mpegvideo.c 94.5 KB
Newer Older
F
Fabrice Bellard 已提交
1 2
/*
 * The simplest mpeg encoder (well, it was the simplest!)
F
Fabrice Bellard 已提交
3
 * Copyright (c) 2000,2001 Fabrice Bellard.
F
Fabrice Bellard 已提交
4
 *
F
Fabrice Bellard 已提交
5 6 7 8
 * This library 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 of the License, or (at your option) any later version.
F
Fabrice Bellard 已提交
9
 *
F
Fabrice Bellard 已提交
10
 * This library is distributed in the hope that it will be useful,
F
Fabrice Bellard 已提交
11
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
F
Fabrice Bellard 已提交
12 13
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
F
Fabrice Bellard 已提交
14
 *
F
Fabrice Bellard 已提交
15 16 17
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
M
Michael Niedermayer 已提交
18
 *
19
 * 4MV & hq & b-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
F
Fabrice Bellard 已提交
20 21 22 23 24
 */
#include "avcodec.h"
#include "dsputil.h"
#include "mpegvideo.h"

25 26 27 28
#ifdef USE_FASTMEMCPY
#include "fastmemcpy.h"
#endif

29 30 31
static void encode_picture(MpegEncContext *s, int picture_number);
static void dct_unquantize_mpeg1_c(MpegEncContext *s, 
                                   DCTELEM *block, int n, int qscale);
32 33
static void dct_unquantize_mpeg2_c(MpegEncContext *s,
                                   DCTELEM *block, int n, int qscale);
34 35
static void dct_unquantize_h263_c(MpegEncContext *s, 
                                  DCTELEM *block, int n, int qscale);
M
Michael Niedermayer 已提交
36
static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w);
37
static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
M
Michael Niedermayer 已提交
38

39
int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow)= dct_quantize_c;
M
Michael Niedermayer 已提交
40 41
void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w)= draw_edges_c;

F
Fabrice Bellard 已提交
42 43 44 45 46 47 48
#define EDGE_WIDTH 16

/* enable all paranoid tests for rounding, overflows, etc... */
//#define PARANOID

//#define DEBUG

49

F
Fabrice Bellard 已提交
50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68
/* for jpeg fast DCT */
#define CONST_BITS 14

static const unsigned short aanscales[64] = {
    /* precomputed values scaled up by 14 bits */
    16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
    22725, 31521, 29692, 26722, 22725, 17855, 12299,  6270,
    21407, 29692, 27969, 25172, 21407, 16819, 11585,  5906,
    19266, 26722, 25172, 22654, 19266, 15137, 10426,  5315,
    16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
    12873, 17855, 16819, 15137, 12873, 10114,  6967,  3552,
    8867, 12299, 11585, 10426,  8867,  6967,  4799,  2446,
    4520,  6270,  5906,  5315,  4520,  3552,  2446,  1247
};

static UINT8 h263_chroma_roundtab[16] = {
    0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2,
};

69
static UINT16 default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
70 71
static UINT8 default_fcode_tab[MAX_MV*2+1];

M
Michael Niedermayer 已提交
72 73
extern UINT8 zigzag_end[64];

74 75 76
/* default motion estimation */
int motion_estimation_method = ME_EPZS;

77 78
static void convert_matrix(int (*qmat)[64], uint16_t (*qmat16)[64], uint16_t (*qmat16_bias)[64],
                           const UINT16 *quant_matrix, int bias)
F
Fabrice Bellard 已提交
79
{
80 81 82 83
    int qscale;

    for(qscale=1; qscale<32; qscale++){
        int i;
84
        if (av_fdct == fdct_ifast) {
85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108
            for(i=0;i<64;i++) {
                const int j= block_permute_op(i);
                /* 16 <= qscale * quant_matrix[i] <= 7905 */
                /* 19952         <= aanscales[i] * qscale * quant_matrix[i]           <= 249205026 */
                /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
                /* 3444240       >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
                
                qmat[qscale][j] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) / 
                                (aanscales[i] * qscale * quant_matrix[j]));
            }
        } else {
            for(i=0;i<64;i++) {
                /* We can safely suppose that 16 <= quant_matrix[i] <= 255
                   So 16           <= qscale * quant_matrix[i]             <= 7905
                   so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
                   so 32768        >= (1<<19) / (qscale * quant_matrix[i]) >= 67
                */
                qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
                qmat16[qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[block_permute_op(i)]);

                if(qmat16[qscale][i]==0 || qmat16[qscale][i]==128*256) qmat16[qscale][i]=128*256-1;

                qmat16_bias[qscale][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][i]);
            }
F
Fabrice Bellard 已提交
109 110 111
        }
    }
}
112 113 114 115 116 117 118 119 120
// move into common.c perhaps 
#define CHECKED_ALLOCZ(p, size)\
{\
    p= av_mallocz(size);\
    if(p==NULL){\
        perror("malloc");\
        goto fail;\
    }\
}
F
Fabrice Bellard 已提交
121 122 123 124 125 126 127

/* init common structure for both encoder and decoder */
int MPV_common_init(MpegEncContext *s)
{
    int c_size, i;
    UINT8 *pict;

128
    s->dct_unquantize_h263 = dct_unquantize_h263_c;
129 130
    s->dct_unquantize_mpeg1 = dct_unquantize_mpeg1_c;
    s->dct_unquantize_mpeg2 = dct_unquantize_mpeg2_c;
131 132 133
        
#ifdef HAVE_MMX
    MPV_common_init_mmx(s);
134 135 136
#endif
#ifdef ARCH_ALPHA
    MPV_common_init_axp(s);
N
Nick Kurshev 已提交
137
#endif
138 139 140 141
    //setup default unquantizers (mpeg4 might change it later)
    if(s->out_format == FMT_H263)
        s->dct_unquantize = s->dct_unquantize_h263;
    else
142
        s->dct_unquantize = s->dct_unquantize_mpeg1;
143
    
F
Fabrice Bellard 已提交
144 145
    s->mb_width = (s->width + 15) / 16;
    s->mb_height = (s->height + 15) / 16;
146
    s->mb_num = s->mb_width * s->mb_height;
147 148
    if(!(s->flags&CODEC_FLAG_DR1)){
      s->linesize = s->mb_width * 16 + 2 * EDGE_WIDTH;
F
Fabrice Bellard 已提交
149

150
      for(i=0;i<3;i++) {
F
Fabrice Bellard 已提交
151 152 153 154 155 156 157 158
        int w, h, shift, pict_start;

        w = s->linesize;
        h = s->mb_height * 16 + 2 * EDGE_WIDTH;
        shift = (i == 0) ? 0 : 1;
        c_size = (w >> shift) * (h >> shift);
        pict_start = (w >> shift) * (EDGE_WIDTH >> shift) + (EDGE_WIDTH >> shift);

159
        CHECKED_ALLOCZ(pict, c_size)
F
Fabrice Bellard 已提交
160 161
        s->last_picture_base[i] = pict;
        s->last_picture[i] = pict + pict_start;
M
Michael Niedermayer 已提交
162
        if(i>0) memset(s->last_picture_base[i], 128, c_size);
F
Fabrice Bellard 已提交
163
    
164
        CHECKED_ALLOCZ(pict, c_size)
F
Fabrice Bellard 已提交
165 166
        s->next_picture_base[i] = pict;
        s->next_picture[i] = pict + pict_start;
M
Michael Niedermayer 已提交
167
        if(i>0) memset(s->next_picture_base[i], 128, c_size);
168
        
