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
32 #include "mpegvideo.h"
36 #include "vc1acdata.h"
37 #include "msmpeg4data.h"
40 #include "vdpau_internal.h"
45 #define MB_INTRA_VLC_BITS 9
49 // offset tables for interlaced picture MVDATA decoding
50 static const int offset_table1[9] = { 0, 1, 2, 4, 8, 16, 32, 64, 128 };
51 static const int offset_table2[9] = { 0, 1, 3, 7, 15, 31, 63, 127, 255 };
53 /***********************************************************************/
55 * @name VC-1 Bitplane decoding
73 /** @} */ //imode defines
76 /** @} */ //Bitplane group
78 static void vc1_put_signed_blocks_clamped(VC1Context *v)
80 MpegEncContext *s = &v->s;
81 int topleft_mb_pos, top_mb_pos;
82 int stride_y, fieldtx;
85 /* The put pixels loop is always one MB row behind the decoding loop,
86 * because we can only put pixels when overlap filtering is done, and
87 * for filtering of the bottom edge of a MB, we need the next MB row
89 * Within the row, the put pixels loop is also one MB col behind the
90 * decoding loop. The reason for this is again, because for filtering
91 * of the right MB edge, we need the next MB present. */
92 if (!s->first_slice_line) {
94 topleft_mb_pos = (s->mb_y - 1) * s->mb_stride + s->mb_x - 1;
95 fieldtx = v->fieldtx_plane[topleft_mb_pos];
96 stride_y = s->linesize << fieldtx;
97 v_dist = (16 - fieldtx) >> (fieldtx == 0);
98 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][0],
99 s->dest[0] - 16 * s->linesize - 16,
101 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][1],
102 s->dest[0] - 16 * s->linesize - 8,
104 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][2],
105 s->dest[0] - v_dist * s->linesize - 16,
107 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][3],
108 s->dest[0] - v_dist * s->linesize - 8,
110 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][4],
111 s->dest[1] - 8 * s->uvlinesize - 8,
113 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][5],
114 s->dest[2] - 8 * s->uvlinesize - 8,
117 if (s->mb_x == s->mb_width - 1) {
118 top_mb_pos = (s->mb_y - 1) * s->mb_stride + s->mb_x;
119 fieldtx = v->fieldtx_plane[top_mb_pos];
120 stride_y = s->linesize << fieldtx;
121 v_dist = fieldtx ? 15 : 8;
122 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][0],
123 s->dest[0] - 16 * s->linesize,
125 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][1],
126 s->dest[0] - 16 * s->linesize + 8,
128 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][2],
129 s->dest[0] - v_dist * s->linesize,
131 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][3],
132 s->dest[0] - v_dist * s->linesize + 8,
134 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][4],
135 s->dest[1] - 8 * s->uvlinesize,
137 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][5],
138 s->dest[2] - 8 * s->uvlinesize,
143 #define inc_blk_idx(idx) do { \
145 if (idx >= v->n_allocated_blks) \
149 inc_blk_idx(v->topleft_blk_idx);
150 inc_blk_idx(v->top_blk_idx);
151 inc_blk_idx(v->left_blk_idx);
152 inc_blk_idx(v->cur_blk_idx);
155 static void vc1_loop_filter_iblk(VC1Context *v, int pq)
157 MpegEncContext *s = &v->s;
159 if (!s->first_slice_line) {
160 v->vc1dsp.vc1_v_loop_filter16(s->dest[0], s->linesize, pq);
162 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
163 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize + 8, s->linesize, pq);
164 for (j = 0; j < 2; j++) {
165 v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1], s->uvlinesize, pq);
167 v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
170 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] + 8 * s->linesize, s->linesize, pq);
172 if (s->mb_y == s->end_mb_y - 1) {
174 v->vc1dsp.vc1_h_loop_filter16(s->dest[0], s->linesize, pq);
175 v->vc1dsp.vc1_h_loop_filter8(s->dest[1], s->uvlinesize, pq);
176 v->vc1dsp.vc1_h_loop_filter8(s->dest[2], s->uvlinesize, pq);
178 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] + 8, s->linesize, pq);
182 static void vc1_loop_filter_iblk_delayed(VC1Context *v, int pq)
184 MpegEncContext *s = &v->s;
187 /* The loopfilter runs 1 row and 1 column behind the overlap filter, which
188 * means it runs two rows/cols behind the decoding loop. */
189 if (!s->first_slice_line) {
191 if (s->mb_y >= s->start_mb_y + 2) {
192 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 16 * s->linesize - 16, s->linesize, pq);
195 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 16, s->linesize, pq);
196 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 8, s->linesize, pq);
197 for (j = 0; j < 2; j++) {
198 v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq);
200 v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 16 * s->uvlinesize - 8, s->uvlinesize, pq);
204 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 8 * s->linesize - 16, s->linesize, pq);
207 if (s->mb_x == s->mb_width - 1) {
208 if (s->mb_y >= s->start_mb_y + 2) {
209 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
212 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize, s->linesize, pq);
213 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize + 8, s->linesize, pq);
214 for (j = 0; j < 2; j++) {
215 v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
217 v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 16 * s->uvlinesize, s->uvlinesize, pq);
221 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 8 * s->linesize, s->linesize, pq);
224 if (s->mb_y == s->end_mb_y) {
227 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 16, s->linesize, pq);
228 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 8, s->linesize, pq);
230 for (j = 0; j < 2; j++) {
231 v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq);
236 if (s->mb_x == s->mb_width - 1) {
238 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
239 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize + 8, s->linesize, pq);
241 for (j = 0; j < 2; j++) {
242 v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
250 static void vc1_smooth_overlap_filter_iblk(VC1Context *v)
252 MpegEncContext *s = &v->s;
255 if (v->condover == CONDOVER_NONE)
258 mb_pos = s->mb_x + s->mb_y * s->mb_stride;
260 /* Within a MB, the horizontal overlap always runs before the vertical.
261 * To accomplish that, we run the H on left and internal borders of the
262 * currently decoded MB. Then, we wait for the next overlap iteration
263 * to do H overlap on the right edge of this MB, before moving over and
264 * running the V overlap. Therefore, the V overlap makes us trail by one
265 * MB col and the H overlap filter makes us trail by one MB row. This
266 * is reflected in the time at which we run the put_pixels loop. */
267 if (v->condover == CONDOVER_ALL || v->pq >= 9 || v->over_flags_plane[mb_pos]) {
268 if (s->mb_x && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
269 v->over_flags_plane[mb_pos - 1])) {
270 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][1],
271 v->block[v->cur_blk_idx][0]);
272 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][3],
273 v->block[v->cur_blk_idx][2]);
274 if (!(s->flags & CODEC_FLAG_GRAY)) {
275 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][4],
276 v->block[v->cur_blk_idx][4]);
277 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][5],
278 v->block[v->cur_blk_idx][5]);
281 v->vc1dsp.vc1_h_s_overlap(v->block[v->cur_blk_idx][0],
282 v->block[v->cur_blk_idx][1]);
283 v->vc1dsp.vc1_h_s_overlap(v->block[v->cur_blk_idx][2],
284 v->block[v->cur_blk_idx][3]);
286 if (s->mb_x == s->mb_width - 1) {
287 if (!s->first_slice_line && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
288 v->over_flags_plane[mb_pos - s->mb_stride])) {
289 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][2],
290 v->block[v->cur_blk_idx][0]);
291 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][3],
292 v->block[v->cur_blk_idx][1]);
293 if (!(s->flags & CODEC_FLAG_GRAY)) {
294 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][4],
295 v->block[v->cur_blk_idx][4]);
296 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][5],
297 v->block[v->cur_blk_idx][5]);
300 v->vc1dsp.vc1_v_s_overlap(v->block[v->cur_blk_idx][0],
301 v->block[v->cur_blk_idx][2]);
302 v->vc1dsp.vc1_v_s_overlap(v->block[v->cur_blk_idx][1],
303 v->block[v->cur_blk_idx][3]);
306 if (s->mb_x && (v->condover == CONDOVER_ALL || v->over_flags_plane[mb_pos - 1])) {
307 if (!s->first_slice_line && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
308 v->over_flags_plane[mb_pos - s->mb_stride - 1])) {
309 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][2],
310 v->block[v->left_blk_idx][0]);
311 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][3],
312 v->block[v->left_blk_idx][1]);
313 if (!(s->flags & CODEC_FLAG_GRAY)) {
314 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][4],
315 v->block[v->left_blk_idx][4]);
316 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][5],
317 v->block[v->left_blk_idx][5]);
320 v->vc1dsp.vc1_v_s_overlap(v->block[v->left_blk_idx][0],
321 v->block[v->left_blk_idx][2]);
322 v->vc1dsp.vc1_v_s_overlap(v->block[v->left_blk_idx][1],
323 v->block[v->left_blk_idx][3]);
327 /** Do motion compensation over 1 macroblock
328 * Mostly adapted hpel_motion and qpel_motion from mpegvideo.c
330 static void vc1_mc_1mv(VC1Context *v, int dir)
332 MpegEncContext *s = &v->s;
333 DSPContext *dsp = &v->s.dsp;
334 uint8_t *srcY, *srcU, *srcV;
335 int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
337 int v_edge_pos = s->v_edge_pos >> v->field_mode;
339 if ((!v->field_mode ||
340 (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
341 !v->s.last_picture.f.data[0])
344 mx = s->mv[dir][0][0];
345 my = s->mv[dir][0][1];
347 // store motion vectors for further use in B frames
348 if (s->pict_type == AV_PICTURE_TYPE_P) {
349 s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][0] = mx;
350 s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][1] = my;
353 uvmx = (mx + ((mx & 3) == 3)) >> 1;
354 uvmy = (my + ((my & 3) == 3)) >> 1;
355 v->luma_mv[s->mb_x][0] = uvmx;
356 v->luma_mv[s->mb_x][1] = uvmy;
359 v->cur_field_type != v->ref_field_type[dir]) {
360 my = my - 2 + 4 * v->cur_field_type;
361 uvmy = uvmy - 2 + 4 * v->cur_field_type;
364 // fastuvmc shall be ignored for interlaced frame picture
365 if (v->fastuvmc && (v->fcm != ILACE_FRAME)) {
366 uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
367 uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
369 if (v->field_mode) { // interlaced field picture
371 if ((v->cur_field_type != v->ref_field_type[dir]) && v->cur_field_type) {
372 srcY = s->current_picture.f.data[0];
373 srcU = s->current_picture.f.data[1];
374 srcV = s->current_picture.f.data[2];
376 srcY = s->last_picture.f.data[0];
377 srcU = s->last_picture.f.data[1];
378 srcV = s->last_picture.f.data[2];
381 srcY = s->next_picture.f.data[0];
382 srcU = s->next_picture.f.data[1];
383 srcV = s->next_picture.f.data[2];
387 srcY = s->last_picture.f.data[0];
388 srcU = s->last_picture.f.data[1];
389 srcV = s->last_picture.f.data[2];
391 srcY = s->next_picture.f.data[0];
392 srcU = s->next_picture.f.data[1];
393 srcV = s->next_picture.f.data[2];
397 src_x = s->mb_x * 16 + (mx >> 2);
398 src_y = s->mb_y * 16 + (my >> 2);
399 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
400 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
402 if (v->profile != PROFILE_ADVANCED) {
403 src_x = av_clip( src_x, -16, s->mb_width * 16);
