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
7 * This file is part of Libav.
9 * Libav 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.
14 * Libav 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.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with Libav; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 * VC-1 and WMV3 block decoding routines
30 #include "h264chroma.h"
32 #include "mpegvideo.h"
35 /** Do motion compensation over 1 macroblock
36 * Mostly adapted hpel_motion and qpel_motion from mpegvideo.c
38 void ff_vc1_mc_1mv(VC1Context *v, int dir)
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;
46 uint8_t (*luty)[256], (*lutuv)[256];
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])
54 mx = s->mv[dir][0][0];
55 my = s->mv[dir][0][1];
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;
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;
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;
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));
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];
87 lutuv = v->curr_lutuv;
88 use_ic = v->curr_use_ic;
90 srcY = s->last_picture.f->data[0];
91 srcU = s->last_picture.f->data[1];
92 srcV = s->last_picture.f->data[2];
94 lutuv = v->last_lutuv;
95 use_ic = v->last_use_ic;
98 srcY = s->next_picture.f->data[0];
99 srcU = s->next_picture.f->data[1];
100 srcV = s->next_picture.f->data[2];
102 lutuv = v->next_lutuv;
103 use_ic = v->next_use_ic;
106 if (!srcY || !srcU) {
107 av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
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);
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);
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);
128 srcY += src_y * s->linesize + src_x;
129 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
130 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
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];
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;
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 *uvbuf = s->edge_emu_buffer + 19 * s->linesize;
150 srcY -= s->mspel * (1 + s->linesize);
151 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY,
152 s->linesize, s->linesize,
153 17 + s->mspel * 2, 17 + s->mspel * 2,
154 src_x - s->mspel, src_y - s->mspel,
155 s->h_edge_pos, v_edge_pos);
156 srcY = s->edge_emu_buffer;
157 s->vdsp.emulated_edge_mc(uvbuf, srcU,
158 s->uvlinesize, s->uvlinesize,
160 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
161 s->vdsp.emulated_edge_mc(uvbuf + 16, srcV,
162 s->uvlinesize, s->uvlinesize,
164 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
167 /* if we deal with range reduction we need to scale source blocks */
168 if (v->rangeredfrm) {
173 for (j = 0; j < 17 + s->mspel * 2; j++) {
174 for (i = 0; i < 17 + s->mspel * 2; i++)
175 src[i] = ((src[i] - 128) >> 1) + 128;
180 for (j = 0; j < 9; j++) {
181 for (i = 0; i < 9; i++) {
182 src[i] = ((src[i] - 128) >> 1) + 128;
183 src2[i] = ((src2[i] - 128) >> 1) + 128;
185 src += s->uvlinesize;
186 src2 += s->uvlinesize;
189 /* if we deal with intensity compensation we need to scale source blocks */
195 for (j = 0; j < 17 + s->mspel * 2; j++) {
196 int f = v->field_mode ? v->ref_field_type[dir] : ((j + src_y - s->mspel) & 1) ;
197 for (i = 0; i < 17 + s->mspel * 2; i++)
198 src[i] = luty[f][src[i]];
203 for (j = 0; j < 9; j++) {
204 int f = v->field_mode ? v->ref_field_type[dir] : ((j + uvsrc_y) & 1);
205 for (i = 0; i < 9; i++) {