169 170 171
        if (s->has_b_frames || s->codec_id==CODEC_ID_MPEG4) {
        /* Note the MPEG4 stuff is here cuz of buggy encoders which dont set the low_delay flag but 
           do low-delay encoding, so we cant allways distinguish b-frame containing streams from low_delay streams */
172
            CHECKED_ALLOCZ(pict, c_size)
F
Fabrice Bellard 已提交
173 174
            s->aux_picture_base[i] = pict;
            s->aux_picture[i] = pict + pict_start;
M
Michael Niedermayer 已提交
175
            if(i>0) memset(s->aux_picture_base[i], 128, c_size);
F
Fabrice Bellard 已提交
176
        }
177
      }
F
Fabrice Bellard 已提交
178
    }
179
    
180 181
    CHECKED_ALLOCZ(s->edge_emu_buffer, (s->width+32)*2*17);
    
182
    if (s->encoding) {
183 184
        int j;
        int mv_table_size= (s->mb_width+2)*(s->mb_height+2);
J
Juanjo 已提交
185
        
186 187
        CHECKED_ALLOCZ(s->mb_var   , s->mb_num * sizeof(INT16))
        CHECKED_ALLOCZ(s->mc_mb_var, s->mb_num * sizeof(INT16))
188 189

        /* Allocate MV tables */
190 191 192 193 194 195 196 197 198 199 200 201 202
        CHECKED_ALLOCZ(s->p_mv_table            , mv_table_size * 2 * sizeof(INT16))
        CHECKED_ALLOCZ(s->b_forw_mv_table       , mv_table_size * 2 * sizeof(INT16))
        CHECKED_ALLOCZ(s->b_back_mv_table       , mv_table_size * 2 * sizeof(INT16))
        CHECKED_ALLOCZ(s->b_bidir_forw_mv_table , mv_table_size * 2 * sizeof(INT16))
        CHECKED_ALLOCZ(s->b_bidir_back_mv_table , mv_table_size * 2 * sizeof(INT16))
        CHECKED_ALLOCZ(s->b_direct_forw_mv_table, mv_table_size * 2 * sizeof(INT16))
        CHECKED_ALLOCZ(s->b_direct_back_mv_table, mv_table_size * 2 * sizeof(INT16))
        CHECKED_ALLOCZ(s->b_direct_mv_table     , mv_table_size * 2 * sizeof(INT16))

        CHECKED_ALLOCZ(s->me_scratchpad,  s->linesize*16*3*sizeof(uint8_t))
        
        CHECKED_ALLOCZ(s->me_map      , ME_MAP_SIZE*sizeof(uint32_t))
        CHECKED_ALLOCZ(s->me_score_map, ME_MAP_SIZE*sizeof(uint16_t))
203

204 205 206 207 208 209 210 211 212 213 214
        if(s->max_b_frames){
            for(j=0; j<REORDER_BUFFER_SIZE; j++){
                int i;
                for(i=0;i<3;i++) {
                    int w, h, shift;

                    w = s->linesize;
                    h = s->mb_height * 16;
                    shift = (i == 0) ? 0 : 1;
                    c_size = (w >> shift) * (h >> shift);

215
                    CHECKED_ALLOCZ(pict, c_size);
216 217 218 219
                    s->picture_buffer[j][i] = pict;
                }
            }
        }
220 221 222 223 224

        if(s->codec_id==CODEC_ID_MPEG4){
            CHECKED_ALLOCZ(s->tex_pb_buffer, PB_BUFFER_SIZE);
            CHECKED_ALLOCZ(   s->pb2_buffer, PB_BUFFER_SIZE);
        }
225 226
    }
    
227
    if (s->out_format == FMT_H263 || s->encoding) {
F
Fabrice Bellard 已提交
228
        int size;
229 230 231
        /* Allocate MB type table */
        CHECKED_ALLOCZ(s->mb_type  , s->mb_num * sizeof(UINT8))

F
Fabrice Bellard 已提交
232 233
        /* MV prediction */
        size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2);
234
        CHECKED_ALLOCZ(s->motion_val, size * 2 * sizeof(INT16));
F
Fabrice Bellard 已提交
235 236
    }

237
    if (s->h263_pred || s->h263_plus) {
F
Fabrice Bellard 已提交
238 239 240 241 242 243 244
        int y_size, c_size, i, size;
        
        /* dc values */

        y_size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2);
        c_size = (s->mb_width + 2) * (s->mb_height + 2);
        size = y_size + 2 * c_size;
245
        CHECKED_ALLOCZ(s->dc_val[0], size * sizeof(INT16));
F
Fabrice Bellard 已提交
246 247 248 249 250 251
        s->dc_val[1] = s->dc_val[0] + y_size;
        s->dc_val[2] = s->dc_val[1] + c_size;
        for(i=0;i<size;i++)
            s->dc_val[0][i] = 1024;

        /* ac values */
252
        CHECKED_ALLOCZ(s->ac_val[0], size * sizeof(INT16) * 16);
F
Fabrice Bellard 已提交
253 254 255 256
        s->ac_val[1] = s->ac_val[0] + y_size;
        s->ac_val[2] = s->ac_val[1] + c_size;
        
        /* cbp values */
257
        CHECKED_ALLOCZ(s->coded_block, y_size);
258 259

        /* which mb is a intra block */
260
        CHECKED_ALLOCZ(s->mbintra_table, s->mb_num);
261
        memset(s->mbintra_table, 1, s->mb_num);
M
Michael Niedermayer 已提交
262 263
        
        /* divx501 bitstream reorder buffer */
264 265 266 267 268 269 270
        CHECKED_ALLOCZ(s->bitstream_buffer, BITSTREAM_BUFFER_SIZE);
        
        /* cbp, ac_pred, pred_dir */
        CHECKED_ALLOCZ(s->cbp_table  , s->mb_num * sizeof(UINT8))
        CHECKED_ALLOCZ(s->pred_dir_table, s->mb_num * sizeof(UINT8))
        
        CHECKED_ALLOCZ(s->qscale_table  , s->mb_num * sizeof(UINT8))
271
    }
F
Fabrice Bellard 已提交
272 273
    /* default structure is frame */
    s->picture_structure = PICT_FRAME;
274
    
275
    /* init macroblock skip table */
276
    CHECKED_ALLOCZ(s->mbskip_table, s->mb_num);
277
    
278
    s->block= s->blocks[0];
279

F
Fabrice Bellard 已提交
280 281 282
    s->context_initialized = 1;
    return 0;
 fail:
283
    MPV_common_end(s);
F
Fabrice Bellard 已提交
284 285 286
    return -1;
}

287 288 289

//extern int sads;

F
Fabrice Bellard 已提交
290 291 292 293 294
/* init common structure for both encoder and decoder */
void MPV_common_end(MpegEncContext *s)
{
    int i;

