har0ke/FFmpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

vc1: Split the decoder in components

Browse Source 
Speed up the overall compilation time.
This commit is contained in:
Luca Barbato 2014-07-16 20:18:20 +02:00
parent 7ae9791b64
commit 04d14c9b68
9 changed files with 5392 additions and 5210 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
libavcodec/Makefile
View File

@ -390,7 +390,9 @@ OBJS-$(CONFIG_V410_ENCODER) += v410enc.o
OBJS-$(CONFIG_V210X_DECODER) += v210x.o OBJS-$(CONFIG_V210X_DECODER) += v210x.o
OBJS-$(CONFIG_VB_DECODER) += vb.o OBJS-$(CONFIG_VB_DECODER) += vb.o
OBJS-$(CONFIG_VBLE_DECODER) += vble.o OBJS-$(CONFIG_VBLE_DECODER) += vble.o
OBJS-$(CONFIG_VC1_DECODER) += vc1dec.o vc1.o vc1data.o vc1dsp.o \ OBJS-$(CONFIG_VC1_DECODER) += vc1dec.o vc1_block.o vc1_loopfilter.o \
vc1_mc.o vc1_pred.o vc1.o vc1data.o \
vc1dsp.o \
msmpeg4dec.o msmpeg4.o msmpeg4data.o msmpeg4dec.o msmpeg4.o msmpeg4data.o
OBJS-$(CONFIG_VCR1_DECODER) += vcr1.o OBJS-$(CONFIG_VCR1_DECODER) += vcr1.o
OBJS-$(CONFIG_VMDAUDIO_DECODER) += vmdaudio.o OBJS-$(CONFIG_VMDAUDIO_DECODER) += vmdaudio.o

12
libavcodec/vc1.h
View File

@ -400,4 +400,16 @@ void ff_vc1_init_transposed_scantables(VC1Context *v);
int ff_vc1_decode_end(AVCodecContext *avctx); int ff_vc1_decode_end(AVCodecContext *avctx);
void ff_vc1_decode_blocks(VC1Context *v); void ff_vc1_decode_blocks(VC1Context *v);
void ff_vc1_loop_filter_iblk(VC1Context *v, int pq);
void ff_vc1_loop_filter_iblk_delayed(VC1Context *v, int pq);
void ff_vc1_smooth_overlap_filter_iblk(VC1Context *v);
void ff_vc1_apply_p_loop_filter(VC1Context *v);
void ff_vc1_mc_1mv(VC1Context *v, int dir);
void ff_vc1_mc_4mv_luma(VC1Context *v, int n, int dir, int avg);
void ff_vc1_mc_4mv_chroma(VC1Context *v, int dir);
void ff_vc1_mc_4mv_chroma4(VC1Context *v, int dir, int dir2, int avg);
void ff_vc1_interp_mc(VC1Context *v);
#endif /* AVCODEC_VC1_H */ #endif /* AVCODEC_VC1_H */

3054
libavcodec/vc1_block.c Normal file
View File

File diff suppressed because it is too large Load Diff

353
libavcodec/vc1_loopfilter.c Normal file
View File

@ -0,0 +1,353 @@
/*
* VC-1 and WMV3 decoder
* Copyright (c) 2011 Mashiat Sarker Shakkhar
* Copyright (c) 2006-2007 Konstantin Shishkov
* Partly based on vc9.c (c) 2005 Anonymous, Alex Beregszaszi, Michael Niedermayer
*
* This file is part of Libav.
*
* Libav is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* Libav is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file
* VC-1 and WMV3 loopfilter
*/
#include "avcodec.h"
#include "mpegvideo.h"
#include "vc1.h"
#include "vc1dsp.h"
void ff_vc1_loop_filter_iblk(VC1Context *v, int pq)
{
MpegEncContext *s = &v->s;
int j;
if (!s->first_slice_line) {
v->vc1dsp.vc1_v_loop_filter16(s->dest[0], s->linesize, pq);
if (s->mb_x)
v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize + 8, s->linesize, pq);
for (j = 0; j < 2; j++) {
v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1], s->uvlinesize, pq);
if (s->mb_x)
v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
}
}
v->vc1dsp.vc1_v_loop_filter16(s->dest[0] + 8 * s->linesize, s->linesize, pq);
if (s->mb_y == s->end_mb_y - 1) {
if (s->mb_x) {
v->vc1dsp.vc1_h_loop_filter16(s->dest[0], s->linesize, pq);
v->vc1dsp.vc1_h_loop_filter8(s->dest[1], s->uvlinesize, pq);
v->vc1dsp.vc1_h_loop_filter8(s->dest[2], s->uvlinesize, pq);
}
v->vc1dsp.vc1_h_loop_filter16(s->dest[0] + 8, s->linesize, pq);
}
}
void ff_vc1_loop_filter_iblk_delayed(VC1Context *v, int pq)
{
MpegEncContext *s = &v->s;
int j;
/* The loopfilter runs 1 row and 1 column behind the overlap filter, which
* means it runs two rows/cols behind the decoding loop. */
if (!s->first_slice_line) {
if (s->mb_x) {
if (s->mb_y >= s->start_mb_y + 2) {
v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 16 * s->linesize - 16, s->linesize, pq);
if (s->mb_x >= 2)
v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 16, s->linesize, pq);
v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 8, s->linesize, pq);
for (j = 0; j < 2; j++) {
v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq);
if (s->mb_x >= 2) {
v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 16 * s->uvlinesize - 8, s->uvlinesize, pq);
}
}
}
v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 8 * s->linesize - 16, s->linesize, pq);
}
if (s->mb_x == s->mb_width - 1) {
if (s->mb_y >= s->start_mb_y + 2) {
v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
if (s->mb_x)
v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize, s->linesize, pq);
v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize + 8, s->linesize, pq);
for (j = 0; j < 2; j++) {
v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
if (s->mb_x >= 2) {
v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 16 * s->uvlinesize, s->uvlinesize, pq);
}
}
}
v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 8 * s->linesize, s->linesize, pq);
}
if (s->mb_y == s->end_mb_y) {
if (s->mb_x) {
if (s->mb_x >= 2)
v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 16, s->linesize, pq);
v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 8, s->linesize, pq);
if (s->mb_x >= 2) {
for (j = 0; j < 2; j++) {
v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq);
}
}
}
if (s->mb_x == s->mb_width - 1) {
if (s->mb_x)
v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize + 8, s->linesize, pq);
if (s->mb_x) {
for (j = 0; j < 2; j++) {
v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
}
}
}
}
}
}
void ff_vc1_smooth_overlap_filter_iblk(VC1Context *v)
{
MpegEncContext *s = &v->s;
int mb_pos;
if (v->condover == CONDOVER_NONE)
return;
mb_pos = s->mb_x + s->mb_y * s->mb_stride;
/* Within a MB, the horizontal overlap always runs before the vertical.
* To accomplish that, we run the H on left and internal borders of the
* currently decoded MB. Then, we wait for the next overlap iteration
* to do H overlap on the right edge of this MB, before moving over and
* running the V overlap. Therefore, the V overlap makes us trail by one
* MB col and the H overlap filter makes us trail by one MB row. This
* is reflected in the time at which we run the put_pixels loop. */
if (v->condover == CONDOVER_ALL || v->pq >= 9 || v->over_flags_plane[mb_pos]) {
if (s->mb_x && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
v->over_flags_plane[mb_pos - 1])) {
v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][1],
v->block[v->cur_blk_idx][0]);
v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][3],
v->block[v->cur_blk_idx][2]);
if (!(s->flags & CODEC_FLAG_GRAY)) {
v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][4],
v->block[v->cur_blk_idx][4]);
v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][5],
v->block[v->cur_blk_idx][5]);
}
}
v->vc1dsp.vc1_h_s_overlap(v->block[v->cur_blk_idx][0],
v->block[v->cur_blk_idx][1]);
v->vc1dsp.vc1_h_s_overlap(v->block[v->cur_blk_idx][2],
v->block[v->cur_blk_idx][3]);
if (s->mb_x == s->mb_width - 1) {
if (!s->first_slice_line && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
v->over_flags_plane[mb_pos - s->mb_stride])) {
v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][2],
v->block[v->cur_blk_idx][0]);
v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][3],
v->block[v->cur_blk_idx][1]);
if (!(s->flags & CODEC_FLAG_GRAY)) {
v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][4],
v->block[v->cur_blk_idx][4]);
v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][5],
v->block[v->cur_blk_idx][5]);
}
}
v->vc1dsp.vc1_v_s_overlap(v->block[v->cur_blk_idx][0],
v->block[v->cur_blk_idx][2]);
v->vc1dsp.vc1_v_s_overlap(v->block[v->cur_blk_idx][1],
v->block[v->cur_blk_idx][3]);
}
}
if (s->mb_x && (v->condover == CONDOVER_ALL || v->over_flags_plane[mb_pos - 1])) {
if (!s->first_slice_line && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
v->over_flags_plane[mb_pos - s->mb_stride - 1])) {
v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][2],
v->block[v->left_blk_idx][0]);
v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][3],
v->block[v->left_blk_idx][1]);
if (!(s->flags & CODEC_FLAG_GRAY)) {
v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][4],
v->block[v->left_blk_idx][4]);
v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][5],
v->block[v->left_blk_idx][5]);
}
}
v->vc1dsp.vc1_v_s_overlap(v->block[v->left_blk_idx][0],
v->block[v->left_blk_idx][2]);
v->vc1dsp.vc1_v_s_overlap(v->block[v->left_blk_idx][1],
v->block[v->left_blk_idx][3]);
}
}
static av_always_inline void vc1_apply_p_v_loop_filter(VC1Context *v, int block_num)
{
MpegEncContext *s = &v->s;
int mb_cbp = v->cbp[s->mb_x - s->mb_stride],
block_cbp = mb_cbp >> (block_num * 4), bottom_cbp,
mb_is_intra = v->is_intra[s->mb_x - s->mb_stride],
block_is_intra = mb_is_intra >> (block_num * 4), bottom_is_intra;
int idx, linesize = block_num > 3 ? s->uvlinesize : s->linesize, ttblk;
uint8_t *dst;
if (block_num > 3) {
dst = s->dest[block_num - 3];
} else {
dst = s->dest[0] + (block_num & 1) * 8 + ((block_num & 2) * 4 - 8) * linesize;
}
if (s->mb_y != s->end_mb_y || block_num < 2) {
int16_t (*mv)[2];
int mv_stride;
if (block_num > 3) {
bottom_cbp = v->cbp[s->mb_x] >> (block_num * 4);
bottom_is_intra = v->is_intra[s->mb_x] >> (block_num * 4);
mv = &v->luma_mv[s->mb_x - s->mb_stride];
mv_stride = s->mb_stride;
} else {
bottom_cbp = (block_num < 2) ? (mb_cbp >> ((block_num + 2) * 4))
: (v->cbp[s->mb_x] >> ((block_num - 2) * 4));
bottom_is_intra = (block_num < 2) ? (mb_is_intra >> ((block_num + 2) * 4))
: (v->is_intra[s->mb_x] >> ((block_num - 2) * 4));
mv_stride = s->b8_stride;
mv = &s->current_picture.motion_val[0][s->block_index[block_num] - 2 * mv_stride];
}
if (bottom_is_intra & 1 || block_is_intra & 1 ||
mv[0][0] != mv[mv_stride][0] || mv[0][1] != mv[mv_stride][1]) {
v->vc1dsp.vc1_v_loop_filter8(dst, linesize, v->pq);
} else {
idx = ((bottom_cbp >> 2) | block_cbp) & 3;
if (idx == 3) {
v->vc1dsp.vc1_v_loop_filter8(dst, linesize, v->pq);
} else if (idx) {
if (idx == 1)
v->vc1dsp.vc1_v_loop_filter4(dst + 4, linesize, v->pq);
else
v->vc1dsp.vc1_v_loop_filter4(dst, linesize, v->pq);
}
}
}
dst -= 4 * linesize;
ttblk = (v->ttblk[s->mb_x - s->mb_stride] >> (block_num * 4)) & 0xF;
if (ttblk == TT_4X4 || ttblk == TT_8X4) {
idx = (block_cbp | (block_cbp >> 2)) & 3;
if (idx == 3) {
v->vc1dsp.vc1_v_loop_filter8(dst, linesize, v->pq);
} else if (idx) {
if (idx == 1)
v->vc1dsp.vc1_v_loop_filter4(dst + 4, linesize, v->pq);
else
v->vc1dsp.vc1_v_loop_filter4(dst, linesize, v->pq);
}
}
}
static av_always_inline void vc1_apply_p_h_loop_filter(VC1Context *v, int block_num)
{
MpegEncContext *s = &v->s;
int mb_cbp = v->cbp[s->mb_x - 1 - s->mb_stride],
block_cbp = mb_cbp >> (block_num * 4), right_cbp,
mb_is_intra = v->is_intra[s->mb_x - 1 - s->mb_stride],
block_is_intra = mb_is_intra >> (block_num * 4), right_is_intra;
int idx, linesize = block_num > 3 ? s->uvlinesize : s->linesize, ttblk;
uint8_t *dst;
if (block_num > 3) {
dst = s->dest[block_num - 3] - 8 * linesize;
} else {
dst = s->dest[0] + (block_num & 1) * 8 + ((block_num & 2) * 4 - 16) * linesize - 8;
}
if (s->mb_x != s->mb_width || !(block_num & 5)) {
int16_t (*mv)[2];
if (block_num > 3) {
right_cbp = v->cbp[s->mb_x - s->mb_stride] >> (block_num * 4);
right_is_intra = v->is_intra[s->mb_x - s->mb_stride] >> (block_num * 4);
mv = &v->luma_mv[s->mb_x - s->mb_stride - 1];
} else {
right_cbp = (block_num & 1) ? (v->cbp[s->mb_x - s->mb_stride] >> ((block_num - 1) * 4))
: (mb_cbp >> ((block_num + 1) * 4));
right_is_intra = (block_num & 1) ? (v->is_intra[s->mb_x - s->mb_stride] >> ((block_num - 1) * 4))
: (mb_is_intra >> ((block_num + 1) * 4));
mv = &s->current_picture.motion_val[0][s->block_index[block_num] - s->b8_stride * 2 - 2];
}
if (block_is_intra & 1 || right_is_intra & 1 || mv[0][0] != mv[1][0] || mv[0][1] != mv[1][1]) {
v->vc1dsp.vc1_h_loop_filter8(dst, linesize, v->pq);
} else {
idx = ((right_cbp >> 1) | block_cbp) & 5; // FIXME check
if (idx == 5) {
v->vc1dsp.vc1_h_loop_filter8(dst, linesize, v->pq);
} else if (idx) {
if (idx == 1)
v->vc1dsp.vc1_h_loop_filter4(dst + 4 * linesize, linesize, v->pq);
else
v->vc1dsp.vc1_h_loop_filter4(dst, linesize, v->pq);
}
}
}
dst -= 4;
ttblk = (v->ttblk[s->mb_x - s->mb_stride - 1] >> (block_num * 4)) & 0xf;
if (ttblk == TT_4X4 || ttblk == TT_4X8) {
idx = (block_cbp | (block_cbp >> 1)) & 5;
if (idx == 5) {
v->vc1dsp.vc1_h_loop_filter8(dst, linesize, v->pq);
} else if (idx) {
if (idx == 1)
v->vc1dsp.vc1_h_loop_filter4(dst + linesize * 4, linesize, v->pq);
else
v->vc1dsp.vc1_h_loop_filter4(dst, linesize, v->pq);
}
}
}
void ff_vc1_apply_p_loop_filter(VC1Context *v)
{
MpegEncContext *s = &v->s;
int i;
for (i = 0; i < 6; i++) {
vc1_apply_p_v_loop_filter(v, i);
}
/* V always precedes H, therefore we run H one MB before V;
* at the end of a row, we catch up to complete the row */
if (s->mb_x) {
for (i = 0; i < 6; i++) {
vc1_apply_p_h_loop_filter(v, i);
}
if (s->mb_x == s->mb_width - 1) {
s->mb_x++;
ff_update_block_index(s);
for (i = 0; i < 6; i++) {
vc1_apply_p_h_loop_filter(v, i);
}
}
}
}