206 src[i] = lutuv[f][src[i]];
207 src2[i] = lutuv[f][src2[i]];
209 src += s->uvlinesize;
210 src2 += s->uvlinesize;
213 srcY += s->mspel * (1 + s->linesize);
217 dxy = ((my & 3) << 2) | (mx & 3);
218 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] , srcY , s->linesize, v->rnd);
219 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8, srcY + 8, s->linesize, v->rnd);
220 srcY += s->linesize * 8;
221 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize , srcY , s->linesize, v->rnd);
222 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
223 } else { // hpel mc - always used for luma
224 dxy = (my & 2) | ((mx & 2) >> 1);
226 s->hdsp.put_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
228 s->hdsp.put_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
231 if (s->flags & CODEC_FLAG_GRAY) return;
232 /* Chroma MC always uses qpel bilinear */
233 uvmx = (uvmx & 3) << 1;
234 uvmy = (uvmy & 3) << 1;
236 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
237 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
239 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
240 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
244 static inline int median4(int a, int b, int c, int d)
247 if (c < d) return (FFMIN(b, d) + FFMAX(a, c)) / 2;
248 else return (FFMIN(b, c) + FFMAX(a, d)) / 2;
250 if (c < d) return (FFMIN(a, d) + FFMAX(b, c)) / 2;
251 else return (FFMIN(a, c) + FFMAX(b, d)) / 2;
255 /** Do motion compensation for 4-MV macroblock - luminance block
257 void ff_vc1_mc_4mv_luma(VC1Context *v, int n, int dir, int avg)
259 MpegEncContext *s = &v->s;
261 int dxy, mx, my, src_x, src_y;
263 int fieldmv = (v->fcm == ILACE_FRAME) ? v->blk_mv_type[s->block_index[n]] : 0;
264 int v_edge_pos = s->v_edge_pos >> v->field_mode;
265 uint8_t (*luty)[256];
268 if ((!v->field_mode ||
269 (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
270 !v->s.last_picture.f->data[0])
273 mx = s->mv[dir][n][0];
274 my = s->mv[dir][n][1];
277 if (v->field_mode && (v->cur_field_type != v->ref_field_type[dir]) && v->second_field) {
278 srcY = s->current_picture.f->data[0];
280 use_ic = v->curr_use_ic;
282 srcY = s->last_picture.f->data[0];
284 use_ic = v->last_use_ic;
287 srcY = s->next_picture.f->data[0];
289 use_ic = v->next_use_ic;
293 av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
298 if (v->cur_field_type != v->ref_field_type[dir])
299 my = my - 2 + 4 * v->cur_field_type;
302 if (s->pict_type == AV_PICTURE_TYPE_P && n == 3 && v->field_mode) {
303 int same_count = 0, opp_count = 0, k;
304 int chosen_mv[2][4][2], f;
306 for (k = 0; k < 4; k++) {
307 f = v->mv_f[0][s->block_index[k] + v->blocks_off];
308 chosen_mv[f][f ? opp_count : same_count][0] = s->mv[0][k][0];
309 chosen_mv[f][f ? opp_count : same_count][1] = s->mv[0][k][1];
313 f = opp_count > same_count;
314 switch (f ? opp_count : same_count) {
316 tx = median4(chosen_mv[f][0][0], chosen_mv[f][1][0],
317 chosen_mv[f][2][0], chosen_mv[f][3][0]);
318 ty = median4(chosen_mv[f][0][1], chosen_mv[f][1][1],
319 chosen_mv[f][2][1], chosen_mv[f][3][1]);
322 tx = mid_pred(chosen_mv[f][0][0], chosen_mv[f][1][0], chosen_mv[f][2][0]);
323 ty = mid_pred(chosen_mv[f][0][1], chosen_mv[f][1][1], chosen_mv[f][2][1]);
326 tx = (chosen_mv[f][0][0] + chosen_mv[f][1][0]) / 2;
327 ty = (chosen_mv[f][0][1] + chosen_mv[f][1][1]) / 2;
330 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx;
331 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty;
332 for (k = 0; k < 4; k++)