295 296
    av_freep(&s->mb_type);
    av_freep(&s->mb_var);
297
    av_freep(&s->mc_mb_var);
298 299 300 301 302 303 304 305 306 307 308 309 310
    av_freep(&s->p_mv_table);
    av_freep(&s->b_forw_mv_table);
    av_freep(&s->b_back_mv_table);
    av_freep(&s->b_bidir_forw_mv_table);
    av_freep(&s->b_bidir_back_mv_table);
    av_freep(&s->b_direct_forw_mv_table);
    av_freep(&s->b_direct_back_mv_table);
    av_freep(&s->b_direct_mv_table);
    av_freep(&s->motion_val);
    av_freep(&s->dc_val[0]);
    av_freep(&s->ac_val[0]);
    av_freep(&s->coded_block);
    av_freep(&s->mbintra_table);
311 312 313
    av_freep(&s->cbp_table);
    av_freep(&s->pred_dir_table);
    av_freep(&s->qscale_table);
314
    av_freep(&s->me_scratchpad);
315 316 317
    av_freep(&s->me_map);
    av_freep(&s->me_score_map);
    
318 319
    av_freep(&s->mbskip_table);
    av_freep(&s->bitstream_buffer);
320 321
    av_freep(&s->tex_pb_buffer);
    av_freep(&s->pb2_buffer);
322 323
    av_freep(&s->edge_emu_buffer);
    
F
Fabrice Bellard 已提交
324
    for(i=0;i<3;i++) {
325
        int j;
326 327 328 329 330 331 332 333 334 335 336 337
        if(!(s->flags&CODEC_FLAG_DR1)){
            av_freep(&s->last_picture_base[i]);
            av_freep(&s->next_picture_base[i]);
            av_freep(&s->aux_picture_base[i]);
        }
        s->last_picture_base[i]=
        s->next_picture_base[i]=
        s->aux_picture_base [i] = NULL;
        s->last_picture[i]=
        s->next_picture[i]=
        s->aux_picture [i] = NULL;

338
        for(j=0; j<REORDER_BUFFER_SIZE; j++){
339
            av_freep(&s->picture_buffer[j][i]);
340
        }
F
Fabrice Bellard 已提交
341 342 343 344 345 346 347 348
    }
    s->context_initialized = 0;
}

/* init video encoder */
int MPV_encode_init(AVCodecContext *avctx)
{
    MpegEncContext *s = avctx->priv_data;
349
    int i;
F
Fabrice Bellard 已提交
350

351 352
    avctx->pix_fmt = PIX_FMT_YUV420P;

F
Fabrice Bellard 已提交
353
    s->bit_rate = avctx->bit_rate;
354
    s->bit_rate_tolerance = avctx->bit_rate_tolerance;
F
Fabrice Bellard 已提交
355 356 357
    s->frame_rate = avctx->frame_rate;
    s->width = avctx->width;
    s->height = avctx->height;
358
    if(avctx->gop_size > 600){
359
        fprintf(stderr, "Warning keyframe interval too large! reducing it ...\n");
360 361
        avctx->gop_size=600;
    }
F
Fabrice Bellard 已提交
362
    s->gop_size = avctx->gop_size;
363 364
    s->rtp_mode = avctx->rtp_mode;
    s->rtp_payload_size = avctx->rtp_payload_size;
365 366
    if (avctx->rtp_callback)
        s->rtp_callback = avctx->rtp_callback;
367 368 369 370 371
    s->qmin= avctx->qmin;
    s->qmax= avctx->qmax;
    s->max_qdiff= avctx->max_qdiff;
    s->qcompress= avctx->qcompress;
    s->qblur= avctx->qblur;
372
    s->b_quant_factor= avctx->b_quant_factor;
373
    s->b_quant_offset= avctx->b_quant_offset;
374
    s->avctx = avctx;
375
    s->aspect_ratio_info= avctx->aspect_ratio_info;
376
    s->flags= avctx->flags;
377
    s->max_b_frames= avctx->max_b_frames;
378 379
    s->rc_strategy= avctx->rc_strategy;
    s->b_frame_strategy= avctx->b_frame_strategy;
380
    s->codec_id= avctx->codec->id;
381 382 383 384
    s->luma_elim_threshold  = avctx->luma_elim_threshold;
    s->chroma_elim_threshold= avctx->chroma_elim_threshold;
    s->strict_std_compliance= avctx->strict_std_compliance;
    s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
385

F
Fabrice Bellard 已提交
386 387 388 389 390 391
    if (s->gop_size <= 1) {
        s->intra_only = 1;
        s->gop_size = 12;
    } else {
        s->intra_only = 0;
    }
392 393
    
    /* ME algorithm */
394 395 396 397 398 399
    if (avctx->me_method == 0)
        /* For compatibility */
        s->me_method = motion_estimation_method;
    else
        s->me_method = avctx->me_method;
        
400
    /* Fixed QSCALE */
F
Fabrice Bellard 已提交
401
    s->fixed_qscale = (avctx->flags & CODEC_FLAG_QSCALE);
402
    
F
Fabrice Bellard 已提交
403 404 405
    switch(avctx->codec->id) {
    case CODEC_ID_MPEG1VIDEO:
        s->out_format = FMT_MPEG1;
M
Michael Niedermayer 已提交
406
        avctx->delay=0; //FIXME not sure, should check the spec
F
Fabrice Bellard 已提交
407 408 409 410
        break;
    case CODEC_ID_MJPEG:
        s->out_format = FMT_MJPEG;
        s->intra_only = 1; /* force intra only for jpeg */
411
        s->mjpeg_write_tables = 1; /* write all tables */
412
	s->mjpeg_data_only_frames = 0; /* write all the needed headers */
413 414 415
        s->mjpeg_vsample[0] = 2; /* set up default sampling factors */
        s->mjpeg_vsample[1] = 1; /* the only currently supported values */
        s->mjpeg_vsample[2] = 1; 
416
        s->mjpeg_hsample[0] = 2;
417 418
        s->mjpeg_hsample[1] = 1; 
        s->mjpeg_hsample[2] = 1; 
F
Fabrice Bellard 已提交
419 420
        if (mjpeg_init(s) < 0)
            return -1;
M
Michael Niedermayer 已提交
421
        avctx->delay=0;
F
Fabrice Bellard 已提交
422 423
        break;
    case CODEC_ID_H263:
424 425
        if (h263_get_picture_format(s->width, s->height) == 7) {
            printf("Input picture size isn't suitable for h263 codec! try h263+\n");
F
Fabrice Bellard 已提交
426
            return -1;
427
        }
F
Fabrice Bellard 已提交
428
        s->out_format = FMT_H263;
M
Michael Niedermayer 已提交
429
        avctx->delay=0;
F
Fabrice Bellard 已提交
430 431 432
        break;
    case CODEC_ID_H263P:
        s->out_format = FMT_H263;
433 434
        s->rtp_mode = 1;
        s->rtp_payload_size = 1200; 
F
Fabrice Bellard 已提交
435
        s->h263_plus = 1;
436
        s->unrestricted_mv = 1;
437
        s->h263_aic = 1;
438 439 440 441
        