948
libavcodec/vc1_mc.c Normal file
View File

@ -0,0 +1,948 @@
/*
* VC-1 and WMV3 decoder
* Copyright (c) 2011 Mashiat Sarker Shakkhar
* Copyright (c) 2006-2007 Konstantin Shishkov
* Partly based on vc9.c (c) 2005 Anonymous, Alex Beregszaszi, Michael Niedermayer
*
* This file is part of Libav.
*
* Libav is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* Libav is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file
* VC-1 and WMV3 block decoding routines
*/
#include "avcodec.h"
#include "h264chroma.h"
#include "mathops.h"
#include "mpegvideo.h"
#include "vc1.h"
/** Do motion compensation over 1 macroblock
* Mostly adapted hpel_motion and qpel_motion from mpegvideo.c
*/
void ff_vc1_mc_1mv(VC1Context *v, int dir)
{
MpegEncContext *s = &v->s;
H264ChromaContext *h264chroma = &v->h264chroma;
uint8_t *srcY, *srcU, *srcV;
int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
int v_edge_pos = s->v_edge_pos >> v->field_mode;
int i;
uint8_t (*luty)[256], (*lutuv)[256];
int use_ic;
if ((!v->field_mode ||
(v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
!v->s.last_picture.f->data[0])
return;
mx = s->mv[dir][0][0];
my = s->mv[dir][0][1];
// store motion vectors for further use in B frames
if (s->pict_type == AV_PICTURE_TYPE_P) {
for (i = 0; i < 4; i++) {
s->current_picture.motion_val[1][s->block_index[i] + v->blocks_off][0] = mx;
s->current_picture.motion_val[1][s->block_index[i] + v->blocks_off][1] = my;
}
}
uvmx = (mx + ((mx & 3) == 3)) >> 1;
uvmy = (my + ((my & 3) == 3)) >> 1;
v->luma_mv[s->mb_x][0] = uvmx;
v->luma_mv[s->mb_x][1] = uvmy;
if (v->field_mode &&
v->cur_field_type != v->ref_field_type[dir]) {
my = my - 2 + 4 * v->cur_field_type;
uvmy = uvmy - 2 + 4 * v->cur_field_type;
}
// fastuvmc shall be ignored for interlaced frame picture
if (v->fastuvmc && (v->fcm != ILACE_FRAME)) {
uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
}
if (!dir) {
if (v->field_mode && (v->cur_field_type != v->ref_field_type[dir]) && v->second_field) {
srcY = s->current_picture.f->data[0];
srcU = s->current_picture.f->data[1];
srcV = s->current_picture.f->data[2];
luty = v->curr_luty;
lutuv = v->curr_lutuv;
use_ic = v->curr_use_ic;
} else {
srcY = s->last_picture.f->data[0];
srcU = s->last_picture.f->data[1];
srcV = s->last_picture.f->data[2];
luty = v->last_luty;
lutuv = v->last_lutuv;
use_ic = v->last_use_ic;
}
} else {
srcY = s->next_picture.f->data[0];
srcU = s->next_picture.f->data[1];
srcV = s->next_picture.f->data[2];
luty = v->next_luty;
lutuv = v->next_lutuv;
use_ic = v->next_use_ic;
}
if (!srcY || !srcU) {
av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
return;
}
src_x = s->mb_x * 16 + (mx >> 2);
src_y = s->mb_y * 16 + (my >> 2);
uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
if (v->profile != PROFILE_ADVANCED) {
src_x = av_clip( src_x, -16, s->mb_width * 16);
src_y = av_clip( src_y, -16, s->mb_height * 16);
uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
} else {
src_x = av_clip( src_x, -17, s->avctx->coded_width);
src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
}
srcY += src_y * s->linesize + src_x;
srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
if (v->field_mode && v->ref_field_type[dir]) {
srcY += s->current_picture_ptr->f->linesize[0];
srcU += s->current_picture_ptr->f->linesize[1];
srcV += s->current_picture_ptr->f->linesize[2];
}
/* for grayscale we should not try to read from unknown area */
if (s->flags & CODEC_FLAG_GRAY) {
srcU = s->edge_emu_buffer + 18 * s->linesize;
srcV = s->edge_emu_buffer + 18 * s->linesize;
}
if (v->rangeredfrm || use_ic
|| s->h_edge_pos < 22 || v_edge_pos < 22
|| (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 16 - s->mspel * 3
|| (unsigned)(src_y - 1) > v_edge_pos - (my&3) - 16 - 3) {
uint8_t *uvbuf = s->edge_emu_buffer + 19 * s->linesize;
srcY -= s->mspel * (1 + s->linesize);
s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY,
s->linesize, s->linesize,
17 + s->mspel * 2, 17 + s->mspel * 2,
src_x - s->mspel, src_y - s->mspel,
s->h_edge_pos, v_edge_pos);
srcY = s->edge_emu_buffer;
s->vdsp.emulated_edge_mc(uvbuf, srcU,
s->uvlinesize, s->uvlinesize,
8 + 1, 8 + 1,
uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
s->vdsp.emulated_edge_mc(uvbuf + 16, srcV,
s->uvlinesize, s->uvlinesize,
8 + 1, 8 + 1,
uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
srcU = uvbuf;
srcV = uvbuf + 16;
/* if we deal with range reduction we need to scale source blocks */
if (v->rangeredfrm) {
int i, j;
uint8_t *src, *src2;
src = srcY;
for (j = 0; j < 17 + s->mspel * 2; j++) {
for (i = 0; i < 17 + s->mspel * 2; i++)
src[i] = ((src[i] - 128) >> 1) + 128;
src += s->linesize;
}
src = srcU;
src2 = srcV;
for (j = 0; j < 9; j++) {
for (i = 0; i < 9; i++) {
src[i] = ((src[i] - 128) >> 1) + 128;
src2[i] = ((src2[i] - 128) >> 1) + 128;
}
src += s->uvlinesize;
src2 += s->uvlinesize;
}
}
/* if we deal with intensity compensation we need to scale source blocks */
if (use_ic) {
int i, j;
uint8_t *src, *src2;
src = srcY;
for (j = 0; j < 17 + s->mspel * 2; j++) {
int f = v->field_mode ? v->ref_field_type[dir] : ((j + src_y - s->mspel) & 1) ;
for (i = 0; i < 17 + s->mspel * 2; i++)
src[i] = luty[f][src[i]];
src += s->linesize;
}
src = srcU;
src2 = srcV;
for (j = 0; j < 9; j++) {
int f = v->field_mode ? v->ref_field_type[dir] : ((j + uvsrc_y) & 1);
for (i = 0; i < 9; i++) {
src[i] = lutuv[f][src[i]];
src2[i] = lutuv[f][src2[i]];
}
src += s->uvlinesize;
src2 += s->uvlinesize;
}
}
srcY += s->mspel * (1 + s->linesize);
}
if (s->mspel) {
dxy = ((my & 3) << 2) | (mx & 3);
v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] , srcY , s->linesize, v->rnd);
v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8, srcY + 8, s->linesize, v->rnd);
srcY += s->linesize * 8;
v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize , srcY , s->linesize, v->rnd);
v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
} else { // hpel mc - always used for luma
dxy = (my & 2) | ((mx & 2) >> 1);
if (!v->rnd)
s->hdsp.put_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
else
s->hdsp.put_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
}
if (s->flags & CODEC_FLAG_GRAY) return;
/* Chroma MC always uses qpel bilinear */
uvmx = (uvmx & 3) << 1;
uvmy = (uvmy & 3) << 1;
if (!v->rnd) {
h264chroma->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
h264chroma->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
} else {
v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
}
}
static inline int median4(int a, int b, int c, int d)
{
if (a < b) {
if (c < d) return (FFMIN(b, d) + FFMAX(a, c)) / 2;
else return (FFMIN(b, c) + FFMAX(a, d)) / 2;
} else {
if (c < d) return (FFMIN(a, d) + FFMAX(b, c)) / 2;
else return (FFMIN(a, c) + FFMAX(b, d)) / 2;
}
}
/** Do motion compensation for 4-MV macroblock - luminance block
*/
void ff_vc1_mc_4mv_luma(VC1Context *v, int n, int dir, int avg)
{
MpegEncContext *s = &v->s;
uint8_t *srcY;
int dxy, mx, my, src_x, src_y;
int off;
int fieldmv = (v->fcm == ILACE_FRAME) ? v->blk_mv_type[s->block_index[n]] : 0;
int v_edge_pos = s->v_edge_pos >> v->field_mode;
uint8_t (*luty)[256];
int use_ic;
if ((!v->field_mode ||
(v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
!v->s.last_picture.f->data[0])
return;
mx = s->mv[dir][n][0];
my = s->mv[dir][n][1];
if (!dir) {
if (v->field_mode && (v->cur_field_type != v->ref_field_type[dir]) && v->second_field) {
srcY = s->current_picture.f->data[0];
luty = v->curr_luty;
use_ic = v->curr_use_ic;
} else {
srcY = s->last_picture.f->data[0];
luty = v->last_luty;
use_ic = v->last_use_ic;
}
} else {
srcY = s->next_picture.f->data[0];
luty = v->next_luty;
use_ic = v->next_use_ic;
}
if (!srcY) {
av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
return;
}
if (v->field_mode) {
if (v->cur_field_type != v->ref_field_type[dir])
my = my - 2 + 4 * v->cur_field_type;
}
if (s->pict_type == AV_PICTURE_TYPE_P && n == 3 && v->field_mode) {
int same_count = 0, opp_count = 0, k;
int chosen_mv[2][4][2], f;
int tx, ty;
for (k = 0; k < 4; k++) {
f = v->mv_f[0][s->block_index[k] + v->blocks_off];
chosen_mv[f][f ? opp_count : same_count][0] = s->mv[0][k][0];
chosen_mv[f][f ? opp_count : same_count][1] = s->mv[0][k][1];
opp_count += f;
same_count += 1 - f;
}
f = opp_count > same_count;
switch (f ? opp_count : same_count) {
case 4:
tx = median4(chosen_mv[f][0][0], chosen_mv[f][1][0],
chosen_mv[f][2][0], chosen_mv[f][3][0]);
ty = median4(chosen_mv[f][0][1], chosen_mv[f][1][1],
chosen_mv[f][2][1], chosen_mv[f][3][1]);
break;
case 3:
tx = mid_pred(chosen_mv[f][0][0], chosen_mv[f][1][0], chosen_mv[f][2][0]);
ty = mid_pred(chosen_mv[f][0][1], chosen_mv[f][1][1], chosen_mv[f][2][1]);
break;
case 2:
tx = (chosen_mv[f][0][0] + chosen_mv[f][1][0]) / 2;