404 src_y = av_clip( src_y, -16, s->mb_height * 16);
405 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
406 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
408 src_x = av_clip( src_x, -17, s->avctx->coded_width);
409 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
410 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
411 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
414 srcY += src_y * s->linesize + src_x;
415 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
416 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
418 if (v->field_mode && v->ref_field_type[dir]) {
419 srcY += s->current_picture_ptr->f.linesize[0];
420 srcU += s->current_picture_ptr->f.linesize[1];
421 srcV += s->current_picture_ptr->f.linesize[2];
424 /* for grayscale we should not try to read from unknown area */
425 if (s->flags & CODEC_FLAG_GRAY) {
426 srcU = s->edge_emu_buffer + 18 * s->linesize;
427 srcV = s->edge_emu_buffer + 18 * s->linesize;
430 if (v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
431 || s->h_edge_pos < 22 || v_edge_pos < 22
432 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 16 - s->mspel * 3
433 || (unsigned)(src_y - 1) > v_edge_pos - (my&3) - 16 - 3) {
434 uint8_t *uvbuf = s->edge_emu_buffer + 19 * s->linesize;
436 srcY -= s->mspel * (1 + s->linesize);
437 s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize,
438 17 + s->mspel * 2, 17 + s->mspel * 2,
439 src_x - s->mspel, src_y - s->mspel,
440 s->h_edge_pos, v_edge_pos);
441 srcY = s->edge_emu_buffer;
442 s->dsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, 8 + 1, 8 + 1,
443 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
444 s->dsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8 + 1, 8 + 1,
445 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
448 /* if we deal with range reduction we need to scale source blocks */
449 if (v->rangeredfrm) {
454 for (j = 0; j < 17 + s->mspel * 2; j++) {
455 for (i = 0; i < 17 + s->mspel * 2; i++)
456 src[i] = ((src[i] - 128) >> 1) + 128;
461 for (j = 0; j < 9; j++) {
462 for (i = 0; i < 9; i++) {
463 src[i] = ((src[i] - 128) >> 1) + 128;
464 src2[i] = ((src2[i] - 128) >> 1) + 128;
466 src += s->uvlinesize;
467 src2 += s->uvlinesize;
470 /* if we deal with intensity compensation we need to scale source blocks */
471 if (v->mv_mode == MV_PMODE_INTENSITY_COMP) {
476 for (j = 0; j < 17 + s->mspel * 2; j++) {
477 for (i = 0; i < 17 + s->mspel * 2; i++)
478 src[i] = v->luty[src[i]];
483 for (j = 0; j < 9; j++) {
484 for (i = 0; i < 9; i++) {
485 src[i] = v->lutuv[src[i]];
486 src2[i] = v->lutuv[src2[i]];
488 src += s->uvlinesize;
489 src2 += s->uvlinesize;
492 srcY += s->mspel * (1 + s->linesize);
495 if (v->field_mode && v->cur_field_type) {
496 off = s->current_picture_ptr->f.linesize[0];
497 off_uv = s->current_picture_ptr->f.linesize[1];
503 dxy = ((my & 3) << 2) | (mx & 3);
504 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off , srcY , s->linesize, v->rnd);
505 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8, srcY + 8, s->linesize, v->rnd);
506 srcY += s->linesize * 8;
507 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize , srcY , s->linesize, v->rnd);
508 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
509 } else { // hpel mc - always used for luma
510 dxy = (my & 2) | ((mx & 2) >> 1);
512 dsp->put_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
514 dsp->put_no_rnd_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
517 if (s->flags & CODEC_FLAG_GRAY) return;
518 /* Chroma MC always uses qpel bilinear */
519 uvmx = (uvmx & 3) << 1;
520 uvmy = (uvmy & 3) << 1;
522 dsp->put_h264_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
523 dsp->put_h264_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
525 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
526 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
530 static inline int median4(int a, int b, int c, int d)
533 if (c < d) return (FFMIN(b, d) + FFMAX(a, c)) / 2;
534 else return (FFMIN(b, c) + FFMAX(a, d)) / 2;
536 if (c < d) return (FFMIN(a, d) + FFMAX(b, c)) / 2;
537 else return (FFMIN(a, c) + FFMAX(b, d)) / 2;
541 /** Do motion compensation for 4-MV macroblock - luminance block
543 static void vc1_mc_4mv_luma(VC1Context *v, int n, int dir)
545 MpegEncContext *s = &v->s;
546 DSPContext *dsp = &v->s.dsp;
548 int dxy, mx, my, src_x, src_y;
550 int fieldmv = (v->fcm == ILACE_FRAME) ? v->blk_mv_type[s->block_index[n]] : 0;
551 int v_edge_pos = s->v_edge_pos >> v->field_mode;
553 if ((!v->field_mode ||
554 (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
555 !v->s.last_picture.f.data[0])
558 mx = s->mv[dir][n][0];
559 my = s->mv[dir][n][1];
563 if ((v->cur_field_type != v->ref_field_type[dir]) && v->cur_field_type)
564 srcY = s->current_picture.f.data[0];
566 srcY = s->last_picture.f.data[0];
568 srcY = s->last_picture.f.data[0];
570 srcY = s->next_picture.f.data[0];
573 if (v->cur_field_type != v->ref_field_type[dir])
574 my = my - 2 + 4 * v->cur_field_type;
577 if (s->pict_type == AV_PICTURE_TYPE_P && n == 3 && v->field_mode) {
578 int same_count = 0, opp_count = 0, k;
579 int chosen_mv[2][4][2], f;
581 for (k = 0; k < 4; k++) {
582 f = v->mv_f[0][s->block_index[k] + v->blocks_off];
583 chosen_mv[f][f ? opp_count : same_count][0] = s->mv[0][k][0];
584 chosen_mv[f][f ? opp_count : same_count][1] = s->mv[0][k][1];
588 f = opp_count > same_count;
589 switch (f ? opp_count : same_count) {
591 tx = median4(chosen_mv[f][0][0], chosen_mv[f][1][0],
592 chosen_mv[f][2][0], chosen_mv[f][3][0]);
593 ty = median4(chosen_mv[f][0][1], chosen_mv[f][1][1],
594 chosen_mv[f][2][1], chosen_mv[f][3][1]);
597 tx = mid_pred(chosen_mv[f][0][0], chosen_mv[f][1][0], chosen_mv[f][2][0]);
598 ty = mid_pred(chosen_mv[f][0][1], chosen_mv[f][1][1], chosen_mv[f][2][1]);
601 tx = (chosen_mv[f][0][0] + chosen_mv[f][1][0]) / 2;
602 ty = (chosen_mv[f][0][1] + chosen_mv[f][1][1]) / 2;
605 s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx;
606 s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty;
607 for (k = 0; k < 4; k++)
608 v->mv_f[1][s->block_index[k] + v->blocks_off] = f;
611 if (v->fcm == ILACE_FRAME) { // not sure if needed for other types of picture
613 int width = s->avctx->coded_width;
614 int height = s->avctx->coded_height >> 1;
615 qx = (s->mb_x * 16) + (mx >> 2);
616 qy = (s->mb_y * 8) + (my >> 3);
621 mx -= 4 * (qx - width);
624 else if (qy > height + 1)
625 my -= 8 * (qy - height - 1);
628 if ((v->fcm == ILACE_FRAME) && fieldmv)
629 off = ((n > 1) ? s->linesize : 0) + (n & 1) * 8;
631 off = s->linesize * 4 * (n & 2) + (n & 1) * 8;
632 if (v->field_mode && v->cur_field_type)
633 off += s->current_picture_ptr->f.linesize[0];
635 src_x = s->mb_x * 16 + (n & 1) * 8 + (mx >> 2);
637 src_y = s->mb_y * 16 + (n & 2) * 4 + (my >> 2);
639 src_y = s->mb_y * 16 + ((n > 1) ? 1 : 0) + (my >> 2);
641 if (v->profile != PROFILE_ADVANCED) {
642 src_x = av_clip(src_x, -16, s->mb_width * 16);
643 src_y = av_clip(src_y, -16, s->mb_height * 16);
645 src_x = av_clip(src_x, -17, s->avctx->coded_width);
646 if (v->fcm == ILACE_FRAME) {
648 src_y = av_clip(src_y, -17, s->avctx->coded_height + 1);
650 src_y = av_clip(src_y, -18, s->avctx->coded_height);
652 src_y = av_clip(src_y, -18, s->avctx->coded_height + 1);
656 srcY += src_y * s->linesize + src_x;
657 if (v->field_mode && v->ref_field_type[dir])
658 srcY += s->current_picture_ptr->f.linesize[0];
660 if (fieldmv && !(src_y & 1))
662 if (fieldmv && (src_y & 1) && src_y < 4)
664 if (v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
665 || s->h_edge_pos < 13 || v_edge_pos < 23
666 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx & 3) - 8 - s->mspel * 2
667 || (unsigned)(src_y - (s->mspel << fieldmv)) > v_edge_pos - (my & 3) - ((8 + s->mspel * 2) << fieldmv)) {
668 srcY -= s->mspel * (1 + (s->linesize << fieldmv));
669 /* check emulate edge stride and offset */
670 s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize,
671 9 + s->mspel * 2, (9 + s->mspel * 2) << fieldmv,
672 src_x - s->mspel, src_y - (s->mspel << fieldmv),
673 s->h_edge_pos, v_edge_pos);
674 srcY = s->edge_emu_buffer;
675 /* if we deal with range reduction we need to scale source blocks */
676 if (v->rangeredfrm) {
681 for (j = 0; j < 9 + s->mspel * 2; j++) {
682 for (i = 0; i < 9 + s->mspel * 2; i++)
683 src[i] = ((src[i] - 128) >> 1) + 128;
684 src += s->linesize << fieldmv;
687 /* if we deal with intensity compensation we need to scale source blocks */
688 if (v->mv_mode == MV_PMODE_INTENSITY_COMP) {
693 for (j = 0; j < 9 + s->mspel * 2; j++) {
694 for (i = 0; i < 9 + s->mspel * 2; i++)
695 src[i] = v->luty[src[i]];
696 src += s->linesize << fieldmv;
699 srcY += s->mspel * (1 + (s->linesize << fieldmv));
703 dxy = ((my & 3) << 2) | (mx & 3);
704 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize << fieldmv, v->rnd);
705 } else { // hpel mc - always used for luma
706 dxy = (my & 2) | ((mx & 2) >> 1);
708 dsp->put_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
710 dsp->put_no_rnd_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
714 static av_always_inline int get_chroma_mv(int *mvx, int *mvy, int *a, int flag, int *tx, int *ty)
717 static const int count[16] = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4};
719 idx = ((a[3] != flag) << 3)
720 | ((a[2] != flag) << 2)
721 | ((a[1] != flag) << 1)
724 *tx = median4(mvx[0], mvx[1], mvx[2], mvx[3]);
725 *ty = median4(mvy[0], mvy[1], mvy[2], mvy[3]);
727 } else if (count[idx] == 1) {
730 *tx = mid_pred(mvx[1], mvx[2], mvx[3]);
731 *ty = mid_pred(mvy[1], mvy[2], mvy[3]);
734 *tx = mid_pred(mvx[0], mvx[2], mvx[3]);
735 *ty = mid_pred(mvy[0], mvy[2], mvy[3]);
738 *tx = mid_pred(mvx[0], mvx[1], mvx[3]);
739 *ty = mid_pred(mvy[0], mvy[1], mvy[3]);
742 *tx = mid_pred(mvx[0], mvx[1], mvx[2]);
743 *ty = mid_pred(mvy[0], mvy[1], mvy[2]);
746 } else if (count[idx] == 2) {
748 for (i = 0; i < 3; i++)
753 for (i = t1 + 1; i < 4; i++)
758 *tx = (mvx[t1] + mvx[t2]) / 2;
759 *ty = (mvy[t1] + mvy[t2]) / 2;
767 /** Do motion compensation for 4-MV macroblock - both chroma blocks
769 static void vc1_mc_4mv_chroma(VC1Context *v, int dir)
771 MpegEncContext *s = &v->s;
772 DSPContext *dsp = &v->s.dsp;
773 uint8_t *srcU, *srcV;
774 int uvmx, uvmy, uvsrc_x, uvsrc_y;
775 int k, tx = 0, ty = 0;
776 int mvx[4], mvy[4], intra[4], mv_f[4];
778 int chroma_ref_type = v->cur_field_type, off = 0;
779 int v_edge_pos = s->v_edge_pos >> v->field_mode;
781 if (!v->field_mode && !v->s.last_picture.f.data[0])
783 if (s->flags & CODEC_FLAG_GRAY)
786 for (k = 0; k < 4; k++) {
787 mvx[k] = s->mv[dir][k][0];
788 mvy[k] = s->mv[dir][k][1];
789 intra[k] = v->mb_type[0][s->block_index[k]];
791 mv_f[k] = v->mv_f[dir][s->block_index[k] + v->blocks_off];
794 /* calculate chroma MV vector from four luma MVs */
795 if (!v->field_mode || (v->field_mode && !v->numref)) {
796 valid_count = get_chroma_mv(mvx, mvy, intra, 0, &tx, &ty);
797 chroma_ref_type = v->reffield;
799 s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][0] = 0;
800 s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][1] = 0;
801 v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
802 return; //no need to do MC for intra blocks
806 if (mv_f[0] + mv_f[1] + mv_f[2] + mv_f[3] > 2)
808 valid_count = get_chroma_mv(mvx, mvy, mv_f, dominant, &tx, &ty);
810 chroma_ref_type = !v->cur_field_type;
