| 1 | /* |
| 2 | * VC-1 and WMV3 decoder |
| 3 | * Copyright (c) 2011 Mashiat Sarker Shakkhar |
| 4 | * Copyright (c) 2006-2007 Konstantin Shishkov |
| 5 | * Partly based on vc9.c (c) 2005 Anonymous, Alex Beregszaszi, Michael Niedermayer |
| 6 | * |
| 7 | * This file is part of FFmpeg. |
| 8 | * |
| 9 | * FFmpeg is free software; you can redistribute it and/or |
| 10 | * modify it under the terms of the GNU Lesser General Public |
| 11 | * License as published by the Free Software Foundation; either |
| 12 | * version 2.1 of the License, or (at your option) any later version. |
| 13 | * |
| 14 | * FFmpeg is distributed in the hope that it will be useful, |
| 15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 17 | * Lesser General Public License for more details. |
| 18 | * |
| 19 | * You should have received a copy of the GNU Lesser General Public |
| 20 | * License along with FFmpeg; if not, write to the Free Software |
| 21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 22 | */ |
| 23 | |
| 24 | /** |
| 25 | * @file |
| 26 | * VC-1 and WMV3 block decoding routines |
| 27 | */ |
| 28 | |
| 29 | #include "avcodec.h" |
| 30 | #include "h264chroma.h" |
| 31 | #include "mathops.h" |
| 32 | #include "mpegvideo.h" |
| 33 | #include "vc1.h" |
| 34 | |
| 35 | /** Do motion compensation over 1 macroblock |
| 36 | * Mostly adapted hpel_motion and qpel_motion from mpegvideo.c |
| 37 | */ |
| 38 | void ff_vc1_mc_1mv(VC1Context *v, int dir) |
| 39 | { |
| 40 | MpegEncContext *s = &v->s; |
| 41 | H264ChromaContext *h264chroma = &v->h264chroma; |
| 42 | uint8_t *srcY, *srcU, *srcV; |
| 43 | int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y; |
| 44 | int v_edge_pos = s->v_edge_pos >> v->field_mode; |
| 45 | int i; |
| 46 | uint8_t (*luty)[256], (*lutuv)[256]; |
| 47 | int use_ic; |
| 48 | |
| 49 | if ((!v->field_mode || |
| 50 | (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) && |
| 51 | !v->s.last_picture.f->data[0]) |
| 52 | return; |
| 53 | |
| 54 | mx = s->mv[dir][0][0]; |
| 55 | my = s->mv[dir][0][1]; |
| 56 | |
| 57 | // store motion vectors for further use in B frames |
| 58 | if (s->pict_type == AV_PICTURE_TYPE_P) { |
| 59 | for (i = 0; i < 4; i++) { |
| 60 | s->current_picture.motion_val[1][s->block_index[i] + v->blocks_off][0] = mx; |
| 61 | s->current_picture.motion_val[1][s->block_index[i] + v->blocks_off][1] = my; |
| 62 | } |
| 63 | } |
| 64 | |
| 65 | uvmx = (mx + ((mx & 3) == 3)) >> 1; |
| 66 | uvmy = (my + ((my & 3) == 3)) >> 1; |
| 67 | v->luma_mv[s->mb_x][0] = uvmx; |
| 68 | v->luma_mv[s->mb_x][1] = uvmy; |
| 69 | |
| 70 | if (v->field_mode && |
| 71 | v->cur_field_type != v->ref_field_type[dir]) { |
| 72 | my = my - 2 + 4 * v->cur_field_type; |
| 73 | uvmy = uvmy - 2 + 4 * v->cur_field_type; |
| 74 | } |
| 75 | |
| 76 | // fastuvmc shall be ignored for interlaced frame picture |
| 77 | if (v->fastuvmc && (v->fcm != ILACE_FRAME)) { |
| 78 | uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1)); |
| 79 | uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1)); |
| 80 | } |
| 81 | if (!dir) { |
| 82 | if (v->field_mode && (v->cur_field_type != v->ref_field_type[dir]) && v->second_field) { |
| 83 | srcY = s->current_picture.f->data[0]; |
| 84 | srcU = s->current_picture.f->data[1]; |
| 85 | srcV = s->current_picture.f->data[2]; |
| 86 | luty = v->curr_luty; |
| 87 | lutuv = v->curr_lutuv; |
| 88 | use_ic = *v->curr_use_ic; |
| 89 | } else { |
| 90 | srcY = s->last_picture.f->data[0]; |
| 91 | srcU = s->last_picture.f->data[1]; |
| 92 | srcV = s->last_picture.f->data[2]; |
| 93 | luty = v->last_luty; |
| 94 | lutuv = v->last_lutuv; |
| 95 | use_ic = v->last_use_ic; |
| 96 | } |
| 97 | } else { |
| 98 | srcY = s->next_picture.f->data[0]; |
| 99 | srcU = s->next_picture.f->data[1]; |
| 100 | srcV = s->next_picture.f->data[2]; |
| 101 | luty = v->next_luty; |
| 102 | lutuv = v->next_lutuv; |
| 103 | use_ic = v->next_use_ic; |
| 104 | } |
| 105 | |
| 106 | if (!srcY || !srcU) { |
| 107 | av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n"); |
| 108 | return; |
| 109 | } |
| 110 | |
| 111 | src_x = s->mb_x * 16 + (mx >> 2); |
| 112 | src_y = s->mb_y * 16 + (my >> 2); |
| 113 | uvsrc_x = s->mb_x * 8 + (uvmx >> 2); |
| 114 | uvsrc_y = s->mb_y * 8 + (uvmy >> 2); |
| 115 | |
| 116 | if (v->profile != PROFILE_ADVANCED) { |
| 117 | src_x = av_clip( src_x, -16, s->mb_width * 16); |
| 118 | src_y = av_clip( src_y, -16, s->mb_height * 16); |
| 119 | uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8); |
| 120 | uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8); |