ty = (chosen_mv[f][0][1] + chosen_mv[f][1][1]) / 2;
break;
}
s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx;
s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty;
for (k = 0; k < 4; k++)
v->mv_f[1][s->block_index[k] + v->blocks_off] = f;
}
if (v->fcm == ILACE_FRAME) { // not sure if needed for other types of picture
int qx, qy;
int width = s->avctx->coded_width;
int height = s->avctx->coded_height >> 1;
if (s->pict_type == AV_PICTURE_TYPE_P) {
s->current_picture.motion_val[1][s->block_index[n] + v->blocks_off][0] = mx;
s->current_picture.motion_val[1][s->block_index[n] + v->blocks_off][1] = my;
}
qx = (s->mb_x * 16) + (mx >> 2);
qy = (s->mb_y * 8) + (my >> 3);
if (qx < -17)
mx -= 4 * (qx + 17);
else if (qx > width)
mx -= 4 * (qx - width);
if (qy < -18)
my -= 8 * (qy + 18);
else if (qy > height + 1)
my -= 8 * (qy - height - 1);
}
if ((v->fcm == ILACE_FRAME) && fieldmv)
off = ((n > 1) ? s->linesize : 0) + (n & 1) * 8;
else
off = s->linesize * 4 * (n & 2) + (n & 1) * 8;
src_x = s->mb_x * 16 + (n & 1) * 8 + (mx >> 2);
if (!fieldmv)
src_y = s->mb_y * 16 + (n & 2) * 4 + (my >> 2);
else
src_y = s->mb_y * 16 + ((n > 1) ? 1 : 0) + (my >> 2);
if (v->profile != PROFILE_ADVANCED) {
src_x = av_clip(src_x, -16, s->mb_width * 16);
src_y = av_clip(src_y, -16, s->mb_height * 16);
} else {
src_x = av_clip(src_x, -17, s->avctx->coded_width);
if (v->fcm == ILACE_FRAME) {
if (src_y & 1)
src_y = av_clip(src_y, -17, s->avctx->coded_height + 1);
else
src_y = av_clip(src_y, -18, s->avctx->coded_height);
} else {
src_y = av_clip(src_y, -18, s->avctx->coded_height + 1);
}
}
srcY += src_y * s->linesize + src_x;
if (v->field_mode && v->ref_field_type[dir])
srcY += s->current_picture_ptr->f->linesize[0];
if (fieldmv && !(src_y & 1))
v_edge_pos--;
if (fieldmv && (src_y & 1) && src_y < 4)
src_y--;
if (v->rangeredfrm || use_ic
|| s->h_edge_pos < 13 || v_edge_pos < 23
|| (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx & 3) - 8 - s->mspel * 2
|| (unsigned)(src_y - (s->mspel << fieldmv)) > v_edge_pos - (my & 3) - ((8 + s->mspel * 2) << fieldmv)) {
srcY -= s->mspel * (1 + (s->linesize << fieldmv));
/* check emulate edge stride and offset */
s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY,
s->linesize, s->linesize,
9 + s->mspel * 2, (9 + s->mspel * 2) << fieldmv,
src_x - s->mspel, src_y - (s->mspel << fieldmv),
s->h_edge_pos, v_edge_pos);
srcY = s->edge_emu_buffer;
/* if we deal with range reduction we need to scale source blocks */
if (v->rangeredfrm) {
int i, j;
uint8_t *src;
src = srcY;
for (j = 0; j < 9 + s->mspel * 2; j++) {
for (i = 0; i < 9 + s->mspel * 2; i++)
src[i] = ((src[i] - 128) >> 1) + 128;
src += s->linesize << fieldmv;
}
}
/* if we deal with intensity compensation we need to scale source blocks */
if (use_ic) {
int i, j;
uint8_t *src;
src = srcY;
for (j = 0; j < 9 + s->mspel * 2; j++) {
int f = v->field_mode ? v->ref_field_type[dir] : (((j<<fieldmv)+src_y - (s->mspel << fieldmv)) & 1);
for (i = 0; i < 9 + s->mspel * 2; i++)
src[i] = luty[f][src[i]];
src += s->linesize << fieldmv;
}
}
srcY += s->mspel * (1 + (s->linesize << fieldmv));
}
if (s->mspel) {
dxy = ((my & 3) << 2) | (mx & 3);
if (avg)
v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize << fieldmv, v->rnd);
else
v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize << fieldmv, v->rnd);
} else { // hpel mc - always used for luma
dxy = (my & 2) | ((mx & 2) >> 1);
if (!v->rnd)
s->hdsp.put_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
else
s->hdsp.put_no_rnd_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
}
}
static av_always_inline int get_chroma_mv(int *mvx, int *mvy, int *a, int flag, int *tx, int *ty)
{
int idx, i;
static const int count[16] = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4};
idx = ((a[3] != flag) << 3)
| ((a[2] != flag) << 2)
| ((a[1] != flag) << 1)
| (a[0] != flag);
if (!idx) {
*tx = median4(mvx[0], mvx[1], mvx[2], mvx[3]);
*ty = median4(mvy[0], mvy[1], mvy[2], mvy[3]);
return 4;
} else if (count[idx] == 1) {
switch (idx) {
case 0x1:
*tx = mid_pred(mvx[1], mvx[2], mvx[3]);
*ty = mid_pred(mvy[1], mvy[2], mvy[3]);
return 3;
case 0x2:
*tx = mid_pred(mvx[0], mvx[2], mvx[3]);
*ty = mid_pred(mvy[0], mvy[2], mvy[3]);
return 3;
case 0x4:
*tx = mid_pred(mvx[0], mvx[1], mvx[3]);
*ty = mid_pred(mvy[0], mvy[1], mvy[3]);
return 3;
case 0x8:
*tx = mid_pred(mvx[0], mvx[1], mvx[2]);
*ty = mid_pred(mvy[0], mvy[1], mvy[2]);
return 3;
}
} else if (count[idx] == 2) {
int t1 = 0, t2 = 0;
for (i = 0; i < 3; i++)
if (!a[i]) {
t1 = i;
break;
}
for (i = t1 + 1; i < 4; i++)
if (!a[i]) {
t2 = i;
break;
}
*tx = (mvx[t1] + mvx[t2]) / 2;
*ty = (mvy[t1] + mvy[t2]) / 2;
return 2;
} else {
return 0;
}
return -1;
}
/** Do motion compensation for 4-MV macroblock - both chroma blocks
*/
void ff_vc1_mc_4mv_chroma(VC1Context *v, int dir)
{
MpegEncContext *s = &v->s;
H264ChromaContext *h264chroma = &v->h264chroma;
uint8_t *srcU, *srcV;
int uvmx, uvmy, uvsrc_x, uvsrc_y;
int k, tx = 0, ty = 0;
int mvx[4], mvy[4], intra[4], mv_f[4];
int valid_count;
int chroma_ref_type = v->cur_field_type;
int v_edge_pos = s->v_edge_pos >> v->field_mode;
uint8_t (*lutuv)[256];
int use_ic;
if (!v->field_mode && !v->s.last_picture.f->data[0])
return;
if (s->flags & CODEC_FLAG_GRAY)
return;
for (k = 0; k < 4; k++) {
mvx[k] = s->mv[dir][k][0];
mvy[k] = s->mv[dir][k][1];
intra[k] = v->mb_type[0][s->block_index[k]];
if (v->field_mode)
mv_f[k] = v->mv_f[dir][s->block_index[k] + v->blocks_off];
}
/* calculate chroma MV vector from four luma MVs */
if (!v->field_mode || (v->field_mode && !v->numref)) {
valid_count = get_chroma_mv(mvx, mvy, intra, 0, &tx, &ty);
chroma_ref_type = v->reffield;
if (!valid_count) {
s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = 0;
s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = 0;
v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
return; //no need to do MC for intra blocks
}
} else {
int dominant = 0;
if (mv_f[0] + mv_f[1] + mv_f[2] + mv_f[3] > 2)
dominant = 1;
valid_count = get_chroma_mv(mvx, mvy, mv_f, dominant, &tx, &ty);
if (dominant)
chroma_ref_type = !v->cur_field_type;
}
if (v->field_mode && chroma_ref_type == 1 && v->cur_field_type == 1 && !v->s.last_picture.f->data[0])
return;
s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx;
s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty;
uvmx = (tx + ((tx & 3) == 3)) >> 1;
uvmy = (ty + ((ty & 3) == 3)) >> 1;
v->luma_mv[s->mb_x][0] = uvmx;
v->luma_mv[s->mb_x][1] = uvmy;
if (v->fastuvmc) {
uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
}
// Field conversion bias
if (v->cur_field_type != chroma_ref_type)
uvmy += 2 - 4 * chroma_ref_type;
uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
if (v->profile != PROFILE_ADVANCED) {
uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
} else {
uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
}
if (!dir) {
if (v->field_mode && (v->cur_field_type != chroma_ref_type) && v->second_field) {
srcU = s->current_picture.f->data[1];
srcV = s->current_picture.f->data[2];
lutuv = v->curr_lutuv;
use_ic = v->curr_use_ic;
} else {
srcU = s->last_picture.f->data[1];
srcV = s->last_picture.f->data[2];
lutuv = v->last_lutuv;
use_ic = v->last_use_ic;
}
} else {
srcU = s->next_picture.f->data[1];
srcV = s->next_picture.f->data[2];
lutuv = v->next_lutuv;
use_ic = v->next_use_ic;
}
if (!srcU) {
av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
return;
}
srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
if (v->field_mode) {
if (chroma_ref_type) {
srcU += s->current_picture_ptr->f->linesize[1];
srcV += s->current_picture_ptr->f->linesize[2];
}
}
if (v->rangeredfrm || use_ic
|| s->h_edge_pos < 18 || v_edge_pos < 18
|| (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 9
|| (unsigned)uvsrc_y > (v_edge_pos >> 1) - 9) {
s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcU,
s->uvlinesize, s->uvlinesize,
8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
s->h_edge_pos >> 1, v_edge_pos >> 1);
s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV,
s->uvlinesize, s->uvlinesize,
8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
s->h_edge_pos >> 1, v_edge_pos >> 1);
srcU = s->edge_emu_buffer;
srcV = s->edge_emu_buffer + 16;
/* if we deal with range reduction we need to scale source blocks */
if (v->rangeredfrm) {
int i, j;
uint8_t *src, *src2;
src = srcU;
src2 = srcV;
for (j = 0; j < 9; j++) {
for (i = 0; i < 9; i++) {
src[i] = ((src[i] - 128) >> 1) + 128;
src2[i] = ((src2[i] - 128) >> 1) + 128;
}
src += s->uvlinesize;
src2 += s->uvlinesize;
}
}
/* if we deal with intensity compensation we need to scale source blocks */
if (use_ic) {
int i, j;
uint8_t *src, *src2;
src = srcU;
src2 = srcV;