        /* These are just to be sure */
        s->umvplus = 0;
        s->umvplus_dec = 0;
M
Michael Niedermayer 已提交
442
        avctx->delay=0;
F
Fabrice Bellard 已提交
443 444 445 446
        break;
    case CODEC_ID_RV10:
        s->out_format = FMT_H263;
        s->h263_rv10 = 1;
M
Michael Niedermayer 已提交
447
        avctx->delay=0;
F
Fabrice Bellard 已提交
448
        break;
F
Fabrice Bellard 已提交
449
    case CODEC_ID_MPEG4:
F
Fabrice Bellard 已提交
450 451 452
        s->out_format = FMT_H263;
        s->h263_pred = 1;
        s->unrestricted_mv = 1;
453
        s->has_b_frames= s->max_b_frames ? 1 : 0;
M
Michael Niedermayer 已提交
454 455
        s->low_delay=0;
        avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1); 
F
Fabrice Bellard 已提交
456
        break;
457
    case CODEC_ID_MSMPEG4V1:
F
Fabrice Bellard 已提交
458 459 460 461
        s->out_format = FMT_H263;
        s->h263_msmpeg4 = 1;
        s->h263_pred = 1;
        s->unrestricted_mv = 1;
462
        s->msmpeg4_version= 1;
M
Michael Niedermayer 已提交
463
        avctx->delay=0;
464 465 466 467 468 469 470
        break;
    case CODEC_ID_MSMPEG4V2:
        s->out_format = FMT_H263;
        s->h263_msmpeg4 = 1;
        s->h263_pred = 1;
        s->unrestricted_mv = 1;
        s->msmpeg4_version= 2;
M
Michael Niedermayer 已提交
471
        avctx->delay=0;
472 473 474 475 476 477 478
        break;
    case CODEC_ID_MSMPEG4V3:
        s->out_format = FMT_H263;
        s->h263_msmpeg4 = 1;
        s->h263_pred = 1;
        s->unrestricted_mv = 1;
        s->msmpeg4_version= 3;
M
Michael Niedermayer 已提交
479
        avctx->delay=0;
F
Fabrice Bellard 已提交
480
        break;
M
Michael Niedermayer 已提交
481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496
    case CODEC_ID_WMV1:
        s->out_format = FMT_H263;
        s->h263_msmpeg4 = 1;
        s->h263_pred = 1;
        s->unrestricted_mv = 1;
        s->msmpeg4_version= 4;
        avctx->delay=0;
        break;
    case CODEC_ID_WMV2:
        s->out_format = FMT_H263;
        s->h263_msmpeg4 = 1;
        s->h263_pred = 1;
        s->unrestricted_mv = 1;
        s->msmpeg4_version= 5;
        avctx->delay=0;
        break;
F
Fabrice Bellard 已提交
497 498 499
    default:
        return -1;
    }
M
Michael Niedermayer 已提交
500
    
501 502 503 504 505
    { /* set up some save defaults, some codecs might override them later */
        static int done=0;
        if(!done){
            int i;
            done=1;
506
            memset(default_mv_penalty, 0, sizeof(UINT16)*(MAX_FCODE+1)*(2*MAX_MV+1));
507 508 509 510 511 512 513 514 515
            memset(default_fcode_tab , 0, sizeof(UINT8)*(2*MAX_MV+1));

            for(i=-16; i<16; i++){
                default_fcode_tab[i + MAX_MV]= 1;
            }
        }
    }
    s->mv_penalty= default_mv_penalty;
    s->fcode_tab= default_fcode_tab;
516 517 518
    s->y_dc_scale_table=
    s->c_dc_scale_table= ff_mpeg1_dc_scale_table;
 
F
Fabrice Bellard 已提交
519
    if (s->out_format == FMT_H263)
520
        h263_encode_init(s);
521
    else if (s->out_format == FMT_MPEG1)
M
Michael Niedermayer 已提交
522 523 524
        ff_mpeg1_encode_init(s);
    if(s->msmpeg4_version)
        ff_msmpeg4_encode_init(s);
F
Fabrice Bellard 已提交
525

526
    /* dont use mv_penalty table for crap MV as it would be confused */
527
    if (s->me_method < ME_EPZS) s->mv_penalty = default_mv_penalty;
528

529 530
    s->encoding = 1;

F
Fabrice Bellard 已提交
531 532 533 534
    /* init */
    if (MPV_common_init(s) < 0)
        return -1;
    
535 536
    /* init default q matrix */
    for(i=0;i<64;i++) {
537
        if(s->out_format == FMT_H263)
538
            s->intra_matrix[i] = ff_mpeg1_default_non_intra_matrix[i];
539
        else
540
            s->intra_matrix[i] = ff_mpeg1_default_intra_matrix[i];
541

542
        s->inter_matrix[i] = ff_mpeg1_default_non_intra_matrix[i];
543 544 545
    }

    /* precompute matrix */
546
    /* for mjpeg, we do include qscale in the matrix */
547 548 549 550 551
    if (s->out_format != FMT_MJPEG) {
        convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->q_intra_matrix16_bias, 
                       s->intra_matrix, s->intra_quant_bias);
        convert_matrix(s->q_inter_matrix, s->q_inter_matrix16, s->q_inter_matrix16_bias, 
                       s->inter_matrix, s->inter_quant_bias);
552 553
    }

554 555
    if(ff_rate_control_init(s) < 0)
        return -1;
F
Fabrice Bellard 已提交
556 557

    s->picture_number = 0;
558
    s->picture_in_gop_number = 0;
F
Fabrice Bellard 已提交
559 560 561
    s->fake_picture_number = 0;
    /* motion detector init */
    s->f_code = 1;
562
    s->b_code = 1;
F
Fabrice Bellard 已提交
563 564 565 566 567 568 569 570 571 572 573

    return 0;
}

int MPV_encode_end(AVCodecContext *avctx)
{
    MpegEncContext *s = avctx->priv_data;

#ifdef STATS
    print_stats();
#endif
574 575 576

    ff_rate_control_uninit(s);

F
Fabrice Bellard 已提交
577 578 579
    MPV_common_end(s);
    if (s->out_format == FMT_MJPEG)
        mjpeg_close(s);
580
      
F
Fabrice Bellard 已提交
581 582 583 584
    return 0;
}

/* draw the edges of width 'w' of an image of size width, height */
M
Michael Niedermayer 已提交
585
static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w)
F
Fabrice Bellard 已提交
586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612
{
    UINT8 *ptr, *last_line;
    int i;

    last_line = buf + (height - 1) * wrap;
    for(i=0;i<w;i++) {
        /* top and bottom */
        memcpy(buf - (i + 1) * wrap, buf, width);
        memcpy(last_line + (i + 1) * wrap, last_line, width);
    }
    /* left and right */
    ptr = buf;
    for(i=0;i<height;i++) {
        memset(ptr - w, ptr[0], w);
        memset(ptr + width, ptr[width-1], w);
        ptr += wrap;
    }
    /* corners */
    for(i=0;i<w;i++) {
        memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */
        memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */
        memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */
        memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */
    }
}

/* generic function for encode/decode called before a frame is coded/decoded */
613
void MPV_frame_start(MpegEncContext *s, AVCodecContext *avctx)
F
Fabrice Bellard 已提交
614 615 616 617
{
    int i;
    UINT8 *tmp;

618
    s->mb_skiped = 0;
619 620
    s->decoding_error=0;

621 622 623 624 625 626 627
    if(avctx->flags&CODEC_FLAG_DR1){
        int i;
        avctx->get_buffer_callback(avctx, s->width, s->height, s->pict_type);

        s->linesize= avctx->dr_stride;
    }
    
F
Fabrice Bellard 已提交
628 629
    if (s->pict_type == B_TYPE) {
        for(i=0;i<3;i++) {
630 631 632
            if(avctx->flags&CODEC_FLAG_DR1)
                s->aux_picture[i]= avctx->dr_buffer[i];