| 121 | } else { |
| 122 | src_x = av_clip( src_x, -17, s->avctx->coded_width); |
| 123 | src_y = av_clip( src_y, -18, s->avctx->coded_height + 1); |
| 124 | uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1); |
| 125 | uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1); |
| 126 | } |
| 127 | |
| 128 | srcY += src_y * s->linesize + src_x; |
| 129 | srcU += uvsrc_y * s->uvlinesize + uvsrc_x; |
| 130 | srcV += uvsrc_y * s->uvlinesize + uvsrc_x; |
| 131 | |
| 132 | if (v->field_mode && v->ref_field_type[dir]) { |
| 133 | srcY += s->current_picture_ptr->f->linesize[0]; |
| 134 | srcU += s->current_picture_ptr->f->linesize[1]; |
| 135 | srcV += s->current_picture_ptr->f->linesize[2]; |
| 136 | } |
| 137 | |
| 138 | /* for grayscale we should not try to read from unknown area */ |
| 139 | if (s->flags & CODEC_FLAG_GRAY) { |
| 140 | srcU = s->edge_emu_buffer + 18 * s->linesize; |
| 141 | srcV = s->edge_emu_buffer + 18 * s->linesize; |
| 142 | } |
| 143 | |
| 144 | if (v->rangeredfrm || use_ic |
| 145 | || s->h_edge_pos < 22 || v_edge_pos < 22 |
| 146 | || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 16 - s->mspel * 3 |
| 147 | || (unsigned)(src_y - 1) > v_edge_pos - (my&3) - 16 - 3) { |
| 148 | uint8_t *ubuf = s->edge_emu_buffer + 19 * s->linesize; |
| 149 | uint8_t *vbuf = ubuf + 9 * s->uvlinesize; |
| 150 | |
| 151 | srcY -= s->mspel * (1 + s->linesize); |
| 152 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, |
| 153 | s->linesize, s->linesize, |
| 154 | 17 + s->mspel * 2, 17 + s->mspel * 2, |
| 155 | src_x - s->mspel, src_y - s->mspel, |
| 156 | s->h_edge_pos, v_edge_pos); |
| 157 | srcY = s->edge_emu_buffer; |
| 158 | s->vdsp.emulated_edge_mc(ubuf, srcU, |
| 159 | s->uvlinesize, s->uvlinesize, |
| 160 | 8 + 1, 8 + 1, |
| 161 | uvsrc_x, uvsrc_y, |
| 162 | s->h_edge_pos >> 1, v_edge_pos >> 1); |
| 163 | s->vdsp.emulated_edge_mc(vbuf, srcV, |
| 164 | s->uvlinesize, s->uvlinesize, |
| 165 | 8 + 1, 8 + 1, |
| 166 | uvsrc_x, uvsrc_y, |
| 167 | s->h_edge_pos >> 1, v_edge_pos >> 1); |
| 168 | srcU = ubuf; |
| 169 | srcV = vbuf; |
| 170 | /* if we deal with range reduction we need to scale source blocks */ |
| 171 | if (v->rangeredfrm) { |
| 172 | int i, j; |
| 173 | uint8_t *src, *src2; |
| 174 | |
| 175 | src = srcY; |
| 176 | for (j = 0; j < 17 + s->mspel * 2; j++) { |
| 177 | for (i = 0; i < 17 + s->mspel * 2; i++) |
| 178 | src[i] = ((src[i] - 128) >> 1) + 128; |
| 179 | src += s->linesize; |
| 180 | } |
| 181 | src = srcU; |
| 182 | src2 = srcV; |
| 183 | for (j = 0; j < 9; j++) { |
| 184 | for (i = 0; i < 9; i++) { |
| 185 | src[i] = ((src[i] - 128) >> 1) + 128; |
| 186 | src2[i] = ((src2[i] - 128) >> 1) + 128; |
| 187 | } |
| 188 | src += s->uvlinesize; |
| 189 | src2 += s->uvlinesize; |
| 190 | } |
| 191 | } |
| 192 | /* if we deal with intensity compensation we need to scale source blocks */ |
| 193 | if (use_ic) { |
| 194 | int i, j; |
| 195 | uint8_t *src, *src2; |
| 196 | |
| 197 | src = srcY; |
| 198 | for (j = 0; j < 17 + s->mspel * 2; j++) { |
| 199 | int f = v->field_mode ? v->ref_field_type[dir] : ((j + src_y - s->mspel) & 1) ; |
| 200 | for (i = 0; i < 17 + s->mspel * 2; i++) |
| 201 | src[i] = luty[f][src[i]]; |
| 202 | src += s->linesize; |
| 203 | } |
| 204 | src = srcU; |
| 205 | src2 = srcV; |
| 206 | for (j = 0; j < 9; j++) { |
| 207 | int f = v->field_mode ? v->ref_field_type[dir] : ((j + uvsrc_y) & 1); |
| 208 | for (i = 0; i < 9; i++) { |
| 209 | src[i] = lutuv[f][src[i]]; |
| 210 | src2[i] = lutuv[f][src2[i]]; |
| 211 | } |
| 212 | src += s->uvlinesize; |
| 213 | src2 += s->uvlinesize; |
| 214 | } |
| 215 | } |
| 216 | srcY += s->mspel * (1 + s->linesize); |
| 217 | } |
| 218 | |
| 219 | if (s->mspel) { |
| 220 | dxy = ((my & 3) << 2) | (mx & 3); |
| 221 | v->vc1dsp.put_vc1_mspel_pixels_tab[0][dxy](s->dest[0] , srcY , s->linesize, v->rnd); |
| 222 | } else { // hpel mc - always used for luma |
| 223 | dxy = (my & 2) | ((mx & 2) >> 1); |
| 224 | if (!v->rnd) |
| 225 | s->hdsp.put_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16); |
| 226 | else |
| 227 | s->hdsp.put_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16); |
| 228 | } |
| 229 | |
| 230 | if (s->flags & CODEC_FLAG_GRAY) return; |
| 231 | /* Chroma MC always uses qpel bilinear */ |
| 232 | uvmx = (uvmx & 3) << 1; |
| 233 | uvmy = (uvmy & 3) << 1; |
| 234 | if (!v->rnd) { |
| 235 | h264chroma->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy); |
| 236 | h264chroma->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy); |