333 v->mv_f[1][s->block_index[k] + v->blocks_off] = f;
336 if (v->fcm == ILACE_FRAME) { // not sure if needed for other types of picture
338 int width = s->avctx->coded_width;
339 int height = s->avctx->coded_height >> 1;
340 if (s->pict_type == AV_PICTURE_TYPE_P) {
341 s->current_picture.motion_val[1][s->block_index[n] + v->blocks_off][0] = mx;
342 s->current_picture.motion_val[1][s->block_index[n] + v->blocks_off][1] = my;
344 qx = (s->mb_x * 16) + (mx >> 2);
345 qy = (s->mb_y * 8) + (my >> 3);
350 mx -= 4 * (qx - width);
353 else if (qy > height + 1)
354 my -= 8 * (qy - height - 1);
357 if ((v->fcm == ILACE_FRAME) && fieldmv)
358 off = ((n > 1) ? s->linesize : 0) + (n & 1) * 8;
360 off = s->linesize * 4 * (n & 2) + (n & 1) * 8;
362 src_x = s->mb_x * 16 + (n & 1) * 8 + (mx >> 2);
364 src_y = s->mb_y * 16 + (n & 2) * 4 + (my >> 2);
366 src_y = s->mb_y * 16 + ((n > 1) ? 1 : 0) + (my >> 2);
368 if (v->profile != PROFILE_ADVANCED) {
369 src_x = av_clip(src_x, -16, s->mb_width * 16);
370 src_y = av_clip(src_y, -16, s->mb_height * 16);
372 src_x = av_clip(src_x, -17, s->avctx->coded_width);
373 if (v->fcm == ILACE_FRAME) {
375 src_y = av_clip(src_y, -17, s->avctx->coded_height + 1);
377 src_y = av_clip(src_y, -18, s->avctx->coded_height);
379 src_y = av_clip(src_y, -18, s->avctx->coded_height + 1);
383 srcY += src_y * s->linesize + src_x;
384 if (v->field_mode && v->ref_field_type[dir])
385 srcY += s->current_picture_ptr->f->linesize[0];
387 if (fieldmv && !(src_y & 1))
389 if (fieldmv && (src_y & 1) && src_y < 4)
391 if (v->rangeredfrm || use_ic
392 || s->h_edge_pos < 13 || v_edge_pos < 23
393 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx & 3) - 8 - s->mspel * 2
394 || (unsigned)(src_y - (s->mspel << fieldmv)) > v_edge_pos - (my & 3) - ((8 + s->mspel * 2) << fieldmv)) {
395 srcY -= s->mspel * (1 + (s->linesize << fieldmv));
396 /* check emulate edge stride and offset */
397 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY,
398 s->linesize, s->linesize,
399 9 + s->mspel * 2, (9 + s->mspel * 2) << fieldmv,
400 src_x - s->mspel, src_y - (s->mspel << fieldmv),
401 s->h_edge_pos, v_edge_pos);
402 srcY = s->edge_emu_buffer;
403 /* if we deal with range reduction we need to scale source blocks */
404 if (v->rangeredfrm) {
409 for (j = 0; j < 9 + s->mspel * 2; j++) {
410 for (i = 0; i < 9 + s->mspel * 2; i++)
411 src[i] = ((src[i] - 128) >> 1) + 128;
412 src += s->linesize << fieldmv;
415 /* if we deal with intensity compensation we need to scale source blocks */
421 for (j = 0; j < 9 + s->mspel * 2; j++) {
422 int f = v->field_mode ? v->ref_field_type[dir] : (((j<<fieldmv)+src_y - (s->mspel << fieldmv)) & 1);
423 for (i = 0; i < 9 + s->mspel * 2; i++)
424 src[i] = luty[f][src[i]];
425 src += s->linesize << fieldmv;
428 srcY += s->mspel * (1 + (s->linesize << fieldmv));
432 dxy = ((my & 3) << 2) | (mx & 3);
434 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize << fieldmv, v->rnd);
436 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize << fieldmv, v->rnd);
437 } else { // hpel mc - always used for luma
438 dxy = (my & 2) | ((mx & 2) >> 1);
440 s->hdsp.put_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
442 s->hdsp.put_no_rnd_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
446 static av_always_inline int get_chroma_mv(int *mvx, int *mvy, int *a, int flag, int *tx, int *ty)
449 static const int count[16] = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4};
451 idx = ((a[3] != flag) << 3)