F
Fabrice Bellard 已提交
633 634 635 636 637
            s->current_picture[i] = s->aux_picture[i];
        }
    } else {
        for(i=0;i<3;i++) {
            /* swap next and last */
638 639 640 641 642
            if(avctx->flags&CODEC_FLAG_DR1)
                tmp= avctx->dr_buffer[i];
            else
                tmp = s->last_picture[i];

F
Fabrice Bellard 已提交
643 644 645
            s->last_picture[i] = s->next_picture[i];
            s->next_picture[i] = tmp;
            s->current_picture[i] = tmp;
646 647 648 649 650 651 652 653

            s->last_dr_opaque= s->next_dr_opaque;
            s->next_dr_opaque= avctx->dr_opaque_frame;

            if(s->has_b_frames && s->last_dr_opaque)
                avctx->dr_opaque_frame= s->last_dr_opaque;
            else
                avctx->dr_opaque_frame= s->next_dr_opaque;
F
Fabrice Bellard 已提交
654 655 656
        }
    }
}
657

F
Fabrice Bellard 已提交
658 659 660
/* generic function for encode/decode called after a frame has been coded/decoded */
void MPV_frame_end(MpegEncContext *s)
{
661 662
//    if((s->picture_number%100)==0 && s->encoding) printf("sads:%d //\n", sads);

F
Fabrice Bellard 已提交
663
    /* draw edge for correct motion prediction if outside */
664
    if (s->pict_type != B_TYPE && !s->intra_only && !(s->flags&CODEC_FLAG_EMU_EDGE)) {
M
Michael Niedermayer 已提交
665
      if(s->avctx==NULL || s->avctx->codec->id!=CODEC_ID_MPEG4 || s->divx_version>=500){
666 667 668
        draw_edges(s->current_picture[0], s->linesize, s->mb_width*16, s->mb_height*16, EDGE_WIDTH);
        draw_edges(s->current_picture[1], s->linesize/2, s->mb_width*8, s->mb_height*8, EDGE_WIDTH/2);
        draw_edges(s->current_picture[2], s->linesize/2, s->mb_width*8, s->mb_height*8, EDGE_WIDTH/2);
669
      }else{
M
Michael Niedermayer 已提交
670
        /* mpeg4? / opendivx / xvid */
F
Fabrice Bellard 已提交
671 672 673
        draw_edges(s->current_picture[0], s->linesize, s->width, s->height, EDGE_WIDTH);
        draw_edges(s->current_picture[1], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2);
        draw_edges(s->current_picture[2], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2);
674
      }
F
Fabrice Bellard 已提交
675
    }
M
Michael Niedermayer 已提交
676
    emms_c();
677 678 679 680
    
    if(s->pict_type!=B_TYPE){
        s->last_non_b_pict_type= s->pict_type;
        s->last_non_b_qscale= s->qscale;
681
        s->last_non_b_mc_mb_var= s->mc_mb_var_sum;
M
Michael Niedermayer 已提交
682 683
        s->num_available_buffers++;
        if(s->num_available_buffers>2) s->num_available_buffers= 2;
684
    }
F
Fabrice Bellard 已提交
685 686
}

687 688 689 690 691 692 693 694 695 696 697 698 699
/* reorder input for encoding */
void reorder_input(MpegEncContext *s, AVPicture *pict)
{
    int i, j, index;
            
    if(s->max_b_frames > FF_MAX_B_FRAMES) s->max_b_frames= FF_MAX_B_FRAMES;

//        delay= s->max_b_frames+1; (or 0 if no b frames cuz decoder diff)

    for(j=0; j<REORDER_BUFFER_SIZE-1; j++){
        s->coded_order[j]= s->coded_order[j+1];
    }
    s->coded_order[j].picture[0]= s->coded_order[j].picture[1]= s->coded_order[j].picture[2]= NULL; //catch uninitalized buffers
M
Michael Niedermayer 已提交
700
    s->coded_order[j].pict_type=0;
701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765

    switch(s->input_pict_type){
    default: 
    case I_TYPE:
    case S_TYPE:
    case P_TYPE:
        index= s->max_b_frames - s->b_frames_since_non_b;
        s->b_frames_since_non_b=0;
        break;            
    case B_TYPE:
        index= s->max_b_frames + 1;
        s->b_frames_since_non_b++;
        break;          
    }
//printf("index:%d type:%d strides: %d %d\n", index, s->input_pict_type, pict->linesize[0], s->linesize);
    if(   (index==0 || (s->flags&CODEC_FLAG_INPUT_PRESERVED))
       && pict->linesize[0] == s->linesize
       && pict->linesize[1] == s->linesize>>1
       && pict->linesize[2] == s->linesize>>1){
//printf("ptr\n");
        for(i=0; i<3; i++){
            s->coded_order[index].picture[i]= pict->data[i];
        }
    }else{
//printf("copy\n");
        for(i=0; i<3; i++){
            uint8_t *src = pict->data[i];
            uint8_t *dest;
            int src_wrap = pict->linesize[i];
            int dest_wrap = s->linesize;
            int w = s->width;
            int h = s->height;

            if(index==0) dest= s->last_picture[i]+16; //is current_picture indeed but the switch hapens after reordering
            else         dest= s->picture_buffer[s->picture_buffer_index][i];

            if (i >= 1) {
                dest_wrap >>= 1;
                w >>= 1;
                h >>= 1;
            }

            s->coded_order[index].picture[i]= dest;
            for(j=0;j<h;j++) {
                memcpy(dest, src, w);
                dest += dest_wrap;
                src += src_wrap;
            }
        }
        if(index!=0){
            s->picture_buffer_index++;
            if(s->picture_buffer_index >= REORDER_BUFFER_SIZE-1) s->picture_buffer_index=0;
        }
    }
    s->coded_order[index].pict_type = s->input_pict_type;
    s->coded_order[index].qscale    = s->input_qscale;
    s->coded_order[index].force_type= s->force_input_type;
    s->coded_order[index].picture_in_gop_number= s->input_picture_in_gop_number;
    s->coded_order[index].picture_number= s->input_picture_number;

    for(i=0; i<3; i++){
        s->new_picture[i]= s->coded_order[0].picture[i];
    }
}

F
Fabrice Bellard 已提交
766 767 768 769 770 771
int MPV_encode_picture(AVCodecContext *avctx,
                       unsigned char *buf, int buf_size, void *data)
{
    MpegEncContext *s = avctx->priv_data;
    AVPicture *pict = data;

772
    s->input_qscale = avctx->quality;
F
Fabrice Bellard 已提交
773 774 775

    init_put_bits(&s->pb, buf, buf_size, NULL, NULL);

776 777 778 779 780 781 782 783 784 785 786 787 788
    if(avctx->flags&CODEC_FLAG_TYPE){
        s->input_pict_type=
        s->force_input_type= avctx->key_frame ? I_TYPE : P_TYPE;
    }else if(s->flags&CODEC_FLAG_PASS2){
        s->input_pict_type=
        s->force_input_type= s->rc_context.entry[s->input_picture_number].new_pict_type;
    }else{
        s->force_input_type=0;
        if (!s->intra_only) {
            /* first picture of GOP is intra */
            if (s->input_picture_in_gop_number % s->gop_size==0){
                s->input_pict_type = I_TYPE;
            }else if(s->max_b_frames==0){
789
                s->input_pict_type = P_TYPE;
790 791 792 793 794 795 796 797
            }else{
                if(s->b_frames_since_non_b < s->max_b_frames) //FIXME more IQ
                    s->input_pict_type = B_TYPE;
                else
                    s->input_pict_type = P_TYPE;
            }
        } else {
            s->input_pict_type = I_TYPE;
798
        }
F
Fabrice Bellard 已提交
799 800
    }