| 237 | } else { |
| 238 | v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy); |
| 239 | v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy); |
| 240 | } |
| 241 | } |
| 242 | |
| 243 | static inline int median4(int a, int b, int c, int d) |
| 244 | { |
| 245 | if (a < b) { |
| 246 | if (c < d) return (FFMIN(b, d) + FFMAX(a, c)) / 2; |
| 247 | else return (FFMIN(b, c) + FFMAX(a, d)) / 2; |
| 248 | } else { |
| 249 | if (c < d) return (FFMIN(a, d) + FFMAX(b, c)) / 2; |
| 250 | else return (FFMIN(a, c) + FFMAX(b, d)) / 2; |
| 251 | } |
| 252 | } |
| 253 | |
| 254 | /** Do motion compensation for 4-MV macroblock - luminance block |
| 255 | */ |
| 256 | void ff_vc1_mc_4mv_luma(VC1Context *v, int n, int dir, int avg) |
| 257 | { |
| 258 | MpegEncContext *s = &v->s; |
| 259 | uint8_t *srcY; |
| 260 | int dxy, mx, my, src_x, src_y; |
| 261 | int off; |
| 262 | int fieldmv = (v->fcm == ILACE_FRAME) ? v->blk_mv_type[s->block_index[n]] : 0; |
| 263 | int v_edge_pos = s->v_edge_pos >> v->field_mode; |
| 264 | uint8_t (*luty)[256]; |
| 265 | int use_ic; |
| 266 | |
| 267 | if ((!v->field_mode || |
| 268 | (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) && |
| 269 | !v->s.last_picture.f->data[0]) |
| 270 | return; |
| 271 | |
| 272 | mx = s->mv[dir][n][0]; |
| 273 | my = s->mv[dir][n][1]; |
| 274 | |
| 275 | if (!dir) { |
| 276 | if (v->field_mode && (v->cur_field_type != v->ref_field_type[dir]) && v->second_field) { |
| 277 | srcY = s->current_picture.f->data[0]; |
| 278 | luty = v->curr_luty; |
| 279 | use_ic = *v->curr_use_ic; |
| 280 | } else { |
| 281 | srcY = s->last_picture.f->data[0]; |
| 282 | luty = v->last_luty; |
| 283 | use_ic = v->last_use_ic; |
| 284 | } |
| 285 | } else { |
| 286 | srcY = s->next_picture.f->data[0]; |
| 287 | luty = v->next_luty; |
| 288 | use_ic = v->next_use_ic; |
| 289 | } |
| 290 | |
| 291 | if (!srcY) { |
| 292 | av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n"); |
| 293 | return; |
| 294 | } |
| 295 | |
| 296 | if (v->field_mode) { |
| 297 | if (v->cur_field_type != v->ref_field_type[dir]) |
| 298 | my = my - 2 + 4 * v->cur_field_type; |
| 299 | } |
| 300 | |
| 301 | if (s->pict_type == AV_PICTURE_TYPE_P && n == 3 && v->field_mode) { |
| 302 | int same_count = 0, opp_count = 0, k; |
| 303 | int chosen_mv[2][4][2], f; |
| 304 | int tx, ty; |
| 305 | for (k = 0; k < 4; k++) { |
| 306 | f = v->mv_f[0][s->block_index[k] + v->blocks_off]; |
| 307 | chosen_mv[f][f ? opp_count : same_count][0] = s->mv[0][k][0]; |
| 308 | chosen_mv[f][f ? opp_count : same_count][1] = s->mv[0][k][1]; |
| 309 | opp_count += f; |
| 310 | same_count += 1 - f; |
| 311 | } |
| 312 | f = opp_count > same_count; |
| 313 | switch (f ? opp_count : same_count) { |
| 314 | case 4: |
| 315 | tx = median4(chosen_mv[f][0][0], chosen_mv[f][1][0], |
| 316 | chosen_mv[f][2][0], chosen_mv[f][3][0]); |
| 317 | ty = median4(chosen_mv[f][0][1], chosen_mv[f][1][1], |
| 318 | chosen_mv[f][2][1], chosen_mv[f][3][1]); |
| 319 | break; |
| 320 | case 3: |
| 321 | tx = mid_pred(chosen_mv[f][0][0], chosen_mv[f][1][0], chosen_mv[f][2][0]); |
| 322 | ty = mid_pred(chosen_mv[f][0][1], chosen_mv[f][1][1], chosen_mv[f][2][1]); |
| 323 | break; |
| 324 | case 2: |
| 325 | tx = (chosen_mv[f][0][0] + chosen_mv[f][1][0]) / 2; |
| 326 | ty = (chosen_mv[f][0][1] + chosen_mv[f][1][1]) / 2; |
| 327 | break; |
| 328 | default: |
| 329 | av_assert0(0); |
| 330 | } |
| 331 | s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx; |
| 332 | s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty; |
| 333 | for (k = 0; k < 4; k++) |
| 334 | v->mv_f[1][s->block_index[k] + v->blocks_off] = f; |
| 335 | } |
| 336 | |
| 337 | if (v->fcm == ILACE_FRAME) { // not sure if needed for other types of picture |
| 338 | int qx, qy; |
| 339 | int width = s->avctx->coded_width; |
| 340 | int height = s->avctx->coded_height >> 1; |
| 341 | if (s->pict_type == AV_PICTURE_TYPE_P) { |
| 342 | s->current_picture.motion_val[1][s->block_index[n] + v->blocks_off][0] = mx; |
| 343 | s->current_picture.motion_val[1][s->block_index[n] + v->blocks_off][1] = my; |
| 344 | } |
| 345 | qx = (s->mb_x * 16) + (mx >> 2); |
| 346 | qy = (s->mb_y * 8) + (my >> 3); |
| 347 | |
| 348 | if (qx < -17) |
| 349 | mx -= 4 * (qx + 17); |
| 350 | else if (qx > width) |
| 351 | mx -= 4 * (qx - width); |
| 352 | if (qy < -18) |
| 353 | my -= 8 * (qy + 18); |
| 354 | else if (qy > height + 1) |
| 355 | my -= 8 * (qy - height - 1); |
| 356 | } |
| 357 | |
| 358 | if ((v->fcm == ILACE_FRAME) && fieldmv) |
| 359 | off = ((n > 1) ? s->linesize : 0) + (n & 1) * 8; |