452 | ((a[2] != flag) << 2)
453 | ((a[1] != flag) << 1)
456 *tx = median4(mvx[0], mvx[1], mvx[2], mvx[3]);
457 *ty = median4(mvy[0], mvy[1], mvy[2], mvy[3]);
459 } else if (count[idx] == 1) {
462 *tx = mid_pred(mvx[1], mvx[2], mvx[3]);
463 *ty = mid_pred(mvy[1], mvy[2], mvy[3]);
466 *tx = mid_pred(mvx[0], mvx[2], mvx[3]);
467 *ty = mid_pred(mvy[0], mvy[2], mvy[3]);
470 *tx = mid_pred(mvx[0], mvx[1], mvx[3]);
471 *ty = mid_pred(mvy[0], mvy[1], mvy[3]);
474 *tx = mid_pred(mvx[0], mvx[1], mvx[2]);
475 *ty = mid_pred(mvy[0], mvy[1], mvy[2]);
478 } else if (count[idx] == 2) {
480 for (i = 0; i < 3; i++)
485 for (i = t1 + 1; i < 4; i++)
490 *tx = (mvx[t1] + mvx[t2]) / 2;
491 *ty = (mvy[t1] + mvy[t2]) / 2;
499 /** Do motion compensation for 4-MV macroblock - both chroma blocks
501 void ff_vc1_mc_4mv_chroma(VC1Context *v, int dir)
503 MpegEncContext *s = &v->s;
504 H264ChromaContext *h264chroma = &v->h264chroma;
505 uint8_t *srcU, *srcV;
506 int uvmx, uvmy, uvsrc_x, uvsrc_y;
507 int k, tx = 0, ty = 0;
508 int mvx[4], mvy[4], intra[4], mv_f[4];
510 int chroma_ref_type = v->cur_field_type;
511 int v_edge_pos = s->v_edge_pos >> v->field_mode;
512 uint8_t (*lutuv)[256];
515 if (!v->field_mode && !v->s.last_picture.f->data[0])
517 if (s->flags & CODEC_FLAG_GRAY)
520 for (k = 0; k < 4; k++) {
521 mvx[k] = s->mv[dir][k][0];
522 mvy[k] = s->mv[dir][k][1];
523 intra[k] = v->mb_type[0][s->block_index[k]];
525 mv_f[k] = v->mv_f[dir][s->block_index[k] + v->blocks_off];
528 /* calculate chroma MV vector from four luma MVs */
529 if (!v->field_mode || (v->field_mode && !v->numref)) {
530 valid_count = get_chroma_mv(mvx, mvy, intra, 0, &tx, &ty);
531 chroma_ref_type = v->reffield;
533 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = 0;
534 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = 0;
535 v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
536 return; //no need to do MC for intra blocks
540 if (mv_f[0] + mv_f[1] + mv_f[2] + mv_f[3] > 2)
542 valid_count = get_chroma_mv(mvx, mvy, mv_f, dominant, &tx, &ty);
544 chroma_ref_type = !v->cur_field_type;
546 if (v->field_mode && chroma_ref_type == 1 && v->cur_field_type == 1 && !v->s.last_picture.f->data[0])
548 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx;
549 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty;
550 uvmx = (tx + ((tx & 3) == 3)) >> 1;
551 uvmy = (ty + ((ty & 3) == 3)) >> 1;
553 v->luma_mv[s->mb_x][0] = uvmx;
554 v->luma_mv[s->mb_x][1] = uvmy;
557 uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
558 uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
560 // Field conversion bias
561 if (v->cur_field_type != chroma_ref_type)
562 uvmy += 2 - 4 * chroma_ref_type;
564 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
565 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
567 if (v->profile != PROFILE_ADVANCED) {
568 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
569 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
571 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
572 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
576 if (v->field_mode && (v->cur_field_type != chroma_ref_type) && v->second_field) {
577 srcU = s->current_picture.f->data[1];
578 srcV = s->current_picture.f->data[2];
579 lutuv = v->curr_lutuv;
580 use_ic = v->curr_use_ic;
582 srcU = s->last_picture.f->data[1];
583 srcV = s->last_picture.f->data[2];
584 lutuv = v->last_lutuv;
585 use_ic = v->last_use_ic;
588 srcU = s->next_picture.f->data[1];