801 802 803
    if(s->input_pict_type==I_TYPE)
        s->input_picture_in_gop_number=0;
    
804 805 806 807 808 809 810 811 812 813 814 815
    reorder_input(s, pict);
    
    /* output? */
    if(s->coded_order[0].picture[0]){

        s->pict_type= s->coded_order[0].pict_type;
        if (s->fixed_qscale) /* the ratecontrol needs the last qscale so we dont touch it for CBR */
            s->qscale= s->coded_order[0].qscale;
        s->force_type= s->coded_order[0].force_type;
        s->picture_in_gop_number= s->coded_order[0].picture_in_gop_number;
        s->picture_number= s->coded_order[0].picture_number;

816
        MPV_frame_start(s, avctx);
817 818

        encode_picture(s, s->picture_number);
J
Juanjo 已提交
819 820 821
        avctx->key_frame   = (s->pict_type == I_TYPE);
        avctx->pict_type   = s->pict_type;
        avctx->real_pict_num  = s->picture_number;
822 823 824 825 826 827 828 829 830 831 832 833 834 835 836
        avctx->header_bits = s->header_bits;
        avctx->mv_bits     = s->mv_bits;
        avctx->misc_bits   = s->misc_bits;
        avctx->i_tex_bits  = s->i_tex_bits;
        avctx->p_tex_bits  = s->p_tex_bits;
        avctx->i_count     = s->i_count;
        avctx->p_count     = s->p_count;
        avctx->skip_count  = s->skip_count;

        MPV_frame_end(s);

        if (s->out_format == FMT_MJPEG)
            mjpeg_picture_trailer(s);

        avctx->quality = s->qscale;
837 838 839
        
        if(s->flags&CODEC_FLAG_PASS1)
            ff_write_pass1_stats(s);
840
    
F
Fabrice Bellard 已提交
841 842
    }

843 844
    s->input_picture_number++;
    s->input_picture_in_gop_number++;
F
Fabrice Bellard 已提交
845 846

    flush_put_bits(&s->pb);
847
    s->frame_bits  = (pbBufPtr(&s->pb) - s->pb.buf) * 8;
848 849 850
    if(s->pict_type==B_TYPE) s->pb_frame_bits+= s->frame_bits;
    else                     s->pb_frame_bits= s->frame_bits;

851
    s->total_bits += s->frame_bits;
852 853 854
    avctx->frame_bits  = s->frame_bits;
//printf("fcode: %d, type: %d, head: %d, mv: %d, misc: %d, frame: %d, itex: %d, ptex: %d\n", 
//s->f_code, avctx->key_frame, s->header_bits, s->mv_bits, s->misc_bits, s->frame_bits, s->i_tex_bits, s->p_tex_bits);
855

856 857 858 859 860
    if (avctx->get_psnr) {
        /* At this point pict->data should have the original frame   */
        /* an s->current_picture should have the coded/decoded frame */
        get_psnr(pict->data, s->current_picture,
                 pict->linesize, s->linesize, avctx);
861
//        printf("%f\n", avctx->psnr_y);
862
    }
863
    return pbBufPtr(&s->pb) - s->pb.buf;
F
Fabrice Bellard 已提交
864 865
}

M
Michael Niedermayer 已提交
866 867 868 869 870 871 872
static inline void gmc1_motion(MpegEncContext *s,
                               UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
                               int dest_offset,
                               UINT8 **ref_picture, int src_offset,
                               int h)
{
    UINT8 *ptr;
873
    int offset, src_x, src_y, linesize;
M
Michael Niedermayer 已提交
874 875
    int motion_x, motion_y;

876
    if(s->real_sprite_warping_points>1) printf("more than 1 warp point isnt supported\n");
M
Michael Niedermayer 已提交
877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918
    motion_x= s->sprite_offset[0][0];
    motion_y= s->sprite_offset[0][1];
    src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1));
    src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1));
    motion_x<<=(3-s->sprite_warping_accuracy);
    motion_y<<=(3-s->sprite_warping_accuracy);
    src_x = clip(src_x, -16, s->width);
    if (src_x == s->width)
        motion_x =0;
    src_y = clip(src_y, -16, s->height);
    if (src_y == s->height)
        motion_y =0;
    
    linesize = s->linesize;
    ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset;

    dest_y+=dest_offset;
    gmc1(dest_y  , ptr  , linesize, h, motion_x&15, motion_y&15, s->no_rounding);
    gmc1(dest_y+8, ptr+8, linesize, h, motion_x&15, motion_y&15, s->no_rounding);

    motion_x= s->sprite_offset[1][0];
    motion_y= s->sprite_offset[1][1];
    src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1));
    src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1));
    motion_x<<=(3-s->sprite_warping_accuracy);
    motion_y<<=(3-s->sprite_warping_accuracy);
    src_x = clip(src_x, -8, s->width>>1);
    if (src_x == s->width>>1)
        motion_x =0;
    src_y = clip(src_y, -8, s->height>>1);
    if (src_y == s->height>>1)
        motion_y =0;

    offset = (src_y * linesize>>1) + src_x + (src_offset>>1);
    ptr = ref_picture[1] + offset;
    gmc1(dest_cb + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding);
    ptr = ref_picture[2] + offset;
    gmc1(dest_cr + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding);
    
    return;
}

919 920 921 922 923 924 925 926
static void emulated_edge_mc(UINT8 *buf, UINT8 *src, int linesize, int block_w, int block_h, 
                                    int src_x, int src_y, int w, int h){
    int x, y;
    int start_y, start_x, end_y, end_x;
    
    if(src_y>= h){
        src+= (h-1-src_y)*linesize;
        src_y=h-1;
M
Michael Niedermayer 已提交
927 928 929
    }else if(src_y<=-block_h){
        src+= (1-block_h-src_y)*linesize;
        src_y=1-block_h;
930 931 932 933
    }
    if(src_x>= w){
        src+= (w-1-src_x);
        src_x=w-1;
M
Michael Niedermayer 已提交
934 935 936
    }else if(src_x<=-block_w){
        src+= (1-block_w-src_x);
        src_x=1-block_w;
937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978
    }

    start_y= MAX(0, -src_y);
    start_x= MAX(0, -src_x);
    end_y= MIN(block_h, h-src_y);
    end_x= MIN(block_w, w-src_x);
    
    // copy existing part
    for(y=start_y; y<end_y; y++){
        for(x=start_x; x<end_x; x++){
            buf[x + y*linesize]= src[x + y*linesize];
        }
    }

    //top
    for(y=0; y<start_y; y++){
        for(x=start_x; x<end_x; x++){
            buf[x + y*linesize]= buf[x + start_y*linesize];
        }
    }

    //bottom
    for(y=end_y; y<block_h; y++){
        for(x=start_x; x<end_x; x++){
            buf[x + y*linesize]= buf[x + (end_y-1)*linesize];
        }
    }
                                    
    for(y=0; y<block_h; y++){
       //left
        for(x=0; x<start_x; x++){
            buf[x + y*linesize]= buf[start_x + y*linesize];
        }
       