812 if (v->field_mode && chroma_ref_type == 1 && v->cur_field_type == 1 && !v->s.last_picture.f.data[0])
814 s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx;
815 s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty;
816 uvmx = (tx + ((tx & 3) == 3)) >> 1;
817 uvmy = (ty + ((ty & 3) == 3)) >> 1;
819 v->luma_mv[s->mb_x][0] = uvmx;
820 v->luma_mv[s->mb_x][1] = uvmy;
823 uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
824 uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
826 // Field conversion bias
827 if (v->cur_field_type != chroma_ref_type)
828 uvmy += 2 - 4 * chroma_ref_type;
830 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
831 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
833 if (v->profile != PROFILE_ADVANCED) {
834 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
835 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
837 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
838 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
843 if ((v->cur_field_type != chroma_ref_type) && v->cur_field_type) {
844 srcU = s->current_picture.f.data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
845 srcV = s->current_picture.f.data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
847 srcU = s->last_picture.f.data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
848 srcV = s->last_picture.f.data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
851 srcU = s->last_picture.f.data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
852 srcV = s->last_picture.f.data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
855 srcU = s->next_picture.f.data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
856 srcV = s->next_picture.f.data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
860 if (chroma_ref_type) {
861 srcU += s->current_picture_ptr->f.linesize[1];
862 srcV += s->current_picture_ptr->f.linesize[2];
864 off = v->cur_field_type ? s->current_picture_ptr->f.linesize[1] : 0;
867 if (v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
868 || s->h_edge_pos < 18 || v_edge_pos < 18
869 || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 9
870 || (unsigned)uvsrc_y > (v_edge_pos >> 1) - 9) {
871 s->dsp.emulated_edge_mc(s->edge_emu_buffer , srcU, s->uvlinesize,
872 8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
873 s->h_edge_pos >> 1, v_edge_pos >> 1);
874 s->dsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV, s->uvlinesize,
875 8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
876 s->h_edge_pos >> 1, v_edge_pos >> 1);
877 srcU = s->edge_emu_buffer;
878 srcV = s->edge_emu_buffer + 16;
880 /* if we deal with range reduction we need to scale source blocks */
881 if (v->rangeredfrm) {
887 for (j = 0; j < 9; j++) {
888 for (i = 0; i < 9; i++) {
889 src[i] = ((src[i] - 128) >> 1) + 128;
890 src2[i] = ((src2[i] - 128) >> 1) + 128;
892 src += s->uvlinesize;
893 src2 += s->uvlinesize;
896 /* if we deal with intensity compensation we need to scale source blocks */
897 if (v->mv_mode == MV_PMODE_INTENSITY_COMP) {
903 for (j = 0; j < 9; j++) {
904 for (i = 0; i < 9; i++) {
905 src[i] = v->lutuv[src[i]];
906 src2[i] = v->lutuv[src2[i]];
908 src += s->uvlinesize;
909 src2 += s->uvlinesize;
914 /* Chroma MC always uses qpel bilinear */
915 uvmx = (uvmx & 3) << 1;
916 uvmy = (uvmy & 3) << 1;
918 dsp->put_h264_chroma_pixels_tab[0](s->dest[1] + off, srcU, s->uvlinesize, 8, uvmx, uvmy);
919 dsp->put_h264_chroma_pixels_tab[0](s->dest[2] + off, srcV, s->uvlinesize, 8, uvmx, uvmy);
921 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1] + off, srcU, s->uvlinesize, 8, uvmx, uvmy);
922 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2] + off, srcV, s->uvlinesize, 8, uvmx, uvmy);
926 /** Do motion compensation for 4-MV field chroma macroblock (both U and V)
928 static void vc1_mc_4mv_chroma4(VC1Context *v)
930 MpegEncContext *s = &v->s;
931 DSPContext *dsp = &v->s.dsp;
932 uint8_t *srcU, *srcV;
933 int uvsrc_x, uvsrc_y;
934 int uvmx_field[4], uvmy_field[4];
936 int fieldmv = v->blk_mv_type[s->block_index[0]];
937 static const int s_rndtblfield[16] = { 0, 0, 1, 2, 4, 4, 5, 6, 2, 2, 3, 8, 6, 6, 7, 12 };
938 int v_dist = fieldmv ? 1 : 4; // vertical offset for lower sub-blocks
939 int v_edge_pos = s->v_edge_pos >> 1;
941 if (!v->s.last_picture.f.data[0])
943 if (s->flags & CODEC_FLAG_GRAY)
946 for (i = 0; i < 4; i++) {
948 uvmx_field[i] = (tx + ((tx & 3) == 3)) >> 1;
951 uvmy_field[i] = (ty >> 4) * 8 + s_rndtblfield[ty & 0xF];
953 uvmy_field[i] = (ty + ((ty & 3) == 3)) >> 1;
956 for (i = 0; i < 4; i++) {
957 off = (i & 1) * 4 + ((i & 2) ? v_dist * s->uvlinesize : 0);
958 uvsrc_x = s->mb_x * 8 + (i & 1) * 4 + (uvmx_field[i] >> 2);
959 uvsrc_y = s->mb_y * 8 + ((i & 2) ? v_dist : 0) + (uvmy_field[i] >> 2);
960 // FIXME: implement proper pull-back (see vc1cropmv.c, vc1CROPMV_ChromaPullBack())
961 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
962 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
963 srcU = s->last_picture.f.data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
964 srcV = s->last_picture.f.data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
965 uvmx_field[i] = (uvmx_field[i] & 3) << 1;
966 uvmy_field[i] = (uvmy_field[i] & 3) << 1;
968 if (fieldmv && !(uvsrc_y & 1))
970 if (fieldmv && (uvsrc_y & 1) && uvsrc_y < 2)
972 if ((v->mv_mode == MV_PMODE_INTENSITY_COMP)
973 || s->h_edge_pos < 10 || v_edge_pos < (5 << fieldmv)
974 || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 5
975 || (unsigned)uvsrc_y > v_edge_pos - (5 << fieldmv)) {
976 s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcU, s->uvlinesize,
977 5, (5 << fieldmv), uvsrc_x, uvsrc_y,
978 s->h_edge_pos >> 1, v_edge_pos);
979 s->dsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV, s->uvlinesize,
980 5, (5 << fieldmv), uvsrc_x, uvsrc_y,
981 s->h_edge_pos >> 1, v_edge_pos);
982 srcU = s->edge_emu_buffer;
983 srcV = s->edge_emu_buffer + 16;
985 /* if we deal with intensity compensation we need to scale source blocks */
986 if (v->mv_mode == MV_PMODE_INTENSITY_COMP) {
992 for (j = 0; j < 5; j++) {
993 for (i = 0; i < 5; i++) {
994 src[i] = v->lutuv[src[i]];
995 src2[i] = v->lutuv[src2[i]];
997 src += s->uvlinesize << 1;
998 src2 += s->uvlinesize << 1;
1003 dsp->put_h264_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
1004 dsp->put_h264_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
1006 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]);
1007 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]);
1012 /***********************************************************************/
1014 * @name VC-1 Block-level functions
1015 * @see 7.1.4, p91 and 8.1.1.7, p(1)04
1021 * @brief Get macroblock-level quantizer scale
1023 #define GET_MQUANT() \
1024 if (v->dquantfrm) { \
1026 if (v->dqprofile == DQPROFILE_ALL_MBS) { \
1027 if (v->dqbilevel) { \
1028 mquant = (get_bits1(gb)) ? v->altpq : v->pq; \
1030 mqdiff = get_bits(gb, 3); \
1032 mquant = v->pq + mqdiff; \
1034 mquant = get_bits(gb, 5); \
1037 if (v->dqprofile == DQPROFILE_SINGLE_EDGE) \
1038 edges = 1 << v->dqsbedge; \
1039 else if (v->dqprofile == DQPROFILE_DOUBLE_EDGES) \
1040 edges = (3 << v->dqsbedge) % 15; \
1041 else if (v->dqprofile == DQPROFILE_FOUR_EDGES) \
1043 if ((edges&1) && !s->mb_x) \
1044 mquant = v->altpq; \
1045 if ((edges&2) && s->first_slice_line) \
1046 mquant = v->altpq; \
1047 if ((edges&4) && s->mb_x == (s->mb_width - 1)) \
1048 mquant = v->altpq; \
1049 if ((edges&8) && s->mb_y == (s->mb_height - 1)) \
1050 mquant = v->altpq; \
1051 if (!mquant || mquant > 31) { \
1052 av_log(v->s.avctx, AV_LOG_ERROR, \
1053 "Overriding invalid mquant %d\n", mquant); \
1059 * @def GET_MVDATA(_dmv_x, _dmv_y)
1060 * @brief Get MV differentials
1061 * @see MVDATA decoding from 8.3.5.2, p(1)20
1062 * @param _dmv_x Horizontal differential for decoded MV
1063 * @param _dmv_y Vertical differential for decoded MV
1065 #define GET_MVDATA(_dmv_x, _dmv_y) \
1066 index = 1 + get_vlc2(gb, ff_vc1_mv_diff_vlc[s->mv_table_index].table, \
1067 VC1_MV_DIFF_VLC_BITS, 2); \
1069 mb_has_coeffs = 1; \
1072 mb_has_coeffs = 0; \
1075 _dmv_x = _dmv_y = 0; \
1076 } else if (index == 35) { \
1077 _dmv_x = get_bits(gb, v->k_x - 1 + s->quarter_sample); \
1078 _dmv_y = get_bits(gb, v->k_y - 1 + s->quarter_sample); \
1079 } else if (index == 36) { \
1084 index1 = index % 6; \
1085 if (!s->quarter_sample && index1 == 5) val = 1; \
1087 if (size_table[index1] - val > 0) \
1088 val = get_bits(gb, size_table[index1] - val); \
1090 sign = 0 - (val&1); \
1091 _dmv_x = (sign ^ ((val>>1) + offset_table[index1])) - sign; \
1093 index1 = index / 6; \
1094 if (!s->quarter_sample && index1 == 5) val = 1; \
1096 if (size_table[index1] - val > 0) \
1097 val = get_bits(gb, size_table[index1] - val); \
1099 sign = 0 - (val & 1); \
1100 _dmv_y = (sign ^ ((val >> 1) + offset_table[index1])) - sign; \
1103 static av_always_inline void get_mvdata_interlaced(VC1Context *v, int *dmv_x,
1104 int *dmv_y, int *pred_flag)
1107 int extend_x = 0, extend_y = 0;
1108 GetBitContext *gb = &v->s.gb;
1111 const int* offs_tab;
1114 bits = VC1_2REF_MVDATA_VLC_BITS;
1117 bits = VC1_1REF_MVDATA_VLC_BITS;
1120 switch (v->dmvrange) {
1128 extend_x = extend_y = 1;
1131 index = get_vlc2(gb, v->imv_vlc->table, bits, 3);
1133 *dmv_x = get_bits(gb, v->k_x);
1134 *dmv_y = get_bits(gb, v->k_y);
1136 *pred_flag = *dmv_y & 1;
1137 *dmv_y = (*dmv_y + *pred_flag) >> 1;
1142 offs_tab = offset_table2;
1144 offs_tab = offset_table1;
1145 index1 = (index + 1) % 9;
1147 val = get_bits(gb, index1 + extend_x);
1148 sign = 0 -(val & 1);
1149 *dmv_x = (sign ^ ((val >> 1) + offs_tab[index1])) - sign;
1153 offs_tab = offset_table2;
1155 offs_tab = offset_table1;
1156 index1 = (index + 1) / 9;
1157 if (index1 > v->numref) {
1158 val = get_bits(gb, (index1 + (extend_y << v->numref)) >> v->numref);
1159 sign = 0 - (val & 1);
1160 *dmv_y = (sign ^ ((val >> 1) + offs_tab[index1 >> v->numref])) - sign;
1164 *pred_flag = index1 & 1;
1168 static av_always_inline int scaleforsame_x(VC1Context *v, int n /* MV */, int dir)
1170 int scaledvalue, refdist;
1171 int scalesame1, scalesame2;
1172 int scalezone1_x, zone1offset_x;
1173 int table_index = dir ^ v->second_field;
1175 if (v->s.pict_type != AV_PICTURE_TYPE_B)
1176 refdist = v->refdist;
1178 refdist = dir ? v->brfd : v->frfd;
1181 scalesame1 = ff_vc1_field_mvpred_scales[table_index][1][refdist];
1182 scalesame2 = ff_vc1_field_mvpred_scales[table_index][2][refdist];
1183 scalezone1_x = ff_vc1_field_mvpred_scales[table_index][3][refdist];
1184 zone1offset_x = ff_vc1_field_mvpred_scales[table_index][5][refdist];
1189 if (FFABS(n) < scalezone1_x)
1190 scaledvalue = (n * scalesame1) >> 8;
1193 scaledvalue = ((n * scalesame2) >> 8) - zone1offset_x;
1195 scaledvalue = ((n * scalesame2) >> 8) + zone1offset_x;
1198 return av_clip(scaledvalue, -v->range_x, v->range_x - 1);
1201 static av_always_inline int scaleforsame_y(VC1Context *v, int i, int n /* MV */, int dir)
1203 int scaledvalue, refdist;
1204 int scalesame1, scalesame2;
1205 int scalezone1_y, zone1offset_y;
1206 int table_index = dir ^ v->second_field;
1208 if (v->s.pict_type != AV_PICTURE_TYPE_B)
1209 refdist = v->refdist;
1211 refdist = dir ? v->brfd : v->frfd;
1214 scalesame1 = ff_vc1_field_mvpred_scales[table_index][1][refdist];
1215 scalesame2 = ff_vc1_field_mvpred_scales[table_index][2][refdist];
1216 scalezone1_y = ff_vc1_field_mvpred_scales[table_index][4][refdist];
