
| 520 : | (uint8_t *) f_refv->y + 32, | ||
| 521 : | (uint8_t *) f_refv->y + 64, | ||
| 522 : | (uint8_t *) f_refv->y + 96, | ||
| 523 : | 16*i + (k&1)*8, 16*j + (k>>1)*8, dx, dy, edged_width, 0); | ||
| 524 : | ptr1 = f_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width; | ||
| 525 : | } else ptr1 = f_ref->y + (16*j + (k>>1)*8 + dy/4)*edged_width + 16*i + (k&1)*8 + dx/4; | ||
| 526 : | chl | 374 | |
| 527 : | syskin | 744 | if (b_dx&3 | b_dy&3) { |
| 528 : | interpolate8x8_quarterpel((uint8_t *) b_refv->y, | ||
| 529 : | (uint8_t *) b_ref->y, | ||
| 530 : | (uint8_t *) f_refv->y + 32, | ||
| 531 : | (uint8_t *) f_refv->y + 64, | ||
| 532 : | (uint8_t *) f_refv->y + 96, | ||
| 533 : | 16*i + (k&1)*8, 16*j + (k>>1)*8, b_dx, b_dy, edged_width, 0); | ||
| 534 : | ptr2 = b_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width; | ||
| 535 : | } else ptr2 = b_ref->y + (16*j + (k>>1)*8 + b_dy/4)*edged_width + 16*i + (k&1)*8 + b_dx/4; | ||
| 536 : | chl | 374 | |
| 537 : | |||
| 538 : | syskin | 658 | transfer_8to16sub2(&dct_codes[k * 64], |
| 539 : | cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, | ||
| 540 : | syskin | 744 | ptr1, ptr2, edged_width); |
| 541 : | syskin | 658 | } |
| 542 : | sum = mb->qmvs[0].y/2 + mb->qmvs[1].y/2 + mb->qmvs[2].y/2 + mb->qmvs[3].y/2; | ||
| 543 : | dy = (sum >> 3) + roundtab_76[sum & 0xf]; | ||
| 544 : | sum = mb->qmvs[0].x/2 + mb->qmvs[1].x/2 + mb->qmvs[2].x/2 + mb->qmvs[3].x/2; | ||
| 545 : | dx = (sum >> 3) + roundtab_76[sum & 0xf]; | ||
| 546 : | chl | 374 | |
| 547 : | syskin | 658 | sum = mb->b_qmvs[0].y/2 + mb->b_qmvs[1].y/2 + mb->b_qmvs[2].y/2 + mb->b_qmvs[3].y/2; |
| 548 : | b_dy = (sum >> 3) + roundtab_76[sum & 0xf]; | ||
| 549 : | sum = mb->b_qmvs[0].x/2 + mb->b_qmvs[1].x/2 + mb->b_qmvs[2].x/2 + mb->b_qmvs[3].x/2; | ||
| 550 : | b_dx = (sum >> 3) + roundtab_76[sum & 0xf]; | ||
| 551 : | chl | 374 | |
| 552 : | syskin | 658 | } else { |
| 553 : | for (k=0;k<4;k++) { | ||
| 554 : | dx = mb->mvs[k].x; | ||
| 555 : | dy = mb->mvs[k].y; | ||
| 556 : | chl | 374 | |
| 557 : | syskin | 658 | b_dx = mb->b_mvs[k].x; |
| 558 : | b_dy = mb->b_mvs[k].y; | ||
| 559 : | chl | 374 | |
| 560 : | syskin | 658 | transfer_8to16sub2(&dct_codes[k * 64], |
| 561 : | cur->y + (i*16 + (k&1)*8) + (j*16 + (k>>1)*8 ) * edged_width, | ||
| 562 : | get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, | ||
| 563 : | 2*i + (k&1), 2*j + (k>>1), 8, dx, dy, | ||
| 564 : | edged_width), | ||
| 565 : | get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, | ||
| 566 : | 2*i + (k&1), 2*j + (k>>1), 8, b_dx, b_dy, | ||
| 567 : | edged_width), | ||
| 568 : | edged_width); | ||
| 569 : | } | ||
| 570 : | chl | 374 | |
| 571 : | syskin | 658 | sum = mb->mvs[0].x + mb->mvs[1].x + mb->mvs[2].x + mb->mvs[3].x; |
| 572 : | dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); | ||
| 573 : | chl | 374 | |
| 574 : | syskin | 658 | sum = mb->mvs[0].y + mb->mvs[1].y + mb->mvs[2].y + mb->mvs[3].y; |
| 575 : | dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); | ||
| 576 : | |||
| 577 : | sum = mb->b_mvs[0].x + mb->b_mvs[1].x + mb->b_mvs[2].x + mb->b_mvs[3].x; | ||
| 578 : | b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); | ||
| 579 : | |||
| 580 : | sum = mb->b_mvs[0].y + mb->b_mvs[1].y + mb->b_mvs[2].y + mb->b_mvs[3].y; | ||
| 581 : | b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); | ||
| 582 : | |||
| 583 : | } | ||
| 584 : | chl | 530 | transfer_8to16sub2(&dct_codes[4 * 64], |
| 585 : | cur->u + (y * 8) * edged_width / 2 + (x * 8), | ||
| 586 : | interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, | ||
| 587 : | b_dx, b_dy, edged_width / 2, 0), | ||
| 588 : | interpolate8x8_switch2(f_refv->u + 8, f_ref->u, 8 * i, 8 * j, dx, dy, | ||
| 589 : | edged_width / 2, 0), | ||
| 590 : | suxen_drol | 449 | edged_width / 2); |
| 591 : | |||
| 592 : | chl | 530 | transfer_8to16sub2(&dct_codes[5 * 64], |
| 593 : | cur->v + (y * 8) * edged_width / 2 + (x * 8), | ||
| 594 : | interpolate8x8_switch2(f_refv->u, b_ref->v, 8 * i, 8 * j, | ||
| 595 : | b_dx, b_dy, edged_width / 2, 0), | ||
| 596 : | interpolate8x8_switch2(f_refv->u + 8, f_ref->v, 8 * i, 8 * j, | ||
| 597 : | dx, dy, edged_width / 2, 0), | ||
| 598 : | suxen_drol | 449 | edged_width / 2); |
| 599 : | chl | 530 | |
| 600 : | |||
| 601 : | suxen_drol | 118 | break; |
| 602 : | } | ||
| 603 : | } |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |