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 decoder
31 #include "error_resilience.h"
32 #include "mpegvideo.h"
34 #include "h264chroma.h"
37 #include "vc1acdata.h"
38 #include "msmpeg4data.h"
41 #include "vdpau_internal.h"
46 #define MB_INTRA_VLC_BITS 9
50 // offset tables for interlaced picture MVDATA decoding
51 static const int offset_table1[9] = { 0, 1, 2, 4, 8, 16, 32, 64, 128 };
52 static const int offset_table2[9] = { 0, 1, 3, 7, 15, 31, 63, 127, 255 };
54 /***********************************************************************/
56 * @name VC-1 Bitplane decoding
74 /** @} */ //imode defines
76 static void init_block_index(VC1Context *v)
78 MpegEncContext *s = &v->s;
79 ff_init_block_index(s);
80 if (v->field_mode && !(v->second_field ^ v->tff)) {
81 s->dest[0] += s->current_picture_ptr->f.linesize[0];
82 s->dest[1] += s->current_picture_ptr->f.linesize[1];
83 s->dest[2] += s->current_picture_ptr->f.linesize[2];
87 /** @} */ //Bitplane group
89 static void vc1_put_signed_blocks_clamped(VC1Context *v)
91 MpegEncContext *s = &v->s;
92 int topleft_mb_pos, top_mb_pos;
93 int stride_y, fieldtx = 0;
96 /* The put pixels loop is always one MB row behind the decoding loop,
97 * because we can only put pixels when overlap filtering is done, and
98 * for filtering of the bottom edge of a MB, we need the next MB row
100 * Within the row, the put pixels loop is also one MB col behind the
101 * decoding loop. The reason for this is again, because for filtering
102 * of the right MB edge, we need the next MB present. */
103 if (!s->first_slice_line) {
105 topleft_mb_pos = (s->mb_y - 1) * s->mb_stride + s->mb_x - 1;
106 if (v->fcm == ILACE_FRAME)
107 fieldtx = v->fieldtx_plane[topleft_mb_pos];
108 stride_y = s->linesize << fieldtx;
109 v_dist = (16 - fieldtx) >> (fieldtx == 0);
110 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][0],
111 s->dest[0] - 16 * s->linesize - 16,
113 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][1],
114 s->dest[0] - 16 * s->linesize - 8,
116 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][2],
117 s->dest[0] - v_dist * s->linesize - 16,
119 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][3],
120 s->dest[0] - v_dist * s->linesize - 8,
122 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][4],
123 s->dest[1] - 8 * s->uvlinesize - 8,
125 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][5],
126 s->dest[2] - 8 * s->uvlinesize - 8,
129 if (s->mb_x == s->mb_width - 1) {
130 top_mb_pos = (s->mb_y - 1) * s->mb_stride + s->mb_x;
131 if (v->fcm == ILACE_FRAME)
132 fieldtx = v->fieldtx_plane[top_mb_pos];
133 stride_y = s->linesize << fieldtx;
134 v_dist = fieldtx ? 15 : 8;
135 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][0],
136 s->dest[0] - 16 * s->linesize,
138 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][1],
139 s->dest[0] - 16 * s->linesize + 8,
141 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][2],
142 s->dest[0] - v_dist * s->linesize,
144 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][3],
145 s->dest[0] - v_dist * s->linesize + 8,
147 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][4],
148 s->dest[1] - 8 * s->uvlinesize,
150 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][5],
151 s->dest[2] - 8 * s->uvlinesize,
156 #define inc_blk_idx(idx) do { \
158 if (idx >= v->n_allocated_blks) \
162 inc_blk_idx(v->topleft_blk_idx);
163 inc_blk_idx(v->top_blk_idx);
164 inc_blk_idx(v->left_blk_idx);
165 inc_blk_idx(v->cur_blk_idx);
168 static void vc1_loop_filter_iblk(VC1Context *v, int pq)
170 MpegEncContext *s = &v->s;
172 if (!s->first_slice_line) {
173 v->vc1dsp.vc1_v_loop_filter16(s->dest[0], s->linesize, pq);
175 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
176 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize + 8, s->linesize, pq);
177 for (j = 0; j < 2; j++) {
178 v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1], s->uvlinesize, pq);
180 v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
183 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] + 8 * s->linesize, s->linesize, pq);
185 if (s->mb_y == s->end_mb_y - 1) {
187 v->vc1dsp.vc1_h_loop_filter16(s->dest[0], s->linesize, pq);
188 v->vc1dsp.vc1_h_loop_filter8(s->dest[1], s->uvlinesize, pq);
189 v->vc1dsp.vc1_h_loop_filter8(s->dest[2], s->uvlinesize, pq);
191 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] + 8, s->linesize, pq);
195 static void vc1_loop_filter_iblk_delayed(VC1Context *v, int pq)
197 MpegEncContext *s = &v->s;
200 /* The loopfilter runs 1 row and 1 column behind the overlap filter, which
201 * means it runs two rows/cols behind the decoding loop. */
202 if (!s->first_slice_line) {
204 if (s->mb_y >= s->start_mb_y + 2) {
205 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 16 * s->linesize - 16, s->linesize, pq);
208 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 16, s->linesize, pq);
209 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 8, s->linesize, pq);
210 for (j = 0; j < 2; j++) {
211 v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq);
213 v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 16 * s->uvlinesize - 8, s->uvlinesize, pq);
217 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 8 * s->linesize - 16, s->linesize, pq);
220 if (s->mb_x == s->mb_width - 1) {
221 if (s->mb_y >= s->start_mb_y + 2) {
222 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
225 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize, s->linesize, pq);
226 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize + 8, s->linesize, pq);
227 for (j = 0; j < 2; j++) {
228 v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
230 v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 16 * s->uvlinesize, s->uvlinesize, pq);
234 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 8 * s->linesize, s->linesize, pq);
237 if (s->mb_y == s->end_mb_y) {
240 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 16, s->linesize, pq);
241 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 8, s->linesize, pq);
243 for (j = 0; j < 2; j++) {
244 v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq);
249 if (s->mb_x == s->mb_width - 1) {
251 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
252 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize + 8, s->linesize, pq);
254 for (j = 0; j < 2; j++) {
255 v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
263 static void vc1_smooth_overlap_filter_iblk(VC1Context *v)
265 MpegEncContext *s = &v->s;
268 if (v->condover == CONDOVER_NONE)
271 mb_pos = s->mb_x + s->mb_y * s->mb_stride;
273 /* Within a MB, the horizontal overlap always runs before the vertical.
274 * To accomplish that, we run the H on left and internal borders of the
275 * currently decoded MB. Then, we wait for the next overlap iteration
276 * to do H overlap on the right edge of this MB, before moving over and
277 * running the V overlap. Therefore, the V overlap makes us trail by one
278 * MB col and the H overlap filter makes us trail by one MB row. This
279 * is reflected in the time at which we run the put_pixels loop. */
280 if (v->condover == CONDOVER_ALL || v->pq >= 9 || v->over_flags_plane[mb_pos]) {
281 if (s->mb_x && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
282 v->over_flags_plane[mb_pos - 1])) {
283 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][1],
284 v->block[v->cur_blk_idx][0]);
285 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][3],
286 v->block[v->cur_blk_idx][2]);
287 if (!(s->flags & CODEC_FLAG_GRAY)) {
288 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][4],
289 v->block[v->cur_blk_idx][4]);
290 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][5],
291 v->block[v->cur_blk_idx][5]);
294 v->vc1dsp.vc1_h_s_overlap(v->block[v->cur_blk_idx][0],
295 v->block[v->cur_blk_idx][1]);
296 v->vc1dsp.vc1_h_s_overlap(v->block[v->cur_blk_idx][2],
297 v->block[v->cur_blk_idx][3]);
299 if (s->mb_x == s->mb_width - 1) {
300 if (!s->first_slice_line && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
301 v->over_flags_plane[mb_pos - s->mb_stride])) {
302 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][2],
303 v->block[v->cur_blk_idx][0]);
304 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][3],
305 v->block[v->cur_blk_idx][1]);
306 if (!(s->flags & CODEC_FLAG_GRAY)) {
307 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][4],
308 v->block[v->cur_blk_idx][4]);
309 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][5],
310 v->block[v->cur_blk_idx][5]);
313 v->vc1dsp.vc1_v_s_overlap(v->block[v->cur_blk_idx][0],
314 v->block[v->cur_blk_idx][2]);
315 v->vc1dsp.vc1_v_s_overlap(v->block[v->cur_blk_idx][1],
316 v->block[v->cur_blk_idx][3]);
319 if (s->mb_x && (v->condover == CONDOVER_ALL || v->over_flags_plane[mb_pos - 1])) {
320 if (!s->first_slice_line && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
321 v->over_flags_plane[mb_pos - s->mb_stride - 1])) {
322 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][2],
323 v->block[v->left_blk_idx][0]);
324 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][3],
325 v->block[v->left_blk_idx][1]);
326 if (!(s->flags & CODEC_FLAG_GRAY)) {
327 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][4],
328 v->block[v->left_blk_idx][4]);
329 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][5],
330 v->block[v->left_blk_idx][5]);
333 v->vc1dsp.vc1_v_s_overlap(v->block[v->left_blk_idx][0],
334 v->block[v->left_blk_idx][2]);
335 v->vc1dsp.vc1_v_s_overlap(v->block[v->left_blk_idx][1],
336 v->block[v->left_blk_idx][3]);
340 /** Do motion compensation over 1 macroblock
341 * Mostly adapted hpel_motion and qpel_motion from mpegvideo.c
343 static void vc1_mc_1mv(VC1Context *v, int dir)
345 MpegEncContext *s = &v->s;
346 H264ChromaContext *h264chroma = &v->h264chroma;
347 uint8_t *srcY, *srcU, *srcV;
348 int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
350 int v_edge_pos = s->v_edge_pos >> v->field_mode;
352 if ((!v->field_mode ||
353 (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
354 !v->s.last_picture.f.data[0])
357 mx = s->mv[dir][0][0];
358 my = s->mv[dir][0][1];
360 // store motion vectors for further use in B frames
361 if (s->pict_type == AV_PICTURE_TYPE_P) {
362 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = mx;
363 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = my;
366 uvmx = (mx + ((mx & 3) == 3)) >> 1;
367 uvmy = (my + ((my & 3) == 3)) >> 1;
368 v->luma_mv[s->mb_x][0] = uvmx;
369 v->luma_mv[s->mb_x][1] = uvmy;
372 v->cur_field_type != v->ref_field_type[dir]) {
373 my = my - 2 + 4 * v->cur_field_type;
374 uvmy = uvmy - 2 + 4 * v->cur_field_type;
377 // fastuvmc shall be ignored for interlaced frame picture
378 if (v->fastuvmc && (v->fcm != ILACE_FRAME)) {
379 uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
380 uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
383 if (v->field_mode && (v->cur_field_type != v->ref_field_type[dir]) && v->second_field) {
384 srcY = s->current_picture.f.data[0];
385 srcU = s->current_picture.f.data[1];
386 srcV = s->current_picture.f.data[2];
388 srcY = s->last_picture.f.data[0];
389 srcU = s->last_picture.f.data[1];
390 srcV = s->last_picture.f.data[2];
393 srcY = s->next_picture.f.data[0];
394 srcU = s->next_picture.f.data[1];
395 srcV = s->next_picture.f.data[2];
398 src_x = s->mb_x * 16 + (mx >> 2);
399 src_y = s->mb_y * 16 + (my >> 2);
400 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
401 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
403 if (v->profile != PROFILE_ADVANCED) {
404 src_x = av_clip( src_x, -16, s->mb_width * 16);
405 src_y = av_clip( src_y, -16, s->mb_height * 16);
406 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
407 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
409 src_x = av_clip( src_x, -17, s->avctx->coded_width);
410 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
411 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
412 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
415 srcY += src_y * s->linesize + src_x;
416 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
417 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
419 if (v->field_mode && v->ref_field_type[dir]) {
420 srcY += s->current_picture_ptr->f.linesize[0];
421 srcU += s->current_picture_ptr->f.linesize[1];
422 srcV += s->current_picture_ptr->f.linesize[2];
425 /* for grayscale we should not try to read from unknown area */
426 if (s->flags & CODEC_FLAG_GRAY) {
427 srcU = s->edge_emu_buffer + 18 * s->linesize;
428 srcV = s->edge_emu_buffer + 18 * s->linesize;
431 if (v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
432 || s->h_edge_pos < 22 || v_edge_pos < 22
433 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 16 - s->mspel * 3
434 || (unsigned)(src_y - 1) > v_edge_pos - (my&3) - 16 - 3) {
435 uint8_t *uvbuf = s->edge_emu_buffer + 19 * s->linesize;
437 srcY -= s->mspel * (1 + s->linesize);
438 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize,
439 17 + s->mspel * 2, 17 + s->mspel * 2,
440 src_x - s->mspel, src_y - s->mspel,
441 s->h_edge_pos, v_edge_pos);
442 srcY = s->edge_emu_buffer;
443 s->vdsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, 8 + 1, 8 + 1,
444 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
445 s->vdsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8 + 1, 8 + 1,
446 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
449 /* if we deal with range reduction we need to scale source blocks */
450 if (v->rangeredfrm) {
455 for (j = 0; j < 17 + s->mspel * 2; j++) {
456 for (i = 0; i < 17 + s->mspel * 2; i++)
457 src[i] = ((src[i] - 128) >> 1) + 128;
462 for (j = 0; j < 9; j++) {
463 for (i = 0; i < 9; i++) {
464 src[i] = ((src[i] - 128) >> 1) + 128;
465 src2[i] = ((src2[i] - 128) >> 1) + 128;
467 src += s->uvlinesize;
468 src2 += s->uvlinesize;
471 /* if we deal with intensity compensation we need to scale source blocks */
472 if (v->mv_mode == MV_PMODE_INTENSITY_COMP) {
477 for (j = 0; j < 17 + s->mspel * 2; j++) {
478 for (i = 0; i < 17 + s->mspel * 2; i++)
479 src[i] = v->luty[src[i]];
484 for (j = 0; j < 9; j++) {
485 for (i = 0; i < 9; i++) {
486 src[i] = v->lutuv[src[i]];
487 src2[i] = v->lutuv[src2[i]];
489 src += s->uvlinesize;
490 src2 += s->uvlinesize;
493 srcY += s->mspel * (1 + s->linesize);
499 dxy = ((my & 3) << 2) | (mx & 3);
500 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off , srcY , s->linesize, v->rnd);
501 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8, srcY + 8, s->linesize, v->rnd);
502 srcY += s->linesize * 8;
503 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize , srcY , s->linesize, v->rnd);
504 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
505 } else { // hpel mc - always used for luma
506 dxy = (my & 2) | ((mx & 2) >> 1);
508 s->hdsp.put_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
510 s->hdsp.put_no_rnd_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
513 if (s->flags & CODEC_FLAG_GRAY) return;
514 /* Chroma MC always uses qpel bilinear */
515 uvmx = (uvmx & 3) << 1;
516 uvmy = (uvmy & 3) << 1;
518 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
519 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
521 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
522 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
526 static inline int median4(int a, int b, int c, int d)
529 if (c < d) return (FFMIN(b, d) + FFMAX(a, c)) / 2;
530 else return (FFMIN(b, c) + FFMAX(a, d)) / 2;
532 if (c < d) return (FFMIN(a, d) + FFMAX(b, c)) / 2;
533 else return (FFMIN(a, c) + FFMAX(b, d)) / 2;
537 /** Do motion compensation for 4-MV macroblock - luminance block
539 static void vc1_mc_4mv_luma(VC1Context *v, int n, int dir)
541 MpegEncContext *s = &v->s;
543 int dxy, mx, my, src_x, src_y;
545 int fieldmv = (v->fcm == ILACE_FRAME) ? v->blk_mv_type[s->block_index[n]] : 0;
546 int v_edge_pos = s->v_edge_pos >> v->field_mode;
548 if ((!v->field_mode ||
549 (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
550 !v->s.last_picture.f.data[0])
553 mx = s->mv[dir][n][0];
554 my = s->mv[dir][n][1];
557 if (v->field_mode && (v->cur_field_type != v->ref_field_type[dir]) && v->second_field) {
558 srcY = s->current_picture.f.data[0];
560 srcY = s->last_picture.f.data[0];
562 srcY = s->next_picture.f.data[0];
565 if (v->cur_field_type != v->ref_field_type[dir])
566 my = my - 2 + 4 * v->cur_field_type;
569 if (s->pict_type == AV_PICTURE_TYPE_P && n == 3 && v->field_mode) {
570 int same_count = 0, opp_count = 0, k;
571 int chosen_mv[2][4][2], f;
573 for (k = 0; k < 4; k++) {
574 f = v->mv_f[0][s->block_index[k] + v->blocks_off];
575 chosen_mv[f][f ? opp_count : same_count][0] = s->mv[0][k][0];
576 chosen_mv[f][f ? opp_count : same_count][1] = s->mv[0][k][1];
580 f = opp_count > same_count;
581 switch (f ? opp_count : same_count) {
583 tx = median4(chosen_mv[f][0][0], chosen_mv[f][1][0],
584 chosen_mv[f][2][0], chosen_mv[f][3][0]);
585 ty = median4(chosen_mv[f][0][1], chosen_mv[f][1][1],
586 chosen_mv[f][2][1], chosen_mv[f][3][1]);
589 tx = mid_pred(chosen_mv[f][0][0], chosen_mv[f][1][0], chosen_mv[f][2][0]);
590 ty = mid_pred(chosen_mv[f][0][1], chosen_mv[f][1][1], chosen_mv[f][2][1]);
593 tx = (chosen_mv[f][0][0] + chosen_mv[f][1][0]) / 2;
594 ty = (chosen_mv[f][0][1] + chosen_mv[f][1][1]) / 2;
597 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx;
598 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty;
599 for (k = 0; k < 4; k++)
600 v->mv_f[1][s->block_index[k] + v->blocks_off] = f;
603 if (v->fcm == ILACE_FRAME) { // not sure if needed for other types of picture
605 int width = s->avctx->coded_width;
606 int height = s->avctx->coded_height >> 1;
607 qx = (s->mb_x * 16) + (mx >> 2);
608 qy = (s->mb_y * 8) + (my >> 3);
613 mx -= 4 * (qx - width);
616 else if (qy > height + 1)
617 my -= 8 * (qy - height - 1);
620 if ((v->fcm == ILACE_FRAME) && fieldmv)
621 off = ((n > 1) ? s->linesize : 0) + (n & 1) * 8;
623 off = s->linesize * 4 * (n & 2) + (n & 1) * 8;
625 src_x = s->mb_x * 16 + (n & 1) * 8 + (mx >> 2);
627 src_y = s->mb_y * 16 + (n & 2) * 4 + (my >> 2);
629 src_y = s->mb_y * 16 + ((n > 1) ? 1 : 0) + (my >> 2);
631 if (v->profile != PROFILE_ADVANCED) {
632 src_x = av_clip(src_x, -16, s->mb_width * 16);
633 src_y = av_clip(src_y, -16, s->mb_height * 16);
635 src_x = av_clip(src_x, -17, s->avctx->coded_width);
636 if (v->fcm == ILACE_FRAME) {
638 src_y = av_clip(src_y, -17, s->avctx->coded_height + 1);
640 src_y = av_clip(src_y, -18, s->avctx->coded_height);
642 src_y = av_clip(src_y, -18, s->avctx->coded_height + 1);
646 srcY += src_y * s->linesize + src_x;
647 if (v->field_mode && v->ref_field_type[dir])
648 srcY += s->current_picture_ptr->f.linesize[0];
650 if (fieldmv && !(src_y & 1))
652 if (fieldmv && (src_y & 1) && src_y < 4)
654 if (v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
655 || s->h_edge_pos < 13 || v_edge_pos < 23
656 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx & 3) - 8 - s->mspel * 2
657 || (unsigned)(src_y - (s->mspel << fieldmv)) > v_edge_pos - (my & 3) - ((8 + s->mspel * 2) << fieldmv)) {
658 srcY -= s->mspel * (1 + (s->linesize << fieldmv));
659 /* check emulate edge stride and offset */
660 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize,
661 9 + s->mspel * 2, (9 + s->mspel * 2) << fieldmv,
662 src_x - s->mspel, src_y - (s->mspel << fieldmv),
663 s->h_edge_pos, v_edge_pos);
664 srcY = s->edge_emu_buffer;
665 /* if we deal with range reduction we need to scale source blocks */
666 if (v->rangeredfrm) {
671 for (j = 0; j < 9 + s->mspel * 2; j++) {
672 for (i = 0; i < 9 + s->mspel * 2; i++)
673 src[i] = ((src[i] - 128) >> 1) + 128;
674 src += s->linesize << fieldmv;
677 /* if we deal with intensity compensation we need to scale source blocks */
678 if (v->mv_mode == MV_PMODE_INTENSITY_COMP) {
683 for (j = 0; j < 9 + s->mspel * 2; j++) {
684 for (i = 0; i < 9 + s->mspel * 2; i++)
685 src[i] = v->luty[src[i]];
686 src += s->linesize << fieldmv;
689 srcY += s->mspel * (1 + (s->linesize << fieldmv));
693 dxy = ((my & 3) << 2) | (mx & 3);
694 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize << fieldmv, v->rnd);
695 } else { // hpel mc - always used for luma
696 dxy = (my & 2) | ((mx & 2) >> 1);
698 s->hdsp.put_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
700 s->hdsp.put_no_rnd_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
704 static av_always_inline int get_chroma_mv(int *mvx, int *mvy, int *a, int flag, int *tx, int *ty)
707 static const int count[16] = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4};
709 idx = ((a[3] != flag) << 3)
710 | ((a[2] != flag) << 2)
711 | ((a[1] != flag) << 1)
714 *tx = median4(mvx[0], mvx[1], mvx[2], mvx[3]);
715 *ty = median4(mvy[0], mvy[1], mvy[2], mvy[3]);
717 } else if (count[idx] == 1) {
720 *tx = mid_pred(mvx[1], mvx[2], mvx[3]);
721 *ty = mid_pred(mvy[1], mvy[2], mvy[3]);
724 *tx = mid_pred(mvx[0], mvx[2], mvx[3]);
725 *ty = mid_pred(mvy[0], mvy[2], mvy[3]);
728 *tx = mid_pred(mvx[0], mvx[1], mvx[3]);
729 *ty = mid_pred(mvy[0], mvy[1], mvy[3]);
732 *tx = mid_pred(mvx[0], mvx[1], mvx[2]);
733 *ty = mid_pred(mvy[0], mvy[1], mvy[2]);
736 } else if (count[idx] == 2) {
738 for (i = 0; i < 3; i++)
743 for (i = t1 + 1; i < 4; i++)
748 *tx = (mvx[t1] + mvx[t2]) / 2;
749 *ty = (mvy[t1] + mvy[t2]) / 2;
757 /** Do motion compensation for 4-MV macroblock - both chroma blocks
759 static void vc1_mc_4mv_chroma(VC1Context *v, int dir)
761 MpegEncContext *s = &v->s;
762 H264ChromaContext *h264chroma = &v->h264chroma;
763 uint8_t *srcU, *srcV;
764 int uvmx, uvmy, uvsrc_x, uvsrc_y;
765 int k, tx = 0, ty = 0;
766 int mvx[4], mvy[4], intra[4], mv_f[4];
768 int chroma_ref_type = v->cur_field_type, off = 0;
769 int v_edge_pos = s->v_edge_pos >> v->field_mode;
771 if (!v->field_mode && !v->s.last_picture.f.data[0])
773 if (s->flags & CODEC_FLAG_GRAY)
776 for (k = 0; k < 4; k++) {
777 mvx[k] = s->mv[dir][k][0];
778 mvy[k] = s->mv[dir][k][1];
779 intra[k] = v->mb_type[0][s->block_index[k]];
781 mv_f[k] = v->mv_f[dir][s->block_index[k] + v->blocks_off];
784 /* calculate chroma MV vector from four luma MVs */
785 if (!v->field_mode || (v->field_mode && !v->numref)) {
786 valid_count = get_chroma_mv(mvx, mvy, intra, 0, &tx, &ty);
787 chroma_ref_type = v->reffield;
789 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = 0;
790 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = 0;
791 v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
792 return; //no need to do MC for intra blocks
796 if (mv_f[0] + mv_f[1] + mv_f[2] + mv_f[3] > 2)
798 valid_count = get_chroma_mv(mvx, mvy, mv_f, dominant, &tx, &ty);
800 chroma_ref_type = !v->cur_field_type;
802 if (v->field_mode && chroma_ref_type == 1 && v->cur_field_type == 1 && !v->s.last_picture.f.data[0])
804 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx;
805 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty;
806 uvmx = (tx + ((tx & 3) == 3)) >> 1;
807 uvmy = (ty + ((ty & 3) == 3)) >> 1;
809 v->luma_mv[s->mb_x][0] = uvmx;
810 v->luma_mv[s->mb_x][1] = uvmy;
813 uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
814 uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
816 // Field conversion bias
817 if (v->cur_field_type != chroma_ref_type)
818 uvmy += 2 - 4 * chroma_ref_type;
820 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
821 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
823 if (v->profile != PROFILE_ADVANCED) {
824 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
825 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
827 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
828 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
832 if (v->field_mode && (v->cur_field_type != chroma_ref_type) && v->second_field) {
833 srcU = s->current_picture.f.data[1];
834 srcV = s->current_picture.f.data[2];
836 srcU = s->last_picture.f.data[1];
837 srcV = s->last_picture.f.data[2];
840 srcU = s->next_picture.f.data[1];
841 srcV = s->next_picture.f.data[2];
844 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
845 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
848 if (chroma_ref_type) {
849 srcU += s->current_picture_ptr->f.linesize[1];
850 srcV += s->current_picture_ptr->f.linesize[2];
855 if (v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
856 || s->h_edge_pos < 18 || v_edge_pos < 18
857 || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 9
858 || (unsigned)uvsrc_y > (v_edge_pos >> 1) - 9) {
859 s->vdsp.emulated_edge_mc(s->edge_emu_buffer , srcU, s->uvlinesize,
860 8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
861 s->h_edge_pos >> 1, v_edge_pos >> 1);
862 s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV, s->uvlinesize,
863 8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
864 s->h_edge_pos >> 1, v_edge_pos >> 1);
865 srcU = s->edge_emu_buffer;
866 srcV = s->edge_emu_buffer + 16;
868 /* if we deal with range reduction we need to scale source blocks */
869 if (v->rangeredfrm) {
875 for (j = 0; j < 9; j++) {
876 for (i = 0; i < 9; i++) {
877 src[i] = ((src[i] - 128) >> 1) + 128;
878 src2[i] = ((src2[i] - 128) >> 1) + 128;
880 src += s->uvlinesize;
881 src2 += s->uvlinesize;
884 /* if we deal with intensity compensation we need to scale source blocks */
885 if (v->mv_mode == MV_PMODE_INTENSITY_COMP) {
891 for (j = 0; j < 9; j++) {
892 for (i = 0; i < 9; i++) {
893 src[i] = v->lutuv[src[i]];
894 src2[i] = v->lutuv[src2[i]];
896 src += s->uvlinesize;
897 src2 += s->uvlinesize;
902 /* Chroma MC always uses qpel bilinear */
903 uvmx = (uvmx & 3) << 1;
904 uvmy = (uvmy & 3) << 1;
906 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[1] + off, srcU, s->uvlinesize, 8, uvmx, uvmy);
907 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[2] + off, srcV, s->uvlinesize, 8, uvmx, uvmy);
909 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1] + off, srcU, s->uvlinesize, 8, uvmx, uvmy);
910 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2] + off, srcV, s->uvlinesize, 8, uvmx, uvmy);
914 /** Do motion compensation for 4-MV field chroma macroblock (both U and V)
916 static void vc1_mc_4mv_chroma4(VC1Context *v)
918 MpegEncContext *s = &v->s;
919 H264ChromaContext *h264chroma = &v->h264chroma;
920 uint8_t *srcU, *srcV;
921 int uvsrc_x, uvsrc_y;
922 int uvmx_field[4], uvmy_field[4];
924 int fieldmv = v->blk_mv_type[s->block_index[0]];
925 static const int s_rndtblfield[16] = { 0, 0, 1, 2, 4, 4, 5, 6, 2, 2, 3, 8, 6, 6, 7, 12 };
926 int v_dist = fieldmv ? 1 : 4; // vertical offset for lower sub-blocks
927 int v_edge_pos = s->v_edge_pos >> 1;
929 if (!v->s.last_picture.f.data[0])
931 if (s->flags & CODEC_FLAG_GRAY)
934 for (i = 0; i < 4; i++) {
936 uvmx_field[i] = (tx + ((tx & 3) == 3)) >> 1;
939 uvmy_field[i] = (ty >> 4) * 8 + s_rndtblfield[ty & 0xF];
941 uvmy_field[i] = (ty + ((ty & 3) == 3)) >> 1;
944 for (i = 0; i < 4; i++) {
945 off = (i & 1) * 4 + ((i & 2) ? v_dist * s->uvlinesize : 0);
946 uvsrc_x = s->mb_x * 8 + (i & 1) * 4 + (uvmx_field[i] >> 2);
947 uvsrc_y = s->mb_y * 8 + ((i & 2) ? v_dist : 0) + (uvmy_field[i] >> 2);
948 // FIXME: implement proper pull-back (see vc1cropmv.c, vc1CROPMV_ChromaPullBack())
949 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
950 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
951 srcU = s->last_picture.f.data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
952 srcV = s->last_picture.f.data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
953 uvmx_field[i] = (uvmx_field[i] & 3) << 1;
954 uvmy_field[i] = (uvmy_field[i] & 3) << 1;
956 if (fieldmv && !(uvsrc_y & 1))
958 if (fieldmv && (uvsrc_y & 1) && uvsrc_y < 2)
960 if ((v->mv_mode == MV_PMODE_INTENSITY_COMP)
961 || s->h_edge_pos < 10 || v_edge_pos < (5 << fieldmv)
962 || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 5
963 || (unsigned)uvsrc_y > v_edge_pos - (5 << fieldmv)) {
964 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcU, s->uvlinesize,
965 5, (5 << fieldmv), uvsrc_x, uvsrc_y,
966 s->h_edge_pos >> 1, v_edge_pos);
967 s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV, s->uvlinesize,
968 5, (5 << fieldmv), uvsrc_x, uvsrc_y,
969 s->h_edge_pos >> 1, v_edge_pos);
970 srcU = s->edge_emu_buffer;
971 srcV = s->edge_emu_buffer + 16;
973 /* if we deal with intensity compensation we need to scale source blocks */
974 if (v->mv_mode == MV_PMODE_INTENSITY_COMP) {
980 for (j = 0; j < 5; j++) {
981 for (i = 0; i < 5; i++) {
982 src[i] = v->lutuv[src[i]];
983 src2[i] = v->lutuv[src2[i]];
985 src += s->uvlinesize << 1;
986 src2 += s->uvlinesize << 1;
991 h264chroma->put_h264_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
992 h264chroma->put_h264_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
994 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]);
995 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]);
1000 /***********************************************************************/
1002 * @name VC-1 Block-level functions
1003 * @see 7.1.4, p91 and 8.1.1.7, p(1)04
1009 * @brief Get macroblock-level quantizer scale
1011 #define GET_MQUANT() \
1012 if (v->dquantfrm) { \
1014 if (v->dqprofile == DQPROFILE_ALL_MBS) { \
1015 if (v->dqbilevel) { \
1016 mquant = (get_bits1(gb)) ? v->altpq : v->pq; \
1018 mqdiff = get_bits(gb, 3); \
1020 mquant = v->pq + mqdiff; \
1022 mquant = get_bits(gb, 5); \
1025 if (v->dqprofile == DQPROFILE_SINGLE_EDGE) \
1026 edges = 1 << v->dqsbedge; \
1027 else if (v->dqprofile == DQPROFILE_DOUBLE_EDGES) \
1028 edges = (3 << v->dqsbedge) % 15; \
1029 else if (v->dqprofile == DQPROFILE_FOUR_EDGES) \
1031 if ((edges&1) && !s->mb_x) \
1032 mquant = v->altpq; \
1033 if ((edges&2) && s->first_slice_line) \
1034 mquant = v->altpq; \
1035 if ((edges&4) && s->mb_x == (s->mb_width - 1)) \
1036 mquant = v->altpq; \
1037 if ((edges&8) && s->mb_y == (s->mb_height - 1)) \
1038 mquant = v->altpq; \
1039 if (!mquant || mquant > 31) { \
1040 av_log(v->s.avctx, AV_LOG_ERROR, \
1041 "Overriding invalid mquant %d\n", mquant); \
1047 * @def GET_MVDATA(_dmv_x, _dmv_y)
1048 * @brief Get MV differentials
1049 * @see MVDATA decoding from 8.3.5.2, p(1)20
1050 * @param _dmv_x Horizontal differential for decoded MV
1051 * @param _dmv_y Vertical differential for decoded MV
1053 #define GET_MVDATA(_dmv_x, _dmv_y) \
1054 index = 1 + get_vlc2(gb, ff_vc1_mv_diff_vlc[s->mv_table_index].table, \
1055 VC1_MV_DIFF_VLC_BITS, 2); \
1057 mb_has_coeffs = 1; \
1060 mb_has_coeffs = 0; \
1063 _dmv_x = _dmv_y = 0; \
1064 } else if (index == 35) { \
1065 _dmv_x = get_bits(gb, v->k_x - 1 + s->quarter_sample); \
1066 _dmv_y = get_bits(gb, v->k_y - 1 + s->quarter_sample); \
1067 } else if (index == 36) { \
1072 index1 = index % 6; \
1073 if (!s->quarter_sample && index1 == 5) val = 1; \
1075 if (size_table[index1] - val > 0) \
1076 val = get_bits(gb, size_table[index1] - val); \
1078 sign = 0 - (val&1); \
1079 _dmv_x = (sign ^ ((val>>1) + offset_table[index1])) - sign; \
1081 index1 = index / 6; \
1082 if (!s->quarter_sample && index1 == 5) val = 1; \
1084 if (size_table[index1] - val > 0) \
1085 val = get_bits(gb, size_table[index1] - val); \
1087 sign = 0 - (val & 1); \
1088 _dmv_y = (sign ^ ((val >> 1) + offset_table[index1])) - sign; \
1091 static av_always_inline void get_mvdata_interlaced(VC1Context *v, int *dmv_x,
1092 int *dmv_y, int *pred_flag)
1095 int extend_x = 0, extend_y = 0;
1096 GetBitContext *gb = &v->s.gb;
1099 const int* offs_tab;
1102 bits = VC1_2REF_MVDATA_VLC_BITS;
1105 bits = VC1_1REF_MVDATA_VLC_BITS;
1108 switch (v->dmvrange) {
1116 extend_x = extend_y = 1;
1119 index = get_vlc2(gb, v->imv_vlc->table, bits, 3);
1121 *dmv_x = get_bits(gb, v->k_x);
1122 *dmv_y = get_bits(gb, v->k_y);
1125 *pred_flag = *dmv_y & 1;
1126 *dmv_y = (*dmv_y + *pred_flag) >> 1;
1128 *dmv_y = (*dmv_y + (*dmv_y & 1)) >> 1;
1134 offs_tab = offset_table2;
1136 offs_tab = offset_table1;
1137 index1 = (index + 1) % 9;
1139 val = get_bits(gb, index1 + extend_x);
1140 sign = 0 -(val & 1);
1141 *dmv_x = (sign ^ ((val >> 1) + offs_tab[index1])) - sign;
1145 offs_tab = offset_table2;
1147 offs_tab = offset_table1;
1148 index1 = (index + 1) / 9;
1149 if (index1 > v->numref) {
1150 val = get_bits(gb, (index1 + (extend_y << v->numref)) >> v->numref);
1151 sign = 0 - (val & 1);
1152 *dmv_y = (sign ^ ((val >> 1) + offs_tab[index1 >> v->numref])) - sign;
1155 if (v->numref && pred_flag)
1156 *pred_flag = index1 & 1;
1160 static av_always_inline int scaleforsame_x(VC1Context *v, int n /* MV */, int dir)
1162 int scaledvalue, refdist;
1163 int scalesame1, scalesame2;
1164 int scalezone1_x, zone1offset_x;
1165 int table_index = dir ^ v->second_field;
1167 if (v->s.pict_type != AV_PICTURE_TYPE_B)
1168 refdist = v->refdist;
1170 refdist = dir ? v->brfd : v->frfd;
1173 scalesame1 = ff_vc1_field_mvpred_scales[table_index][1][refdist];
1174 scalesame2 = ff_vc1_field_mvpred_scales[table_index][2][refdist];
1175 scalezone1_x = ff_vc1_field_mvpred_scales[table_index][3][refdist];
1176 zone1offset_x = ff_vc1_field_mvpred_scales[table_index][5][refdist];
1181 if (FFABS(n) < scalezone1_x)
1182 scaledvalue = (n * scalesame1) >> 8;
1185 scaledvalue = ((n * scalesame2) >> 8) - zone1offset_x;
1187 scaledvalue = ((n * scalesame2) >> 8) + zone1offset_x;
1190 return av_clip(scaledvalue, -v->range_x, v->range_x - 1);
1193 static av_always_inline int scaleforsame_y(VC1Context *v, int i, int n /* MV */, int dir)
1195 int scaledvalue, refdist;
1196 int scalesame1, scalesame2;
1197 int scalezone1_y, zone1offset_y;
1198 int table_index = dir ^ v->second_field;
1200 if (v->s.pict_type != AV_PICTURE_TYPE_B)
1201 refdist = v->refdist;
1203 refdist = dir ? v->brfd : v->frfd;
1206 scalesame1 = ff_vc1_field_mvpred_scales[table_index][1][refdist];
1207 scalesame2 = ff_vc1_field_mvpred_scales[table_index][2][refdist];
1208 scalezone1_y = ff_vc1_field_mvpred_scales[table_index][4][refdist];
1209 zone1offset_y = ff_vc1_field_mvpred_scales[table_index][6][refdist];
1214 if (FFABS(n) < scalezone1_y)
1215 scaledvalue = (n * scalesame1) >> 8;
1218 scaledvalue = ((n * scalesame2) >> 8) - zone1offset_y;
1220 scaledvalue = ((n * scalesame2) >> 8) + zone1offset_y;
1224 if (v->cur_field_type && !v->ref_field_type[dir])
1225 return av_clip(scaledvalue, -v->range_y / 2 + 1, v->range_y / 2);
1227 return av_clip(scaledvalue, -v->range_y / 2, v->range_y / 2 - 1);
1230 static av_always_inline int scaleforopp_x(VC1Context *v, int n /* MV */)
1232 int scalezone1_x, zone1offset_x;
1233 int scaleopp1, scaleopp2, brfd;
1236 brfd = FFMIN(v->brfd, 3);
1237 scalezone1_x = ff_vc1_b_field_mvpred_scales[3][brfd];
1238 zone1offset_x = ff_vc1_b_field_mvpred_scales[5][brfd];
1239 scaleopp1 = ff_vc1_b_field_mvpred_scales[1][brfd];
1240 scaleopp2 = ff_vc1_b_field_mvpred_scales[2][brfd];
1245 if (FFABS(n) < scalezone1_x)
1246 scaledvalue = (n * scaleopp1) >> 8;
1249 scaledvalue = ((n * scaleopp2) >> 8) - zone1offset_x;
1251 scaledvalue = ((n * scaleopp2) >> 8) + zone1offset_x;
1254 return av_clip(scaledvalue, -v->range_x, v->range_x - 1);
1257 static av_always_inline int scaleforopp_y(VC1Context *v, int n /* MV */, int dir)
1259 int scalezone1_y, zone1offset_y;
1260 int scaleopp1, scaleopp2, brfd;
1263 brfd = FFMIN(v->brfd, 3);
1264 scalezone1_y = ff_vc1_b_field_mvpred_scales[4][brfd];
1265 zone1offset_y = ff_vc1_b_field_mvpred_scales[6][brfd];
1266 scaleopp1 = ff_vc1_b_field_mvpred_scales[1][brfd];
1267 scaleopp2 = ff_vc1_b_field_mvpred_scales[2][brfd];
1272 if (FFABS(n) < scalezone1_y)
1273 scaledvalue = (n * scaleopp1) >> 8;
1276 scaledvalue = ((n * scaleopp2) >> 8) - zone1offset_y;
1278 scaledvalue = ((n * scaleopp2) >> 8) + zone1offset_y;
1281 if (v->cur_field_type && !v->ref_field_type[dir]) {
1282 return av_clip(scaledvalue, -v->range_y / 2 + 1, v->range_y / 2);
1284 return av_clip(scaledvalue, -v->range_y / 2, v->range_y / 2 - 1);
1288 static av_always_inline int scaleforsame(VC1Context *v, int i, int n /* MV */,
1291 int brfd, scalesame;
1292 int hpel = 1 - v->s.quarter_sample;
1295 if (v->s.pict_type != AV_PICTURE_TYPE_B || v->second_field || !dir) {
1297 n = scaleforsame_y(v, i, n, dir) << hpel;
1299 n = scaleforsame_x(v, n, dir) << hpel;
1302 brfd = FFMIN(v->brfd, 3);
1303 scalesame = ff_vc1_b_field_mvpred_scales[0][brfd];
1305 n = (n * scalesame >> 8) << hpel;
1309 static av_always_inline int scaleforopp(VC1Context *v, int n /* MV */,
1312 int refdist, scaleopp;
1313 int hpel = 1 - v->s.quarter_sample;
1316 if (v->s.pict_type == AV_PICTURE_TYPE_B && !v->second_field && dir == 1) {
1318 n = scaleforopp_y(v, n, dir) << hpel;
1320 n = scaleforopp_x(v, n) << hpel;
1323 if (v->s.pict_type != AV_PICTURE_TYPE_B)
1324 refdist = FFMIN(v->refdist, 3);
1326 refdist = dir ? v->brfd : v->frfd;
1327 scaleopp = ff_vc1_field_mvpred_scales[dir ^ v->second_field][0][refdist];
1329 n = (n * scaleopp >> 8) << hpel;
1333 /** Predict and set motion vector
1335 static inline void vc1_pred_mv(VC1Context *v, int n, int dmv_x, int dmv_y,
1336 int mv1, int r_x, int r_y, uint8_t* is_intra,
1337 int pred_flag, int dir)
1339 MpegEncContext *s = &v->s;
1340 int xy, wrap, off = 0;
1344 int mixedmv_pic, num_samefield = 0, num_oppfield = 0;
1345 int opposite, a_f, b_f, c_f;
1346 int16_t field_predA[2];
1347 int16_t field_predB[2];
1348 int16_t field_predC[2];
1349 int a_valid, b_valid, c_valid;
1350 int hybridmv_thresh, y_bias = 0;
1352 if (v->mv_mode == MV_PMODE_MIXED_MV ||
1353 ((v->mv_mode == MV_PMODE_INTENSITY_COMP) && (v->mv_mode2 == MV_PMODE_MIXED_MV)))
1357 /* scale MV difference to be quad-pel */
1358 dmv_x <<= 1 - s->quarter_sample;
1359 dmv_y <<= 1 - s->quarter_sample;
1361 wrap = s->b8_stride;
1362 xy = s->block_index[n];
1365 s->mv[0][n][0] = s->current_picture.motion_val[0][xy + v->blocks_off][0] = 0;
1366 s->mv[0][n][1] = s->current_picture.motion_val[0][xy + v->blocks_off][1] = 0;
1367 s->current_picture.motion_val[1][xy + v->blocks_off][0] = 0;
1368 s->current_picture.motion_val[1][xy + v->blocks_off][1] = 0;
1369 if (mv1) { /* duplicate motion data for 1-MV block */
1370 s->current_picture.motion_val[0][xy + 1 + v->blocks_off][0] = 0;
1371 s->current_picture.motion_val[0][xy + 1 + v->blocks_off][1] = 0;
1372 s->current_picture.motion_val[0][xy + wrap + v->blocks_off][0] = 0;
1373 s->current_picture.motion_val[0][xy + wrap + v->blocks_off][1] = 0;
1374 s->current_picture.motion_val[0][xy + wrap + 1 + v->blocks_off][0] = 0;
1375 s->current_picture.motion_val[0][xy + wrap + 1 + v->blocks_off][1] = 0;
1376 v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
1377 s->current_picture.motion_val[1][xy + 1 + v->blocks_off][0] = 0;
1378 s->current_picture.motion_val[1][xy + 1 + v->blocks_off][1] = 0;
1379 s->current_picture.motion_val[1][xy + wrap][0] = 0;
1380 s->current_picture.motion_val[1][xy + wrap + v->blocks_off][1] = 0;
1381 s->current_picture.motion_val[1][xy + wrap + 1 + v->blocks_off][0] = 0;
1382 s->current_picture.motion_val[1][xy + wrap + 1 + v->blocks_off][1] = 0;
1387 C = s->current_picture.motion_val[dir][xy - 1 + v->blocks_off];
1388 A = s->current_picture.motion_val[dir][xy - wrap + v->blocks_off];
1390 if (v->field_mode && mixedmv_pic)
1391 off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
1393 off = (s->mb_x == (s->mb_width - 1)) ? -1 : 2;
1395 //in 4-MV mode different blocks have different B predictor position
1398 off = (s->mb_x > 0) ? -1 : 1;
1401 off = (s->mb_x == (s->mb_width - 1)) ? -1 : 1;
1410 B = s->current_picture.motion_val[dir][xy - wrap + off + v->blocks_off];
1412 a_valid = !s->first_slice_line || (n == 2 || n == 3);
1413 b_valid = a_valid && (s->mb_width > 1);
1414 c_valid = s->mb_x || (n == 1 || n == 3);
1415 if (v->field_mode) {
1416 a_valid = a_valid && !is_intra[xy - wrap];
1417 b_valid = b_valid && !is_intra[xy - wrap + off];
1418 c_valid = c_valid && !is_intra[xy - 1];
1422 a_f = v->mv_f[dir][xy - wrap + v->blocks_off];
1423 num_oppfield += a_f;
1424 num_samefield += 1 - a_f;
1425 field_predA[0] = A[0];
1426 field_predA[1] = A[1];
1428 field_predA[0] = field_predA[1] = 0;
1432 b_f = v->mv_f[dir][xy - wrap + off + v->blocks_off];
1433 num_oppfield += b_f;
1434 num_samefield += 1 - b_f;
1435 field_predB[0] = B[0];
1436 field_predB[1] = B[1];
1438 field_predB[0] = field_predB[1] = 0;
1442 c_f = v->mv_f[dir][xy - 1 + v->blocks_off];
1443 num_oppfield += c_f;
1444 num_samefield += 1 - c_f;
1445 field_predC[0] = C[0];
1446 field_predC[1] = C[1];
1448 field_predC[0] = field_predC[1] = 0;
1452 if (v->field_mode) {
1454 // REFFIELD determines if the last field or the second-last field is
1455 // to be used as reference
1456 opposite = 1 - v->reffield;
1458 if (num_samefield <= num_oppfield)
1459 opposite = 1 - pred_flag;
1461 opposite = pred_flag;
1466 if (a_valid && !a_f) {
1467 field_predA[0] = scaleforopp(v, field_predA[0], 0, dir);
1468 field_predA[1] = scaleforopp(v, field_predA[1], 1, dir);
1470 if (b_valid && !b_f) {
1471 field_predB[0] = scaleforopp(v, field_predB[0], 0, dir);
1472 field_predB[1] = scaleforopp(v, field_predB[1], 1, dir);
1474 if (c_valid && !c_f) {
1475 field_predC[0] = scaleforopp(v, field_predC[0], 0, dir);
1476 field_predC[1] = scaleforopp(v, field_predC[1], 1, dir);
1478 v->mv_f[dir][xy + v->blocks_off] = 1;
1479 v->ref_field_type[dir] = !v->cur_field_type;
1481 if (a_valid && a_f) {
1482 field_predA[0] = scaleforsame(v, n, field_predA[0], 0, dir);
1483 field_predA[1] = scaleforsame(v, n, field_predA[1], 1, dir);
1485 if (b_valid && b_f) {
1486 field_predB[0] = scaleforsame(v, n, field_predB[0], 0, dir);
1487 field_predB[1] = scaleforsame(v, n, field_predB[1], 1, dir);
1489 if (c_valid && c_f) {
1490 field_predC[0] = scaleforsame(v, n, field_predC[0], 0, dir);
1491 field_predC[1] = scaleforsame(v, n, field_predC[1], 1, dir);
1493 v->mv_f[dir][xy + v->blocks_off] = 0;
1494 v->ref_field_type[dir] = v->cur_field_type;
1498 px = field_predA[0];
1499 py = field_predA[1];
1500 } else if (c_valid) {
1501 px = field_predC[0];
1502 py = field_predC[1];
1503 } else if (b_valid) {
1504 px = field_predB[0];
1505 py = field_predB[1];
1511 if (num_samefield + num_oppfield > 1) {
1512 px = mid_pred(field_predA[0], field_predB[0], field_predC[0]);
1513 py = mid_pred(field_predA[1], field_predB[1], field_predC[1]);
1516 /* Pullback MV as specified in 8.3.5.3.4 */
1517 if (!v->field_mode) {
1519 qx = (s->mb_x << 6) + ((n == 1 || n == 3) ? 32 : 0);
1520 qy = (s->mb_y << 6) + ((n == 2 || n == 3) ? 32 : 0);
1521 X = (s->mb_width << 6) - 4;
1522 Y = (s->mb_height << 6) - 4;
1524 if (qx + px < -60) px = -60 - qx;
1525 if (qy + py < -60) py = -60 - qy;
1527 if (qx + px < -28) px = -28 - qx;
1528 if (qy + py < -28) py = -28 - qy;
1530 if (qx + px > X) px = X - qx;
1531 if (qy + py > Y) py = Y - qy;
1534 if (!v->field_mode || s->pict_type != AV_PICTURE_TYPE_B) {
1535 /* Calculate hybrid prediction as specified in 8.3.5.3.5 (also 10.3.5.4.3.5) */
1536 hybridmv_thresh = 32;
1537 if (a_valid && c_valid) {
1538 if (is_intra[xy - wrap])
1539 sum = FFABS(px) + FFABS(py);
1541 sum = FFABS(px - field_predA[0]) + FFABS(py - field_predA[1]);
1542 if (sum > hybridmv_thresh) {
1543 if (get_bits1(&s->gb)) { // read HYBRIDPRED bit
1544 px = field_predA[0];
1545 py = field_predA[1];
1547 px = field_predC[0];
1548 py = field_predC[1];
1551 if (is_intra[xy - 1])
1552 sum = FFABS(px) + FFABS(py);
1554 sum = FFABS(px - field_predC[0]) + FFABS(py - field_predC[1]);
1555 if (sum > hybridmv_thresh) {
1556 if (get_bits1(&s->gb)) {
1557 px = field_predA[0];
1558 py = field_predA[1];
1560 px = field_predC[0];
1561 py = field_predC[1];
1568 if (v->field_mode && v->numref)
1570 if (v->field_mode && v->cur_field_type && v->ref_field_type[dir] == 0)
1572 /* store MV using signed modulus of MV range defined in 4.11 */
1573 s->mv[dir][n][0] = s->current_picture.motion_val[dir][xy + v->blocks_off][0] = ((px + dmv_x + r_x) & ((r_x << 1) - 1)) - r_x;
1574 s->mv[dir][n][1] = s->current_picture.motion_val[dir][xy + v->blocks_off][1] = ((py + dmv_y + r_y - y_bias) & ((r_y << 1) - 1)) - r_y + y_bias;
1575 if (mv1) { /* duplicate motion data for 1-MV block */
1576 s->current_picture.motion_val[dir][xy + 1 + v->blocks_off][0] = s->current_picture.motion_val[dir][xy + v->blocks_off][0];
1577 s->current_picture.motion_val[dir][xy + 1 + v->blocks_off][1] = s->current_picture.motion_val[dir][xy + v->blocks_off][1];
1578 s->current_picture.motion_val[dir][xy + wrap + v->blocks_off][0] = s->current_picture.motion_val[dir][xy + v->blocks_off][0];
1579 s->current_picture.motion_val[dir][xy + wrap + v->blocks_off][1] = s->current_picture.motion_val[dir][xy + v->blocks_off][1];
1580 s->current_picture.motion_val[dir][xy + wrap + 1 + v->blocks_off][0] = s->current_picture.motion_val[dir][xy + v->blocks_off][0];
1581 s->current_picture.motion_val[dir][xy + wrap + 1 + v->blocks_off][1] = s->current_picture.motion_val[dir][xy + v->blocks_off][1];
1582 v->mv_f[dir][xy + 1 + v->blocks_off] = v->mv_f[dir][xy + v->blocks_off];
1583 v->mv_f[dir][xy + wrap + v->blocks_off] = v->mv_f[dir][xy + wrap + 1 + v->blocks_off] = v->mv_f[dir][xy + v->blocks_off];
1587 /** Predict and set motion vector for interlaced frame picture MBs
1589 static inline void vc1_pred_mv_intfr(VC1Context *v, int n, int dmv_x, int dmv_y,
1590 int mvn, int r_x, int r_y, uint8_t* is_intra)
1592 MpegEncContext *s = &v->s;
1593 int xy, wrap, off = 0;
1594 int A[2], B[2], C[2];
1596 int a_valid = 0, b_valid = 0, c_valid = 0;
1597 int field_a, field_b, field_c; // 0: same, 1: opposit
1598 int total_valid, num_samefield, num_oppfield;
1599 int pos_c, pos_b, n_adj;
1601 wrap = s->b8_stride;
1602 xy = s->block_index[n];
1605 s->mv[0][n][0] = s->current_picture.motion_val[0][xy][0] = 0;
1606 s->mv[0][n][1] = s->current_picture.motion_val[0][xy][1] = 0;
1607 s->current_picture.motion_val[1][xy][0] = 0;
1608 s->current_picture.motion_val[1][xy][1] = 0;
1609 if (mvn == 1) { /* duplicate motion data for 1-MV block */
1610 s->current_picture.motion_val[0][xy + 1][0] = 0;
1611 s->current_picture.motion_val[0][xy + 1][1] = 0;
1612 s->current_picture.motion_val[0][xy + wrap][0] = 0;
1613 s->current_picture.motion_val[0][xy + wrap][1] = 0;
1614 s->current_picture.motion_val[0][xy + wrap + 1][0] = 0;
1615 s->current_picture.motion_val[0][xy + wrap + 1][1] = 0;
1616 v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
1617 s->current_picture.motion_val[1][xy + 1][0] = 0;
1618 s->current_picture.motion_val[1][xy + 1][1] = 0;
1619 s->current_picture.motion_val[1][xy + wrap][0] = 0;
1620 s->current_picture.motion_val[1][xy + wrap][1] = 0;
1621 s->current_picture.motion_val[1][xy + wrap + 1][0] = 0;
1622 s->current_picture.motion_val[1][xy + wrap + 1][1] = 0;
1627 off = ((n == 0) || (n == 1)) ? 1 : -1;
1629 if (s->mb_x || (n == 1) || (n == 3)) {
1630 if ((v->blk_mv_type[xy]) // current block (MB) has a field MV
1631 || (!v->blk_mv_type[xy] && !v->blk_mv_type[xy - 1])) { // or both have frame MV
1632 A[0] = s->current_picture.motion_val[0][xy - 1][0];
1633 A[1] = s->current_picture.motion_val[0][xy - 1][1];
1635 } else { // current block has frame mv and cand. has field MV (so average)
1636 A[0] = (s->current_picture.motion_val[0][xy - 1][0]
1637 + s->current_picture.motion_val[0][xy - 1 + off * wrap][0] + 1) >> 1;
1638 A[1] = (s->current_picture.motion_val[0][xy - 1][1]
1639 + s->current_picture.motion_val[0][xy - 1 + off * wrap][1] + 1) >> 1;
1642 if (!(n & 1) && v->is_intra[s->mb_x - 1]) {
1648 /* Predict B and C */
1649 B[0] = B[1] = C[0] = C[1] = 0;
1650 if (n == 0 || n == 1 || v->blk_mv_type[xy]) {
1651 if (!s->first_slice_line) {
1652 if (!v->is_intra[s->mb_x - s->mb_stride]) {
1655 pos_b = s->block_index[n_adj] - 2 * wrap;
1656 if (v->blk_mv_type[pos_b] && v->blk_mv_type[xy]) {
1657 n_adj = (n & 2) | (n & 1);
1659 B[0] = s->current_picture.motion_val[0][s->block_index[n_adj] - 2 * wrap][0];
1660 B[1] = s->current_picture.motion_val[0][s->block_index[n_adj] - 2 * wrap][1];
1661 if (v->blk_mv_type[pos_b] && !v->blk_mv_type[xy]) {
1662 B[0] = (B[0] + s->current_picture.motion_val[0][s->block_index[n_adj ^ 2] - 2 * wrap][0] + 1) >> 1;
1663 B[1] = (B[1] + s->current_picture.motion_val[0][s->block_index[n_adj ^ 2] - 2 * wrap][1] + 1) >> 1;
1666 if (s->mb_width > 1) {
1667 if (!v->is_intra[s->mb_x - s->mb_stride + 1]) {
1670 pos_c = s->block_index[2] - 2 * wrap + 2;
1671 if (v->blk_mv_type[pos_c] && v->blk_mv_type[xy]) {
1674 C[0] = s->current_picture.motion_val[0][s->block_index[n_adj] - 2 * wrap + 2][0];
1675 C[1] = s->current_picture.motion_val[0][s->block_index[n_adj] - 2 * wrap + 2][1];
1676 if (v->blk_mv_type[pos_c] && !v->blk_mv_type[xy]) {
1677 C[0] = (1 + C[0] + (s->current_picture.motion_val[0][s->block_index[n_adj ^ 2] - 2 * wrap + 2][0])) >> 1;
1678 C[1] = (1 + C[1] + (s->current_picture.motion_val[0][s->block_index[n_adj ^ 2] - 2 * wrap + 2][1])) >> 1;
1680 if (s->mb_x == s->mb_width - 1) {
1681 if (!v->is_intra[s->mb_x - s->mb_stride - 1]) {
1684 pos_c = s->block_index[3] - 2 * wrap - 2;
1685 if (v->blk_mv_type[pos_c] && v->blk_mv_type[xy]) {
1688 C[0] = s->current_picture.motion_val[0][s->block_index[n_adj] - 2 * wrap - 2][0];
1689 C[1] = s->current_picture.motion_val[0][s->block_index[n_adj] - 2 * wrap - 2][1];
1690 if (v->blk_mv_type[pos_c] && !v->blk_mv_type[xy]) {
1691 C[0] = (1 + C[0] + s->current_picture.motion_val[0][s->block_index[1] - 2 * wrap - 2][0]) >> 1;
1692 C[1] = (1 + C[1] + s->current_picture.motion_val[0][s->block_index[1] - 2 * wrap - 2][1]) >> 1;
1701 pos_b = s->block_index[1];
1703 B[0] = s->current_picture.motion_val[0][pos_b][0];
1704 B[1] = s->current_picture.motion_val[0][pos_b][1];
1705 pos_c = s->block_index[0];
1707 C[0] = s->current_picture.motion_val[0][pos_c][0];
1708 C[1] = s->current_picture.motion_val[0][pos_c][1];
1711 total_valid = a_valid + b_valid + c_valid;
1712 // check if predictor A is out of bounds
1713 if (!s->mb_x && !(n == 1 || n == 3)) {
1716 // check if predictor B is out of bounds
1717 if ((s->first_slice_line && v->blk_mv_type[xy]) || (s->first_slice_line && !(n & 2))) {
1718 B[0] = B[1] = C[0] = C[1] = 0;
1720 if (!v->blk_mv_type[xy]) {
1721 if (s->mb_width == 1) {
1725 if (total_valid >= 2) {
1726 px = mid_pred(A[0], B[0], C[0]);
1727 py = mid_pred(A[1], B[1], C[1]);
1728 } else if (total_valid) {
1729 if (a_valid) { px = A[0]; py = A[1]; }
1730 if (b_valid) { px = B[0]; py = B[1]; }
1731 if (c_valid) { px = C[0]; py = C[1]; }
1737 field_a = (A[1] & 4) ? 1 : 0;
1741 field_b = (B[1] & 4) ? 1 : 0;
1745 field_c = (C[1] & 4) ? 1 : 0;
1749 num_oppfield = field_a + field_b + field_c;
1750 num_samefield = total_valid - num_oppfield;
1751 if (total_valid == 3) {
1752 if ((num_samefield == 3) || (num_oppfield == 3)) {
1753 px = mid_pred(A[0], B[0], C[0]);
1754 py = mid_pred(A[1], B[1], C[1]);
1755 } else if (num_samefield >= num_oppfield) {
1756 /* take one MV from same field set depending on priority
1757 the check for B may not be necessary */
1758 px = !field_a ? A[0] : B[0];
1759 py = !field_a ? A[1] : B[1];
1761 px = field_a ? A[0] : B[0];
1762 py = field_a ? A[1] : B[1];
1764 } else if (total_valid == 2) {
1765 if (num_samefield >= num_oppfield) {
1766 if (!field_a && a_valid) {
1769 } else if (!field_b && b_valid) {
1772 } else if (c_valid) {
1777 if (field_a && a_valid) {
1780 } else if (field_b && b_valid) {
1783 } else if (c_valid) {
1788 } else if (total_valid == 1) {
1789 px = (a_valid) ? A[0] : ((b_valid) ? B[0] : C[0]);
1790 py = (a_valid) ? A[1] : ((b_valid) ? B[1] : C[1]);
1795 /* store MV using signed modulus of MV range defined in 4.11 */
1796 s->mv[0][n][0] = s->current_picture.motion_val[0][xy][0] = ((px + dmv_x + r_x) & ((r_x << 1) - 1)) - r_x;
1797 s->mv[0][n][1] = s->current_picture.motion_val[0][xy][1] = ((py + dmv_y + r_y) & ((r_y << 1) - 1)) - r_y;
1798 if (mvn == 1) { /* duplicate motion data for 1-MV block */
1799 s->current_picture.motion_val[0][xy + 1 ][0] = s->current_picture.motion_val[0][xy][0];
1800 s->current_picture.motion_val[0][xy + 1 ][1] = s->current_picture.motion_val[0][xy][1];
1801 s->current_picture.motion_val[0][xy + wrap ][0] = s->current_picture.motion_val[0][xy][0];
1802 s->current_picture.motion_val[0][xy + wrap ][1] = s->current_picture.motion_val[0][xy][1];
1803 s->current_picture.motion_val[0][xy + wrap + 1][0] = s->current_picture.motion_val[0][xy][0];
1804 s->current_picture.motion_val[0][xy + wrap + 1][1] = s->current_picture.motion_val[0][xy][1];
1805 } else if (mvn == 2) { /* duplicate motion data for 2-Field MV block */
1806 s->current_picture.motion_val[0][xy + 1][0] = s->current_picture.motion_val[0][xy][0];
1807 s->current_picture.motion_val[0][xy + 1][1] = s->current_picture.motion_val[0][xy][1];
1808 s->mv[0][n + 1][0] = s->mv[0][n][0];
1809 s->mv[0][n + 1][1] = s->mv[0][n][1];
1813 /** Motion compensation for direct or interpolated blocks in B-frames
1815 static void vc1_interp_mc(VC1Context *v)
1817 MpegEncContext *s = &v->s;
1818 H264ChromaContext *h264chroma = &v->h264chroma;
1819 uint8_t *srcY, *srcU, *srcV;
1820 int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
1822 int v_edge_pos = s->v_edge_pos >> v->field_mode;
1824 if (!v->field_mode && !v->s.next_picture.f.data[0])
1827 mx = s->mv[1][0][0];
1828 my = s->mv[1][0][1];
1829 uvmx = (mx + ((mx & 3) == 3)) >> 1;
1830 uvmy = (my + ((my & 3) == 3)) >> 1;
1831 if (v->field_mode) {
1832 if (v->cur_field_type != v->ref_field_type[1])
1833 my = my - 2 + 4 * v->cur_field_type;
1834 uvmy = uvmy - 2 + 4 * v->cur_field_type;
1837 uvmx = uvmx + ((uvmx < 0) ? -(uvmx & 1) : (uvmx & 1));
1838 uvmy = uvmy + ((uvmy < 0) ? -(uvmy & 1) : (uvmy & 1));
1840 srcY = s->next_picture.f.data[0];
1841 srcU = s->next_picture.f.data[1];
1842 srcV = s->next_picture.f.data[2];
1844 src_x = s->mb_x * 16 + (mx >> 2);
1845 src_y = s->mb_y * 16 + (my >> 2);
1846 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
1847 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
1849 if (v->profile != PROFILE_ADVANCED) {
1850 src_x = av_clip( src_x, -16, s->mb_width * 16);
1851 src_y = av_clip( src_y, -16, s->mb_height * 16);
1852 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
1853 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
1855 src_x = av_clip( src_x, -17, s->avctx->coded_width);
1856 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
1857 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
1858 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
1861 srcY += src_y * s->linesize + src_x;
1862 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
1863 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
1865 if (v->field_mode && v->ref_field_type[1]) {
1866 srcY += s->current_picture_ptr->f.linesize[0];
1867 srcU += s->current_picture_ptr->f.linesize[1];
1868 srcV += s->current_picture_ptr->f.linesize[2];
1871 /* for grayscale we should not try to read from unknown area */
1872 if (s->flags & CODEC_FLAG_GRAY) {
1873 srcU = s->edge_emu_buffer + 18 * s->linesize;
1874 srcV = s->edge_emu_buffer + 18 * s->linesize;
1877 if (v->rangeredfrm || s->h_edge_pos < 22 || v_edge_pos < 22
1878 || (unsigned)(src_x - 1) > s->h_edge_pos - (mx & 3) - 16 - 3
1879 || (unsigned)(src_y - 1) > v_edge_pos - (my & 3) - 16 - 3) {
1880 uint8_t *uvbuf = s->edge_emu_buffer + 19 * s->linesize;
1882 srcY -= s->mspel * (1 + s->linesize);
1883 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize,
1884 17 + s->mspel * 2, 17 + s->mspel * 2,
1885 src_x - s->mspel, src_y - s->mspel,
1886 s->h_edge_pos, v_edge_pos);
1887 srcY = s->edge_emu_buffer;
1888 s->vdsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, 8 + 1, 8 + 1,
1889 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
1890 s->vdsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8 + 1, 8 + 1,
1891 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
1894 /* if we deal with range reduction we need to scale source blocks */
1895 if (v->rangeredfrm) {
1897 uint8_t *src, *src2;
1900 for (j = 0; j < 17 + s->mspel * 2; j++) {
1901 for (i = 0; i < 17 + s->mspel * 2; i++)
1902 src[i] = ((src[i] - 128) >> 1) + 128;
1907 for (j = 0; j < 9; j++) {
1908 for (i = 0; i < 9; i++) {
1909 src[i] = ((src[i] - 128) >> 1) + 128;
1910 src2[i] = ((src2[i] - 128) >> 1) + 128;
1912 src += s->uvlinesize;
1913 src2 += s->uvlinesize;
1916 srcY += s->mspel * (1 + s->linesize);
1923 dxy = ((my & 3) << 2) | (mx & 3);
1924 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off , srcY , s->linesize, v->rnd);
1925 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8, srcY + 8, s->linesize, v->rnd);
1926 srcY += s->linesize * 8;
1927 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize , srcY , s->linesize, v->rnd);
1928 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
1930 dxy = (my & 2) | ((mx & 2) >> 1);
1933 s->hdsp.avg_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
1935 s->hdsp.avg_no_rnd_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize, 16);
1938 if (s->flags & CODEC_FLAG_GRAY) return;
1939 /* Chroma MC always uses qpel blilinear */
1940 uvmx = (uvmx & 3) << 1;
1941 uvmy = (uvmy & 3) << 1;
1943 h264chroma->avg_h264_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
1944 h264chroma->avg_h264_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
1946 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
1947 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
1951 static av_always_inline int scale_mv(int value, int bfrac, int inv, int qs)
1955 #if B_FRACTION_DEN==256
1959 return 2 * ((value * n + 255) >> 9);
1960 return (value * n + 128) >> 8;
1963 n -= B_FRACTION_DEN;
1965 return 2 * ((value * n + B_FRACTION_DEN - 1) / (2 * B_FRACTION_DEN));
1966 return (value * n + B_FRACTION_DEN/2) / B_FRACTION_DEN;
1970 /** Reconstruct motion vector for B-frame and do motion compensation
1972 static inline void vc1_b_mc(VC1Context *v, int dmv_x[2], int dmv_y[2],
1973 int direct, int mode)
1976 v->mv_mode2 = v->mv_mode;
1977 v->mv_mode = MV_PMODE_INTENSITY_COMP;
1983 v->mv_mode = v->mv_mode2;
1986 if (mode == BMV_TYPE_INTERPOLATED) {
1990 v->mv_mode = v->mv_mode2;
1994 if (v->use_ic && (mode == BMV_TYPE_BACKWARD))
1995 v->mv_mode = v->mv_mode2;
1996 vc1_mc_1mv(v, (mode == BMV_TYPE_BACKWARD));
1998 v->mv_mode = v->mv_mode2;
2001 static inline void vc1_pred_b_mv(VC1Context *v, int dmv_x[2], int dmv_y[2],
2002 int direct, int mvtype)
2004 MpegEncContext *s = &v->s;
2005 int xy, wrap, off = 0;
2010 const uint8_t *is_intra = v->mb_type[0];
2014 /* scale MV difference to be quad-pel */
2015 dmv_x[0] <<= 1 - s->quarter_sample;
2016 dmv_y[0] <<= 1 - s->quarter_sample;
2017 dmv_x[1] <<= 1 - s->quarter_sample;
2018 dmv_y[1] <<= 1 - s->quarter_sample;
2020 wrap = s->b8_stride;
2021 xy = s->block_index[0];
2024 s->current_picture.motion_val[0][xy + v->blocks_off][0] =
2025 s->current_picture.motion_val[0][xy + v->blocks_off][1] =
2026 s->current_picture.motion_val[1][xy + v->blocks_off][0] =
2027 s->current_picture.motion_val[1][xy + v->blocks_off][1] = 0;
2030 if (!v->field_mode) {
2031 s->mv[0][0][0] = scale_mv(s->next_picture.motion_val[1][xy][0], v->bfraction, 0, s->quarter_sample);
2032 s->mv[0][0][1] = scale_mv(s->next_picture.motion_val[1][xy][1], v->bfraction, 0, s->quarter_sample);
2033 s->mv[1][0][0] = scale_mv(s->next_picture.motion_val[1][xy][0], v->bfraction, 1, s->quarter_sample);
2034 s->mv[1][0][1] = scale_mv(s->next_picture.motion_val[1][xy][1], v->bfraction, 1, s->quarter_sample);
2036 /* Pullback predicted motion vectors as specified in 8.4.5.4 */
2037 s->mv[0][0][0] = av_clip(s->mv[0][0][0], -60 - (s->mb_x << 6), (s->mb_width << 6) - 4 - (s->mb_x << 6));
2038 s->mv[0][0][1] = av_clip(s->mv[0][0][1], -60 - (s->mb_y << 6), (s->mb_height << 6) - 4 - (s->mb_y << 6));
2039 s->mv[1][0][0] = av_clip(s->mv[1][0][0], -60 - (s->mb_x << 6), (s->mb_width << 6) - 4 - (s->mb_x << 6));
2040 s->mv[1][0][1] = av_clip(s->mv[1][0][1], -60 - (s->mb_y << 6), (s->mb_height << 6) - 4 - (s->mb_y << 6));
2043 s->current_picture.motion_val[0][xy + v->blocks_off][0] = s->mv[0][0][0];
2044 s->current_picture.motion_val[0][xy + v->blocks_off][1] = s->mv[0][0][1];
2045 s->current_picture.motion_val[1][xy + v->blocks_off][0] = s->mv[1][0][0];
2046 s->current_picture.motion_val[1][xy + v->blocks_off][1] = s->mv[1][0][1];
2050 if ((mvtype == BMV_TYPE_FORWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
2051 C = s->current_picture.motion_val[0][xy - 2];
2052 A = s->current_picture.motion_val[0][xy - wrap * 2];
2053 off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
2054 B = s->current_picture.motion_val[0][xy - wrap * 2 + off];
2056 if (!s->mb_x) C[0] = C[1] = 0;
2057 if (!s->first_slice_line) { // predictor A is not out of bounds
2058 if (s->mb_width == 1) {
2062 px = mid_pred(A[0], B[0], C[0]);
2063 py = mid_pred(A[1], B[1], C[1]);
2065 } else if (s->mb_x) { // predictor C is not out of bounds
2071 /* Pullback MV as specified in 8.3.5.3.4 */
2074 if (v->profile < PROFILE_ADVANCED) {
2075 qx = (s->mb_x << 5);
2076 qy = (s->mb_y << 5);
2077 X = (s->mb_width << 5) - 4;
2078 Y = (s->mb_height << 5) - 4;
2079 if (qx + px < -28) px = -28 - qx;
2080 if (qy + py < -28) py = -28 - qy;
2081 if (qx + px > X) px = X - qx;
2082 if (qy + py > Y) py = Y - qy;
2084 qx = (s->mb_x << 6);
2085 qy = (s->mb_y << 6);
2086 X = (s->mb_width << 6) - 4;
2087 Y = (s->mb_height << 6) - 4;
2088 if (qx + px < -60) px = -60 - qx;
2089 if (qy + py < -60) py = -60 - qy;
2090 if (qx + px > X) px = X - qx;
2091 if (qy + py > Y) py = Y - qy;
2094 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
2095 if (0 && !s->first_slice_line && s->mb_x) {
2096 if (is_intra[xy - wrap])
2097 sum = FFABS(px) + FFABS(py);
2099 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
2101 if (get_bits1(&s->gb)) {
2109 if (is_intra[xy - 2])
2110 sum = FFABS(px) + FFABS(py);
2112 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
2114 if (get_bits1(&s->gb)) {
2124 /* store MV using signed modulus of MV range defined in 4.11 */
2125 s->mv[0][0][0] = ((px + dmv_x[0] + r_x) & ((r_x << 1) - 1)) - r_x;
2126 s->mv[0][0][1] = ((py + dmv_y[0] + r_y) & ((r_y << 1) - 1)) - r_y;
2128 if ((mvtype == BMV_TYPE_BACKWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
2129 C = s->current_picture.motion_val[1][xy - 2];
2130 A = s->current_picture.motion_val[1][xy - wrap * 2];
2131 off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
2132 B = s->current_picture.motion_val[1][xy - wrap * 2 + off];
2136 if (!s->first_slice_line) { // predictor A is not out of bounds
2137 if (s->mb_width == 1) {
2141 px = mid_pred(A[0], B[0], C[0]);
2142 py = mid_pred(A[1], B[1], C[1]);
2144 } else if (s->mb_x) { // predictor C is not out of bounds
2150 /* Pullback MV as specified in 8.3.5.3.4 */
2153 if (v->profile < PROFILE_ADVANCED) {
2154 qx = (s->mb_x << 5);
2155 qy = (s->mb_y << 5);
2156 X = (s->mb_width << 5) - 4;
2157 Y = (s->mb_height << 5) - 4;
2158 if (qx + px < -28) px = -28 - qx;
2159 if (qy + py < -28) py = -28 - qy;
2160 if (qx + px > X) px = X - qx;
2161 if (qy + py > Y) py = Y - qy;
2163 qx = (s->mb_x << 6);
2164 qy = (s->mb_y << 6);
2165 X = (s->mb_width << 6) - 4;
2166 Y = (s->mb_height << 6) - 4;
2167 if (qx + px < -60) px = -60 - qx;
2168 if (qy + py < -60) py = -60 - qy;
2169 if (qx + px > X) px = X - qx;
2170 if (qy + py > Y) py = Y - qy;
2173 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
2174 if (0 && !s->first_slice_line && s->mb_x) {
2175 if (is_intra[xy - wrap])
2176 sum = FFABS(px) + FFABS(py);
2178 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
2180 if (get_bits1(&s->gb)) {
2188 if (is_intra[xy - 2])
2189 sum = FFABS(px) + FFABS(py);
2191 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
2193 if (get_bits1(&s->gb)) {
2203 /* store MV using signed modulus of MV range defined in 4.11 */
2205 s->mv[1][0][0] = ((px + dmv_x[1] + r_x) & ((r_x << 1) - 1)) - r_x;
2206 s->mv[1][0][1] = ((py + dmv_y[1] + r_y) & ((r_y << 1) - 1)) - r_y;
2208 s->current_picture.motion_val[0][xy][0] = s->mv[0][0][0];
2209 s->current_picture.motion_val[0][xy][1] = s->mv[0][0][1];
2210 s->current_picture.motion_val[1][xy][0] = s->mv[1][0][0];
2211 s->current_picture.motion_val[1][xy][1] = s->mv[1][0][1];
2214 static inline void vc1_pred_b_mv_intfi(VC1Context *v, int n, int *dmv_x, int *dmv_y, int mv1, int *pred_flag)
2216 int dir = (v->bmvtype == BMV_TYPE_BACKWARD) ? 1 : 0;
2217 MpegEncContext *s = &v->s;
2218 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2220 if (v->bmvtype == BMV_TYPE_DIRECT) {
2221 int total_opp, k, f;
2222 if (s->next_picture.mb_type[mb_pos + v->mb_off] != MB_TYPE_INTRA) {
2223 s->mv[0][0][0] = scale_mv(s->next_picture.motion_val[1][s->block_index[0] + v->blocks_off][0],
2224 v->bfraction, 0, s->quarter_sample);
2225 s->mv[0][0][1] = scale_mv(s->next_picture.motion_val[1][s->block_index[0] + v->blocks_off][1],
2226 v->bfraction, 0, s->quarter_sample);
2227 s->mv[1][0][0] = scale_mv(s->next_picture.motion_val[1][s->block_index[0] + v->blocks_off][0],
2228 v->bfraction, 1, s->quarter_sample);
2229 s->mv[1][0][1] = scale_mv(s->next_picture.motion_val[1][s->block_index[0] + v->blocks_off][1],
2230 v->bfraction, 1, s->quarter_sample);
2232 total_opp = v->mv_f_next[0][s->block_index[0] + v->blocks_off]
2233 + v->mv_f_next[0][s->block_index[1] + v->blocks_off]
2234 + v->mv_f_next[0][s->block_index[2] + v->blocks_off]
2235 + v->mv_f_next[0][s->block_index[3] + v->blocks_off];
2236 f = (total_opp > 2) ? 1 : 0;
2238 s->mv[0][0][0] = s->mv[0][0][1] = 0;
2239 s->mv[1][0][0] = s->mv[1][0][1] = 0;
2242 v->ref_field_type[0] = v->ref_field_type[1] = v->cur_field_type ^ f;
2243 for (k = 0; k < 4; k++) {
2244 s->current_picture.motion_val[0][s->block_index[k] + v->blocks_off][0] = s->mv[0][0][0];
2245 s->current_picture.motion_val[0][s->block_index[k] + v->blocks_off][1] = s->mv[0][0][1];
2246 s->current_picture.motion_val[1][s->block_index[k] + v->blocks_off][0] = s->mv[1][0][0];
2247 s->current_picture.motion_val[1][s->block_index[k] + v->blocks_off][1] = s->mv[1][0][1];
2248 v->mv_f[0][s->block_index[k] + v->blocks_off] = f;
2249 v->mv_f[1][s->block_index[k] + v->blocks_off] = f;
2253 if (v->bmvtype == BMV_TYPE_INTERPOLATED) {
2254 vc1_pred_mv(v, 0, dmv_x[0], dmv_y[0], 1, v->range_x, v->range_y, v->mb_type[0], pred_flag[0], 0);
2255 vc1_pred_mv(v, 0, dmv_x[1], dmv_y[1], 1, v->range_x, v->range_y, v->mb_type[0], pred_flag[1], 1);
2258 if (dir) { // backward
2259 vc1_pred_mv(v, n, dmv_x[1], dmv_y[1], mv1, v->range_x, v->range_y, v->mb_type[0], pred_flag[1], 1);
2260 if (n == 3 || mv1) {
2261 vc1_pred_mv(v, 0, dmv_x[0], dmv_y[0], 1, v->range_x, v->range_y, v->mb_type[0], 0, 0);
2264 vc1_pred_mv(v, n, dmv_x[0], dmv_y[0], mv1, v->range_x, v->range_y, v->mb_type[0], pred_flag[0], 0);
2265 if (n == 3 || mv1) {
2266 vc1_pred_mv(v, 0, dmv_x[1], dmv_y[1], 1, v->range_x, v->range_y, v->mb_type[0], 0, 1);
2271 /** Get predicted DC value for I-frames only
2272 * prediction dir: left=0, top=1
2273 * @param s MpegEncContext
2274 * @param overlap flag indicating that overlap filtering is used
2275 * @param pq integer part of picture quantizer
2276 * @param[in] n block index in the current MB
2277 * @param dc_val_ptr Pointer to DC predictor
2278 * @param dir_ptr Prediction direction for use in AC prediction
2280 static inline int vc1_i_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
2281 int16_t **dc_val_ptr, int *dir_ptr)
2283 int a, b, c, wrap, pred, scale;
2285 static const uint16_t dcpred[32] = {
2286 -1, 1024, 512, 341, 256, 205, 171, 146, 128,
2287 114, 102, 93, 85, 79, 73, 68, 64,
2288 60, 57, 54, 51, 49, 47, 45, 43,
2289 41, 39, 38, 37, 35, 34, 33
2292 /* find prediction - wmv3_dc_scale always used here in fact */
2293 if (n < 4) scale = s->y_dc_scale;
2294 else scale = s->c_dc_scale;
2296 wrap = s->block_wrap[n];
2297 dc_val = s->dc_val[0] + s->block_index[n];
2303 b = dc_val[ - 1 - wrap];
2304 a = dc_val[ - wrap];
2306 if (pq < 9 || !overlap) {
2307 /* Set outer values */
2308 if (s->first_slice_line && (n != 2 && n != 3))
2309 b = a = dcpred[scale];
2310 if (s->mb_x == 0 && (n != 1 && n != 3))
2311 b = c = dcpred[scale];
2313 /* Set outer values */
2314 if (s->first_slice_line && (n != 2 && n != 3))
2316 if (s->mb_x == 0 && (n != 1 && n != 3))
2320 if (abs(a - b) <= abs(b - c)) {
2322 *dir_ptr = 1; // left
2325 *dir_ptr = 0; // top
2328 /* update predictor */
2329 *dc_val_ptr = &dc_val[0];
2334 /** Get predicted DC value
2335 * prediction dir: left=0, top=1
2336 * @param s MpegEncContext
2337 * @param overlap flag indicating that overlap filtering is used
2338 * @param pq integer part of picture quantizer
2339 * @param[in] n block index in the current MB
2340 * @param a_avail flag indicating top block availability
2341 * @param c_avail flag indicating left block availability
2342 * @param dc_val_ptr Pointer to DC predictor
2343 * @param dir_ptr Prediction direction for use in AC prediction
2345 static inline int vc1_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
2346 int a_avail, int c_avail,
2347 int16_t **dc_val_ptr, int *dir_ptr)
2349 int a, b, c, wrap, pred;
2351 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2355 wrap = s->block_wrap[n];
2356 dc_val = s->dc_val[0] + s->block_index[n];
2362 b = dc_val[ - 1 - wrap];
2363 a = dc_val[ - wrap];
2364 /* scale predictors if needed */
2365 q1 = s->current_picture.qscale_table[mb_pos];
2366 dqscale_index = s->y_dc_scale_table[q1] - 1;
2367 if (dqscale_index < 0)
2369 if (c_avail && (n != 1 && n != 3)) {
2370 q2 = s->current_picture.qscale_table[mb_pos - 1];
2372 c = (c * s->y_dc_scale_table[q2] * ff_vc1_dqscale[dqscale_index] + 0x20000) >> 18;
2374 if (a_avail && (n != 2 && n != 3)) {
2375 q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
2377 a = (a * s->y_dc_scale_table[q2] * ff_vc1_dqscale[dqscale_index] + 0x20000) >> 18;
2379 if (a_avail && c_avail && (n != 3)) {
2384 off -= s->mb_stride;
2385 q2 = s->current_picture.qscale_table[off];
2387 b = (b * s->y_dc_scale_table[q2] * ff_vc1_dqscale[dqscale_index] + 0x20000) >> 18;
2390 if (a_avail && c_avail) {
2391 if (abs(a - b) <= abs(b - c)) {
2393 *dir_ptr = 1; // left
2396 *dir_ptr = 0; // top
2398 } else if (a_avail) {
2400 *dir_ptr = 0; // top
2401 } else if (c_avail) {
2403 *dir_ptr = 1; // left
2406 *dir_ptr = 1; // left
2409 /* update predictor */
2410 *dc_val_ptr = &dc_val[0];
2414 /** @} */ // Block group
2417 * @name VC1 Macroblock-level functions in Simple/Main Profiles
2418 * @see 7.1.4, p91 and 8.1.1.7, p(1)04
2422 static inline int vc1_coded_block_pred(MpegEncContext * s, int n,
2423 uint8_t **coded_block_ptr)
2425 int xy, wrap, pred, a, b, c;
2427 xy = s->block_index[n];
2428 wrap = s->b8_stride;
2433 a = s->coded_block[xy - 1 ];
2434 b = s->coded_block[xy - 1 - wrap];
2435 c = s->coded_block[xy - wrap];
2444 *coded_block_ptr = &s->coded_block[xy];
2450 * Decode one AC coefficient
2451 * @param v The VC1 context
2452 * @param last Last coefficient
2453 * @param skip How much zero coefficients to skip
2454 * @param value Decoded AC coefficient value
2455 * @param codingset set of VLC to decode data
2458 static void vc1_decode_ac_coeff(VC1Context *v, int *last, int *skip,
2459 int *value, int codingset)
2461 GetBitContext *gb = &v->s.gb;
2462 int index, escape, run = 0, level = 0, lst = 0;
2464 index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
2465 if (index != ff_vc1_ac_sizes[codingset] - 1) {
2466 run = vc1_index_decode_table[codingset][index][0];
2467 level = vc1_index_decode_table[codingset][index][1];
2468 lst = index >= vc1_last_decode_table[codingset] || get_bits_left(gb) < 0;
2472 escape = decode210(gb);
2474 index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
2475 run = vc1_index_decode_table[codingset][index][0];
2476 level = vc1_index_decode_table[codingset][index][1];
2477 lst = index >= vc1_last_decode_table[codingset];
2480 level += vc1_last_delta_level_table[codingset][run];
2482 level += vc1_delta_level_table[codingset][run];
2485 run += vc1_last_delta_run_table[codingset][level] + 1;
2487 run += vc1_delta_run_table[codingset][level] + 1;
2493 lst = get_bits1(gb);
2494 if (v->s.esc3_level_length == 0) {
2495 if (v->pq < 8 || v->dquantfrm) { // table 59
2496 v->s.esc3_level_length = get_bits(gb, 3);
2497 if (!v->s.esc3_level_length)
2498 v->s.esc3_level_length = get_bits(gb, 2) + 8;
2499 } else { // table 60
2500 v->s.esc3_level_length = get_unary(gb, 1, 6) + 2;
2502 v->s.esc3_run_length = 3 + get_bits(gb, 2);
2504 run = get_bits(gb, v->s.esc3_run_length);
2505 sign = get_bits1(gb);
2506 level = get_bits(gb, v->s.esc3_level_length);
2517 /** Decode intra block in intra frames - should be faster than decode_intra_block
2518 * @param v VC1Context
2519 * @param block block to decode
2520 * @param[in] n subblock index
2521 * @param coded are AC coeffs present or not
2522 * @param codingset set of VLC to decode data
2524 static int vc1_decode_i_block(VC1Context *v, int16_t block[64], int n,
2525 int coded, int codingset)
2527 GetBitContext *gb = &v->s.gb;
2528 MpegEncContext *s = &v->s;
2529 int dc_pred_dir = 0; /* Direction of the DC prediction used */
2532 int16_t *ac_val, *ac_val2;
2535 /* Get DC differential */
2537 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2539 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2542 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
2546 if (dcdiff == 119 /* ESC index value */) {
2547 /* TODO: Optimize */
2548 if (v->pq == 1) dcdiff = get_bits(gb, 10);
2549 else if (v->pq == 2) dcdiff = get_bits(gb, 9);
2550 else dcdiff = get_bits(gb, 8);
2553 dcdiff = (dcdiff << 2) + get_bits(gb, 2) - 3;
2554 else if (v->pq == 2)
2555 dcdiff = (dcdiff << 1) + get_bits1(gb) - 1;
2562 dcdiff += vc1_i_pred_dc(&v->s, v->overlap, v->pq, n, &dc_val, &dc_pred_dir);
2565 /* Store the quantized DC coeff, used for prediction */
2567 block[0] = dcdiff * s->y_dc_scale;
2569 block[0] = dcdiff * s->c_dc_scale;
2580 int last = 0, skip, value;
2581 const uint8_t *zz_table;
2585 scale = v->pq * 2 + v->halfpq;
2589 zz_table = v->zz_8x8[2];
2591 zz_table = v->zz_8x8[3];
2593 zz_table = v->zz_8x8[1];
2595 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
2597 if (dc_pred_dir) // left
2600 ac_val -= 16 * s->block_wrap[n];
2603 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
2607 block[zz_table[i++]] = value;
2610 /* apply AC prediction if needed */
2612 if (dc_pred_dir) { // left
2613 for (k = 1; k < 8; k++)
2614 block[k << v->left_blk_sh] += ac_val[k];
2616 for (k = 1; k < 8; k++)
2617 block[k << v->top_blk_sh] += ac_val[k + 8];
2620 /* save AC coeffs for further prediction */
2621 for (k = 1; k < 8; k++) {
2622 ac_val2[k] = block[k << v->left_blk_sh];
2623 ac_val2[k + 8] = block[k << v->top_blk_sh];
2626 /* scale AC coeffs */
2627 for (k = 1; k < 64; k++)
2631 block[k] += (block[k] < 0) ? -v->pq : v->pq;
2634 if (s->ac_pred) i = 63;
2640 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
2644 scale = v->pq * 2 + v->halfpq;
2645 memset(ac_val2, 0, 16 * 2);
2646 if (dc_pred_dir) { // left
2649 memcpy(ac_val2, ac_val, 8 * 2);
2651 ac_val -= 16 * s->block_wrap[n];
2653 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
2656 /* apply AC prediction if needed */
2658 if (dc_pred_dir) { //left
2659 for (k = 1; k < 8; k++) {
2660 block[k << v->left_blk_sh] = ac_val[k] * scale;
2661 if (!v->pquantizer && block[k << v->left_blk_sh])
2662 block[k << v->left_blk_sh] += (block[k << v->left_blk_sh] < 0) ? -v->pq : v->pq;
2665 for (k = 1; k < 8; k++) {
2666 block[k << v->top_blk_sh] = ac_val[k + 8] * scale;
2667 if (!v->pquantizer && block[k << v->top_blk_sh])
2668 block[k << v->top_blk_sh] += (block[k << v->top_blk_sh] < 0) ? -v->pq : v->pq;
2674 s->block_last_index[n] = i;
2679 /** Decode intra block in intra frames - should be faster than decode_intra_block
2680 * @param v VC1Context
2681 * @param block block to decode
2682 * @param[in] n subblock number
2683 * @param coded are AC coeffs present or not
2684 * @param codingset set of VLC to decode data
2685 * @param mquant quantizer value for this macroblock
2687 static int vc1_decode_i_block_adv(VC1Context *v, int16_t block[64], int n,
2688 int coded, int codingset, int mquant)
2690 GetBitContext *gb = &v->s.gb;
2691 MpegEncContext *s = &v->s;
2692 int dc_pred_dir = 0; /* Direction of the DC prediction used */
2695 int16_t *ac_val, *ac_val2;
2697 int a_avail = v->a_avail, c_avail = v->c_avail;
2698 int use_pred = s->ac_pred;
2701 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2703 /* Get DC differential */
2705 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2707 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2710 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
2714 if (dcdiff == 119 /* ESC index value */) {
2715 /* TODO: Optimize */
2716 if (mquant == 1) dcdiff = get_bits(gb, 10);
2717 else if (mquant == 2) dcdiff = get_bits(gb, 9);
2718 else dcdiff = get_bits(gb, 8);
2721 dcdiff = (dcdiff << 2) + get_bits(gb, 2) - 3;
2722 else if (mquant == 2)
2723 dcdiff = (dcdiff << 1) + get_bits1(gb) - 1;
2730 dcdiff += vc1_pred_dc(&v->s, v->overlap, mquant, n, v->a_avail, v->c_avail, &dc_val, &dc_pred_dir);
2733 /* Store the quantized DC coeff, used for prediction */
2735 block[0] = dcdiff * s->y_dc_scale;
2737 block[0] = dcdiff * s->c_dc_scale;
2743 /* check if AC is needed at all */
2744 if (!a_avail && !c_avail)
2746 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
2749 scale = mquant * 2 + ((mquant == v->pq) ? v->halfpq : 0);
2751 if (dc_pred_dir) // left
2754 ac_val -= 16 * s->block_wrap[n];
2756 q1 = s->current_picture.qscale_table[mb_pos];
2757 if ( dc_pred_dir && c_avail && mb_pos)
2758 q2 = s->current_picture.qscale_table[mb_pos - 1];
2759 if (!dc_pred_dir && a_avail && mb_pos >= s->mb_stride)
2760 q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
2761 if ( dc_pred_dir && n == 1)
2763 if (!dc_pred_dir && n == 2)
2769 int last = 0, skip, value;
2770 const uint8_t *zz_table;
2774 if (!use_pred && v->fcm == ILACE_FRAME) {
2775 zz_table = v->zzi_8x8;
2777 if (!dc_pred_dir) // top
2778 zz_table = v->zz_8x8[2];
2780 zz_table = v->zz_8x8[3];
2783 if (v->fcm != ILACE_FRAME)
2784 zz_table = v->zz_8x8[1];
2786 zz_table = v->zzi_8x8;
2790 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
2794 block[zz_table[i++]] = value;
2797 /* apply AC prediction if needed */
2799 /* scale predictors if needed*/
2800 if (q2 && q1 != q2) {
2801 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2802 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2805 return AVERROR_INVALIDDATA;
2806 if (dc_pred_dir) { // left
2807 for (k = 1; k < 8; k++)
2808 block[k << v->left_blk_sh] += (ac_val[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2810 for (k = 1; k < 8; k++)
2811 block[k << v->top_blk_sh] += (ac_val[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2814 if (dc_pred_dir) { //left
2815 for (k = 1; k < 8; k++)
2816 block[k << v->left_blk_sh] += ac_val[k];
2818 for (k = 1; k < 8; k++)
2819 block[k << v->top_blk_sh] += ac_val[k + 8];
2823 /* save AC coeffs for further prediction */
2824 for (k = 1; k < 8; k++) {
2825 ac_val2[k ] = block[k << v->left_blk_sh];
2826 ac_val2[k + 8] = block[k << v->top_blk_sh];
2829 /* scale AC coeffs */
2830 for (k = 1; k < 64; k++)
2834 block[k] += (block[k] < 0) ? -mquant : mquant;
2837 if (use_pred) i = 63;
2838 } else { // no AC coeffs
2841 memset(ac_val2, 0, 16 * 2);
2842 if (dc_pred_dir) { // left
2844 memcpy(ac_val2, ac_val, 8 * 2);
2845 if (q2 && q1 != q2) {
2846 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2847 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2849 return AVERROR_INVALIDDATA;
2850 for (k = 1; k < 8; k++)
2851 ac_val2[k] = (ac_val2[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2856 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
2857 if (q2 && q1 != q2) {
2858 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2859 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2861 return AVERROR_INVALIDDATA;
2862 for (k = 1; k < 8; k++)