1217 zone1offset_y = ff_vc1_field_mvpred_scales[table_index][6][refdist];
1222 if (FFABS(n) < scalezone1_y)
1223 scaledvalue = (n * scalesame1) >> 8;
1226 scaledvalue = ((n * scalesame2) >> 8) - zone1offset_y;
1228 scaledvalue = ((n * scalesame2) >> 8) + zone1offset_y;
1232 if (v->cur_field_type && !v->ref_field_type[dir])
1233 return av_clip(scaledvalue, -v->range_y / 2 + 1, v->range_y / 2);
1235 return av_clip(scaledvalue, -v->range_y / 2, v->range_y / 2 - 1);
1238 static av_always_inline int scaleforopp_x(VC1Context *v, int n /* MV */)
1240 int scalezone1_x, zone1offset_x;
1241 int scaleopp1, scaleopp2, brfd;
1244 brfd = FFMIN(v->brfd, 3);
1245 scalezone1_x = ff_vc1_b_field_mvpred_scales[3][brfd];
1246 zone1offset_x = ff_vc1_b_field_mvpred_scales[5][brfd];
1247 scaleopp1 = ff_vc1_b_field_mvpred_scales[1][brfd];
1248 scaleopp2 = ff_vc1_b_field_mvpred_scales[2][brfd];
1253 if (FFABS(n) < scalezone1_x)
1254 scaledvalue = (n * scaleopp1) >> 8;
1257 scaledvalue = ((n * scaleopp2) >> 8) - zone1offset_x;
1259 scaledvalue = ((n * scaleopp2) >> 8) + zone1offset_x;
1262 return av_clip(scaledvalue, -v->range_x, v->range_x - 1);
1265 static av_always_inline int scaleforopp_y(VC1Context *v, int n /* MV */, int dir)
1267 int scalezone1_y, zone1offset_y;
1268 int scaleopp1, scaleopp2, brfd;
1271 brfd = FFMIN(v->brfd, 3);
1272 scalezone1_y = ff_vc1_b_field_mvpred_scales[4][brfd];
1273 zone1offset_y = ff_vc1_b_field_mvpred_scales[6][brfd];
1274 scaleopp1 = ff_vc1_b_field_mvpred_scales[1][brfd];
1275 scaleopp2 = ff_vc1_b_field_mvpred_scales[2][brfd];
1280 if (FFABS(n) < scalezone1_y)
1281 scaledvalue = (n * scaleopp1) >> 8;
1284 scaledvalue = ((n * scaleopp2) >> 8) - zone1offset_y;
1286 scaledvalue = ((n * scaleopp2) >> 8) + zone1offset_y;
1289 if (v->cur_field_type && !v->ref_field_type[dir]) {
1290 return av_clip(scaledvalue, -v->range_y / 2 + 1, v->range_y / 2);
1292 return av_clip(scaledvalue, -v->range_y / 2, v->range_y / 2 - 1);
1296 static av_always_inline int scaleforsame(VC1Context *v, int i, int n /* MV */,
1299 int brfd, scalesame;
1300 int hpel = 1 - v->s.quarter_sample;
1303 if (v->s.pict_type != AV_PICTURE_TYPE_B || v->second_field || !dir) {
1305 n = scaleforsame_y(v, i, n, dir) << hpel;
1307 n = scaleforsame_x(v, n, dir) << hpel;
1310 brfd = FFMIN(v->brfd, 3);
1311 scalesame = ff_vc1_b_field_mvpred_scales[0][brfd];
1313 n = (n * scalesame >> 8) << hpel;
1317 static av_always_inline int scaleforopp(VC1Context *v, int n /* MV */,
1320 int refdist, scaleopp;
1321 int hpel = 1 - v->s.quarter_sample;
1324 if (v->s.pict_type == AV_PICTURE_TYPE_B && !v->second_field && dir == 1) {
1326 n = scaleforopp_y(v, n, dir) << hpel;
1328 n = scaleforopp_x(v, n) << hpel;
1331 if (v->s.pict_type != AV_PICTURE_TYPE_B)
1332 refdist = FFMIN(v->refdist, 3);
1334 refdist = dir ? v->brfd : v->frfd;
1335 scaleopp = ff_vc1_field_mvpred_scales[dir ^ v->second_field][0][refdist];
1337 n = (n * scaleopp >> 8) << hpel;
1341 /** Predict and set motion vector
1343 static inline void vc1_pred_mv(VC1Context *v, int n, int dmv_x, int dmv_y,
1344 int mv1, int r_x, int r_y, uint8_t* is_intra,
1345 int pred_flag, int dir)
1347 MpegEncContext *s = &v->s;
1348 int xy, wrap, off = 0;
1352 int mixedmv_pic, num_samefield = 0, num_oppfield = 0;
1353 int opposite, a_f, b_f, c_f;
1354 int16_t field_predA[2];
1355 int16_t field_predB[2];
1356 int16_t field_predC[2];
1357 int a_valid, b_valid, c_valid;
1358 int hybridmv_thresh, y_bias = 0;
1360 if (v->mv_mode == MV_PMODE_MIXED_MV ||
1361 ((v->mv_mode == MV_PMODE_INTENSITY_COMP) && (v->mv_mode2 == MV_PMODE_MIXED_MV)))
1365 /* scale MV difference to be quad-pel */
1366 dmv_x <<= 1 - s->quarter_sample;
1367 dmv_y <<= 1 - s->quarter_sample;
1369 wrap = s->b8_stride;
1370 xy = s->block_index[n];
1373 s->mv[0][n][0] = s->current_picture.f.motion_val[0][xy + v->blocks_off][0] = 0;
1374 s->mv[0][n][1] = s->current_picture.f.motion_val[0][xy + v->blocks_off][1] = 0;
1375 s->current_picture.f.motion_val[1][xy + v->blocks_off][0] = 0;
1376 s->current_picture.f.motion_val[1][xy + v->blocks_off][1] = 0;
1377 if (mv1) { /* duplicate motion data for 1-MV block */
1378 s->current_picture.f.motion_val[0][xy + 1 + v->blocks_off][0] = 0;
1379 s->current_picture.f.motion_val[0][xy + 1 + v->blocks_off][1] = 0;
1380 s->current_picture.f.motion_val[0][xy + wrap + v->blocks_off][0] = 0;
1381 s->current_picture.f.motion_val[0][xy + wrap + v->blocks_off][1] = 0;
1382 s->current_picture.f.motion_val[0][xy + wrap + 1 + v->blocks_off][0] = 0;
1383 s->current_picture.f.motion_val[0][xy + wrap + 1 + v->blocks_off][1] = 0;
1384 v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
1385 s->current_picture.f.motion_val[1][xy + 1 + v->blocks_off][0] = 0;
1386 s->current_picture.f.motion_val[1][xy + 1 + v->blocks_off][1] = 0;
1387 s->current_picture.f.motion_val[1][xy + wrap][0] = 0;
1388 s->current_picture.f.motion_val[1][xy + wrap + v->blocks_off][1] = 0;
1389 s->current_picture.f.motion_val[1][xy + wrap + 1 + v->blocks_off][0] = 0;
1390 s->current_picture.f.motion_val[1][xy + wrap + 1 + v->blocks_off][1] = 0;
1395 C = s->current_picture.f.motion_val[dir][xy - 1 + v->blocks_off];
1396 A = s->current_picture.f.motion_val[dir][xy - wrap + v->blocks_off];
1398 if (v->field_mode && mixedmv_pic)
1399 off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
1401 off = (s->mb_x == (s->mb_width - 1)) ? -1 : 2;
1403 //in 4-MV mode different blocks have different B predictor position
1406 off = (s->mb_x > 0) ? -1 : 1;
1409 off = (s->mb_x == (s->mb_width - 1)) ? -1 : 1;
1418 B = s->current_picture.f.motion_val[dir][xy - wrap + off + v->blocks_off];
1420 a_valid = !s->first_slice_line || (n == 2 || n == 3);
1421 b_valid = a_valid && (s->mb_width > 1);
1422 c_valid = s->mb_x || (n == 1 || n == 3);
1423 if (v->field_mode) {
1424 a_valid = a_valid && !is_intra[xy - wrap];
1425 b_valid = b_valid && !is_intra[xy - wrap + off];
1426 c_valid = c_valid && !is_intra[xy - 1];
1430 a_f = v->mv_f[dir][xy - wrap + v->blocks_off];
1431 num_oppfield += a_f;
1432 num_samefield += 1 - a_f;
1433 field_predA[0] = A[0];
1434 field_predA[1] = A[1];
1436 field_predA[0] = field_predA[1] = 0;
1440 b_f = v->mv_f[dir][xy - wrap + off + v->blocks_off];
1441 num_oppfield += b_f;
1442 num_samefield += 1 - b_f;
1443 field_predB[0] = B[0];
1444 field_predB[1] = B[1];
1446 field_predB[0] = field_predB[1] = 0;
1450 c_f = v->mv_f[dir][xy - 1 + v->blocks_off];
1451 num_oppfield += c_f;
1452 num_samefield += 1 - c_f;
1453 field_predC[0] = C[0];
1454 field_predC[1] = C[1];
1456 field_predC[0] = field_predC[1] = 0;
1460 if (v->field_mode) {
1462 // REFFIELD determines if the last field or the second-last field is
1463 // to be used as reference
1464 opposite = 1 - v->reffield;
1466 if (num_samefield <= num_oppfield)
1467 opposite = 1 - pred_flag;
1469 opposite = pred_flag;
1474 if (a_valid && !a_f) {
1475 field_predA[0] = scaleforopp(v, field_predA[0], 0, dir);
1476 field_predA[1] = scaleforopp(v, field_predA[1], 1, dir);
1478 if (b_valid && !b_f) {
1479 field_predB[0] = scaleforopp(v, field_predB[0], 0, dir);
1480 field_predB[1] = scaleforopp(v, field_predB[1], 1, dir);
1482 if (c_valid && !c_f) {
1483 field_predC[0] = scaleforopp(v, field_predC[0], 0, dir);
1484 field_predC[1] = scaleforopp(v, field_predC[1], 1, dir);
1486 v->mv_f[dir][xy + v->blocks_off] = 1;
1487 v->ref_field_type[dir] = !v->cur_field_type;
1489 if (a_valid && a_f) {
1490 field_predA[0] = scaleforsame(v, n, field_predA[0], 0, dir);
1491 field_predA[1] = scaleforsame(v, n, field_predA[1], 1, dir);
1493 if (b_valid && b_f) {
1494 field_predB[0] = scaleforsame(v, n, field_predB[0], 0, dir);
1495 field_predB[1] = scaleforsame(v, n, field_predB[1], 1, dir);
1497 if (c_valid && c_f) {
1498 field_predC[0] = scaleforsame(v, n, field_predC[0], 0, dir);
1499 field_predC[1] = scaleforsame(v, n, field_predC[1], 1, dir);
1501 v->mv_f[dir][xy + v->blocks_off] = 0;
1502 v->ref_field_type[dir] = v->cur_field_type;
1506 px = field_predA[0];
1507 py = field_predA[1];
1508 } else if (c_valid) {
1509 px = field_predC[0];
1510 py = field_predC[1];
1511 } else if (b_valid) {
1512 px = field_predB[0];
1513 py = field_predB[1];
1519 if (num_samefield + num_oppfield > 1) {
1520 px = mid_pred(field_predA[0], field_predB[0], field_predC[0]);
1521 py = mid_pred(field_predA[1], field_predB[1], field_predC[1]);
1524 /* Pullback MV as specified in 8.3.5.3.4 */
1525 if (!v->field_mode) {
1527 qx = (s->mb_x << 6) + ((n == 1 || n == 3) ? 32 : 0);
1528 qy = (s->mb_y << 6) + ((n == 2 || n == 3) ? 32 : 0);
1529 X = (s->mb_width << 6) - 4;
1530 Y = (s->mb_height << 6) - 4;
1532 if (qx + px < -60) px = -60 - qx;
1533 if (qy + py < -60) py = -60 - qy;
1535 if (qx + px < -28) px = -28 - qx;
1536 if (qy + py < -28) py = -28 - qy;
1538 if (qx + px > X) px = X - qx;
1539 if (qy + py > Y) py = Y - qy;
1542 if (!v->field_mode || s->pict_type != AV_PICTURE_TYPE_B) {
1543 /* Calculate hybrid prediction as specified in 8.3.5.3.5 (also 10.3.5.4.3.5) */
1544 hybridmv_thresh = 32;
1545 if (a_valid && c_valid) {
1546 if (is_intra[xy - wrap])
1547 sum = FFABS(px) + FFABS(py);
1549 sum = FFABS(px - field_predA[0]) + FFABS(py - field_predA[1]);
1550 if (sum > hybridmv_thresh) {
1551 if (get_bits1(&s->gb)) { // read HYBRIDPRED bit
1552 px = field_predA[0];
1553 py = field_predA[1];
1555 px = field_predC[0];
1556 py = field_predC[1];
1559 if (is_intra[xy - 1])
1560 sum = FFABS(px) + FFABS(py);
1562 sum = FFABS(px - field_predC[0]) + FFABS(py - field_predC[1]);
1563 if (sum > hybridmv_thresh) {
1564 if (get_bits1(&s->gb)) {
1565 px = field_predA[0];
1566 py = field_predA[1];
1568 px = field_predC[0];
1569 py = field_predC[1];
1576 if (v->field_mode && v->numref)
1578 if (v->field_mode && v->cur_field_type && v->ref_field_type[dir] == 0)
1580 /* store MV using signed modulus of MV range defined in 4.11 */
1581 s->mv[dir][n][0] = s->current_picture.f.motion_val[dir][xy + v->blocks_off][0] = ((px + dmv_x + r_x) & ((r_x << 1) - 1)) - r_x;
1582 s->mv[dir][n][1] = s->current_picture.f.motion_val[dir][xy + v->blocks_off][1] = ((py + dmv_y + r_y - y_bias) & ((r_y << 1) - 1)) - r_y + y_bias;
1583 if (mv1) { /* duplicate motion data for 1-MV block */
1584 s->current_picture.f.motion_val[dir][xy + 1 + v->blocks_off][0] = s->current_picture.f.motion_val[dir][xy + v->blocks_off][0];
1585 s->current_picture.f.motion_val[dir][xy + 1 + v->blocks_off][1] = s->current_picture.f.motion_val[dir][xy + v->blocks_off][1];
1586 s->current_picture.f.motion_val[dir][xy + wrap + v->blocks_off][0] = s->current_picture.f.motion_val[dir][xy + v->blocks_off][0];
1587 s->current_picture.f.motion_val[dir][xy + wrap + v->blocks_off][1] = s->current_picture.f.motion_val[dir][xy + v->blocks_off][1];
1588 s->current_picture.f.motion_val[dir][xy + wrap + 1 + v->blocks_off][0] = s->current_picture.f.motion_val[dir][xy + v->blocks_off][0];
1589 s->current_picture.f.motion_val[dir][xy + wrap + 1 + v->blocks_off][1] = s->current_picture.f.motion_val[dir][xy + v->blocks_off][1];
1590 v->mv_f[dir][xy + 1 + v->blocks_off] = v->mv_f[dir][xy + v->blocks_off];
1591 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];
1595 /** Predict and set motion vector for interlaced frame picture MBs
1597 static inline void vc1_pred_mv_intfr(VC1Context *v, int n, int dmv_x, int dmv_y,
1598 int mvn, int r_x, int r_y, uint8_t* is_intra)
1600 MpegEncContext *s = &v->s;