for (j = 0; j < 9; j++) {
int f = v->field_mode ? chroma_ref_type : ((j + uvsrc_y) & 1);
for (i = 0; i < 9; i++) {
src[i] = lutuv[f][src[i]];
src2[i] = lutuv[f][src2[i]];
}
src += s->uvlinesize;
src2 += s->uvlinesize;
}
}
}
/* Chroma MC always uses qpel bilinear */
uvmx = (uvmx & 3) << 1;
uvmy = (uvmy & 3) << 1;
if (!v->rnd) {
h264chroma->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
h264chroma->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
} else {
v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
}
}
/** Do motion compensation for 4-MV interlaced frame chroma macroblock (both U and V)
*/
void ff_vc1_mc_4mv_chroma4(VC1Context *v, int dir, int dir2, int avg)
{
MpegEncContext *s = &v->s;
H264ChromaContext *h264chroma = &v->h264chroma;
uint8_t *srcU, *srcV;
int uvsrc_x, uvsrc_y;
int uvmx_field[4], uvmy_field[4];
int i, off, tx, ty;
int fieldmv = v->blk_mv_type[s->block_index[0]];
static const int s_rndtblfield[16] = { 0, 0, 1, 2, 4, 4, 5, 6, 2, 2, 3, 8, 6, 6, 7, 12 };
int v_dist = fieldmv ? 1 : 4; // vertical offset for lower sub-blocks
int v_edge_pos = s->v_edge_pos >> 1;
int use_ic;
uint8_t (*lutuv)[256];
if (s->flags & CODEC_FLAG_GRAY)
return;
for (i = 0; i < 4; i++) {
int d = i < 2 ? dir: dir2;
tx = s->mv[d][i][0];
uvmx_field[i] = (tx + ((tx & 3) == 3)) >> 1;
ty = s->mv[d][i][1];
if (fieldmv)
uvmy_field[i] = (ty >> 4) * 8 + s_rndtblfield[ty & 0xF];
else
uvmy_field[i] = (ty + ((ty & 3) == 3)) >> 1;
}
for (i = 0; i < 4; i++) {
off = (i & 1) * 4 + ((i & 2) ? v_dist * s->uvlinesize : 0);
uvsrc_x = s->mb_x * 8 + (i & 1) * 4 + (uvmx_field[i] >> 2);
uvsrc_y = s->mb_y * 8 + ((i & 2) ? v_dist : 0) + (uvmy_field[i] >> 2);
// FIXME: implement proper pull-back (see vc1cropmv.c, vc1CROPMV_ChromaPullBack())
uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
if (i < 2 ? dir : dir2) {
srcU = s->next_picture.f->data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
srcV = s->next_picture.f->data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
lutuv = v->next_lutuv;
use_ic = v->next_use_ic;
} else {
srcU = s->last_picture.f->data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
srcV = s->last_picture.f->data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
lutuv = v->last_lutuv;
use_ic = v->last_use_ic;
}
uvmx_field[i] = (uvmx_field[i] & 3) << 1;
uvmy_field[i] = (uvmy_field[i] & 3) << 1;
if (fieldmv && !(uvsrc_y & 1))
v_edge_pos--;
if (fieldmv && (uvsrc_y & 1) && uvsrc_y < 2)
uvsrc_y--;
if (use_ic
|| s->h_edge_pos < 10 || v_edge_pos < (5 << fieldmv)
|| (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 5
|| (unsigned)uvsrc_y > v_edge_pos - (5 << fieldmv)) {
s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcU,
s->uvlinesize, s->uvlinesize,
5, (5 << fieldmv), uvsrc_x, uvsrc_y,
s->h_edge_pos >> 1, v_edge_pos);
s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV,
s->uvlinesize, s->uvlinesize,
5, (5 << fieldmv), uvsrc_x, uvsrc_y,
s->h_edge_pos >> 1, v_edge_pos);
srcU = s->edge_emu_buffer;
srcV = s->edge_emu_buffer + 16;
/* if we deal with intensity compensation we need to scale source blocks */
if (use_ic) {
int i, j;
uint8_t *src, *src2;
src = srcU;
src2 = srcV;
for (j = 0; j < 5; j++) {
int f = (uvsrc_y + (j << fieldmv)) & 1;
for (i = 0; i < 5; i++) {
src[i] = lutuv[f][src[i]];
src2[i] = lutuv[f][src2[i]];
}
src += s->uvlinesize << fieldmv;
src2 += s->uvlinesize << fieldmv;
}
}
}
if (avg) {
if (!v->rnd) {
h264chroma->avg_h264_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
h264chroma->avg_h264_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
} else {
v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
}
} else {
if (!v->rnd) {
h264chroma->put_h264_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
h264chroma->put_h264_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
} else {
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]);
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]);
}
}
}
}
/** Motion compensation for direct or interpolated blocks in B-frames
*/
void ff_vc1_interp_mc(VC1Context *v)
{
MpegEncContext *s = &v->s;
H264ChromaContext *h264chroma = &v->h264chroma;
uint8_t *srcY, *srcU, *srcV;
int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
int off, off_uv;
int v_edge_pos = s->v_edge_pos >> v->field_mode;
int use_ic = v->next_use_ic;
if (!v->field_mode && !v->s.next_picture.f->data[0])
return;
mx = s->mv[1][0][0];
my = s->mv[1][0][1];
uvmx = (mx + ((mx & 3) == 3)) >> 1;
uvmy = (my + ((my & 3) == 3)) >> 1;
if (v->field_mode) {
if (v->cur_field_type != v->ref_field_type[1])
my = my - 2 + 4 * v->cur_field_type;
uvmy = uvmy - 2 + 4 * v->cur_field_type;
}
if (v->fastuvmc) {
uvmx = uvmx + ((uvmx < 0) ? -(uvmx & 1) : (uvmx & 1));
uvmy = uvmy + ((uvmy < 0) ? -(uvmy & 1) : (uvmy & 1));
}
srcY = s->next_picture.f->data[0];
srcU = s->next_picture.f->data[1];
srcV = s->next_picture.f->data[2];
src_x = s->mb_x * 16 + (mx >> 2);
src_y = s->mb_y * 16 + (my >> 2);
uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
if (v->profile != PROFILE_ADVANCED) {
src_x = av_clip( src_x, -16, s->mb_width * 16);
src_y = av_clip( src_y, -16, s->mb_height * 16);
uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
} else {
src_x = av_clip( src_x, -17, s->avctx->coded_width);
src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
}
srcY += src_y * s->linesize + src_x;
srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
if (v->field_mode && v->ref_field_type[1]) {
srcY += s->current_picture_ptr->f->linesize[0];
srcU += s->current_picture_ptr->f->linesize[1];
srcV += s->current_picture_ptr->f->linesize[2];
}
/* for grayscale we should not try to read from unknown area */
if (s->flags & CODEC_FLAG_GRAY) {
srcU = s->edge_emu_buffer + 18 * s->linesize;
srcV = s->edge_emu_buffer + 18 * s->linesize;
}
if (v->rangeredfrm || s->h_edge_pos < 22 || v_edge_pos < 22 || use_ic
|| (unsigned)(src_x - 1) > s->h_edge_pos - (mx & 3) - 16 - 3
|| (unsigned)(src_y - 1) > v_edge_pos - (my & 3) - 16 - 3) {
uint8_t *uvbuf = s->edge_emu_buffer + 19 * s->linesize;
srcY -= s->mspel * (1 + s->linesize);
s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY,
s->linesize, s->linesize,
17 + s->mspel * 2, 17 + s->mspel * 2,
src_x - s->mspel, src_y - s->mspel,
s->h_edge_pos, v_edge_pos);
srcY = s->edge_emu_buffer;
s->vdsp.emulated_edge_mc(uvbuf, srcU,
s->uvlinesize, s->uvlinesize,
8 + 1, 8 + 1,
uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
s->vdsp.emulated_edge_mc(uvbuf + 16, srcV,
s->uvlinesize, s->uvlinesize,
8 + 1, 8 + 1,
uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
srcU = uvbuf;
srcV = uvbuf + 16;
/* if we deal with range reduction we need to scale source blocks */
if (v->rangeredfrm) {
int i, j;
uint8_t *src, *src2;
src = srcY;
for (j = 0; j < 17 + s->mspel * 2; j++) {
for (i = 0; i < 17 + s->mspel * 2; i++)
src[i] = ((src[i] - 128) >> 1) + 128;
src += s->linesize;
}
src = srcU;
src2 = srcV;
for (j = 0; j < 9; j++) {
for (i = 0; i < 9; i++) {
src[i] = ((src[i] - 128) >> 1) + 128;
src2[i] = ((src2[i] - 128) >> 1) + 128;
}
src += s->uvlinesize;
src2 += s->uvlinesize;
}
}
if (use_ic) {
uint8_t (*luty )[256] = v->next_luty;
uint8_t (*lutuv)[256] = v->next_lutuv;
int i, j;
uint8_t *src, *src2;
src = srcY;
for (j = 0; j < 17 + s->mspel * 2; j++) {
int f = v->field_mode ? v->ref_field_type[1] : ((j+src_y - s->mspel) & 1);
for (i = 0; i < 17 + s->mspel * 2; i++)
src[i] = luty[f][src[i]];
src += s->linesize;
}
src = srcU;
src2 = srcV;
for (j = 0; j < 9; j++) {
int f = v->field_mode ? v->ref_field_type[1] : ((j+uvsrc_y) & 1);
for (i = 0; i < 9; i++) {
src[i] = lutuv[f][src[i]];
src2[i] = lutuv[f][src2[i]];
}
src += s->uvlinesize;
src2 += s->uvlinesize;
}
}
srcY += s->mspel * (1 + s->linesize);
}
off = 0;
off_uv = 0;
if (s->mspel) {
dxy = ((my & 3) << 2) | (mx & 3);
v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off , srcY , s->linesize, v->rnd);
v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8, srcY + 8, s->linesize, v->rnd);
srcY += s->linesize * 8;
v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize , srcY , s->linesize, v->rnd);
v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
} else { // hpel mc
dxy = (my & 2) | ((mx & 2) >> 1);
if (!v->rnd)
s->hdsp.avg_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
else
s->hdsp.avg_no_rnd_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize, 16);
}
if (s->flags & CODEC_FLAG_GRAY) return;
/* Chroma MC always uses qpel blilinear */
uvmx = (uvmx & 3) << 1;
uvmy = (uvmy & 3) << 1;
if (!v->rnd) {
h264chroma->avg_h264_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
h264chroma->avg_h264_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
} else {
v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
}
}