| 360 | else |
| 361 | off = s->linesize * 4 * (n & 2) + (n & 1) * 8; |
| 362 | |
| 363 | src_x = s->mb_x * 16 + (n & 1) * 8 + (mx >> 2); |
| 364 | if (!fieldmv) |
| 365 | src_y = s->mb_y * 16 + (n & 2) * 4 + (my >> 2); |
| 366 | else |
| 367 | src_y = s->mb_y * 16 + ((n > 1) ? 1 : 0) + (my >> 2); |
| 368 | |
| 369 | if (v->profile != PROFILE_ADVANCED) { |
| 370 | src_x = av_clip(src_x, -16, s->mb_width * 16); |
| 371 | src_y = av_clip(src_y, -16, s->mb_height * 16); |
| 372 | } else { |
| 373 | src_x = av_clip(src_x, -17, s->avctx->coded_width); |
| 374 | if (v->fcm == ILACE_FRAME) { |
| 375 | if (src_y & 1) |
| 376 | src_y = av_clip(src_y, -17, s->avctx->coded_height + 1); |
| 377 | else |
| 378 | src_y = av_clip(src_y, -18, s->avctx->coded_height); |
| 379 | } else { |
| 380 | src_y = av_clip(src_y, -18, s->avctx->coded_height + 1); |
| 381 | } |
| 382 | } |
| 383 | |
| 384 | srcY += src_y * s->linesize + src_x; |
| 385 | if (v->field_mode && v->ref_field_type[dir]) |
| 386 | srcY += s->current_picture_ptr->f->linesize[0]; |
| 387 | |
| 388 | if (fieldmv && !(src_y & 1)) |
| 389 | v_edge_pos--; |
| 390 | if (fieldmv && (src_y & 1) && src_y < 4) |
| 391 | src_y--; |
| 392 | if (v->rangeredfrm || use_ic |
| 393 | || s->h_edge_pos < 13 || v_edge_pos < 23 |
| 394 | || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx & 3) - 8 - s->mspel * 2 |
| 395 | || (unsigned)(src_y - (s->mspel << fieldmv)) > v_edge_pos - (my & 3) - ((8 + s->mspel * 2) << fieldmv)) { |
| 396 | srcY -= s->mspel * (1 + (s->linesize << fieldmv)); |
| 397 | /* check emulate edge stride and offset */ |
| 398 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, |
| 399 | s->linesize, s->linesize, |
| 400 | 9 + s->mspel * 2, (9 + s->mspel * 2) << fieldmv, |
| 401 | src_x - s->mspel, src_y - (s->mspel << fieldmv), |
| 402 | s->h_edge_pos, v_edge_pos); |
| 403 | srcY = s->edge_emu_buffer; |
| 404 | /* if we deal with range reduction we need to scale source blocks */ |
| 405 | if (v->rangeredfrm) { |
| 406 | int i, j; |
| 407 | uint8_t *src; |
| 408 | |
| 409 | src = srcY; |
| 410 | for (j = 0; j < 9 + s->mspel * 2; j++) { |
| 411 | for (i = 0; i < 9 + s->mspel * 2; i++) |
| 412 | src[i] = ((src[i] - 128) >> 1) + 128; |
| 413 | src += s->linesize << fieldmv; |
| 414 | } |
| 415 | } |
| 416 | /* if we deal with intensity compensation we need to scale source blocks */ |
| 417 | if (use_ic) { |
| 418 | int i, j; |
| 419 | uint8_t *src; |
| 420 | |
| 421 | src = srcY; |
| 422 | for (j = 0; j < 9 + s->mspel * 2; j++) { |
| 423 | int f = v->field_mode ? v->ref_field_type[dir] : (((j<<fieldmv)+src_y - (s->mspel << fieldmv)) & 1); |
| 424 | for (i = 0; i < 9 + s->mspel * 2; i++) |
| 425 | src[i] = luty[f][src[i]]; |
| 426 | src += s->linesize << fieldmv; |
| 427 | } |
| 428 | } |
| 429 | srcY += s->mspel * (1 + (s->linesize << fieldmv)); |
| 430 | } |
| 431 | |
| 432 | if (s->mspel) { |
| 433 | dxy = ((my & 3) << 2) | (mx & 3); |
| 434 | if (avg) |
| 435 | v->vc1dsp.avg_vc1_mspel_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize << fieldmv, v->rnd); |
| 436 | else |
| 437 | v->vc1dsp.put_vc1_mspel_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize << fieldmv, v->rnd); |
| 438 | } else { // hpel mc - always used for luma |
| 439 | dxy = (my & 2) | ((mx & 2) >> 1); |
| 440 | if (!v->rnd) |
| 441 | s->hdsp.put_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8); |
| 442 | else |
| 443 | s->hdsp.put_no_rnd_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8); |
| 444 | } |
| 445 | } |
| 446 | |
| 447 | static av_always_inline int get_chroma_mv(int *mvx, int *mvy, int *a, int flag, int *tx, int *ty) |
| 448 | { |
| 449 | int idx, i; |
| 450 | static const int count[16] = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4}; |
| 451 | |
| 452 | idx = ((a[3] != flag) << 3) |
| 453 | | ((a[2] != flag) << 2) |
| 454 | | ((a[1] != flag) << 1) |
| 455 | | (a[0] != flag); |
| 456 | if (!idx) { |
| 457 | *tx = median4(mvx[0], mvx[1], mvx[2], mvx[3]); |
| 458 | *ty = median4(mvy[0], mvy[1], mvy[2], mvy[3]); |
| 459 | return 4; |
| 460 | } else if (count[idx] == 1) { |
| 461 | switch (idx) { |
| 462 | case 0x1: |
| 463 | *tx = mid_pred(mvx[1], mvx[2], mvx[3]); |
| 464 | *ty = mid_pred(mvy[1], mvy[2], mvy[3]); |
| 465 | return 3; |
| 466 | case 0x2: |
| 467 | *tx = mid_pred(mvx[0], mvx[2], mvx[3]); |
| 468 | *ty = mid_pred(mvy[0], mvy[2], mvy[3]); |
| 469 | return 3; |
| 470 | case 0x4: |
| 471 | *tx = mid_pred(mvx[0], mvx[1], mvx[3]); |
| 472 | *ty = mid_pred(mvy[0], mvy[1], mvy[3]); |