1601 int xy, wrap, off = 0;
1602 int A[2], B[2], C[2];
1604 int a_valid = 0, b_valid = 0, c_valid = 0;
1605 int field_a, field_b, field_c; // 0: same, 1: opposit
1606 int total_valid, num_samefield, num_oppfield;
1607 int pos_c, pos_b, n_adj;
1609 wrap = s->b8_stride;
1610 xy = s->block_index[n];
1613 s->mv[0][n][0] = s->current_picture.f.motion_val[0][xy][0] = 0;
1614 s->mv[0][n][1] = s->current_picture.f.motion_val[0][xy][1] = 0;
1615 s->current_picture.f.motion_val[1][xy][0] = 0;
1616 s->current_picture.f.motion_val[1][xy][1] = 0;
1617 if (mvn == 1) { /* duplicate motion data for 1-MV block */
1618 s->current_picture.f.motion_val[0][xy + 1][0] = 0;
1619 s->current_picture.f.motion_val[0][xy + 1][1] = 0;
1620 s->current_picture.f.motion_val[0][xy + wrap][0] = 0;
1621 s->current_picture.f.motion_val[0][xy + wrap][1] = 0;
1622 s->current_picture.f.motion_val[0][xy + wrap + 1][0] = 0;
1623 s->current_picture.f.motion_val[0][xy + wrap + 1][1] = 0;
1624 v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
1625 s->current_picture.f.motion_val[1][xy + 1][0] = 0;
1626 s->current_picture.f.motion_val[1][xy + 1][1] = 0;
1627 s->current_picture.f.motion_val[1][xy + wrap][0] = 0;
1628 s->current_picture.f.motion_val[1][xy + wrap][1] = 0;
1629 s->current_picture.f.motion_val[1][xy + wrap + 1][0] = 0;
1630 s->current_picture.f.motion_val[1][xy + wrap + 1][1] = 0;
1635 off = ((n == 0) || (n == 1)) ? 1 : -1;
1637 if (s->mb_x || (n == 1) || (n == 3)) {
1638 if ((v->blk_mv_type[xy]) // current block (MB) has a field MV
1639 || (!v->blk_mv_type[xy] && !v->blk_mv_type[xy - 1])) { // or both have frame MV
1640 A[0] = s->current_picture.f.motion_val[0][xy - 1][0];
1641 A[1] = s->current_picture.f.motion_val[0][xy - 1][1];
1643 } else { // current block has frame mv and cand. has field MV (so average)
1644 A[0] = (s->current_picture.f.motion_val[0][xy - 1][0]
1645 + s->current_picture.f.motion_val[0][xy - 1 + off * wrap][0] + 1) >> 1;
1646 A[1] = (s->current_picture.f.motion_val[0][xy - 1][1]
1647 + s->current_picture.f.motion_val[0][xy - 1 + off * wrap][1] + 1) >> 1;
1650 if (!(n & 1) && v->is_intra[s->mb_x - 1]) {
1656 /* Predict B and C */
1657 B[0] = B[1] = C[0] = C[1] = 0;
1658 if (n == 0 || n == 1 || v->blk_mv_type[xy]) {
1659 if (!s->first_slice_line) {
1660 if (!v->is_intra[s->mb_x - s->mb_stride]) {
1663 pos_b = s->block_index[n_adj] - 2 * wrap;
1664 if (v->blk_mv_type[pos_b] && v->blk_mv_type[xy]) {
1665 n_adj = (n & 2) | (n & 1);
1667 B[0] = s->current_picture.f.motion_val[0][s->block_index[n_adj] - 2 * wrap][0];
1668 B[1] = s->current_picture.f.motion_val[0][s->block_index[n_adj] - 2 * wrap][1];
1669 if (v->blk_mv_type[pos_b] && !v->blk_mv_type[xy]) {
1670 B[0] = (B[0] + s->current_picture.f.motion_val[0][s->block_index[n_adj ^ 2] - 2 * wrap][0] + 1) >> 1;
1671 B[1] = (B[1] + s->current_picture.f.motion_val[0][s->block_index[n_adj ^ 2] - 2 * wrap][1] + 1) >> 1;
1674 if (s->mb_width > 1) {
1675 if (!v->is_intra[s->mb_x - s->mb_stride + 1]) {
1678 pos_c = s->block_index[2] - 2 * wrap + 2;
1679 if (v->blk_mv_type[pos_c] && v->blk_mv_type[xy]) {
1682 C[0] = s->current_picture.f.motion_val[0][s->block_index[n_adj] - 2 * wrap + 2][0];
1683 C[1] = s->current_picture.f.motion_val[0][s->block_index[n_adj] - 2 * wrap + 2][1];
1684 if (v->blk_mv_type[pos_c] && !v->blk_mv_type[xy]) {
1685 C[0] = (1 + C[0] + (s->current_picture.f.motion_val[0][s->block_index[n_adj ^ 2] - 2 * wrap + 2][0])) >> 1;
1686 C[1] = (1 + C[1] + (s->current_picture.f.motion_val[0][s->block_index[n_adj ^ 2] - 2 * wrap + 2][1])) >> 1;
1688 if (s->mb_x == s->mb_width - 1) {
1689 if (!v->is_intra[s->mb_x - s->mb_stride - 1]) {
1692 pos_c = s->block_index[3] - 2 * wrap - 2;
1693 if (v->blk_mv_type[pos_c] && v->blk_mv_type[xy]) {
1696 C[0] = s->current_picture.f.motion_val[0][s->block_index[n_adj] - 2 * wrap - 2][0];
1697 C[1] = s->current_picture.f.motion_val[0][s->block_index[n_adj] - 2 * wrap - 2][1];
1698 if (v->blk_mv_type[pos_c] && !v->blk_mv_type[xy]) {
1699 C[0] = (1 + C[0] + s->current_picture.f.motion_val[0][s->block_index[1] - 2 * wrap - 2][0]) >> 1;
1700 C[1] = (1 + C[1] + s->current_picture.f.motion_val[0][s->block_index[1] - 2 * wrap - 2][1]) >> 1;
1709 pos_b = s->block_index[1];
1711 B[0] = s->current_picture.f.motion_val[0][pos_b][0];
1712 B[1] = s->current_picture.f.motion_val[0][pos_b][1];
1713 pos_c = s->block_index[0];
1715 C[0] = s->current_picture.f.motion_val[0][pos_c][0];
1716 C[1] = s->current_picture.f.motion_val[0][pos_c][1];
1719 total_valid = a_valid + b_valid + c_valid;
1720 // check if predictor A is out of bounds
1721 if (!s->mb_x && !(n == 1 || n == 3)) {
1724 // check if predictor B is out of bounds
1725 if ((s->first_slice_line && v->blk_mv_type[xy]) || (s->first_slice_line && !(n & 2))) {
1726 B[0] = B[1] = C[0] = C[1] = 0;
1728 if (!v->blk_mv_type[xy]) {
1729 if (s->mb_width == 1) {
1733 if (total_valid >= 2) {
1734 px = mid_pred(A[0], B[0], C[0]);
1735 py = mid_pred(A[1], B[1], C[1]);
1736 } else if (total_valid) {
1737 if (a_valid) { px = A[0]; py = A[1]; }
1738 if (b_valid) { px = B[0]; py = B[1]; }
1739 if (c_valid) { px = C[0]; py = C[1]; }
1745 field_a = (A[1] & 4) ? 1 : 0;
1749 field_b = (B[1] & 4) ? 1 : 0;
1753 field_c = (C[1] & 4) ? 1 : 0;
1757 num_oppfield = field_a + field_b + field_c;
1758 num_samefield = total_valid - num_oppfield;
1759 if (total_valid == 3) {
1760 if ((num_samefield == 3) || (num_oppfield == 3)) {
1761 px = mid_pred(A[0], B[0], C[0]);
1762 py = mid_pred(A[1], B[1], C[1]);
1763 } else if (num_samefield >= num_oppfield) {
1764 /* take one MV from same field set depending on priority
1765 the check for B may not be necessary */
1766 px = !field_a ? A[0] : B[0];
1767 py = !field_a ? A[1] : B[1];
1769 px = field_a ? A[0] : B[0];
1770 py = field_a ? A[1] : B[1];
1772 } else if (total_valid == 2) {
1773 if (num_samefield >= num_oppfield) {
1774 if (!field_a && a_valid) {
1777 } else if (!field_b && b_valid) {
1780 } else if (c_valid) {
1785 if (field_a && a_valid) {
1788 } else if (field_b && b_valid) {
1791 } else if (c_valid) {
1796 } else if (total_valid == 1) {
1797 px = (a_valid) ? A[0] : ((b_valid) ? B[0] : C[0]);
1798 py = (a_valid) ? A[1] : ((b_valid) ? B[1] : C[1]);
1803 /* store MV using signed modulus of MV range defined in 4.11 */
1804 s->mv[0][n][0] = s->current_picture.f.motion_val[0][xy][0] = ((px + dmv_x + r_x) & ((r_x << 1) - 1)) - r_x;
1805 s->mv[0][n][1] = s->current_picture.f.motion_val[0][xy][1] = ((py + dmv_y + r_y) & ((r_y << 1) - 1)) - r_y;
1806 if (mvn == 1) { /* duplicate motion data for 1-MV block */
1807 s->current_picture.f.motion_val[0][xy + 1 ][0] = s->current_picture.f.motion_val[0][xy][0];
1808 s->current_picture.f.motion_val[0][xy + 1 ][1] = s->current_picture.f.motion_val[0][xy][1];
1809 s->current_picture.f.motion_val[0][xy + wrap ][0] = s->current_picture.f.motion_val[0][xy][0];
1810 s->current_picture.f.motion_val[0][xy + wrap ][1] = s->current_picture.f.motion_val[0][xy][1];
1811 s->current_picture.f.motion_val[0][xy + wrap + 1][0] = s->current_picture.f.motion_val[0][xy][0];
1812 s->current_picture.f.motion_val[0][xy + wrap + 1][1] = s->current_picture.f.motion_val[0][xy][1];
1813 } else if (mvn == 2) { /* duplicate motion data for 2-Field MV block */
1814 s->current_picture.f.motion_val[0][xy + 1][0] = s->current_picture.f.motion_val[0][xy][0];
1815 s->current_picture.f.motion_val[0][xy + 1][1] = s->current_picture.f.motion_val[0][xy][1];
1816 s->mv[0][n + 1][0] = s->mv[0][n][0];
1817 s->mv[0][n + 1][1] = s->mv[0][n][1];
1821 /** Motion compensation for direct or interpolated blocks in B-frames
1823 static void vc1_interp_mc(VC1Context *v)
1825 MpegEncContext *s = &v->s;
1826 DSPContext *dsp = &v->s.dsp;
1827 uint8_t *srcY, *srcU, *srcV;
1828 int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
1830 int v_edge_pos = s->v_edge_pos >> v->field_mode;
1832 if (!v->field_mode && !v->s.next_picture.f.data[0])
1835 mx = s->mv[1][0][0];
1836 my = s->mv[1][0][1];
1837 uvmx = (mx + ((mx & 3) == 3)) >> 1;
1838 uvmy = (my + ((my & 3) == 3)) >> 1;
1839 if (v->field_mode) {
1840 if (v->cur_field_type != v->ref_field_type[1])
1841 my = my - 2 + 4 * v->cur_field_type;
1842 uvmy = uvmy - 2 + 4 * v->cur_field_type;
1845 uvmx = uvmx + ((uvmx < 0) ? -(uvmx & 1) : (uvmx & 1));
1846 uvmy = uvmy + ((uvmy < 0) ? -(uvmy & 1) : (uvmy & 1));
1848 srcY = s->next_picture.f.data[0];
1849 srcU = s->next_picture.f.data[1];
1850 srcV = s->next_picture.f.data[2];
1852 src_x = s->mb_x * 16 + (mx >> 2);
1853 src_y = s->mb_y * 16 + (my >> 2);
1854 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
1855 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
1857 if (v->profile != PROFILE_ADVANCED) {
1858 src_x = av_clip( src_x, -16, s->mb_width * 16);
1859 src_y = av_clip( src_y, -16, s->mb_height * 16);
1860 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
1861 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
1863 src_x = av_clip( src_x, -17, s->avctx->coded_width);
1864 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
1865 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
1866 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
1869 srcY += src_y * s->linesize + src_x;
1870 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
1871 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
1873 if (v->field_mode && v->ref_field_type[1]) {
1874 srcY += s->current_picture_ptr->f.linesize[0];
1875 srcU += s->current_picture_ptr->f.linesize[1];
1876 srcV += s->current_picture_ptr->f.linesize[2];
1879 /* for grayscale we should not try to read from unknown area */
1880 if (s->flags & CODEC_FLAG_GRAY) {
1881 srcU = s->edge_emu_buffer + 18 * s->linesize;
1882 srcV = s->edge_emu_buffer + 18 * s->linesize;
1885 if (v->rangeredfrm || s->h_edge_pos < 22 || v_edge_pos < 22
1886 || (unsigned)(src_x - 1) > s->h_edge_pos - (mx & 3) - 16 - 3
1887 || (unsigned)(src_y - 1) > v_edge_pos - (my & 3) - 16 - 3) {
1888 uint8_t *uvbuf = s->edge_emu_buffer + 19 * s->linesize;
1890 srcY -= s->mspel * (1 + s->linesize);
1891 s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize,
1892 17 + s->mspel * 2, 17 + s->mspel * 2,
1893 src_x - s->mspel, src_y - s->mspel,
1894 s->h_edge_pos, v_edge_pos);
1895 srcY = s->edge_emu_buffer;
1896 s->dsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, 8 + 1, 8 + 1,
1897 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
1898 s->dsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8 + 1, 8 + 1,
1899 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
1902 /* if we deal with range reduction we need to scale source blocks */
1903 if (v->rangeredfrm) {
1905 uint8_t *src, *src2;
1908 for (j = 0; j < 17 + s->mspel * 2; j++) {
1909 for (i = 0; i < 17 + s->mspel * 2; i++)
1910 src[i] = ((src[i] - 128) >> 1) + 128;
1915 for (j = 0; j < 9; j++) {
1916 for (i = 0; i < 9; i++) {
1917 src[i] = ((src[i] - 128) >> 1) + 128;
1918 src2[i] = ((src2[i] - 128) >> 1) + 128;
1920 src += s->uvlinesize;
1921 src2 += s->uvlinesize;
1924 srcY += s->mspel * (1 + s->linesize);
1927 if (v->field_mode && v->cur_field_type) {
1928 off = s->current_picture_ptr->f.linesize[0];
1929 off_uv = s->current_picture_ptr->f.linesize[1];
1936 dxy = ((my & 3) << 2) | (mx & 3);
1937 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off , srcY , s->linesize, v->rnd);
1938 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8, srcY + 8, s->linesize, v->rnd);