959
libavcodec/vc1_pred.c Normal file
View File

@ -0,0 +1,959 @@
/*
* VC-1 and WMV3 decoder
* Copyright (c) 2011 Mashiat Sarker Shakkhar
* Copyright (c) 2006-2007 Konstantin Shishkov
* Partly based on vc9.c (c) 2005 Anonymous, Alex Beregszaszi, Michael Niedermayer
*
* This file is part of Libav.
*
* Libav is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* Libav is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file
* VC-1 and WMV3 block decoding routines
*/
#include "mathops.h"
#include "mpegutils.h"
#include "mpegvideo.h"
#include "vc1.h"
#include "vc1_pred.h"
#include "vc1data.h"
static av_always_inline int scaleforsame_x(VC1Context *v, int n /* MV */, int dir)
{
int scaledvalue, refdist;
int scalesame1, scalesame2;
int scalezone1_x, zone1offset_x;
int table_index = dir ^ v->second_field;
if (v->s.pict_type != AV_PICTURE_TYPE_B)
refdist = v->refdist;
else
refdist = dir ? v->brfd : v->frfd;
if (refdist > 3)
refdist = 3;
scalesame1 = ff_vc1_field_mvpred_scales[table_index][1][refdist];
scalesame2 = ff_vc1_field_mvpred_scales[table_index][2][refdist];
scalezone1_x = ff_vc1_field_mvpred_scales[table_index][3][refdist];
zone1offset_x = ff_vc1_field_mvpred_scales[table_index][5][refdist];
if (FFABS(n) > 255)
scaledvalue = n;
else {
if (FFABS(n) < scalezone1_x)
scaledvalue = (n * scalesame1) >> 8;
else {
if (n < 0)
scaledvalue = ((n * scalesame2) >> 8) - zone1offset_x;
else
scaledvalue = ((n * scalesame2) >> 8) + zone1offset_x;
}
}
return av_clip(scaledvalue, -v->range_x, v->range_x - 1);
}
static av_always_inline int scaleforsame_y(VC1Context *v, int i, int n /* MV */, int dir)
{
int scaledvalue, refdist;
int scalesame1, scalesame2;
int scalezone1_y, zone1offset_y;
int table_index = dir ^ v->second_field;
if (v->s.pict_type != AV_PICTURE_TYPE_B)
refdist = v->refdist;
else
refdist = dir ? v->brfd : v->frfd;
if (refdist > 3)
refdist = 3;
scalesame1 = ff_vc1_field_mvpred_scales[table_index][1][refdist];
scalesame2 = ff_vc1_field_mvpred_scales[table_index][2][refdist];
scalezone1_y = ff_vc1_field_mvpred_scales[table_index][4][refdist];
zone1offset_y = ff_vc1_field_mvpred_scales[table_index][6][refdist];
if (FFABS(n) > 63)
scaledvalue = n;
else {
if (FFABS(n) < scalezone1_y)
scaledvalue = (n * scalesame1) >> 8;
else {
if (n < 0)
scaledvalue = ((n * scalesame2) >> 8) - zone1offset_y;
else
scaledvalue = ((n * scalesame2) >> 8) + zone1offset_y;
}
}
if (v->cur_field_type && !v->ref_field_type[dir])
return av_clip(scaledvalue, -v->range_y / 2 + 1, v->range_y / 2);
else
return av_clip(scaledvalue, -v->range_y / 2, v->range_y / 2 - 1);
}
static av_always_inline int scaleforopp_x(VC1Context *v, int n /* MV */)
{
int scalezone1_x, zone1offset_x;
int scaleopp1, scaleopp2, brfd;
int scaledvalue;
brfd = FFMIN(v->brfd, 3);
scalezone1_x = ff_vc1_b_field_mvpred_scales[3][brfd];
zone1offset_x = ff_vc1_b_field_mvpred_scales[5][brfd];
scaleopp1 = ff_vc1_b_field_mvpred_scales[1][brfd];
scaleopp2 = ff_vc1_b_field_mvpred_scales[2][brfd];
if (FFABS(n) > 255)
scaledvalue = n;
else {
if (FFABS(n) < scalezone1_x)
scaledvalue = (n * scaleopp1) >> 8;
else {
if (n < 0)
scaledvalue = ((n * scaleopp2) >> 8) - zone1offset_x;
else
scaledvalue = ((n * scaleopp2) >> 8) + zone1offset_x;
}
}
return av_clip(scaledvalue, -v->range_x, v->range_x - 1);
}
static av_always_inline int scaleforopp_y(VC1Context *v, int n /* MV */, int dir)
{
int scalezone1_y, zone1offset_y;
int scaleopp1, scaleopp2, brfd;
int scaledvalue;
brfd = FFMIN(v->brfd, 3);
scalezone1_y = ff_vc1_b_field_mvpred_scales[4][brfd];
zone1offset_y = ff_vc1_b_field_mvpred_scales[6][brfd];
scaleopp1 = ff_vc1_b_field_mvpred_scales[1][brfd];
scaleopp2 = ff_vc1_b_field_mvpred_scales[2][brfd];
if (FFABS(n) > 63)
scaledvalue = n;
else {
if (FFABS(n) < scalezone1_y)
scaledvalue = (n * scaleopp1) >> 8;
else {
if (n < 0)
scaledvalue = ((n * scaleopp2) >> 8) - zone1offset_y;
else
scaledvalue = ((n * scaleopp2) >> 8) + zone1offset_y;
}
}
if (v->cur_field_type && !v->ref_field_type[dir]) {
return av_clip(scaledvalue, -v->range_y / 2 + 1, v->range_y / 2);
} else {
return av_clip(scaledvalue, -v->range_y / 2, v->range_y / 2 - 1);
}
}
static av_always_inline int scaleforsame(VC1Context *v, int i, int n /* MV */,
int dim, int dir)
{
int brfd, scalesame;
int hpel = 1 - v->s.quarter_sample;
n >>= hpel;
if (v->s.pict_type != AV_PICTURE_TYPE_B || v->second_field || !dir) {
if (dim)
n = scaleforsame_y(v, i, n, dir) << hpel;
else
n = scaleforsame_x(v, n, dir) << hpel;
return n;
}
brfd = FFMIN(v->brfd, 3);
scalesame = ff_vc1_b_field_mvpred_scales[0][brfd];
n = (n * scalesame >> 8) << hpel;
return n;
}
static av_always_inline int scaleforopp(VC1Context *v, int n /* MV */,
int dim, int dir)
{
int refdist, scaleopp;
int hpel = 1 - v->s.quarter_sample;
n >>= hpel;
if (v->s.pict_type == AV_PICTURE_TYPE_B && !v->second_field && dir == 1) {
if (dim)
n = scaleforopp_y(v, n, dir) << hpel;
else
n = scaleforopp_x(v, n) << hpel;
return n;
}
if (v->s.pict_type != AV_PICTURE_TYPE_B)
refdist = FFMIN(v->refdist, 3);
else
refdist = dir ? v->brfd : v->frfd;
scaleopp = ff_vc1_field_mvpred_scales[dir ^ v->second_field][0][refdist];
n = (n * scaleopp >> 8) << hpel;
return n;
}
/** Predict and set motion vector
*/
void ff_vc1_pred_mv(VC1Context *v, int n, int dmv_x, int dmv_y,
int mv1, int r_x, int r_y, uint8_t* is_intra,
int pred_flag, int dir)
{
MpegEncContext *s = &v->s;
int xy, wrap, off = 0;
int16_t *A, *B, *C;
int px, py;
int sum;
int mixedmv_pic, num_samefield = 0, num_oppfield = 0;
int opposite, a_f, b_f, c_f;
int16_t field_predA[2];
int16_t field_predB[2];
int16_t field_predC[2];
int a_valid, b_valid, c_valid;
int hybridmv_thresh, y_bias = 0;
if (v->mv_mode == MV_PMODE_MIXED_MV ||
((v->mv_mode == MV_PMODE_INTENSITY_COMP) && (v->mv_mode2 == MV_PMODE_MIXED_MV)))
mixedmv_pic = 1;
else
mixedmv_pic = 0;
/* scale MV difference to be quad-pel */
dmv_x <<= 1 - s->quarter_sample;
dmv_y <<= 1 - s->quarter_sample;
wrap = s->b8_stride;
xy = s->block_index[n];
if (s->mb_intra) {
s->mv[0][n][0] = s->current_picture.motion_val[0][xy + v->blocks_off][0] = 0;
s->mv[0][n][1] = s->current_picture.motion_val[0][xy + v->blocks_off][1] = 0;
s->current_picture.motion_val[1][xy + v->blocks_off][0] = 0;
s->current_picture.motion_val[1][xy + v->blocks_off][1] = 0;
if (mv1) { /* duplicate motion data for 1-MV block */
s->current_picture.motion_val[0][xy + 1 + v->blocks_off][0] = 0;
s->current_picture.motion_val[0][xy + 1 + v->blocks_off][1] = 0;
s->current_picture.motion_val[0][xy + wrap + v->blocks_off][0] = 0;
s->current_picture.motion_val[0][xy + wrap + v->blocks_off][1] = 0;
s->current_picture.motion_val[0][xy + wrap + 1 + v->blocks_off][0] = 0;
s->current_picture.motion_val[0][xy + wrap + 1 + v->blocks_off][1] = 0;
v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
s->current_picture.motion_val[1][xy + 1 + v->blocks_off][0] = 0;
s->current_picture.motion_val[1][xy + 1 + v->blocks_off][1] = 0;
s->current_picture.motion_val[1][xy + wrap][0] = 0;
s->current_picture.motion_val[1][xy + wrap + v->blocks_off][1] = 0;
s->current_picture.motion_val[1][xy + wrap + 1 + v->blocks_off][0] = 0;
s->current_picture.motion_val[1][xy + wrap + 1 + v->blocks_off][1] = 0;
}
return;
}
C = s->current_picture.motion_val[dir][xy - 1 + v->blocks_off];
A = s->current_picture.motion_val[dir][xy - wrap + v->blocks_off];
if (mv1) {
if (v->field_mode && mixedmv_pic)
off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
else
off = (s->mb_x == (s->mb_width - 1)) ? -1 : 2;
} else {
//in 4-MV mode different blocks have different B predictor position
switch (n) {
case 0:
off = (s->mb_x > 0) ? -1 : 1;
break;
case 1:
off = (s->mb_x == (s->mb_width - 1)) ? -1 : 1;
break;
case 2:
off = 1;
break;
case 3:
off = -1;
}
}
B = s->current_picture.motion_val[dir][xy - wrap + off + v->blocks_off];
a_valid = !s->first_slice_line || (n == 2 || n == 3);
b_valid = a_valid && (s->mb_width > 1);
c_valid = s->mb_x || (n == 1 || n == 3);
if (v->field_mode) {
a_valid = a_valid && !is_intra[xy - wrap];
b_valid = b_valid && !is_intra[xy - wrap + off];
c_valid = c_valid && !is_intra[xy - 1];
}
if (a_valid) {
a_f = v->mv_f[dir][xy - wrap + v->blocks_off];
num_oppfield += a_f;
num_samefield += 1 - a_f;
field_predA[0] = A[0];
field_predA[1] = A[1];
} else {
field_predA[0] = field_predA[1] = 0;
a_f = 0;
}
if (b_valid) {
b_f = v->mv_f[dir][xy - wrap + off + v->blocks_off];
num_oppfield += b_f;
num_samefield += 1 - b_f;
field_predB[0] = B[0];
field_predB[1] = B[1];
} else {
field_predB[0] = field_predB[1] = 0;
b_f = 0;
}
if (c_valid) {
c_f = v->mv_f[dir][xy - 1 + v->blocks_off];
num_oppfield += c_f;
num_samefield += 1 - c_f;
field_predC[0] = C[0];
field_predC[1] = C[1];
} else {
field_predC[0] = field_predC[1] = 0;
c_f = 0;
}
if (v->field_mode) {
if (!v->numref)
// REFFIELD determines if the last field or the second-last field is
// to be used as reference
opposite = 1 - v->reffield;
else {
if (num_samefield <= num_oppfield)
opposite = 1 - pred_flag;
else
opposite = pred_flag;
}
} else
opposite = 0;
if (opposite) {
if (a_valid && !a_f) {
field_predA[0] = scaleforopp(v, field_predA[0], 0, dir);
field_predA[1] = scaleforopp(v, field_predA[1], 1, dir);
}
if (b_valid && !b_f) {
field_predB[0] = scaleforopp(v, field_predB[0], 0, dir);
field_predB[1] = scaleforopp(v, field_predB[1], 1, dir);
}
if (c_valid && !c_f) {
field_predC[0] = scaleforopp(v, field_predC[0], 0, dir);
field_predC[1] = scaleforopp(v, field_predC[1], 1, dir);
}
v->mv_f[dir][xy + v->blocks_off] = 1;
v->ref_field_type[dir] = !v->cur_field_type;
} else {
if (a_valid && a_f) {
field_predA[0] = scaleforsame(v, n, field_predA[0], 0, dir);
field_predA[1] = scaleforsame(v, n, field_predA[1], 1, dir);
}
if (b_valid && b_f) {