| 473 | return 3; |
| 474 | case 0x8: |
| 475 | *tx = mid_pred(mvx[0], mvx[1], mvx[2]); |
| 476 | *ty = mid_pred(mvy[0], mvy[1], mvy[2]); |
| 477 | return 3; |
| 478 | } |
| 479 | } else if (count[idx] == 2) { |
| 480 | int t1 = 0, t2 = 0; |
| 481 | for (i = 0; i < 3; i++) |
| 482 | if (!a[i]) { |
| 483 | t1 = i; |
| 484 | break; |
| 485 | } |
| 486 | for (i = t1 + 1; i < 4; i++) |
| 487 | if (!a[i]) { |
| 488 | t2 = i; |
| 489 | break; |
| 490 | } |
| 491 | *tx = (mvx[t1] + mvx[t2]) / 2; |
| 492 | *ty = (mvy[t1] + mvy[t2]) / 2; |
| 493 | return 2; |
| 494 | } else { |
| 495 | return 0; |
| 496 | } |
| 497 | return -1; |
| 498 | } |
| 499 | |
| 500 | /** Do motion compensation for 4-MV macroblock - both chroma blocks |
| 501 | */ |
| 502 | void ff_vc1_mc_4mv_chroma(VC1Context *v, int dir) |
| 503 | { |
| 504 | MpegEncContext *s = &v->s; |
| 505 | H264ChromaContext *h264chroma = &v->h264chroma; |
| 506 | uint8_t *srcU, *srcV; |
| 507 | int uvmx, uvmy, uvsrc_x, uvsrc_y; |
| 508 | int k, tx = 0, ty = 0; |
| 509 | int mvx[4], mvy[4], intra[4], mv_f[4]; |
| 510 | int valid_count; |
| 511 | int chroma_ref_type = v->cur_field_type; |
| 512 | int v_edge_pos = s->v_edge_pos >> v->field_mode; |
| 513 | uint8_t (*lutuv)[256]; |
| 514 | int use_ic; |
| 515 | |
| 516 | if (!v->field_mode && !v->s.last_picture.f->data[0]) |
| 517 | return; |
| 518 | if (s->flags & CODEC_FLAG_GRAY) |
| 519 | return; |
| 520 | |
| 521 | for (k = 0; k < 4; k++) { |
| 522 | mvx[k] = s->mv[dir][k][0]; |
| 523 | mvy[k] = s->mv[dir][k][1]; |
| 524 | intra[k] = v->mb_type[0][s->block_index[k]]; |
| 525 | if (v->field_mode) |
| 526 | mv_f[k] = v->mv_f[dir][s->block_index[k] + v->blocks_off]; |
| 527 | } |
| 528 | |
| 529 | /* calculate chroma MV vector from four luma MVs */ |
| 530 | if (!v->field_mode || (v->field_mode && !v->numref)) { |
| 531 | valid_count = get_chroma_mv(mvx, mvy, intra, 0, &tx, &ty); |
| 532 | chroma_ref_type = v->reffield; |
| 533 | if (!valid_count) { |
| 534 | s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = 0; |
| 535 | s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = 0; |
| 536 | v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0; |
| 537 | return; //no need to do MC for intra blocks |
| 538 | } |
| 539 | } else { |
| 540 | int dominant = 0; |
| 541 | if (mv_f[0] + mv_f[1] + mv_f[2] + mv_f[3] > 2) |
| 542 | dominant = 1; |
| 543 | valid_count = get_chroma_mv(mvx, mvy, mv_f, dominant, &tx, &ty); |
| 544 | if (dominant) |
| 545 | chroma_ref_type = !v->cur_field_type; |
| 546 | } |
| 547 | if (v->field_mode && chroma_ref_type == 1 && v->cur_field_type == 1 && !v->s.last_picture.f->data[0]) |
| 548 | return; |
| 549 | s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx; |
| 550 | s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty; |
| 551 | uvmx = (tx + ((tx & 3) == 3)) >> 1; |
| 552 | uvmy = (ty + ((ty & 3) == 3)) >> 1; |
| 553 | |
| 554 | v->luma_mv[s->mb_x][0] = uvmx; |
| 555 | v->luma_mv[s->mb_x][1] = uvmy; |
| 556 | |
| 557 | if (v->fastuvmc) { |
| 558 | uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1)); |
| 559 | uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1)); |
| 560 | } |
| 561 | // Field conversion bias |
| 562 | if (v->cur_field_type != chroma_ref_type) |
| 563 | uvmy += 2 - 4 * chroma_ref_type; |
| 564 | |
| 565 | uvsrc_x = s->mb_x * 8 + (uvmx >> 2); |
| 566 | uvsrc_y = s->mb_y * 8 + (uvmy >> 2); |
| 567 | |
| 568 | if (v->profile != PROFILE_ADVANCED) { |
| 569 | uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8); |
| 570 | uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8); |
| 571 | } else { |
| 572 | uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1); |
| 573 | uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1); |
| 574 | } |
| 575 | |
| 576 | if (!dir) { |
| 577 | if (v->field_mode && (v->cur_field_type != chroma_ref_type) && v->second_field) { |
| 578 | srcU = s->current_picture.f->data[1]; |
| 579 | srcV = s->current_picture.f->data[2]; |
| 580 | lutuv = v->curr_lutuv; |
| 581 | use_ic = *v->curr_use_ic; |
| 582 | } else { |
| 583 | srcU = s->last_picture.f->data[1]; |
| 584 | srcV = s->last_picture.f->data[2]; |
| 585 | lutuv = v->last_lutuv; |
| 586 | use_ic = v->last_use_ic; |
| 587 | } |
| 588 | } else { |
| 589 | srcU = s->next_picture.f->data[1]; |
| 590 | srcV = s->next_picture.f->data[2]; |
| 591 | lutuv = v->next_lutuv; |
| 592 | use_ic = v->next_use_ic; |
| 593 | } |
| 594 | |
| 595 | if (!srcU) { |
| 596 | av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n"); |
| 597 | return; |
| 598 | } |