589 srcV = s->next_picture.f->data[2];
590 lutuv = v->next_lutuv;
591 use_ic = v->next_use_ic;
595 av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
599 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
600 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
603 if (chroma_ref_type) {
604 srcU += s->current_picture_ptr->f->linesize[1];
605 srcV += s->current_picture_ptr->f->linesize[2];
609 if (v->rangeredfrm || use_ic
610 || s->h_edge_pos < 18 || v_edge_pos < 18
611 || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 9
612 || (unsigned)uvsrc_y > (v_edge_pos >> 1) - 9) {
613 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcU,
614 s->uvlinesize, s->uvlinesize,
615 8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
616 s->h_edge_pos >> 1, v_edge_pos >> 1);
617 s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV,
618 s->uvlinesize, s->uvlinesize,
619 8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
620 s->h_edge_pos >> 1, v_edge_pos >> 1);
621 srcU = s->edge_emu_buffer;
622 srcV = s->edge_emu_buffer + 16;
624 /* if we deal with range reduction we need to scale source blocks */
625 if (v->rangeredfrm) {
631 for (j = 0; j < 9; j++) {
632 for (i = 0; i < 9; i++) {
633 src[i] = ((src[i] - 128) >> 1) + 128;
634 src2[i] = ((src2[i] - 128) >> 1) + 128;
636 src += s->uvlinesize;
637 src2 += s->uvlinesize;
640 /* if we deal with intensity compensation we need to scale source blocks */
647 for (j = 0; j < 9; j++) {
648 int f = v->field_mode ? chroma_ref_type : ((j + uvsrc_y) & 1);
649 for (i = 0; i < 9; i++) {
650 src[i] = lutuv[f][src[i]];
651 src2[i] = lutuv[f][src2[i]];
653 src += s->uvlinesize;
654 src2 += s->uvlinesize;
659 /* Chroma MC always uses qpel bilinear */
660 uvmx = (uvmx & 3) << 1;
661 uvmy = (uvmy & 3) << 1;
663 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
664 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
666 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
667 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
671 /** Do motion compensation for 4-MV interlaced frame chroma macroblock (both U and V)
673 void ff_vc1_mc_4mv_chroma4(VC1Context *v, int dir, int dir2, int avg)
675 MpegEncContext *s = &v->s;
676 H264ChromaContext *h264chroma = &v->h264chroma;
677 uint8_t *srcU, *srcV;
678 int uvsrc_x, uvsrc_y;
679 int uvmx_field[4], uvmy_field[4];
681 int fieldmv = v->blk_mv_type[s->block_index[0]];
682 static const int s_rndtblfield[16] = { 0, 0, 1, 2, 4, 4, 5, 6, 2, 2, 3, 8, 6, 6, 7, 12 };
683 int v_dist = fieldmv ? 1 : 4; // vertical offset for lower sub-blocks
684 int v_edge_pos = s->v_edge_pos >> 1;
686 uint8_t (*lutuv)[256];
688 if (s->flags & CODEC_FLAG_GRAY)
691 for (i = 0; i < 4; i++) {
692 int d = i < 2 ? dir: dir2;
694 uvmx_field[i] = (tx + ((tx & 3) == 3)) >> 1;
697 uvmy_field[i] = (ty >> 4) * 8 + s_rndtblfield[ty & 0xF];
699 uvmy_field[i] = (ty + ((ty & 3) == 3)) >> 1;
702 for (i = 0; i < 4; i++) {
703 off = (i & 1) * 4 + ((i & 2) ? v_dist * s->uvlinesize : 0);
704 uvsrc_x = s->mb_x * 8 + (i & 1) * 4 + (uvmx_field[i] >> 2);
705 uvsrc_y = s->mb_y * 8 + ((i & 2) ? v_dist : 0) + (uvmy_field[i] >> 2);
706 // FIXME: implement proper pull-back (see vc1cropmv.c, vc1CROPMV_ChromaPullBack())
707 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
708 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
709 if (i < 2 ? dir : dir2) {
710 srcU = s->next_picture.f->data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
711 srcV = s->next_picture.f->data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