field_predB[0] = scaleforsame(v, n, field_predB[0], 0, dir);
field_predB[1] = scaleforsame(v, n, field_predB[1], 1, dir);
}
if (c_valid && c_f) {
field_predC[0] = scaleforsame(v, n, field_predC[0], 0, dir);
field_predC[1] = scaleforsame(v, n, field_predC[1], 1, dir);
}
v->mv_f[dir][xy + v->blocks_off] = 0;
v->ref_field_type[dir] = v->cur_field_type;
}
if (a_valid) {
px = field_predA[0];
py = field_predA[1];
} else if (c_valid) {
px = field_predC[0];
py = field_predC[1];
} else if (b_valid) {
px = field_predB[0];
py = field_predB[1];
} else {
px = 0;
py = 0;
}
if (num_samefield + num_oppfield > 1) {
px = mid_pred(field_predA[0], field_predB[0], field_predC[0]);
py = mid_pred(field_predA[1], field_predB[1], field_predC[1]);
}
/* Pullback MV as specified in 8.3.5.3.4 */
if (!v->field_mode) {
int qx, qy, X, Y;
qx = (s->mb_x << 6) + ((n == 1 || n == 3) ? 32 : 0);
qy = (s->mb_y << 6) + ((n == 2 || n == 3) ? 32 : 0);
X = (s->mb_width << 6) - 4;
Y = (s->mb_height << 6) - 4;
if (mv1) {
if (qx + px < -60) px = -60 - qx;
if (qy + py < -60) py = -60 - qy;
} else {
if (qx + px < -28) px = -28 - qx;
if (qy + py < -28) py = -28 - qy;
}
if (qx + px > X) px = X - qx;
if (qy + py > Y) py = Y - qy;
}
if (!v->field_mode || s->pict_type != AV_PICTURE_TYPE_B) {
/* Calculate hybrid prediction as specified in 8.3.5.3.5 (also 10.3.5.4.3.5) */
hybridmv_thresh = 32;
if (a_valid && c_valid) {
if (is_intra[xy - wrap])
sum = FFABS(px) + FFABS(py);
else
sum = FFABS(px - field_predA[0]) + FFABS(py - field_predA[1]);
if (sum > hybridmv_thresh) {
if (get_bits1(&s->gb)) { // read HYBRIDPRED bit
px = field_predA[0];
py = field_predA[1];
} else {
px = field_predC[0];
py = field_predC[1];
}
} else {
if (is_intra[xy - 1])
sum = FFABS(px) + FFABS(py);
else
sum = FFABS(px - field_predC[0]) + FFABS(py - field_predC[1]);
if (sum > hybridmv_thresh) {
if (get_bits1(&s->gb)) {
px = field_predA[0];
py = field_predA[1];
} else {
px = field_predC[0];
py = field_predC[1];
}
}
}
}
}
if (v->field_mode && v->numref)
r_y >>= 1;
if (v->field_mode && v->cur_field_type && v->ref_field_type[dir] == 0)
y_bias = 1;
/* store MV using signed modulus of MV range defined in 4.11 */
s->mv[dir][n][0] = s->current_picture.motion_val[dir][xy + v->blocks_off][0] = ((px + dmv_x + r_x) & ((r_x << 1) - 1)) - r_x;
s->mv[dir][n][1] = s->current_picture.motion_val[dir][xy + v->blocks_off][1] = ((py + dmv_y + r_y - y_bias) & ((r_y << 1) - 1)) - r_y + y_bias;
if (mv1) { /* duplicate motion data for 1-MV block */
s->current_picture.motion_val[dir][xy + 1 + v->blocks_off][0] = s->current_picture.motion_val[dir][xy + v->blocks_off][0];
s->current_picture.motion_val[dir][xy + 1 + v->blocks_off][1] = s->current_picture.motion_val[dir][xy + v->blocks_off][1];
s->current_picture.motion_val[dir][xy + wrap + v->blocks_off][0] = s->current_picture.motion_val[dir][xy + v->blocks_off][0];
s->current_picture.motion_val[dir][xy + wrap + v->blocks_off][1] = s->current_picture.motion_val[dir][xy + v->blocks_off][1];
s->current_picture.motion_val[dir][xy + wrap + 1 + v->blocks_off][0] = s->current_picture.motion_val[dir][xy + v->blocks_off][0];
s->current_picture.motion_val[dir][xy + wrap + 1 + v->blocks_off][1] = s->current_picture.motion_val[dir][xy + v->blocks_off][1];
v->mv_f[dir][xy + 1 + v->blocks_off] = v->mv_f[dir][xy + v->blocks_off];
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];
}
}
/** Predict and set motion vector for interlaced frame picture MBs
*/
void ff_vc1_pred_mv_intfr(VC1Context *v, int n, int dmv_x, int dmv_y,
int mvn, int r_x, int r_y, uint8_t* is_intra, int dir)
{
MpegEncContext *s = &v->s;
int xy, wrap, off = 0;
int A[2], B[2], C[2];
int px, py;
int a_valid = 0, b_valid = 0, c_valid = 0;
int field_a, field_b, field_c; // 0: same, 1: opposit
int total_valid, num_samefield, num_oppfield;
int pos_c, pos_b, n_adj;
wrap = s->b8_stride;
xy = s->block_index[n];
if (s->mb_intra) {
s->mv[0][n][0] = s->current_picture.motion_val[0][xy][0] = 0;
s->mv[0][n][1] = s->current_picture.motion_val[0][xy][1] = 0;
s->current_picture.motion_val[1][xy][0] = 0;
s->current_picture.motion_val[1][xy][1] = 0;
if (mvn == 1) { /* duplicate motion data for 1-MV block */
s->current_picture.motion_val[0][xy + 1][0] = 0;
s->current_picture.motion_val[0][xy + 1][1] = 0;
s->current_picture.motion_val[0][xy + wrap][0] = 0;
s->current_picture.motion_val[0][xy + wrap][1] = 0;
s->current_picture.motion_val[0][xy + wrap + 1][0] = 0;
s->current_picture.motion_val[0][xy + wrap + 1][1] = 0;
v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
s->current_picture.motion_val[1][xy + 1][0] = 0;
s->current_picture.motion_val[1][xy + 1][1] = 0;
s->current_picture.motion_val[1][xy + wrap][0] = 0;
s->current_picture.motion_val[1][xy + wrap][1] = 0;
s->current_picture.motion_val[1][xy + wrap + 1][0] = 0;
s->current_picture.motion_val[1][xy + wrap + 1][1] = 0;
}
return;
}
off = ((n == 0) || (n == 1)) ? 1 : -1;
/* predict A */
if (s->mb_x || (n == 1) || (n == 3)) {
if ((v->blk_mv_type[xy]) // current block (MB) has a field MV
|| (!v->blk_mv_type[xy] && !v->blk_mv_type[xy - 1])) { // or both have frame MV
A[0] = s->current_picture.motion_val[dir][xy - 1][0];
A[1] = s->current_picture.motion_val[dir][xy - 1][1];
a_valid = 1;
} else { // current block has frame mv and cand. has field MV (so average)
A[0] = (s->current_picture.motion_val[dir][xy - 1][0]
+ s->current_picture.motion_val[dir][xy - 1 + off * wrap][0] + 1) >> 1;
A[1] = (s->current_picture.motion_val[dir][xy - 1][1]
+ s->current_picture.motion_val[dir][xy - 1 + off * wrap][1] + 1) >> 1;
a_valid = 1;
}
if (!(n & 1) && v->is_intra[s->mb_x - 1]) {
a_valid = 0;
A[0] = A[1] = 0;
}
} else
A[0] = A[1] = 0;
/* Predict B and C */
B[0] = B[1] = C[0] = C[1] = 0;
if (n == 0 || n == 1 || v->blk_mv_type[xy]) {
if (!s->first_slice_line) {
if (!v->is_intra[s->mb_x - s->mb_stride]) {
b_valid = 1;
n_adj = n | 2;
pos_b = s->block_index[n_adj] - 2 * wrap;
if (v->blk_mv_type[pos_b] && v->blk_mv_type[xy]) {
n_adj = (n & 2) | (n & 1);
}
B[0] = s->current_picture.motion_val[dir][s->block_index[n_adj] - 2 * wrap][0];
B[1] = s->current_picture.motion_val[dir][s->block_index[n_adj] - 2 * wrap][1];
if (v->blk_mv_type[pos_b] && !v->blk_mv_type[xy]) {
B[0] = (B[0] + s->current_picture.motion_val[dir][s->block_index[n_adj ^ 2] - 2 * wrap][0] + 1) >> 1;
B[1] = (B[1] + s->current_picture.motion_val[dir][s->block_index[n_adj ^ 2] - 2 * wrap][1] + 1) >> 1;
}
}
if (s->mb_width > 1) {
if (!v->is_intra[s->mb_x - s->mb_stride + 1]) {
c_valid = 1;
n_adj = 2;
pos_c = s->block_index[2] - 2 * wrap + 2;
if (v->blk_mv_type[pos_c] && v->blk_mv_type[xy]) {
n_adj = n & 2;
}
C[0] = s->current_picture.motion_val[dir][s->block_index[n_adj] - 2 * wrap + 2][0];
C[1] = s->current_picture.motion_val[dir][s->block_index[n_adj] - 2 * wrap + 2][1];
if (v->blk_mv_type[pos_c] && !v->blk_mv_type[xy]) {
C[0] = (1 + C[0] + (s->current_picture.motion_val[dir][s->block_index[n_adj ^ 2] - 2 * wrap + 2][0])) >> 1;
C[1] = (1 + C[1] + (s->current_picture.motion_val[dir][s->block_index[n_adj ^ 2] - 2 * wrap + 2][1])) >> 1;
}
if (s->mb_x == s->mb_width - 1) {
if (!v->is_intra[s->mb_x - s->mb_stride - 1]) {
c_valid = 1;
n_adj = 3;
pos_c = s->block_index[3] - 2 * wrap - 2;
if (v->blk_mv_type[pos_c] && v->blk_mv_type[xy]) {
n_adj = n | 1;
}
C[0] = s->current_picture.motion_val[dir][s->block_index[n_adj] - 2 * wrap - 2][0];
C[1] = s->current_picture.motion_val[dir][s->block_index[n_adj] - 2 * wrap - 2][1];
if (v->blk_mv_type[pos_c] && !v->blk_mv_type[xy]) {
C[0] = (1 + C[0] + s->current_picture.motion_val[dir][s->block_index[1] - 2 * wrap - 2][0]) >> 1;
C[1] = (1 + C[1] + s->current_picture.motion_val[dir][s->block_index[1] - 2 * wrap - 2][1]) >> 1;
}
} else
c_valid = 0;
}
}
}
}
} else {
pos_b = s->block_index[1];
b_valid = 1;
B[0] = s->current_picture.motion_val[dir][pos_b][0];
B[1] = s->current_picture.motion_val[dir][pos_b][1];
pos_c = s->block_index[0];
c_valid = 1;
C[0] = s->current_picture.motion_val[dir][pos_c][0];
C[1] = s->current_picture.motion_val[dir][pos_c][1];
}
total_valid = a_valid + b_valid + c_valid;
// check if predictor A is out of bounds
if (!s->mb_x && !(n == 1 || n == 3)) {
A[0] = A[1] = 0;
}
// check if predictor B is out of bounds
if ((s->first_slice_line && v->blk_mv_type[xy]) || (s->first_slice_line && !(n & 2))) {
B[0] = B[1] = C[0] = C[1] = 0;
}
if (!v->blk_mv_type[xy]) {
if (s->mb_width == 1) {
px = B[0];
py = B[1];
} else {
if (total_valid >= 2) {
px = mid_pred(A[0], B[0], C[0]);
py = mid_pred(A[1], B[1], C[1]);
} else if (total_valid) {
if (a_valid) { px = A[0]; py = A[1]; }
if (b_valid) { px = B[0]; py = B[1]; }
if (c_valid) { px = C[0]; py = C[1]; }
} else
px = py = 0;
}
} else {
if (a_valid)
field_a = (A[1] & 4) ? 1 : 0;
else
field_a = 0;
if (b_valid)
field_b = (B[1] & 4) ? 1 : 0;
else
field_b = 0;
if (c_valid)
field_c = (C[1] & 4) ? 1 : 0;
else
field_c = 0;
num_oppfield = field_a + field_b + field_c;
num_samefield = total_valid - num_oppfield;
if (total_valid == 3) {
if ((num_samefield == 3) || (num_oppfield == 3)) {
px = mid_pred(A[0], B[0], C[0]);
py = mid_pred(A[1], B[1], C[1]);
} else if (num_samefield >= num_oppfield) {
/* take one MV from same field set depending on priority
the check for B may not be necessary */
px = !field_a ? A[0] : B[0];
py = !field_a ? A[1] : B[1];
} else {
px = field_a ? A[0] : B[0];
py = field_a ? A[1] : B[1];
}
} else if (total_valid == 2) {
if (num_samefield >= num_oppfield) {
if (!field_a && a_valid) {
px = A[0];
py = A[1];
} else if (!field_b && b_valid) {
px = B[0];
py = B[1];
} else if (c_valid) {
px = C[0];
py = C[1];
} else px = py = 0;
} else {
if (field_a && a_valid) {
px = A[0];
py = A[1];
} else if (field_b && b_valid) {
px = B[0];
py = B[1];