| 599 | |
| 600 | srcU += uvsrc_y * s->uvlinesize + uvsrc_x; |
| 601 | srcV += uvsrc_y * s->uvlinesize + uvsrc_x; |
| 602 | |
| 603 | if (v->field_mode) { |
| 604 | if (chroma_ref_type) { |
| 605 | srcU += s->current_picture_ptr->f->linesize[1]; |
| 606 | srcV += s->current_picture_ptr->f->linesize[2]; |
| 607 | } |
| 608 | } |
| 609 | |
| 610 | if (v->rangeredfrm || use_ic |
| 611 | || s->h_edge_pos < 18 || v_edge_pos < 18 |
| 612 | || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 9 |
| 613 | || (unsigned)uvsrc_y > (v_edge_pos >> 1) - 9) { |
| 614 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcU, |
| 615 | s->uvlinesize, s->uvlinesize, |
| 616 | 8 + 1, 8 + 1, uvsrc_x, uvsrc_y, |
| 617 | s->h_edge_pos >> 1, v_edge_pos >> 1); |
| 618 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV, |
| 619 | s->uvlinesize, s->uvlinesize, |
| 620 | 8 + 1, 8 + 1, uvsrc_x, uvsrc_y, |
| 621 | s->h_edge_pos >> 1, v_edge_pos >> 1); |
| 622 | srcU = s->edge_emu_buffer; |
| 623 | srcV = s->edge_emu_buffer + 16; |
| 624 | |
| 625 | /* if we deal with range reduction we need to scale source blocks */ |
| 626 | if (v->rangeredfrm) { |
| 627 | int i, j; |
| 628 | uint8_t *src, *src2; |
| 629 | |
| 630 | src = srcU; |
| 631 | src2 = srcV; |
| 632 | for (j = 0; j < 9; j++) { |
| 633 | for (i = 0; i < 9; i++) { |
| 634 | src[i] = ((src[i] - 128) >> 1) + 128; |
| 635 | src2[i] = ((src2[i] - 128) >> 1) + 128; |
| 636 | } |
| 637 | src += s->uvlinesize; |
| 638 | src2 += s->uvlinesize; |
| 639 | } |
| 640 | } |
| 641 | /* if we deal with intensity compensation we need to scale source blocks */ |
| 642 | if (use_ic) { |
| 643 | int i, j; |
| 644 | uint8_t *src, *src2; |
| 645 | |
| 646 | src = srcU; |
| 647 | src2 = srcV; |
| 648 | for (j = 0; j < 9; j++) { |
| 649 | int f = v->field_mode ? chroma_ref_type : ((j + uvsrc_y) & 1); |
| 650 | for (i = 0; i < 9; i++) { |
| 651 | src[i] = lutuv[f][src[i]]; |
| 652 | src2[i] = lutuv[f][src2[i]]; |
| 653 | } |
| 654 | src += s->uvlinesize; |
| 655 | src2 += s->uvlinesize; |
| 656 | } |
| 657 | } |
| 658 | } |
| 659 | |
| 660 | /* Chroma MC always uses qpel bilinear */ |
| 661 | uvmx = (uvmx & 3) << 1; |
| 662 | uvmy = (uvmy & 3) << 1; |
| 663 | if (!v->rnd) { |
| 664 | h264chroma->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy); |
| 665 | h264chroma->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy); |
| 666 | } else { |
| 667 | v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy); |
| 668 | v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy); |
| 669 | } |
| 670 | } |
| 671 | |
| 672 | /** Do motion compensation for 4-MV interlaced frame chroma macroblock (both U and V) |
| 673 | */ |
| 674 | void ff_vc1_mc_4mv_chroma4(VC1Context *v, int dir, int dir2, int avg) |
| 675 | { |
| 676 | MpegEncContext *s = &v->s; |
| 677 | H264ChromaContext *h264chroma = &v->h264chroma; |
| 678 | uint8_t *srcU, *srcV; |
| 679 | int uvsrc_x, uvsrc_y; |
| 680 | int uvmx_field[4], uvmy_field[4]; |
| 681 | int i, off, tx, ty; |
| 682 | int fieldmv = v->blk_mv_type[s->block_index[0]]; |
| 683 | static const int s_rndtblfield[16] = { 0, 0, 1, 2, 4, 4, 5, 6, 2, 2, 3, 8, 6, 6, 7, 12 }; |
| 684 | int v_dist = fieldmv ? 1 : 4; // vertical offset for lower sub-blocks |
| 685 | int v_edge_pos = s->v_edge_pos >> 1; |
| 686 | int use_ic; |
| 687 | uint8_t (*lutuv)[256]; |
| 688 | |
| 689 | if (s->flags & CODEC_FLAG_GRAY) |
| 690 | return; |
| 691 | |
| 692 | for (i = 0; i < 4; i++) { |
| 693 | int d = i < 2 ? dir: dir2; |
| 694 | tx = s->mv[d][i][0]; |
| 695 | uvmx_field[i] = (tx + ((tx & 3) == 3)) >> 1; |
| 696 | ty = s->mv[d][i][1]; |
| 697 | if (fieldmv) |
| 698 | uvmy_field[i] = (ty >> 4) * 8 + s_rndtblfield[ty & 0xF]; |
| 699 | else |
| 700 | uvmy_field[i] = (ty + ((ty & 3) == 3)) >> 1; |
| 701 | } |
| 702 | |
| 703 | for (i = 0; i < 4; i++) { |
| 704 | off = (i & 1) * 4 + ((i & 2) ? v_dist * s->uvlinesize : 0); |
| 705 | uvsrc_x = s->mb_x * 8 + (i & 1) * 4 + (uvmx_field[i] >> 2); |
| 706 | uvsrc_y = s->mb_y * 8 + ((i & 2) ? v_dist : 0) + (uvmy_field[i] >> 2); |
| 707 | // FIXME: implement proper pull-back (see vc1cropmv.c, vc1CROPMV_ChromaPullBack()) |
| 708 | uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1); |
| 709 | uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1); |
| 710 | if (i < 2 ? dir : dir2) { |
| 711 | srcU = s->next_picture.f->data[1]; |
| 712 | srcV = s->next_picture.f->data[2]; |
| 713 | lutuv = v->next_lutuv; |
| 714 | use_ic = v->next_use_ic; |
| 715 | } else { |
| 716 | srcU = s->last_picture.f->data[1]; |