2863 ac_val2[k + 8] = (ac_val2[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2868 /* apply AC prediction if needed */
2870 if (dc_pred_dir) { // left
2871 for (k = 1; k < 8; k++) {
2872 block[k << v->left_blk_sh] = ac_val2[k] * scale;
2873 if (!v->pquantizer && block[k << v->left_blk_sh])
2874 block[k << v->left_blk_sh] += (block[k << v->left_blk_sh] < 0) ? -mquant : mquant;
2877 for (k = 1; k < 8; k++) {
2878 block[k << v->top_blk_sh] = ac_val2[k + 8] * scale;
2879 if (!v->pquantizer && block[k << v->top_blk_sh])
2880 block[k << v->top_blk_sh] += (block[k << v->top_blk_sh] < 0) ? -mquant : mquant;
2886 s->block_last_index[n] = i;
2891 /** Decode intra block in inter frames - more generic version than vc1_decode_i_block
2892 * @param v VC1Context
2893 * @param block block to decode
2894 * @param[in] n subblock index
2895 * @param coded are AC coeffs present or not
2896 * @param mquant block quantizer
2897 * @param codingset set of VLC to decode data
2899 static int vc1_decode_intra_block(VC1Context *v, int16_t block[64], int n,
2900 int coded, int mquant, int codingset)
2902 GetBitContext *gb = &v->s.gb;
2903 MpegEncContext *s = &v->s;
2904 int dc_pred_dir = 0; /* Direction of the DC prediction used */
2907 int16_t *ac_val, *ac_val2;
2909 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2910 int a_avail = v->a_avail, c_avail = v->c_avail;
2911 int use_pred = s->ac_pred;
2915 s->dsp.clear_block(block);
2917 /* XXX: Guard against dumb values of mquant */
2918 mquant = (mquant < 1) ? 0 : ((mquant > 31) ? 31 : mquant);
2920 /* Set DC scale - y and c use the same */
2921 s->y_dc_scale = s->y_dc_scale_table[mquant];
2922 s->c_dc_scale = s->c_dc_scale_table[mquant];
2924 /* Get DC differential */
2926 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2928 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2931 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
2935 if (dcdiff == 119 /* ESC index value */) {
2936 /* TODO: Optimize */
2937 if (mquant == 1) dcdiff = get_bits(gb, 10);
2938 else if (mquant == 2) dcdiff = get_bits(gb, 9);
2939 else dcdiff = get_bits(gb, 8);
2942 dcdiff = (dcdiff << 2) + get_bits(gb, 2) - 3;
2943 else if (mquant == 2)
2944 dcdiff = (dcdiff << 1) + get_bits1(gb) - 1;
2951 dcdiff += vc1_pred_dc(&v->s, v->overlap, mquant, n, a_avail, c_avail, &dc_val, &dc_pred_dir);
2954 /* Store the quantized DC coeff, used for prediction */
2957 block[0] = dcdiff * s->y_dc_scale;
2959 block[0] = dcdiff * s->c_dc_scale;
2965 /* check if AC is needed at all and adjust direction if needed */
2966 if (!a_avail) dc_pred_dir = 1;
2967 if (!c_avail) dc_pred_dir = 0;
2968 if (!a_avail && !c_avail) use_pred = 0;
2969 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
2972 scale = mquant * 2 + v->halfpq;
2974 if (dc_pred_dir) //left
2977 ac_val -= 16 * s->block_wrap[n];
2979 q1 = s->current_picture.qscale_table[mb_pos];
2980 if (dc_pred_dir && c_avail && mb_pos)
2981 q2 = s->current_picture.qscale_table[mb_pos - 1];
2982 if (!dc_pred_dir && a_avail && mb_pos >= s->mb_stride)
2983 q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
2984 if ( dc_pred_dir && n == 1)
2986 if (!dc_pred_dir && n == 2)
2988 if (n == 3) q2 = q1;
2991 int last = 0, skip, value;
2995 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
2999 if (v->fcm == PROGRESSIVE)
3000 block[v->zz_8x8[0][i++]] = value;
3002 if (use_pred && (v->fcm == ILACE_FRAME)) {
3003 if (!dc_pred_dir) // top
3004 block[v->zz_8x8[2][i++]] = value;
3006 block[v->zz_8x8[3][i++]] = value;
3008 block[v->zzi_8x8[i++]] = value;
3013 /* apply AC prediction if needed */
3015 /* scale predictors if needed*/
3016 if (q2 && q1 != q2) {
3017 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
3018 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
3021 return AVERROR_INVALIDDATA;
3022 if (dc_pred_dir) { // left
3023 for (k = 1; k < 8; k++)
3024 block[k << v->left_blk_sh] += (ac_val[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
3026 for (k = 1; k < 8; k++)
3027 block[k << v->top_blk_sh] += (ac_val[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
3030 if (dc_pred_dir) { // left
3031 for (k = 1; k < 8; k++)
3032 block[k << v->left_blk_sh] += ac_val[k];
3034 for (k = 1; k < 8; k++)
3035 block[k << v->top_blk_sh] += ac_val[k + 8];
3039 /* save AC coeffs for further prediction */
3040 for (k = 1; k < 8; k++) {
3041 ac_val2[k ] = block[k << v->left_blk_sh];
3042 ac_val2[k + 8] = block[k << v->top_blk_sh];
3045 /* scale AC coeffs */
3046 for (k = 1; k < 64; k++)
3050 block[k] += (block[k] < 0) ? -mquant : mquant;
3053 if (use_pred) i = 63;
3054 } else { // no AC coeffs
3057 memset(ac_val2, 0, 16 * 2);
3058 if (dc_pred_dir) { // left
3060 memcpy(ac_val2, ac_val, 8 * 2);
3061 if (q2 && q1 != q2) {
3062 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
3063 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
3065 return AVERROR_INVALIDDATA;
3066 for (k = 1; k < 8; k++)
3067 ac_val2[k] = (ac_val2[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
3072 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
3073 if (q2 && q1 != q2) {
3074 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
3075 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
3077 return AVERROR_INVALIDDATA;
3078 for (k = 1; k < 8; k++)
3079 ac_val2[k + 8] = (ac_val2[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
3084 /* apply AC prediction if needed */
3086 if (dc_pred_dir) { // left
3087 for (k = 1; k < 8; k++) {
3088 block[k << v->left_blk_sh] = ac_val2[k] * scale;
3089 if (!v->pquantizer && block[k << v->left_blk_sh])
3090 block[k << v->left_blk_sh] += (block[k << v->left_blk_sh] < 0) ? -mquant : mquant;
3093 for (k = 1; k < 8; k++) {
3094 block[k << v->top_blk_sh] = ac_val2[k + 8] * scale;
3095 if (!v->pquantizer && block[k << v->top_blk_sh])
3096 block[k << v->top_blk_sh] += (block[k << v->top_blk_sh] < 0) ? -mquant : mquant;
3102 s->block_last_index[n] = i;
3109 static int vc1_decode_p_block(VC1Context *v, int16_t block[64], int n,
3110 int mquant, int ttmb, int first_block,
3111 uint8_t *dst, int linesize, int skip_block,
3114 MpegEncContext *s = &v->s;
3115 GetBitContext *gb = &s->gb;
3118 int scale, off, idx, last, skip, value;
3119 int ttblk = ttmb & 7;
3122 s->dsp.clear_block(block);
3125 ttblk = ff_vc1_ttblk_to_tt[v->tt_index][get_vlc2(gb, ff_vc1_ttblk_vlc[v->tt_index].table, VC1_TTBLK_VLC_BITS, 1)];
3127 if (ttblk == TT_4X4) {
3128 subblkpat = ~(get_vlc2(gb, ff_vc1_subblkpat_vlc[v->tt_index].table, VC1_SUBBLKPAT_VLC_BITS, 1) + 1);
3130 if ((ttblk != TT_8X8 && ttblk != TT_4X4)
3131 && ((v->ttmbf || (ttmb != -1 && (ttmb & 8) && !first_block))
3132 || (!v->res_rtm_flag && !first_block))) {
3133 subblkpat = decode012(gb);
3135 subblkpat ^= 3; // swap decoded pattern bits
3136 if (ttblk == TT_8X4_TOP || ttblk == TT_8X4_BOTTOM)
3138 if (ttblk == TT_4X8_RIGHT || ttblk == TT_4X8_LEFT)
3141 scale = 2 * mquant + ((v->pq == mquant) ? v->halfpq : 0);
3143 // convert transforms like 8X4_TOP to generic TT and SUBBLKPAT
3144 if (ttblk == TT_8X4_TOP || ttblk == TT_8X4_BOTTOM) {
3145 subblkpat = 2 - (ttblk == TT_8X4_TOP);
3148 if (ttblk == TT_4X8_RIGHT || ttblk == TT_4X8_LEFT) {
3149 subblkpat = 2 - (ttblk == TT_4X8_LEFT);
3158 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
3163 idx = v->zz_8x8[0][i++];
3165 idx = v->zzi_8x8[i++];
3166 block[idx] = value * scale;
3168 block[idx] += (block[idx] < 0) ? -mquant : mquant;
3172 v->vc1dsp.vc1_inv_trans_8x8_dc(dst, linesize, block);
3174 v->vc1dsp.vc1_inv_trans_8x8(block);
3175 s->dsp.add_pixels_clamped(block, dst, linesize);
3180 pat = ~subblkpat & 0xF;
3181 for (j = 0; j < 4; j++) {
3182 last = subblkpat & (1 << (3 - j));
3184 off = (j & 1) * 4 + (j & 2) * 16;
3186 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
3191 idx = ff_vc1_simple_progressive_4x4_zz[i++];
3193 idx = ff_vc1_adv_interlaced_4x4_zz[i++];
3194 block[idx + off] = value * scale;
3196 block[idx + off] += (block[idx + off] < 0) ? -mquant : mquant;
3198 if (!(subblkpat & (1 << (3 - j))) && !skip_block) {
3200 v->vc1dsp.vc1_inv_trans_4x4_dc(dst + (j & 1) * 4 + (j & 2) * 2 * linesize, linesize, block + off);
3202 v->vc1dsp.vc1_inv_trans_4x4(dst + (j & 1) * 4 + (j & 2) * 2 * linesize, linesize, block + off);
3207 pat = ~((subblkpat & 2) * 6 + (subblkpat & 1) * 3) & 0xF;
3208 for (j = 0; j < 2; j++) {
3209 last = subblkpat & (1 << (1 - j));
3213 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
3218 idx = v->zz_8x4[i++] + off;
3220 idx = ff_vc1_adv_interlaced_8x4_zz[i++] + off;
3221 block[idx] = value * scale;
3223 block[idx] += (block[idx] < 0) ? -mquant : mquant;
3225 if (!(subblkpat & (1 << (1 - j))) && !skip_block) {
3227 v->vc1dsp.vc1_inv_trans_8x4_dc(dst + j * 4 * linesize, linesize, block + off);
3229 v->vc1dsp.vc1_inv_trans_8x4(dst + j * 4 * linesize, linesize, block + off);
3234 pat = ~(subblkpat * 5) & 0xF;
3235 for (j = 0; j < 2; j++) {
3236 last = subblkpat & (1 << (1 - j));
3240 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
3245 idx = v->zz_4x8[i++] + off;
3247 idx = ff_vc1_adv_interlaced_4x8_zz[i++] + off;
3248 block[idx] = value * scale;
3250 block[idx] += (block[idx] < 0) ? -mquant : mquant;
3252 if (!(subblkpat & (1 << (1 - j))) && !skip_block) {
3254 v->vc1dsp.vc1_inv_trans_4x8_dc(dst + j * 4, linesize, block + off);
3256 v->vc1dsp.vc1_inv_trans_4x8(dst + j*4, linesize, block + off);
3262 *ttmb_out |= ttblk << (n * 4);
3266 /** @} */ // Macroblock group
3268 static const int size_table [6] = { 0, 2, 3, 4, 5, 8 };
3269 static const int offset_table[6] = { 0, 1, 3, 7, 15, 31 };
3271 static av_always_inline void vc1_apply_p_v_loop_filter(VC1Context *v, int block_num)
3273 MpegEncContext *s = &v->s;
3274 int mb_cbp = v->cbp[s->mb_x - s->mb_stride],
3275 block_cbp = mb_cbp >> (block_num * 4), bottom_cbp,
3276 mb_is_intra = v->is_intra[s->mb_x - s->mb_stride],
3277 block_is_intra = mb_is_intra >> (block_num * 4), bottom_is_intra;
3278 int idx, linesize = block_num > 3 ? s->uvlinesize : s->linesize, ttblk;
3281 if (block_num > 3) {
3282 dst = s->dest[block_num - 3];
3284 dst = s->dest[0] + (block_num & 1) * 8 + ((block_num & 2) * 4 - 8) * linesize;
3286 if (s->mb_y != s->end_mb_y || block_num < 2) {
3290 if (block_num > 3) {
3291 bottom_cbp = v->cbp[s->mb_x] >> (block_num * 4);
3292 bottom_is_intra = v->is_intra[s->mb_x] >> (block_num * 4);
3293 mv = &v->luma_mv[s->mb_x - s->mb_stride];
3294 mv_stride = s->mb_stride;
3296 bottom_cbp = (block_num < 2) ? (mb_cbp >> ((block_num + 2) * 4))
3297 : (v->cbp[s->mb_x] >> ((block_num - 2) * 4));
3298 bottom_is_intra = (block_num < 2) ? (mb_is_intra >> ((block_num + 2) * 4))
3299 : (v->is_intra[s->mb_x] >> ((block_num - 2) * 4));
3300 mv_stride = s->b8_stride;
3301 mv = &s->current_picture.motion_val[0][s->block_index[block_num] - 2 * mv_stride];
3304 if (bottom_is_intra & 1 || block_is_intra & 1 ||
3305 mv[0][0] != mv[mv_stride][0] || mv[0][1] != mv[mv_stride][1]) {
3306 v->vc1dsp.vc1_v_loop_filter8(dst, linesize, v->pq);
3308 idx = ((bottom_cbp >> 2) | block_cbp) & 3;
3310 v->vc1dsp.vc1_v_loop_filter8(dst, linesize, v->pq);
3313 v->vc1dsp.vc1_v_loop_filter4(dst + 4, linesize, v->pq);
3315 v->vc1dsp.vc1_v_loop_filter4(dst, linesize, v->pq);
3320 dst -= 4 * linesize;
3321 ttblk = (v->ttblk[s->mb_x - s->mb_stride] >> (block_num * 4)) & 0xF;
3322 if (ttblk == TT_4X4 || ttblk == TT_8X4) {
3323 idx = (block_cbp | (block_cbp >> 2)) & 3;
3325 v->vc1dsp.vc1_v_loop_filter8(dst, linesize, v->pq);