712 lutuv = v->next_lutuv;
713 use_ic = v->next_use_ic;
715 srcU = s->last_picture.f->data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
716 srcV = s->last_picture.f->data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
717 lutuv = v->last_lutuv;
718 use_ic = v->last_use_ic;
720 uvmx_field[i] = (uvmx_field[i] & 3) << 1;
721 uvmy_field[i] = (uvmy_field[i] & 3) << 1;
723 if (fieldmv && !(uvsrc_y & 1))
725 if (fieldmv && (uvsrc_y & 1) && uvsrc_y < 2)
728 || s->h_edge_pos < 10 || v_edge_pos < (5 << fieldmv)
729 || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 5
730 || (unsigned)uvsrc_y > v_edge_pos - (5 << fieldmv)) {
731 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcU,
732 s->uvlinesize, s->uvlinesize,
733 5, (5 << fieldmv), uvsrc_x, uvsrc_y,
734 s->h_edge_pos >> 1, v_edge_pos);
735 s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV,
736 s->uvlinesize, s->uvlinesize,
737 5, (5 << fieldmv), uvsrc_x, uvsrc_y,
738 s->h_edge_pos >> 1, v_edge_pos);
739 srcU = s->edge_emu_buffer;
740 srcV = s->edge_emu_buffer + 16;
742 /* if we deal with intensity compensation we need to scale source blocks */
749 for (j = 0; j < 5; j++) {
750 int f = (uvsrc_y + (j << fieldmv)) & 1;
751 for (i = 0; i < 5; i++) {
752 src[i] = lutuv[f][src[i]];
753 src2[i] = lutuv[f][src2[i]];
755 src += s->uvlinesize << fieldmv;
756 src2 += s->uvlinesize << fieldmv;
762 h264chroma->avg_h264_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
763 h264chroma->avg_h264_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
765 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]);
766 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]);
770 h264chroma->put_h264_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
771 h264chroma->put_h264_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
773 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]);
774 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]);
780 /** Motion compensation for direct or interpolated blocks in B-frames
782 void ff_vc1_interp_mc(VC1Context *v)
784 MpegEncContext *s = &v->s;
785 H264ChromaContext *h264chroma = &v->h264chroma;
786 uint8_t *srcY, *srcU, *srcV;
787 int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
789 int v_edge_pos = s->v_edge_pos >> v->field_mode;
790 int use_ic = v->next_use_ic;
792 if (!v->field_mode && !v->s.next_picture.f->data[0])
797 uvmx = (mx + ((mx & 3) == 3)) >> 1;
798 uvmy = (my + ((my & 3) == 3)) >> 1;
799 if (v->field_mode && v->cur_field_type != v->ref_field_type[1]) {
800 my = my - 2 + 4 * v->cur_field_type;
801 uvmy = uvmy - 2 + 4 * v->cur_field_type;
804 uvmx = uvmx + ((uvmx < 0) ? -(uvmx & 1) : (uvmx & 1));
805 uvmy = uvmy + ((uvmy < 0) ? -(uvmy & 1) : (uvmy & 1));
807 srcY = s->next_picture.f->data[0];
808 srcU = s->next_picture.f->data[1];
809 srcV = s->next_picture.f->data[2];
811 src_x = s->mb_x * 16 + (mx >> 2);
812 src_y = s->mb_y * 16 + (my >> 2);
813 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
814 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
816 if (v->profile != PROFILE_ADVANCED) {
817 src_x = av_clip( src_x, -16, s->mb_width * 16);
818 src_y = av_clip( src_y, -16, s->mb_height * 16);
819 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
820 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
822 src_x = av_clip( src_x, -17, s->avctx->coded_width);
823 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
824 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