1939 srcY += s->linesize * 8;
1940 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize , srcY , s->linesize, v->rnd);
1941 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
1943 dxy = (my & 2) | ((mx & 2) >> 1);
1946 dsp->avg_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
1948 dsp->avg_no_rnd_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
1951 if (s->flags & CODEC_FLAG_GRAY) return;
1952 /* Chroma MC always uses qpel blilinear */
1953 uvmx = (uvmx & 3) << 1;
1954 uvmy = (uvmy & 3) << 1;
1956 dsp->avg_h264_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
1957 dsp->avg_h264_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
1959 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
1960 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
1964 static av_always_inline int scale_mv(int value, int bfrac, int inv, int qs)
1968 #if B_FRACTION_DEN==256
1972 return 2 * ((value * n + 255) >> 9);
1973 return (value * n + 128) >> 8;
1976 n -= B_FRACTION_DEN;
1978 return 2 * ((value * n + B_FRACTION_DEN - 1) / (2 * B_FRACTION_DEN));
1979 return (value * n + B_FRACTION_DEN/2) / B_FRACTION_DEN;
1983 /** Reconstruct motion vector for B-frame and do motion compensation
1985 static inline void vc1_b_mc(VC1Context *v, int dmv_x[2], int dmv_y[2],
1986 int direct, int mode)
1989 v->mv_mode2 = v->mv_mode;
1990 v->mv_mode = MV_PMODE_INTENSITY_COMP;
1996 v->mv_mode = v->mv_mode2;
1999 if (mode == BMV_TYPE_INTERPOLATED) {
2003 v->mv_mode = v->mv_mode2;
2007 if (v->use_ic && (mode == BMV_TYPE_BACKWARD))
2008 v->mv_mode = v->mv_mode2;
2009 vc1_mc_1mv(v, (mode == BMV_TYPE_BACKWARD));
2011 v->mv_mode = v->mv_mode2;
2014 static inline void vc1_pred_b_mv(VC1Context *v, int dmv_x[2], int dmv_y[2],
2015 int direct, int mvtype)
2017 MpegEncContext *s = &v->s;
2018 int xy, wrap, off = 0;
2023 const uint8_t *is_intra = v->mb_type[0];
2027 /* scale MV difference to be quad-pel */
2028 dmv_x[0] <<= 1 - s->quarter_sample;
2029 dmv_y[0] <<= 1 - s->quarter_sample;
2030 dmv_x[1] <<= 1 - s->quarter_sample;
2031 dmv_y[1] <<= 1 - s->quarter_sample;
2033 wrap = s->b8_stride;
2034 xy = s->block_index[0];
2037 s->current_picture.f.motion_val[0][xy + v->blocks_off][0] =
2038 s->current_picture.f.motion_val[0][xy + v->blocks_off][1] =
2039 s->current_picture.f.motion_val[1][xy + v->blocks_off][0] =
2040 s->current_picture.f.motion_val[1][xy + v->blocks_off][1] = 0;
2043 if (!v->field_mode) {
2044 s->mv[0][0][0] = scale_mv(s->next_picture.f.motion_val[1][xy][0], v->bfraction, 0, s->quarter_sample);
2045 s->mv[0][0][1] = scale_mv(s->next_picture.f.motion_val[1][xy][1], v->bfraction, 0, s->quarter_sample);
2046 s->mv[1][0][0] = scale_mv(s->next_picture.f.motion_val[1][xy][0], v->bfraction, 1, s->quarter_sample);
2047 s->mv[1][0][1] = scale_mv(s->next_picture.f.motion_val[1][xy][1], v->bfraction, 1, s->quarter_sample);
2049 /* Pullback predicted motion vectors as specified in 8.4.5.4 */
2050 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));
2051 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));
2052 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));
2053 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));
2056 s->current_picture.f.motion_val[0][xy + v->blocks_off][0] = s->mv[0][0][0];
2057 s->current_picture.f.motion_val[0][xy + v->blocks_off][1] = s->mv[0][0][1];
2058 s->current_picture.f.motion_val[1][xy + v->blocks_off][0] = s->mv[1][0][0];
2059 s->current_picture.f.motion_val[1][xy + v->blocks_off][1] = s->mv[1][0][1];
2063 if ((mvtype == BMV_TYPE_FORWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
2064 C = s->current_picture.f.motion_val[0][xy - 2];
2065 A = s->current_picture.f.motion_val[0][xy - wrap * 2];
2066 off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
2067 B = s->current_picture.f.motion_val[0][xy - wrap * 2 + off];
2069 if (!s->mb_x) C[0] = C[1] = 0;
2070 if (!s->first_slice_line) { // predictor A is not out of bounds
2071 if (s->mb_width == 1) {
2075 px = mid_pred(A[0], B[0], C[0]);
2076 py = mid_pred(A[1], B[1], C[1]);
2078 } else if (s->mb_x) { // predictor C is not out of bounds
2084 /* Pullback MV as specified in 8.3.5.3.4 */
2087 if (v->profile < PROFILE_ADVANCED) {
2088 qx = (s->mb_x << 5);
2089 qy = (s->mb_y << 5);
2090 X = (s->mb_width << 5) - 4;
2091 Y = (s->mb_height << 5) - 4;
2092 if (qx + px < -28) px = -28 - qx;
2093 if (qy + py < -28) py = -28 - qy;
2094 if (qx + px > X) px = X - qx;
2095 if (qy + py > Y) py = Y - qy;
2097 qx = (s->mb_x << 6);
2098 qy = (s->mb_y << 6);
2099 X = (s->mb_width << 6) - 4;
2100 Y = (s->mb_height << 6) - 4;
2101 if (qx + px < -60) px = -60 - qx;
2102 if (qy + py < -60) py = -60 - qy;
2103 if (qx + px > X) px = X - qx;
2104 if (qy + py > Y) py = Y - qy;
2107 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
2108 if (0 && !s->first_slice_line && s->mb_x) {
2109 if (is_intra[xy - wrap])
2110 sum = FFABS(px) + FFABS(py);
2112 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
2114 if (get_bits1(&s->gb)) {
2122 if (is_intra[xy - 2])
2123 sum = FFABS(px) + FFABS(py);
2125 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
2127 if (get_bits1(&s->gb)) {
2137 /* store MV using signed modulus of MV range defined in 4.11 */
2138 s->mv[0][0][0] = ((px + dmv_x[0] + r_x) & ((r_x << 1) - 1)) - r_x;
2139 s->mv[0][0][1] = ((py + dmv_y[0] + r_y) & ((r_y << 1) - 1)) - r_y;
2141 if ((mvtype == BMV_TYPE_BACKWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
2142 C = s->current_picture.f.motion_val[1][xy - 2];
2143 A = s->current_picture.f.motion_val[1][xy - wrap * 2];
2144 off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
2145 B = s->current_picture.f.motion_val[1][xy - wrap * 2 + off];
2149 if (!s->first_slice_line) { // predictor A is not out of bounds
2150 if (s->mb_width == 1) {
2154 px = mid_pred(A[0], B[0], C[0]);
2155 py = mid_pred(A[1], B[1], C[1]);
2157 } else if (s->mb_x) { // predictor C is not out of bounds
2163 /* Pullback MV as specified in 8.3.5.3.4 */
2166 if (v->profile < PROFILE_ADVANCED) {
2167 qx = (s->mb_x << 5);
2168 qy = (s->mb_y << 5);
2169 X = (s->mb_width << 5) - 4;
2170 Y = (s->mb_height << 5) - 4;
2171 if (qx + px < -28) px = -28 - qx;
2172 if (qy + py < -28) py = -28 - qy;
2173 if (qx + px > X) px = X - qx;
2174 if (qy + py > Y) py = Y - qy;
2176 qx = (s->mb_x << 6);
2177 qy = (s->mb_y << 6);
2178 X = (s->mb_width << 6) - 4;
2179 Y = (s->mb_height << 6) - 4;
2180 if (qx + px < -60) px = -60 - qx;
2181 if (qy + py < -60) py = -60 - qy;
2182 if (qx + px > X) px = X - qx;
2183 if (qy + py > Y) py = Y - qy;
2186 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
2187 if (0 && !s->first_slice_line && s->mb_x) {
2188 if (is_intra[xy - wrap])
2189 sum = FFABS(px) + FFABS(py);
2191 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
2193 if (get_bits1(&s->gb)) {
2201 if (is_intra[xy - 2])
2202 sum = FFABS(px) + FFABS(py);
2204 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
2206 if (get_bits1(&s->gb)) {
2216 /* store MV using signed modulus of MV range defined in 4.11 */
2218 s->mv[1][0][0] = ((px + dmv_x[1] + r_x) & ((r_x << 1) - 1)) - r_x;
2219 s->mv[1][0][1] = ((py + dmv_y[1] + r_y) & ((r_y << 1) - 1)) - r_y;
2221 s->current_picture.f.motion_val[0][xy][0] = s->mv[0][0][0];
2222 s->current_picture.f.motion_val[0][xy][1] = s->mv[0][0][1];
2223 s->current_picture.f.motion_val[1][xy][0] = s->mv[1][0][0];
2224 s->current_picture.f.motion_val[1][xy][1] = s->mv[1][0][1];
2227 static inline void vc1_pred_b_mv_intfi(VC1Context *v, int n, int *dmv_x, int *dmv_y, int mv1, int *pred_flag)
2229 int dir = (v->bmvtype == BMV_TYPE_BACKWARD) ? 1 : 0;
2230 MpegEncContext *s = &v->s;
2231 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2233 if (v->bmvtype == BMV_TYPE_DIRECT) {
2234 int total_opp, k, f;
2235 if (s->next_picture.f.mb_type[mb_pos + v->mb_off] != MB_TYPE_INTRA) {
2236 s->mv[0][0][0] = scale_mv(s->next_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][0],
2237 v->bfraction, 0, s->quarter_sample);
2238 s->mv[0][0][1] = scale_mv(s->next_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][1],
2239 v->bfraction, 0, s->quarter_sample);
2240 s->mv[1][0][0] = scale_mv(s->next_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][0],
2241 v->bfraction, 1, s->quarter_sample);
2242 s->mv[1][0][1] = scale_mv(s->next_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][1],
2243 v->bfraction, 1, s->quarter_sample);
2245 total_opp = v->mv_f_next[0][s->block_index[0] + v->blocks_off]
2246 + v->mv_f_next[0][s->block_index[1] + v->blocks_off]
2247 + v->mv_f_next[0][s->block_index[2] + v->blocks_off]
2248 + v->mv_f_next[0][s->block_index[3] + v->blocks_off];
2249 f = (total_opp > 2) ? 1 : 0;
2251 s->mv[0][0][0] = s->mv[0][0][1] = 0;
2252 s->mv[1][0][0] = s->mv[1][0][1] = 0;
2255 v->ref_field_type[0] = v->ref_field_type[1] = v->cur_field_type ^ f;
2256 for (k = 0; k < 4; k++) {
2257 s->current_picture.f.motion_val[0][s->block_index[k] + v->blocks_off][0] = s->mv[0][0][0];
2258 s->current_picture.f.motion_val[0][s->block_index[k] + v->blocks_off][1] = s->mv[0][0][1];
2259 s->current_picture.f.motion_val[1][s->block_index[k] + v->blocks_off][0] = s->mv[1][0][0];
2260 s->current_picture.f.motion_val[1][s->block_index[k] + v->blocks_off][1] = s->mv[1][0][1];
2261 v->mv_f[0][s->block_index[k] + v->blocks_off] = f;
2262 v->mv_f[1][s->block_index[k] + v->blocks_off] = f;
2266 if (v->bmvtype == BMV_TYPE_INTERPOLATED) {
2267 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);
2268 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);
2271 if (dir) { // backward
2272 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);
2273 if (n == 3 || mv1) {
2274 vc1_pred_mv(v, 0, dmv_x[0], dmv_y[0], 1, v->range_x, v->range_y, v->mb_type[0], 0, 0);
2277 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);
2278 if (n == 3 || mv1) {
2279 vc1_pred_mv(v, 0, dmv_x[1], dmv_y[1], 1, v->range_x, v->range_y, v->mb_type[0], 0, 1);
2284 /** Get predicted DC value for I-frames only
2285 * prediction dir: left=0, top=1
2286 * @param s MpegEncContext
2287 * @param overlap flag indicating that overlap filtering is used
2288 * @param pq integer part of picture quantizer
2289 * @param[in] n block index in the current MB
2290 * @param dc_val_ptr Pointer to DC predictor
2291 * @param dir_ptr Prediction direction for use in AC prediction
2293 static inline int vc1_i_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
2294 int16_t **dc_val_ptr, int *dir_ptr)
2296 int a, b, c, wrap, pred, scale;
2298 static const uint16_t dcpred[32] = {
2299 -1, 1024, 512, 341, 256, 205, 171, 146, 128,
2300 114, 102, 93, 85, 79, 73, 68, 64,
2301 60, 57, 54, 51, 49, 47, 45, 43,
2302 41, 39, 38, 37, 35, 34, 33
2305 /* find prediction - wmv3_dc_scale always used here in fact */
2306 if (n < 4) scale = s->y_dc_scale;
2307 else scale = s->c_dc_scale;
2309 wrap = s->block_wrap[n];
2310 dc_val = s->dc_val[0] + s->block_index[n];
2316 b = dc_val[ - 1 - wrap];
2317 a = dc_val[ - wrap];
2319 if (pq < 9 || !overlap) {
2320 /* Set outer values */
2321 if (s->first_slice_line && (n != 2 && n != 3))
2322 b = a = dcpred[scale];
2323 if (s->mb_x == 0 && (n != 1 && n != 3))
2324 b = c = dcpred[scale];
2326 /* Set outer values */
2327 if (s->first_slice_line && (n != 2 && n != 3))
2329 if (s->mb_x == 0 && (n != 1 && n != 3))
2333 if (abs(a - b) <= abs(b - c)) {
2335 *dir_ptr = 1; // left
2338 *dir_ptr = 0; // top