       //right
        for(x=end_x; x<block_w; x++){
            buf[x + y*linesize]= buf[end_x - 1 + y*linesize];
        }
    }
}


F
Fabrice Bellard 已提交
979 980 981 982 983 984 985 986 987 988
/* apply one mpeg motion vector to the three components */
static inline void mpeg_motion(MpegEncContext *s,
                               UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
                               int dest_offset,
                               UINT8 **ref_picture, int src_offset,
                               int field_based, op_pixels_func *pix_op,
                               int motion_x, int motion_y, int h)
{
    UINT8 *ptr;
    int dxy, offset, mx, my, src_x, src_y, height, linesize;
989 990
    int emu=0;
    
M
Michael Niedermayer 已提交
991 992 993 994 995
if(s->quarter_sample)
{
    motion_x>>=1;
    motion_y>>=1;
}
F
Fabrice Bellard 已提交
996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010
    dxy = ((motion_y & 1) << 1) | (motion_x & 1);
    src_x = s->mb_x * 16 + (motion_x >> 1);
    src_y = s->mb_y * (16 >> field_based) + (motion_y >> 1);
                
    /* WARNING: do no forget half pels */
    height = s->height >> field_based;
    src_x = clip(src_x, -16, s->width);
    if (src_x == s->width)
        dxy &= ~1;
    src_y = clip(src_y, -16, height);
    if (src_y == height)
        dxy &= ~2;
    linesize = s->linesize << field_based;
    ptr = ref_picture[0] + (src_y * linesize) + (src_x) + src_offset;
    dest_y += dest_offset;
1011 1012 1013 1014 1015 1016 1017 1018 1019

    if(s->flags&CODEC_FLAG_EMU_EDGE){
        if(src_x<0 || src_y<0 || src_x + (motion_x&1) + 16 > s->width
                              || src_y + (motion_y&1) + h  > height){
            emulated_edge_mc(s->edge_emu_buffer, ptr, linesize, 17, h+1, src_x, src_y, s->width, height);
            ptr= s->edge_emu_buffer;
            emu=1;
        }
    }
F
Fabrice Bellard 已提交
1020 1021 1022
    pix_op[dxy](dest_y, ptr, linesize, h);
    pix_op[dxy](dest_y + 8, ptr + 8, linesize, h);

M
Michael Niedermayer 已提交
1023 1024
    if(s->flags&CODEC_FLAG_GRAY) return;

F
Fabrice Bellard 已提交
1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051
    if (s->out_format == FMT_H263) {
        dxy = 0;
        if ((motion_x & 3) != 0)
            dxy |= 1;
        if ((motion_y & 3) != 0)
            dxy |= 2;
        mx = motion_x >> 2;
        my = motion_y >> 2;
    } else {
        mx = motion_x / 2;
        my = motion_y / 2;
        dxy = ((my & 1) << 1) | (mx & 1);
        mx >>= 1;
        my >>= 1;
    }
    
    src_x = s->mb_x * 8 + mx;
    src_y = s->mb_y * (8 >> field_based) + my;
    src_x = clip(src_x, -8, s->width >> 1);
    if (src_x == (s->width >> 1))
        dxy &= ~1;
    src_y = clip(src_y, -8, height >> 1);
    if (src_y == (height >> 1))
        dxy &= ~2;

    offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1);
    ptr = ref_picture[1] + offset;
1052 1053 1054 1055
    if(emu){
        emulated_edge_mc(s->edge_emu_buffer, ptr, linesize>>1, 9, (h>>1)+1, src_x, src_y, s->width>>1, height>>1);
        ptr= s->edge_emu_buffer;
    }
F
Fabrice Bellard 已提交
1056
    pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1);
1057

F
Fabrice Bellard 已提交
1058
    ptr = ref_picture[2] + offset;
1059 1060 1061 1062
    if(emu){
        emulated_edge_mc(s->edge_emu_buffer, ptr, linesize>>1, 9, (h>>1)+1, src_x, src_y, s->width>>1, height>>1);
        ptr= s->edge_emu_buffer;
    }
F
Fabrice Bellard 已提交
1063 1064 1065
    pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1);
}

M
Michael Niedermayer 已提交
1066 1067 1068 1069 1070 1071 1072 1073 1074 1075
static inline void qpel_motion(MpegEncContext *s,
                               UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
                               int dest_offset,
                               UINT8 **ref_picture, int src_offset,
                               int field_based, op_pixels_func *pix_op,
                               qpel_mc_func *qpix_op,
                               int motion_x, int motion_y, int h)
{
    UINT8 *ptr;
    int dxy, offset, mx, my, src_x, src_y, height, linesize;
M
Michael Niedermayer 已提交
1076
    int emu=0;
M
Michael Niedermayer 已提交
1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092

    dxy = ((motion_y & 3) << 2) | (motion_x & 3);
    src_x = s->mb_x * 16 + (motion_x >> 2);
    src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);

    height = s->height >> field_based;
    src_x = clip(src_x, -16, s->width);
    if (src_x == s->width)
        dxy &= ~3;
    src_y = clip(src_y, -16, height);
    if (src_y == height)
        dxy &= ~12;
    linesize = s->linesize << field_based;
    ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset;
    dest_y += dest_offset;
//printf("%d %d %d\n", src_x, src_y, dxy);
M
Michael Niedermayer 已提交
1093 1094 1095 1096 1097 1098 1099 1100 1101
    
    if(s->flags&CODEC_FLAG_EMU_EDGE){
        if(src_x<0 || src_y<0 || src_x + (motion_x&3) + 16 > s->width
                              || src_y + (motion_y&3) + h  > height){
            emulated_edge_mc(s->edge_emu_buffer, ptr, linesize, 17, h+1, src_x, src_y, s->width, height);
            ptr= s->edge_emu_buffer;
            emu=1;
        }
    }
M
Michael Niedermayer 已提交
1102 1103 1104 1105 1106
    qpix_op[dxy](dest_y                 , ptr                 , linesize, linesize, motion_x&3, motion_y&3);
    qpix_op[dxy](dest_y              + 8, ptr              + 8, linesize, linesize, motion_x&3, motion_y&3);
    qpix_op[dxy](dest_y + linesize*8    , ptr + linesize*8    , linesize, linesize, motion_x&3, motion_y&3);
    qpix_op[dxy](dest_y + linesize*8 + 8, ptr + linesize*8 + 8, linesize, linesize, motion_x&3, motion_y&3);
    
M
Michael Niedermayer 已提交
1107 1108
    if(s->flags&CODEC_FLAG_GRAY) return;

M
Michael Niedermayer 已提交
1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130
    mx= (motion_x>>1) | (motion_x&1);
    my= (motion_y>>1) | (motion_y&1);

    dxy = 0;
    if ((mx & 3) != 0)
        dxy |= 1;
    if ((my & 3) != 0)
        dxy |= 2;
    mx = mx >> 2;
    my = my >> 2;
    
    src_x = s->mb_x * 8 + mx;
    src_y = s->mb_y * (8 >> field_based) + my;
    src_x = clip(src_x, -8, s->width >> 1);
    if (src_x == (s->width >> 1))
        dxy &= ~1;
    src_y = clip(src_y, -8, height >> 1);
    if (src_y == (height >> 1))
        dxy &= ~2;

    offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1);
    ptr = ref_picture[1] + offset;
M
Michael Niedermayer 已提交
1131 1132 1133 1134
    if(emu){
        emulated_edge_mc(s->edge_emu_buffer, ptr, linesize>>1, 9, (h>>1)+1, src_x, src_y, s->width>>1, height>>1);
        ptr= s->edge_emu_buffer;
    }
M
Michael Niedermayer 已提交
1135
    pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1);
M
Michael Niedermayer 已提交
1136
    