825 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
828 srcY += src_y * s->linesize + src_x;
829 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
830 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
832 if (v->field_mode && v->ref_field_type[1]) {
833 srcY += s->current_picture_ptr->f->linesize[0];
834 srcU += s->current_picture_ptr->f->linesize[1];
835 srcV += s->current_picture_ptr->f->linesize[2];
838 /* for grayscale we should not try to read from unknown area */
839 if (s->flags & CODEC_FLAG_GRAY) {
840 srcU = s->edge_emu_buffer + 18 * s->linesize;
841 srcV = s->edge_emu_buffer + 18 * s->linesize;
844 if (v->rangeredfrm || s->h_edge_pos < 22 || v_edge_pos < 22 || use_ic
845 || (unsigned)(src_x - 1) > s->h_edge_pos - (mx & 3) - 16 - 3
846 || (unsigned)(src_y - 1) > v_edge_pos - (my & 3) - 16 - 3) {
847 uint8_t *uvbuf = s->edge_emu_buffer + 19 * s->linesize;
849 srcY -= s->mspel * (1 + s->linesize);
850 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY,
851 s->linesize, s->linesize,
852 17 + s->mspel * 2, 17 + s->mspel * 2,
853 src_x - s->mspel, src_y - s->mspel,
854 s->h_edge_pos, v_edge_pos);
855 srcY = s->edge_emu_buffer;
856 s->vdsp.emulated_edge_mc(uvbuf, srcU,
857 s->uvlinesize, s->uvlinesize,
859 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
860 s->vdsp.emulated_edge_mc(uvbuf + 16, srcV,
861 s->uvlinesize, s->uvlinesize,
863 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
866 /* if we deal with range reduction we need to scale source blocks */
867 if (v->rangeredfrm) {
872 for (j = 0; j < 17 + s->mspel * 2; j++) {
873 for (i = 0; i < 17 + s->mspel * 2; i++)
874 src[i] = ((src[i] - 128) >> 1) + 128;
879 for (j = 0; j < 9; j++) {
880 for (i = 0; i < 9; i++) {
881 src[i] = ((src[i] - 128) >> 1) + 128;
882 src2[i] = ((src2[i] - 128) >> 1) + 128;
884 src += s->uvlinesize;
885 src2 += s->uvlinesize;
890 uint8_t (*luty )[256] = v->next_luty;
891 uint8_t (*lutuv)[256] = v->next_lutuv;
896 for (j = 0; j < 17 + s->mspel * 2; j++) {
897 int f = v->field_mode ? v->ref_field_type[1] : ((j+src_y - s->mspel) & 1);
898 for (i = 0; i < 17 + s->mspel * 2; i++)
899 src[i] = luty[f][src[i]];
904 for (j = 0; j < 9; j++) {
905 int f = v->field_mode ? v->ref_field_type[1] : ((j+uvsrc_y) & 1);
906 for (i = 0; i < 9; i++) {
907 src[i] = lutuv[f][src[i]];
908 src2[i] = lutuv[f][src2[i]];
910 src += s->uvlinesize;
911 src2 += s->uvlinesize;
914 srcY += s->mspel * (1 + s->linesize);
921 dxy = ((my & 3) << 2) | (mx & 3);
922 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off , srcY , s->linesize, v->rnd);
923 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8, srcY + 8, s->linesize, v->rnd);
924 srcY += s->linesize * 8;
925 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize , srcY , s->linesize, v->rnd);
926 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
928 dxy = (my & 2) | ((mx & 2) >> 1);
931 s->hdsp.avg_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
933 s->hdsp.avg_no_rnd_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize, 16);
936 if (s->flags & CODEC_FLAG_GRAY) return;
937 /* Chroma MC always uses qpel blilinear */
938 uvmx = (uvmx & 3) << 1;
939 uvmy = (uvmy & 3) << 1;
941 h264chroma->avg_h264_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
942 h264chroma->avg_h264_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
944 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
945 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);