} else if (c_valid) {
px = C[0];
py = C[1];
} else
px = py = 0;
}
} else if (total_valid == 1) {
px = (a_valid) ? A[0] : ((b_valid) ? B[0] : C[0]);
py = (a_valid) ? A[1] : ((b_valid) ? B[1] : C[1]);
} else
px = py = 0;
}
/* store MV using signed modulus of MV range defined in 4.11 */
s->mv[dir][n][0] = s->current_picture.motion_val[dir][xy][0] = ((px + dmv_x + r_x) & ((r_x << 1) - 1)) - r_x;
s->mv[dir][n][1] = s->current_picture.motion_val[dir][xy][1] = ((py + dmv_y + r_y) & ((r_y << 1) - 1)) - r_y;
if (mvn == 1) { /* duplicate motion data for 1-MV block */
s->current_picture.motion_val[dir][xy + 1 ][0] = s->current_picture.motion_val[dir][xy][0];
s->current_picture.motion_val[dir][xy + 1 ][1] = s->current_picture.motion_val[dir][xy][1];
s->current_picture.motion_val[dir][xy + wrap ][0] = s->current_picture.motion_val[dir][xy][0];
s->current_picture.motion_val[dir][xy + wrap ][1] = s->current_picture.motion_val[dir][xy][1];
s->current_picture.motion_val[dir][xy + wrap + 1][0] = s->current_picture.motion_val[dir][xy][0];
s->current_picture.motion_val[dir][xy + wrap + 1][1] = s->current_picture.motion_val[dir][xy][1];
} else if (mvn == 2) { /* duplicate motion data for 2-Field MV block */
s->current_picture.motion_val[dir][xy + 1][0] = s->current_picture.motion_val[dir][xy][0];
s->current_picture.motion_val[dir][xy + 1][1] = s->current_picture.motion_val[dir][xy][1];
s->mv[dir][n + 1][0] = s->mv[dir][n][0];
s->mv[dir][n + 1][1] = s->mv[dir][n][1];
}
}
void ff_vc1_pred_b_mv(VC1Context *v, int dmv_x[2], int dmv_y[2],
int direct, int mvtype)
{
MpegEncContext *s = &v->s;
int xy, wrap, off = 0;
int16_t *A, *B, *C;
int px, py;
int sum;
int r_x, r_y;
const uint8_t *is_intra = v->mb_type[0];
r_x = v->range_x;
r_y = v->range_y;
/* scale MV difference to be quad-pel */
dmv_x[0] <<= 1 - s->quarter_sample;
dmv_y[0] <<= 1 - s->quarter_sample;
dmv_x[1] <<= 1 - s->quarter_sample;
dmv_y[1] <<= 1 - s->quarter_sample;
wrap = s->b8_stride;
xy = s->block_index[0];
if (s->mb_intra) {
s->current_picture.motion_val[0][xy + v->blocks_off][0] =
s->current_picture.motion_val[0][xy + v->blocks_off][1] =
s->current_picture.motion_val[1][xy + v->blocks_off][0] =
s->current_picture.motion_val[1][xy + v->blocks_off][1] = 0;
return;
}
if (!v->field_mode) {
s->mv[0][0][0] = scale_mv(s->next_picture.motion_val[1][xy][0], v->bfraction, 0, s->quarter_sample);
s->mv[0][0][1] = scale_mv(s->next_picture.motion_val[1][xy][1], v->bfraction, 0, s->quarter_sample);
s->mv[1][0][0] = scale_mv(s->next_picture.motion_val[1][xy][0], v->bfraction, 1, s->quarter_sample);
s->mv[1][0][1] = scale_mv(s->next_picture.motion_val[1][xy][1], v->bfraction, 1, s->quarter_sample);
/* Pullback predicted motion vectors as specified in 8.4.5.4 */
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));
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));
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));
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));
}
if (direct) {
s->current_picture.motion_val[0][xy + v->blocks_off][0] = s->mv[0][0][0];
s->current_picture.motion_val[0][xy + v->blocks_off][1] = s->mv[0][0][1];
s->current_picture.motion_val[1][xy + v->blocks_off][0] = s->mv[1][0][0];
s->current_picture.motion_val[1][xy + v->blocks_off][1] = s->mv[1][0][1];
return;
}
if ((mvtype == BMV_TYPE_FORWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
C = s->current_picture.motion_val[0][xy - 2];
A = s->current_picture.motion_val[0][xy - wrap * 2];
off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
B = s->current_picture.motion_val[0][xy - wrap * 2 + off];
if (!s->mb_x) C[0] = C[1] = 0;
if (!s->first_slice_line) { // predictor A is not out of bounds
if (s->mb_width == 1) {
px = A[0];
py = A[1];
} else {
px = mid_pred(A[0], B[0], C[0]);
py = mid_pred(A[1], B[1], C[1]);
}
} else if (s->mb_x) { // predictor C is not out of bounds
px = C[0];
py = C[1];
} else {
px = py = 0;
}
/* Pullback MV as specified in 8.3.5.3.4 */
{
int qx, qy, X, Y;
if (v->profile < PROFILE_ADVANCED) {
qx = (s->mb_x << 5);
qy = (s->mb_y << 5);
X = (s->mb_width << 5) - 4;
Y = (s->mb_height << 5) - 4;
if (qx + px < -28) px = -28 - qx;
if (qy + py < -28) py = -28 - qy;
if (qx + px > X) px = X - qx;
if (qy + py > Y) py = Y - qy;
} else {
qx = (s->mb_x << 6);
qy = (s->mb_y << 6);
X = (s->mb_width << 6) - 4;
Y = (s->mb_height << 6) - 4;
if (qx + px < -60) px = -60 - qx;
if (qy + py < -60) py = -60 - qy;
if (qx + px > X) px = X - qx;
if (qy + py > Y) py = Y - qy;
}
}
/* Calculate hybrid prediction as specified in 8.3.5.3.5 */
if (0 && !s->first_slice_line && s->mb_x) {
if (is_intra[xy - wrap])
sum = FFABS(px) + FFABS(py);
else
sum = FFABS(px - A[0]) + FFABS(py - A[1]);
if (sum > 32) {
if (get_bits1(&s->gb)) {
px = A[0];
py = A[1];
} else {
px = C[0];
py = C[1];
}
} else {
if (is_intra[xy - 2])
sum = FFABS(px) + FFABS(py);
else
sum = FFABS(px - C[0]) + FFABS(py - C[1]);
if (sum > 32) {
if (get_bits1(&s->gb)) {
px = A[0];
py = A[1];
} else {
px = C[0];
py = C[1];
}
}
}
}
/* store MV using signed modulus of MV range defined in 4.11 */
s->mv[0][0][0] = ((px + dmv_x[0] + r_x) & ((r_x << 1) - 1)) - r_x;
s->mv[0][0][1] = ((py + dmv_y[0] + r_y) & ((r_y << 1) - 1)) - r_y;
}
if ((mvtype == BMV_TYPE_BACKWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
C = s->current_picture.motion_val[1][xy - 2];
A = s->current_picture.motion_val[1][xy - wrap * 2];
off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
B = s->current_picture.motion_val[1][xy - wrap * 2 + off];
if (!s->mb_x)
C[0] = C[1] = 0;
if (!s->first_slice_line) { // predictor A is not out of bounds
if (s->mb_width == 1) {
px = A[0];
py = A[1];
} else {
px = mid_pred(A[0], B[0], C[0]);
py = mid_pred(A[1], B[1], C[1]);
}
} else if (s->mb_x) { // predictor C is not out of bounds
px = C[0];
py = C[1];
} else {
px = py = 0;
}
/* Pullback MV as specified in 8.3.5.3.4 */
{
int qx, qy, X, Y;
if (v->profile < PROFILE_ADVANCED) {
qx = (s->mb_x << 5);
qy = (s->mb_y << 5);
X = (s->mb_width << 5) - 4;
Y = (s->mb_height << 5) - 4;
if (qx + px < -28) px = -28 - qx;
if (qy + py < -28) py = -28 - qy;
if (qx + px > X) px = X - qx;
if (qy + py > Y) py = Y - qy;
} else {
qx = (s->mb_x << 6);
qy = (s->mb_y << 6);
X = (s->mb_width << 6) - 4;
Y = (s->mb_height << 6) - 4;
if (qx + px < -60) px = -60 - qx;
if (qy + py < -60) py = -60 - qy;
if (qx + px > X) px = X - qx;
if (qy + py > Y) py = Y - qy;
}
}
/* Calculate hybrid prediction as specified in 8.3.5.3.5 */
if (0 && !s->first_slice_line && s->mb_x) {
if (is_intra[xy - wrap])
sum = FFABS(px) + FFABS(py);
else
sum = FFABS(px - A[0]) + FFABS(py - A[1]);
if (sum > 32) {
if (get_bits1(&s->gb)) {
px = A[0];
py = A[1];
} else {
px = C[0];
py = C[1];
}
} else {
if (is_intra[xy - 2])
sum = FFABS(px) + FFABS(py);
else
sum = FFABS(px - C[0]) + FFABS(py - C[1]);
if (sum > 32) {
if (get_bits1(&s->gb)) {
px = A[0];
py = A[1];
} else {
px = C[0];
py = C[1];
}
}
}
}
/* store MV using signed modulus of MV range defined in 4.11 */
s->mv[1][0][0] = ((px + dmv_x[1] + r_x) & ((r_x << 1) - 1)) - r_x;
s->mv[1][0][1] = ((py + dmv_y[1] + r_y) & ((r_y << 1) - 1)) - r_y;
}
s->current_picture.motion_val[0][xy][0] = s->mv[0][0][0];
s->current_picture.motion_val[0][xy][1] = s->mv[0][0][1];
s->current_picture.motion_val[1][xy][0] = s->mv[1][0][0];
s->current_picture.motion_val[1][xy][1] = s->mv[1][0][1];
}
void ff_vc1_pred_b_mv_intfi(VC1Context *v, int n, int *dmv_x, int *dmv_y,
int mv1, int *pred_flag)
{
int dir = (v->bmvtype == BMV_TYPE_BACKWARD) ? 1 : 0;
MpegEncContext *s = &v->s;
int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
if (v->bmvtype == BMV_TYPE_DIRECT) {
int total_opp, k, f;
if (s->next_picture.mb_type[mb_pos + v->mb_off] != MB_TYPE_INTRA) {
s->mv[0][0][0] = scale_mv(s->next_picture.motion_val[1][s->block_index[0] + v->blocks_off][0],
v->bfraction, 0, s->quarter_sample);
s->mv[0][0][1] = scale_mv(s->next_picture.motion_val[1][s->block_index[0] + v->blocks_off][1],
v->bfraction, 0, s->quarter_sample);
s->mv[1][0][0] = scale_mv(s->next_picture.motion_val[1][s->block_index[0] + v->blocks_off][0],
v->bfraction, 1, s->quarter_sample);
s->mv[1][0][1] = scale_mv(s->next_picture.motion_val[1][s->block_index[0] + v->blocks_off][1],
v->bfraction, 1, s->quarter_sample);
total_opp = v->mv_f_next[0][s->block_index[0] + v->blocks_off]
+ v->mv_f_next[0][s->block_index[1] + v->blocks_off]
+ v->mv_f_next[0][s->block_index[2] + v->blocks_off]
+ v->mv_f_next[0][s->block_index[3] + v->blocks_off];
f = (total_opp > 2) ? 1 : 0;
} else {
s->mv[0][0][0] = s->mv[0][0][1] = 0;
s->mv[1][0][0] = s->mv[1][0][1] = 0;
f = 0;
}
v->ref_field_type[0] = v->ref_field_type[1] = v->cur_field_type ^ f;
for (k = 0; k < 4; k++) {
s->current_picture.motion_val[0][s->block_index[k] + v->blocks_off][0] = s->mv[0][0][0];
s->current_picture.motion_val[0][s->block_index[k] + v->blocks_off][1] = s->mv[0][0][1];
s->current_picture.motion_val[1][s->block_index[k] + v->blocks_off][0] = s->mv[1][0][0];
s->current_picture.motion_val[1][s->block_index[k] + v->blocks_off][1] = s->mv[1][0][1];
v->mv_f[0][s->block_index[k] + v->blocks_off] = f;
v->mv_f[1][s->block_index[k] + v->blocks_off] = f;
}
return;
}
if (v->bmvtype == BMV_TYPE_INTERPOLATED) {
ff_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);
ff_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);
return;
}
if (dir) { // backward
ff_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);
if (n == 3 || mv1) {
ff_vc1_pred_mv(v, 0, dmv_x[0], dmv_y[0], 1, v->range_x, v->range_y, v->mb_type[0], 0, 0);
}
} else { // forward
ff_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);
if (n == 3 || mv1) {
ff_vc1_pred_mv(v, 0, dmv_x[1], dmv_y[1], 1, v->range_x, v->range_y, v->mb_type[0], 0, 1);
}
}
}