| 717 | srcV = s->last_picture.f->data[2]; |
| 718 | lutuv = v->last_lutuv; |
| 719 | use_ic = v->last_use_ic; |
| 720 | } |
| 721 | if (!srcU) |
| 722 | return; |
| 723 | srcU += uvsrc_y * s->uvlinesize + uvsrc_x; |
| 724 | srcV += uvsrc_y * s->uvlinesize + uvsrc_x; |
| 725 | uvmx_field[i] = (uvmx_field[i] & 3) << 1; |
| 726 | uvmy_field[i] = (uvmy_field[i] & 3) << 1; |
| 727 | |
| 728 | if (fieldmv && !(uvsrc_y & 1)) |
| 729 | v_edge_pos = (s->v_edge_pos >> 1) - 1; |
| 730 | |
| 731 | if (fieldmv && (uvsrc_y & 1) && uvsrc_y < 2) |
| 732 | uvsrc_y--; |
| 733 | if (use_ic |
| 734 | || s->h_edge_pos < 10 || v_edge_pos < (5 << fieldmv) |
| 735 | || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 5 |
| 736 | || (unsigned)uvsrc_y > v_edge_pos - (5 << fieldmv)) { |
| 737 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcU, |
| 738 | s->uvlinesize, s->uvlinesize, |
| 739 | 5, (5 << fieldmv), uvsrc_x, uvsrc_y, |
| 740 | s->h_edge_pos >> 1, v_edge_pos); |
| 741 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV, |
| 742 | s->uvlinesize, s->uvlinesize, |
| 743 | 5, (5 << fieldmv), uvsrc_x, uvsrc_y, |
| 744 | s->h_edge_pos >> 1, v_edge_pos); |
| 745 | srcU = s->edge_emu_buffer; |
| 746 | srcV = s->edge_emu_buffer + 16; |
| 747 | |
| 748 | /* if we deal with intensity compensation we need to scale source blocks */ |
| 749 | if (use_ic) { |
| 750 | int i, j; |
| 751 | uint8_t *src, *src2; |
| 752 | |
| 753 | src = srcU; |
| 754 | src2 = srcV; |
| 755 | for (j = 0; j < 5; j++) { |
| 756 | int f = (uvsrc_y + (j << fieldmv)) & 1; |
| 757 | for (i = 0; i < 5; i++) { |
| 758 | src[i] = lutuv[f][src[i]]; |
| 759 | src2[i] = lutuv[f][src2[i]]; |
| 760 | } |
| 761 | src += s->uvlinesize << fieldmv; |
| 762 | src2 += s->uvlinesize << fieldmv; |
| 763 | } |
| 764 | } |
| 765 | } |
| 766 | if (avg) { |
| 767 | if (!v->rnd) { |
| 768 | h264chroma->avg_h264_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]); |
| 769 | h264chroma->avg_h264_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]); |
| 770 | } else { |
| 771 | v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]); |
| 772 | v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]); |
| 773 | } |
| 774 | } else { |
| 775 | if (!v->rnd) { |
| 776 | h264chroma->put_h264_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]); |
| 777 | h264chroma->put_h264_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]); |
| 778 | } else { |
| 779 | v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]); |
| 780 | v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]); |
| 781 | } |
| 782 | } |
| 783 | } |
| 784 | } |
| 785 | |
| 786 | /** Motion compensation for direct or interpolated blocks in B-frames |
| 787 | */ |
| 788 | void ff_vc1_interp_mc(VC1Context *v) |
| 789 | { |
| 790 | MpegEncContext *s = &v->s; |
| 791 | H264ChromaContext *h264chroma = &v->h264chroma; |
| 792 | uint8_t *srcY, *srcU, *srcV; |
| 793 | int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y; |
| 794 | int off, off_uv; |
| 795 | int v_edge_pos = s->v_edge_pos >> v->field_mode; |
| 796 | int use_ic = v->next_use_ic; |
| 797 | |
| 798 | if (!v->field_mode && !v->s.next_picture.f->data[0]) |
| 799 | return; |
| 800 | |
| 801 | mx = s->mv[1][0][0]; |
| 802 | my = s->mv[1][0][1]; |
| 803 | uvmx = (mx + ((mx & 3) == 3)) >> 1; |
| 804 | uvmy = (my + ((my & 3) == 3)) >> 1; |
| 805 | if (v->field_mode && v->cur_field_type != v->ref_field_type[1]) { |
| 806 | my = my - 2 + 4 * v->cur_field_type; |
| 807 | uvmy = uvmy - 2 + 4 * v->cur_field_type; |
| 808 | } |
| 809 | if (v->fastuvmc) { |
| 810 | uvmx = uvmx + ((uvmx < 0) ? -(uvmx & 1) : (uvmx & 1)); |
| 811 | uvmy = uvmy + ((uvmy < 0) ? -(uvmy & 1) : (uvmy & 1)); |
| 812 | } |
| 813 | srcY = s->next_picture.f->data[0]; |
| 814 | srcU = s->next_picture.f->data[1]; |
| 815 | srcV = s->next_picture.f->data[2]; |
| 816 | |
| 817 | src_x = s->mb_x * 16 + (mx >> 2); |
| 818 | src_y = s->mb_y * 16 + (my >> 2); |
| 819 | uvsrc_x = s->mb_x * 8 + (uvmx >> 2); |
| 820 | uvsrc_y = s->mb_y * 8 + (uvmy >> 2); |
| 821 | |
| 822 | if (v->profile != PROFILE_ADVANCED) { |
| 823 | src_x = av_clip( src_x, -16, s->mb_width * 16); |
| 824 | src_y = av_clip( src_y, -16, s->mb_height * 16); |
| 825 | uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8); |
| 826 | uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8); |
| 827 | } else { |
| 828 | src_x = av_clip( src_x, -17, s->avctx->coded_width); |
| 829 | src_y = av_clip( src_y, -18, s->avctx->coded_height + 1); |
| 830 | uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1); |