2341 /* update predictor */
2342 *dc_val_ptr = &dc_val[0];
2347 /** Get predicted DC value
2348 * prediction dir: left=0, top=1
2349 * @param s MpegEncContext
2350 * @param overlap flag indicating that overlap filtering is used
2351 * @param pq integer part of picture quantizer
2352 * @param[in] n block index in the current MB
2353 * @param a_avail flag indicating top block availability
2354 * @param c_avail flag indicating left block availability
2355 * @param dc_val_ptr Pointer to DC predictor
2356 * @param dir_ptr Prediction direction for use in AC prediction
2358 static inline int vc1_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
2359 int a_avail, int c_avail,
2360 int16_t **dc_val_ptr, int *dir_ptr)
2362 int a, b, c, wrap, pred;
2364 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2368 wrap = s->block_wrap[n];
2369 dc_val = s->dc_val[0] + s->block_index[n];
2375 b = dc_val[ - 1 - wrap];
2376 a = dc_val[ - wrap];
2377 /* scale predictors if needed */
2378 q1 = s->current_picture.f.qscale_table[mb_pos];
2379 dqscale_index = s->y_dc_scale_table[q1] - 1;
2380 if (dqscale_index < 0)
2382 if (c_avail && (n != 1 && n != 3)) {
2383 q2 = s->current_picture.f.qscale_table[mb_pos - 1];
2385 c = (c * s->y_dc_scale_table[q2] * ff_vc1_dqscale[dqscale_index] + 0x20000) >> 18;
2387 if (a_avail && (n != 2 && n != 3)) {
2388 q2 = s->current_picture.f.qscale_table[mb_pos - s->mb_stride];
2390 a = (a * s->y_dc_scale_table[q2] * ff_vc1_dqscale[dqscale_index] + 0x20000) >> 18;
2392 if (a_avail && c_avail && (n != 3)) {
2397 off -= s->mb_stride;
2398 q2 = s->current_picture.f.qscale_table[off];
2400 b = (b * s->y_dc_scale_table[q2] * ff_vc1_dqscale[dqscale_index] + 0x20000) >> 18;
2403 if (a_avail && c_avail) {
2404 if (abs(a - b) <= abs(b - c)) {
2406 *dir_ptr = 1; // left
2409 *dir_ptr = 0; // top
2411 } else if (a_avail) {
2413 *dir_ptr = 0; // top
2414 } else if (c_avail) {
2416 *dir_ptr = 1; // left
2419 *dir_ptr = 1; // left
2422 /* update predictor */
2423 *dc_val_ptr = &dc_val[0];
2427 /** @} */ // Block group
2430 * @name VC1 Macroblock-level functions in Simple/Main Profiles
2431 * @see 7.1.4, p91 and 8.1.1.7, p(1)04
2435 static inline int vc1_coded_block_pred(MpegEncContext * s, int n,
2436 uint8_t **coded_block_ptr)
2438 int xy, wrap, pred, a, b, c;
2440 xy = s->block_index[n];
2441 wrap = s->b8_stride;
2446 a = s->coded_block[xy - 1 ];
2447 b = s->coded_block[xy - 1 - wrap];
2448 c = s->coded_block[xy - wrap];
2457 *coded_block_ptr = &s->coded_block[xy];
2463 * Decode one AC coefficient
2464 * @param v The VC1 context
2465 * @param last Last coefficient
2466 * @param skip How much zero coefficients to skip
2467 * @param value Decoded AC coefficient value
2468 * @param codingset set of VLC to decode data
2471 static void vc1_decode_ac_coeff(VC1Context *v, int *last, int *skip,
2472 int *value, int codingset)
2474 GetBitContext *gb = &v->s.gb;
2475 int index, escape, run = 0, level = 0, lst = 0;
2477 index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
2478 if (index != ff_vc1_ac_sizes[codingset] - 1) {
2479 run = vc1_index_decode_table[codingset][index][0];
2480 level = vc1_index_decode_table[codingset][index][1];
2481 lst = index >= vc1_last_decode_table[codingset] || get_bits_left(gb) < 0;
2485 escape = decode210(gb);
2487 index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
2488 run = vc1_index_decode_table[codingset][index][0];
2489 level = vc1_index_decode_table[codingset][index][1];
2490 lst = index >= vc1_last_decode_table[codingset];
2493 level += vc1_last_delta_level_table[codingset][run];
2495 level += vc1_delta_level_table[codingset][run];
2498 run += vc1_last_delta_run_table[codingset][level] + 1;
2500 run += vc1_delta_run_table[codingset][level] + 1;
2506 lst = get_bits1(gb);
2507 if (v->s.esc3_level_length == 0) {
2508 if (v->pq < 8 || v->dquantfrm) { // table 59
2509 v->s.esc3_level_length = get_bits(gb, 3);
2510 if (!v->s.esc3_level_length)
2511 v->s.esc3_level_length = get_bits(gb, 2) + 8;
2512 } else { // table 60
2513 v->s.esc3_level_length = get_unary(gb, 1, 6) + 2;
2515 v->s.esc3_run_length = 3 + get_bits(gb, 2);
2517 run = get_bits(gb, v->s.esc3_run_length);
2518 sign = get_bits1(gb);
2519 level = get_bits(gb, v->s.esc3_level_length);
2530 /** Decode intra block in intra frames - should be faster than decode_intra_block
2531 * @param v VC1Context
2532 * @param block block to decode
2533 * @param[in] n subblock index
2534 * @param coded are AC coeffs present or not
2535 * @param codingset set of VLC to decode data
2537 static int vc1_decode_i_block(VC1Context *v, DCTELEM block[64], int n,
2538 int coded, int codingset)
2540 GetBitContext *gb = &v->s.gb;
2541 MpegEncContext *s = &v->s;
2542 int dc_pred_dir = 0; /* Direction of the DC prediction used */
2545 int16_t *ac_val, *ac_val2;
2548 /* Get DC differential */
2550 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2552 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2555 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
2559 if (dcdiff == 119 /* ESC index value */) {
2560 /* TODO: Optimize */
2561 if (v->pq == 1) dcdiff = get_bits(gb, 10);
2562 else if (v->pq == 2) dcdiff = get_bits(gb, 9);
2563 else dcdiff = get_bits(gb, 8);
2566 dcdiff = (dcdiff << 2) + get_bits(gb, 2) - 3;
2567 else if (v->pq == 2)
2568 dcdiff = (dcdiff << 1) + get_bits1(gb) - 1;
2575 dcdiff += vc1_i_pred_dc(&v->s, v->overlap, v->pq, n, &dc_val, &dc_pred_dir);
2578 /* Store the quantized DC coeff, used for prediction */
2580 block[0] = dcdiff * s->y_dc_scale;
2582 block[0] = dcdiff * s->c_dc_scale;
2593 int last = 0, skip, value;
2594 const uint8_t *zz_table;
2598 scale = v->pq * 2 + v->halfpq;
2602 zz_table = v->zz_8x8[2];
2604 zz_table = v->zz_8x8[3];
2606 zz_table = v->zz_8x8[1];
2608 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
2610 if (dc_pred_dir) // left
2613 ac_val -= 16 * s->block_wrap[n];
2616 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
2620 block[zz_table[i++]] = value;
2623 /* apply AC prediction if needed */
2625 if (dc_pred_dir) { // left
2626 for (k = 1; k < 8; k++)
2627 block[k << v->left_blk_sh] += ac_val[k];
2629 for (k = 1; k < 8; k++)
2630 block[k << v->top_blk_sh] += ac_val[k + 8];
2633 /* save AC coeffs for further prediction */
2634 for (k = 1; k < 8; k++) {
2635 ac_val2[k] = block[k << v->left_blk_sh];
2636 ac_val2[k + 8] = block[k << v->top_blk_sh];
2639 /* scale AC coeffs */
2640 for (k = 1; k < 64; k++)
2644 block[k] += (block[k] < 0) ? -v->pq : v->pq;
2647 if (s->ac_pred) i = 63;
2653 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
2657 scale = v->pq * 2 + v->halfpq;
2658 memset(ac_val2, 0, 16 * 2);
2659 if (dc_pred_dir) { // left
2662 memcpy(ac_val2, ac_val, 8 * 2);
2664 ac_val -= 16 * s->block_wrap[n];
2666 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
2669 /* apply AC prediction if needed */
2671 if (dc_pred_dir) { //left
2672 for (k = 1; k < 8; k++) {
2673 block[k << v->left_blk_sh] = ac_val[k] * scale;
2674 if (!v->pquantizer && block[k << v->left_blk_sh])
2675 block[k << v->left_blk_sh] += (block[k << v->left_blk_sh] < 0) ? -v->pq : v->pq;
2678 for (k = 1; k < 8; k++) {
2679 block[k << v->top_blk_sh] = ac_val[k + 8] * scale;
2680 if (!v->pquantizer && block[k << v->top_blk_sh])
2681 block[k << v->top_blk_sh] += (block[k << v->top_blk_sh] < 0) ? -v->pq : v->pq;
2687 s->block_last_index[n] = i;
2692 /** Decode intra block in intra frames - should be faster than decode_intra_block
2693 * @param v VC1Context
2694 * @param block block to decode
2695 * @param[in] n subblock number
2696 * @param coded are AC coeffs present or not
2697 * @param codingset set of VLC to decode data
2698 * @param mquant quantizer value for this macroblock
2700 static int vc1_decode_i_block_adv(VC1Context *v, DCTELEM block[64], int n,
2701 int coded, int codingset, int mquant)
2703 GetBitContext *gb = &v->s.gb;
2704 MpegEncContext *s = &v->s;
2705 int dc_pred_dir = 0; /* Direction of the DC prediction used */
2708 int16_t *ac_val, *ac_val2;
2710 int a_avail = v->a_avail, c_avail = v->c_avail;
2711 int use_pred = s->ac_pred;
2714 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2716 /* Get DC differential */
2718 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2720 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2723 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
2727 if (dcdiff == 119 /* ESC index value */) {
2728 /* TODO: Optimize */
2729 if (mquant == 1) dcdiff = get_bits(gb, 10);
2730 else if (mquant == 2) dcdiff = get_bits(gb, 9);
2731 else dcdiff = get_bits(gb, 8);
2734 dcdiff = (dcdiff << 2) + get_bits(gb, 2) - 3;
2735 else if (mquant == 2)
2736 dcdiff = (dcdiff << 1) + get_bits1(gb) - 1;
2743 dcdiff += vc1_pred_dc(&v->s, v->overlap, mquant, n, v->a_avail, v->c_avail, &dc_val, &dc_pred_dir);
2746 /* Store the quantized DC coeff, used for prediction */
2748 block[0] = dcdiff * s->y_dc_scale;
2750 block[0] = dcdiff * s->c_dc_scale;
2756 /* check if AC is needed at all */
2757 if (!a_avail && !c_avail)
2759 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
2762 scale = mquant * 2 + ((mquant == v->pq) ? v->halfpq : 0);
2764 if (dc_pred_dir) // left
2767 ac_val -= 16 * s->block_wrap[n];
2769 q1 = s->current_picture.f.qscale_table[mb_pos];
2770 if ( dc_pred_dir && c_avail && mb_pos)
2771 q2 = s->current_picture.f.qscale_table[mb_pos - 1];
2772 if (!dc_pred_dir && a_avail && mb_pos >= s->mb_stride)
2773 q2 = s->current_picture.f.qscale_table[mb_pos - s->mb_stride];
2774 if ( dc_pred_dir && n == 1)
2776 if (!dc_pred_dir && n == 2)
2782 int last = 0, skip, value;
2783 const uint8_t *zz_table;
2787 if (!use_pred && v->fcm == ILACE_FRAME) {
2788 zz_table = v->zzi_8x8;
2790 if (!dc_pred_dir) // top
2791 zz_table = v->zz_8x8[2];
2793 zz_table = v->zz_8x8[3];
2796 if (v->fcm != ILACE_FRAME)
2797 zz_table = v->zz_8x8[1];
2799 zz_table = v->zzi_8x8;
2803 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
2807 block[zz_table[i++]] = value;
2810 /* apply AC prediction if needed */
2812 /* scale predictors if needed*/
2813 if (q2 && q1 != q2) {
2814 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2815 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2818 return AVERROR_INVALIDDATA;
2819 if (dc_pred_dir) { // left
2820 for (k = 1; k < 8; k++)
2821 block[k << v->left_blk_sh] += (ac_val[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2823 for (k = 1; k < 8; k++)
2824 block[k << v->top_blk_sh] += (ac_val[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2827 if (dc_pred_dir) { //left
2828 for (k = 1; k < 8; k++)
2829 block[k << v->left_blk_sh] += ac_val[k];
2831 for (k = 1; k < 8; k++)
2832 block[k << v->top_blk_sh] += ac_val[k + 8];
2836 /* save AC coeffs for further prediction */
2837 for (k = 1; k < 8; k++) {
2838 ac_val2[k ] = block[k << v->left_blk_sh];
2839 ac_val2[k + 8] = block[k << v->top_blk_sh];
2842 /* scale AC coeffs */
2843 for (k = 1; k < 64; k++)
2847 block[k] += (block[k] < 0) ? -mquant : mquant;
2850 if (use_pred) i = 63;
2851 } else { // no AC coeffs
2854 memset(ac_val2, 0, 16 * 2);
2855 if (dc_pred_dir) { // left
2857 memcpy(ac_val2, ac_val, 8 * 2);
2858 if (q2 && q1 != q2) {
2859 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2860 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2862 return AVERROR_INVALIDDATA;
2863 for (k = 1; k < 8; k++)