59
libavcodec/vc1_pred.h Normal file
View File

@ -0,0 +1,59 @@
/*
* VC-1 and WMV3 decoder
* Copyright (c) 2006-2007 Konstantin Shishkov
* Partly based on vc9.c (c) 2005 Anonymous, Alex Beregszaszi, Michael Niedermayer
*
* This file is part of Libav.
*
* Libav is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* Libav is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#ifndef AVCODEC_VC1_PRED_H
#define AVCODEC_VC1_PRED_H
#include "vc1.h"
#include "vc1data.h"
void ff_vc1_pred_mv(VC1Context *v, int n, int dmv_x, int dmv_y,
int mv1, int r_x, int r_y, uint8_t* is_intra,
int pred_flag, int dir);
void ff_vc1_pred_mv_intfr(VC1Context *v, int n, int dmv_x, int dmv_y,
int mvn, int r_x, int r_y, uint8_t* is_intra,
int dir);
void ff_vc1_pred_b_mv(VC1Context *v, int dmv_x[2], int dmv_y[2],
int direct, int mvtype);
void ff_vc1_pred_b_mv_intfi(VC1Context *v, int n, int *dmv_x, int *dmv_y,
int mv1, int *pred_flag);
static av_always_inline int scale_mv(int value, int bfrac, int inv, int qs)
{
int n = bfrac;
#if B_FRACTION_DEN==256
if (inv)
n -= 256;
if (!qs)
return 2 * ((value * n + 255) >> 9);
return (value * n + 128) >> 8;
#else
if (inv)
n -= B_FRACTION_DEN;
if (!qs)
return 2 * ((value * n + B_FRACTION_DEN - 1) / (2 * B_FRACTION_DEN));
return (value * n + B_FRACTION_DEN/2) / B_FRACTION_DEN;
#endif
}
#endif /* AVCODEC_VC1_PRED_H */

2
libavcodec/vc1data.h
View File

@ -94,8 +94,6 @@ extern VLC ff_vc1_ac_coeff_table[8];
#define VC1_IF_MBMODE_VLC_BITS 5 #define VC1_IF_MBMODE_VLC_BITS 5
//@} //@}
/* Denominator used for ff_vc1_bfraction_lut */
#define B_FRACTION_DEN 256 #define B_FRACTION_DEN 256
/* pre-computed scales for all bfractions and base=256 */ /* pre-computed scales for all bfractions and base=256 */

5211
libavcodec/vc1dec.c
View File

File diff suppressed because it is too large Load Diff