| 831 | uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1); |
| 832 | } |
| 833 | |
| 834 | srcY += src_y * s->linesize + src_x; |
| 835 | srcU += uvsrc_y * s->uvlinesize + uvsrc_x; |
| 836 | srcV += uvsrc_y * s->uvlinesize + uvsrc_x; |
| 837 | |
| 838 | if (v->field_mode && v->ref_field_type[1]) { |
| 839 | srcY += s->current_picture_ptr->f->linesize[0]; |
| 840 | srcU += s->current_picture_ptr->f->linesize[1]; |
| 841 | srcV += s->current_picture_ptr->f->linesize[2]; |
| 842 | } |
| 843 | |
| 844 | /* for grayscale we should not try to read from unknown area */ |
| 845 | if (s->flags & CODEC_FLAG_GRAY) { |
| 846 | srcU = s->edge_emu_buffer + 18 * s->linesize; |
| 847 | srcV = s->edge_emu_buffer + 18 * s->linesize; |
| 848 | } |
| 849 | |
| 850 | if (v->rangeredfrm || s->h_edge_pos < 22 || v_edge_pos < 22 || use_ic |
| 851 | || (unsigned)(src_x - 1) > s->h_edge_pos - (mx & 3) - 16 - 3 |
| 852 | || (unsigned)(src_y - 1) > v_edge_pos - (my & 3) - 16 - 3) { |
| 853 | uint8_t *ubuf = s->edge_emu_buffer + 19 * s->linesize; |
| 854 | uint8_t *vbuf = ubuf + 9 * s->uvlinesize; |
| 855 | |
| 856 | srcY -= s->mspel * (1 + s->linesize); |
| 857 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, |
| 858 | s->linesize, s->linesize, |
| 859 | 17 + s->mspel * 2, 17 + s->mspel * 2, |
| 860 | src_x - s->mspel, src_y - s->mspel, |
| 861 | s->h_edge_pos, v_edge_pos); |
| 862 | srcY = s->edge_emu_buffer; |
| 863 | s->vdsp.emulated_edge_mc(ubuf, srcU, |
| 864 | s->uvlinesize, s->uvlinesize, |
| 865 | 8 + 1, 8 + 1, |
| 866 | uvsrc_x, uvsrc_y, |
| 867 | s->h_edge_pos >> 1, v_edge_pos >> 1); |
| 868 | s->vdsp.emulated_edge_mc(vbuf, srcV, |
| 869 | s->uvlinesize, s->uvlinesize, |
| 870 | 8 + 1, 8 + 1, |
| 871 | uvsrc_x, uvsrc_y, |
| 872 | s->h_edge_pos >> 1, v_edge_pos >> 1); |
| 873 | srcU = ubuf; |
| 874 | srcV = vbuf; |
| 875 | /* if we deal with range reduction we need to scale source blocks */ |
| 876 | if (v->rangeredfrm) { |
| 877 | int i, j; |
| 878 | uint8_t *src, *src2; |
| 879 | |
| 880 | src = srcY; |
| 881 | for (j = 0; j < 17 + s->mspel * 2; j++) { |
| 882 | for (i = 0; i < 17 + s->mspel * 2; i++) |
| 883 | src[i] = ((src[i] - 128) >> 1) + 128; |
| 884 | src += s->linesize; |
| 885 | } |
| 886 | src = srcU; |
| 887 | src2 = srcV; |
| 888 | for (j = 0; j < 9; j++) { |
| 889 | for (i = 0; i < 9; i++) { |
| 890 | src[i] = ((src[i] - 128) >> 1) + 128; |
| 891 | src2[i] = ((src2[i] - 128) >> 1) + 128; |
| 892 | } |
| 893 | src += s->uvlinesize; |
| 894 | src2 += s->uvlinesize; |
| 895 | } |
| 896 | } |
| 897 | |
| 898 | if (use_ic) { |
| 899 | uint8_t (*luty )[256] = v->next_luty; |
| 900 | uint8_t (*lutuv)[256] = v->next_lutuv; |
| 901 | int i, j; |
| 902 | uint8_t *src, *src2; |
| 903 | |
| 904 | src = srcY; |
| 905 | for (j = 0; j < 17 + s->mspel * 2; j++) { |
| 906 | int f = v->field_mode ? v->ref_field_type[1] : ((j+src_y - s->mspel) & 1); |
| 907 | for (i = 0; i < 17 + s->mspel * 2; i++) |
| 908 | src[i] = luty[f][src[i]]; |
| 909 | src += s->linesize; |
| 910 | } |
| 911 | src = srcU; |
| 912 | src2 = srcV; |
| 913 | for (j = 0; j < 9; j++) { |
| 914 | int f = v->field_mode ? v->ref_field_type[1] : ((j+uvsrc_y) & 1); |
| 915 | for (i = 0; i < 9; i++) { |
| 916 | src[i] = lutuv[f][src[i]]; |
| 917 | src2[i] = lutuv[f][src2[i]]; |
| 918 | } |
| 919 | src += s->uvlinesize; |
| 920 | src2 += s->uvlinesize; |
| 921 | } |
| 922 | } |
| 923 | srcY += s->mspel * (1 + s->linesize); |
| 924 | } |
| 925 | |
| 926 | off = 0; |
| 927 | off_uv = 0; |
| 928 | |
| 929 | if (s->mspel) { |
| 930 | dxy = ((my & 3) << 2) | (mx & 3); |
| 931 | v->vc1dsp.avg_vc1_mspel_pixels_tab[0][dxy](s->dest[0] + off , srcY , s->linesize, v->rnd); |
| 932 | } else { // hpel mc |
| 933 | dxy = (my & 2) | ((mx & 2) >> 1); |
| 934 | |
| 935 | if (!v->rnd) |
| 936 | s->hdsp.avg_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16); |
| 937 | else |
| 938 | s->hdsp.avg_no_rnd_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize, 16); |
| 939 | } |
| 940 | |
| 941 | if (s->flags & CODEC_FLAG_GRAY) return; |
| 942 | /* Chroma MC always uses qpel blilinear */ |
| 943 | uvmx = (uvmx & 3) << 1; |
| 944 | uvmy = (uvmy & 3) << 1; |
| 945 | if (!v->rnd) { |
| 946 | h264chroma->avg_h264_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy); |
| 947 | h264chroma->avg_h264_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy); |
| 948 | } else { |
| 949 | v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy); |
| 950 | v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy); |
| 951 | } |
| 952 | } |