2864 ac_val2[k] = (ac_val2[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2869 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
2870 if (q2 && q1 != q2) {
2871 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2872 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2874 return AVERROR_INVALIDDATA;
2875 for (k = 1; k < 8; k++)
2876 ac_val2[k + 8] = (ac_val2[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2881 /* apply AC prediction if needed */
2883 if (dc_pred_dir) { // left
2884 for (k = 1; k < 8; k++) {
2885 block[k << v->left_blk_sh] = ac_val2[k] * scale;
2886 if (!v->pquantizer && block[k << v->left_blk_sh])
2887 block[k << v->left_blk_sh] += (block[k << v->left_blk_sh] < 0) ? -mquant : mquant;
2890 for (k = 1; k < 8; k++) {
2891 block[k << v->top_blk_sh] = ac_val2[k + 8] * scale;
2892 if (!v->pquantizer && block[k << v->top_blk_sh])
2893 block[k << v->top_blk_sh] += (block[k << v->top_blk_sh] < 0) ? -mquant : mquant;
2899 s->block_last_index[n] = i;
2904 /** Decode intra block in inter frames - more generic version than vc1_decode_i_block
2905 * @param v VC1Context
2906 * @param block block to decode
2907 * @param[in] n subblock index
2908 * @param coded are AC coeffs present or not
2909 * @param mquant block quantizer
2910 * @param codingset set of VLC to decode data
2912 static int vc1_decode_intra_block(VC1Context *v, DCTELEM block[64], int n,
2913 int coded, int mquant, int codingset)
2915 GetBitContext *gb = &v->s.gb;
2916 MpegEncContext *s = &v->s;
2917 int dc_pred_dir = 0; /* Direction of the DC prediction used */
2920 int16_t *ac_val, *ac_val2;
2922 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2923 int a_avail = v->a_avail, c_avail = v->c_avail;
2924 int use_pred = s->ac_pred;
2928 s->dsp.clear_block(block);
2930 /* XXX: Guard against dumb values of mquant */
2931 mquant = (mquant < 1) ? 0 : ((mquant > 31) ? 31 : mquant);
2933 /* Set DC scale - y and c use the same */
2934 s->y_dc_scale = s->y_dc_scale_table[mquant];
2935 s->c_dc_scale = s->c_dc_scale_table[mquant];
2937 /* Get DC differential */
2939 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2941 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2944 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
2948 if (dcdiff == 119 /* ESC index value */) {
2949 /* TODO: Optimize */
2950 if (mquant == 1) dcdiff = get_bits(gb, 10);
2951 else if (mquant == 2) dcdiff = get_bits(gb, 9);
2952 else dcdiff = get_bits(gb, 8);
2955 dcdiff = (dcdiff << 2) + get_bits(gb, 2) - 3;
2956 else if (mquant == 2)
2957 dcdiff = (dcdiff << 1) + get_bits1(gb) - 1;
2964 dcdiff += vc1_pred_dc(&v->s, v->overlap, mquant, n, a_avail, c_avail, &dc_val, &dc_pred_dir);
2967 /* Store the quantized DC coeff, used for prediction */
2970 block[0] = dcdiff * s->y_dc_scale;
2972 block[0] = dcdiff * s->c_dc_scale;
2978 /* check if AC is needed at all and adjust direction if needed */
2979 if (!a_avail) dc_pred_dir = 1;
2980 if (!c_avail) dc_pred_dir = 0;
2981 if (!a_avail && !c_avail) use_pred = 0;
2982 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
2985 scale = mquant * 2 + v->halfpq;
2987 if (dc_pred_dir) //left
2990 ac_val -= 16 * s->block_wrap[n];
2992 q1 = s->current_picture.f.qscale_table[mb_pos];
2993 if (dc_pred_dir && c_avail && mb_pos)
2994 q2 = s->current_picture.f.qscale_table[mb_pos - 1];
2995 if (!dc_pred_dir && a_avail && mb_pos >= s->mb_stride)
2996 q2 = s->current_picture.f.qscale_table[mb_pos - s->mb_stride];
2997 if ( dc_pred_dir && n == 1)
2999 if (!dc_pred_dir && n == 2)
3001 if (n == 3) q2 = q1;
3004 int last = 0, skip, value;
3008 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
3012 if (v->fcm == PROGRESSIVE)
3013 block[v->zz_8x8[0][i++]] = value;
3015 if (use_pred && (v->fcm == ILACE_FRAME)) {
3016 if (!dc_pred_dir) // top
3017 block[v->zz_8x8[2][i++]] = value;
3019 block[v->zz_8x8[3][i++]] = value;
3021 block[v->zzi_8x8[i++]] = value;
3026 /* apply AC prediction if needed */
3028 /* scale predictors if needed*/
3029 if (q2 && q1 != q2) {
3030 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
3031 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
3034 return AVERROR_INVALIDDATA;
3035 if (dc_pred_dir) { // left
3036 for (k = 1; k < 8; k++)
3037 block[k << v->left_blk_sh] += (ac_val[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
3039 for (k = 1; k < 8; k++)
3040 block[k << v->top_blk_sh] += (ac_val[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
3043 if (dc_pred_dir) { // left
3044 for (k = 1; k < 8; k++)
3045 block[k << v->left_blk_sh] += ac_val[k];
3047 for (k = 1; k < 8; k++)
3048 block[k << v->top_blk_sh] += ac_val[k + 8];
3052 /* save AC coeffs for further prediction */
3053 for (k = 1; k < 8; k++) {
3054 ac_val2[k ] = block[k << v->left_blk_sh];
3055 ac_val2[k + 8] = block[k << v->top_blk_sh];
3058 /* scale AC coeffs */
3059 for (k = 1; k < 64; k++)
3063 block[k] += (block[k] < 0) ? -mquant : mquant;
3066 if (use_pred) i = 63;
3067 } else { // no AC coeffs
3070 memset(ac_val2, 0, 16 * 2);
3071 if (dc_pred_dir) { // left
3073 memcpy(ac_val2, ac_val, 8 * 2);
3074 if (q2 && q1 != q2) {
3075 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
3076 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
3078 return AVERROR_INVALIDDATA;
3079 for (k = 1; k < 8; k++)
3080 ac_val2[k] = (ac_val2[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
3085 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
3086 if (q2 && q1 != q2) {
3087 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
3088 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
3090 return AVERROR_INVALIDDATA;
3091 for (k = 1; k < 8; k++)
3092 ac_val2[k + 8] = (ac_val2[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
3097 /* apply AC prediction if needed */
3099 if (dc_pred_dir) { // left
3100 for (k = 1; k < 8; k++) {
3101 block[k << v->left_blk_sh] = ac_val2[k] * scale;
3102 if (!v->pquantizer && block[k << v->left_blk_sh])
3103 block[k << v->left_blk_sh] += (block[k << v->left_blk_sh] < 0) ? -mquant : mquant;
3106 for (k = 1; k < 8; k++) {
3107 block[k << v->top_blk_sh] = ac_val2[k + 8] * scale;
3108 if (!v->pquantizer && block[k << v->top_blk_sh])
3109 block[k << v->top_blk_sh] += (block[k << v->top_blk_sh] < 0) ? -mquant : mquant;
3115 s->block_last_index[n] = i;
3122 static int vc1_decode_p_block(VC1Context *v, DCTELEM block[64], int n,
3123 int mquant, int ttmb, int first_block,
3124 uint8_t *dst, int linesize, int skip_block,
3127 MpegEncContext *s = &v->s;
3128 GetBitContext *gb = &s->gb;
3131 int scale, off, idx, last, skip, value;
3132 int ttblk = ttmb & 7;
3135 s->dsp.clear_block(block);
3138 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)];
3140 if (ttblk == TT_4X4) {
3141 subblkpat = ~(get_vlc2(gb, ff_vc1_subblkpat_vlc[v->tt_index].table, VC1_SUBBLKPAT_VLC_BITS, 1) + 1);
3143 if ((ttblk != TT_8X8 && ttblk != TT_4X4)
3144 && ((v->ttmbf || (ttmb != -1 && (ttmb & 8) && !first_block))
3145 || (!v->res_rtm_flag && !first_block))) {
3146 subblkpat = decode012(gb);
3148 subblkpat ^= 3; // swap decoded pattern bits
3149 if (ttblk == TT_8X4_TOP || ttblk == TT_8X4_BOTTOM)
3151 if (ttblk == TT_4X8_RIGHT || ttblk == TT_4X8_LEFT)
3154 scale = 2 * mquant + ((v->pq == mquant) ? v->halfpq : 0);
3156 // convert transforms like 8X4_TOP to generic TT and SUBBLKPAT
3157 if (ttblk == TT_8X4_TOP || ttblk == TT_8X4_BOTTOM) {
3158 subblkpat = 2 - (ttblk == TT_8X4_TOP);
3161 if (ttblk == TT_4X8_RIGHT || ttblk == TT_4X8_LEFT) {
3162 subblkpat = 2 - (ttblk == TT_4X8_LEFT);
3171 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
3176 idx = v->zz_8x8[0][i++];
3178 idx = v->zzi_8x8[i++];
3179 block[idx] = value * scale;
3181 block[idx] += (block[idx] < 0) ? -mquant : mquant;
3185 v->vc1dsp.vc1_inv_trans_8x8_dc(dst, linesize, block);
3187 v->vc1dsp.vc1_inv_trans_8x8(block);
3188 s->dsp.add_pixels_clamped(block, dst, linesize);
3193 pat = ~subblkpat & 0xF;
3194 for (j = 0; j < 4; j++) {
3195 last = subblkpat & (1 << (3 - j));
3197 off = (j & 1) * 4 + (j & 2) * 16;
3199 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
3204 idx = ff_vc1_simple_progressive_4x4_zz[i++];
3206 idx = ff_vc1_adv_interlaced_4x4_zz[i++];
3207 block[idx + off] = value * scale;
3209 block[idx + off] += (block[idx + off] < 0) ? -mquant : mquant;
3211 if (!(subblkpat & (1 << (3 - j))) && !skip_block) {
3213 v->vc1dsp.vc1_inv_trans_4x4_dc(dst + (j & 1) * 4 + (j & 2) * 2 * linesize, linesize, block + off);
3215 v->vc1dsp.vc1_inv_trans_4x4(dst + (j & 1) * 4 + (j & 2) * 2 * linesize, linesize, block + off);
3220 pat = ~((subblkpat & 2) * 6 + (subblkpat & 1) * 3) & 0xF;
3221 for (j = 0; j < 2; j++) {
3222 last = subblkpat & (1 << (1 - j));
3226 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
3231 idx = v->zz_8x4[i++] + off;
3233 idx = ff_vc1_adv_interlaced_8x4_zz[i++] + off;
3234 block[idx] = value * scale;
3236 block[idx] += (block[idx] < 0) ? -mquant : mquant;
3238 if (!(subblkpat & (1 << (1 - j))) && !skip_block) {
3240 v->vc1dsp.vc1_inv_trans_8x4_dc(dst + j * 4 * linesize, linesize, block + off);
3242 v->vc1dsp.vc1_inv_trans_8x4(dst + j * 4 * linesize, linesize, block + off);
3247 pat = ~(subblkpat * 5) & 0xF;
3248 for (j = 0; j < 2; j++) {
3249 last = subblkpat & (1 << (1 - j));
3253 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
3258 idx = v->zz_4x8[i++] + off;
3260 idx = ff_vc1_adv_interlaced_4x8_zz[i++] + off;
3261 block[idx] = value * scale;
3263 block[idx] += (block[idx] < 0) ? -mquant : mquant;
3265 if (!(subblkpat & (1 << (1 - j))) && !skip_block) {
3267 v->vc1dsp.vc1_inv_trans_4x8_dc(dst + j * 4, linesize, block + off);
3269 v->vc1dsp.vc1_inv_trans_4x8(dst + j*4, linesize, block + off);
3275 *ttmb_out |= ttblk << (n * 4);
3279 /** @} */ // Macroblock group
3281 static const int size_table [6] = { 0, 2, 3, 4, 5, 8 };
3282 static const int offset_table[6] = { 0, 1, 3, 7, 15, 31 };
3284 static av_always_inline void vc1_apply_p_v_loop_filter(VC1Context *v, int block_num)
3286 MpegEncContext *s = &v->s;
3287 int mb_cbp = v->cbp[s->mb_x - s->mb_stride],
3288 block_cbp = mb_cbp >> (block_num * 4), bottom_cbp,
3289 mb_is_intra = v->is_intra[s->mb_x - s->mb_stride],
3290 block_is_intra = mb_is_intra >> (block_num * 4), bottom_is_intra;
3291 int idx, linesize = block_num > 3 ? s->uvlinesize : s->linesize, ttblk;
3294 if (block_num > 3) {
3295 dst = s->dest[block_num - 3];
3297 dst = s->dest[0] + (block_num & 1) * 8 + ((block_num & 2) * 4 - 8) * linesize;
3299 if (s->mb_y != s->end_mb_y || block_num < 2) {
3303 if (block_num > 3) {
3304 bottom_cbp = v->cbp[s->mb_x] >> (block_num * 4);
3305 bottom_is_intra = v->is_intra[s->mb_x] >> (block_num * 4);
3306 mv = &v->luma_mv[s->mb_x - s->mb_stride];
3307 mv_stride = s->mb_stride;
3309 bottom_cbp = (block_num < 2) ? (mb_cbp >> ((block_num + 2) * 4))
3310 : (v->cbp[s->mb_x] >> ((block_num - 2) * 4));
3311 bottom_is_intra = (block_num < 2) ? (mb_is_intra >> ((block_num + 2) * 4))
3312 : (v->is_intra[s->mb_x] >> ((block_num - 2) * 4));
3313 mv_stride = s->b8_stride;
3314 mv = &s->current_picture.f.motion_val[0][s->block_index[block_num] - 2 * mv_stride];
3317 if (bottom_is_intra & 1 || block_is_intra & 1 ||
3318 mv[0][0] != mv[mv_stride][0] || mv[0][1] != mv[mv_stride][1]) {
3319 v->vc1dsp.vc1_v_loop_filter8(dst, linesize, v->pq);
3321 idx = ((bottom_cbp >> 2) | block_cbp) & 3;
3323 v->vc1dsp.vc1_v_loop_filter8(dst, linesize, v->pq);
3326 v->vc1dsp.vc1_v_loop_filter4(dst + 4, linesize, v->pq);