M
Michael Niedermayer 已提交
1137
    ptr = ref_picture[2] + offset;
M
Michael Niedermayer 已提交
1138 1139 1140 1141
    if(emu){
        emulated_edge_mc(s->edge_emu_buffer, ptr, linesize>>1, 9, (h>>1)+1, src_x, src_y, s->width>>1, height>>1);
        ptr= s->edge_emu_buffer;
    }
M
Michael Niedermayer 已提交
1142 1143 1144 1145
    pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1);
}


F
Fabrice Bellard 已提交
1146 1147 1148
static inline void MPV_motion(MpegEncContext *s, 
                              UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
                              int dir, UINT8 **ref_picture, 
M
Michael Niedermayer 已提交
1149
                              op_pixels_func *pix_op, qpel_mc_func *qpix_op)
F
Fabrice Bellard 已提交
1150 1151 1152 1153
{
    int dxy, offset, mx, my, src_x, src_y, motion_x, motion_y;
    int mb_x, mb_y, i;
    UINT8 *ptr, *dest;
M
Michael Niedermayer 已提交
1154
    int emu=0;
F
Fabrice Bellard 已提交
1155 1156 1157 1158 1159 1160

    mb_x = s->mb_x;
    mb_y = s->mb_y;

    switch(s->mv_type) {
    case MV_TYPE_16X16:
M
Michael Niedermayer 已提交
1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173
        if(s->mcsel){
#if 0
            mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
                        ref_picture, 0,
                        0, pix_op,
                        s->sprite_offset[0][0]>>3,
                        s->sprite_offset[0][1]>>3,
                        16);
#else
            gmc1_motion(s, dest_y, dest_cb, dest_cr, 0,
                        ref_picture, 0,
                        16);
#endif
M
Michael Niedermayer 已提交
1174
        }else if(s->quarter_sample && dir==0){ //FIXME
M
Michael Niedermayer 已提交
1175 1176 1177 1178 1179 1180 1181 1182 1183 1184
            qpel_motion(s, dest_y, dest_cb, dest_cr, 0,
                        ref_picture, 0,
                        0, pix_op, qpix_op,
                        s->mv[dir][0][0], s->mv[dir][0][1], 16);
        }else{
            mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
                        ref_picture, 0,
                        0, pix_op,
                        s->mv[dir][0][0], s->mv[dir][0][1], 16);
        }           
F
Fabrice Bellard 已提交
1185 1186 1187 1188 1189 1190 1191 1192
        break;
    case MV_TYPE_8X8:
        for(i=0;i<4;i++) {
            motion_x = s->mv[dir][i][0];
            motion_y = s->mv[dir][i][1];

            dxy = ((motion_y & 1) << 1) | (motion_x & 1);
            src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8;
M
Michael Niedermayer 已提交
1193
            src_y = mb_y * 16 + (motion_y >> 1) + (i >>1) * 8;
F
Fabrice Bellard 已提交
1194 1195 1196 1197 1198 1199 1200 1201 1202 1203
                    
            /* WARNING: do no forget half pels */
            src_x = clip(src_x, -16, s->width);
            if (src_x == s->width)
                dxy &= ~1;
            src_y = clip(src_y, -16, s->height);
            if (src_y == s->height)
                dxy &= ~2;
                    
            ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
M
Michael Niedermayer 已提交
1204 1205 1206 1207 1208 1209 1210
            if(s->flags&CODEC_FLAG_EMU_EDGE){
                if(src_x<0 || src_y<0 || src_x + (motion_x&1) + 8 > s->width
                                      || src_y + (motion_y&1) + 8 > s->height){
                    emulated_edge_mc(s->edge_emu_buffer, ptr, s->linesize, 9, 9, src_x, src_y, s->width, s->height);
                    ptr= s->edge_emu_buffer;
                }
            }
F
Fabrice Bellard 已提交
1211 1212 1213
            dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
            pix_op[dxy](dest, ptr, s->linesize, 8);
        }
M
Michael Niedermayer 已提交
1214 1215
    
        if(s->flags&CODEC_FLAG_GRAY) break;
F
Fabrice Bellard 已提交
1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250
        /* In case of 8X8, we construct a single chroma motion vector
           with a special rounding */
        mx = 0;
        my = 0;
        for(i=0;i<4;i++) {
            mx += s->mv[dir][i][0];
            my += s->mv[dir][i][1];
        }
        if (mx >= 0)
            mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1));
        else {
            mx = -mx;
            mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1));
        }
        if (my >= 0)
            my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1));
        else {
            my = -my;
            my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1));
        }
        dxy = ((my & 1) << 1) | (mx & 1);
        mx >>= 1;
        my >>= 1;

        src_x = mb_x * 8 + mx;
        src_y = mb_y * 8 + my;
        src_x = clip(src_x, -8, s->width/2);
        if (src_x == s->width/2)
            dxy &= ~1;
        src_y = clip(src_y, -8, s->height/2);
        if (src_y == s->height/2)
            dxy &= ~2;
        
        offset = (src_y * (s->linesize >> 1)) + src_x;
        ptr = ref_picture[1] + offset;
M
Michael Niedermayer 已提交
1251 1252 1253 1254 1255 1256 1257 1258
        if(s->flags&CODEC_FLAG_EMU_EDGE){
                if(src_x<0 || src_y<0 || src_x + (dxy &1) + 8 > s->width >>1
                                      || src_y + (dxy>>1) + 8 > s->height>>1){
                    emulated_edge_mc(s->edge_emu_buffer, ptr, s->linesize>>1, 9, 9, src_x, src_y, s->width>>1, s->height>>1);
                    ptr= s->edge_emu_buffer;
                    emu=1;
                }
            }
F
Fabrice Bellard 已提交
1259
        pix_op[dxy](dest_cb, ptr, s->linesize >> 1, 8);
M
Michael Niedermayer 已提交
1260

F
Fabrice Bellard 已提交
1261
        ptr = ref_picture[2] + offset;
M
Michael Niedermayer 已提交
1262 1263 1264 1265
        if(emu){
            emulated_edge_mc(s->edge_emu_buffer, ptr, s->linesize>>1, 9, 9, src_x, src_y, s->width>>1, s->height>>1);
            ptr= s->edge_emu_buffer;
        }
F
Fabrice Bellard 已提交
1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293
        pix_op[dxy](dest_cr, ptr, s->linesize >> 1, 8);
        break;
    case MV_TYPE_FIELD:
        if (s->picture_structure == PICT_FRAME) {
            /* top field */
            mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
                        ref_picture, s->field_select[dir][0] ? s->linesize : 0,
                        1, pix_op,
                        s->mv[dir][0][0], s->mv[dir][0][1], 8);
            /* bottom field */
            mpeg_motion(s, dest_y, dest_cb, dest_cr, s->linesize,
                        ref_picture, s->field_select[dir][1] ? s->linesize : 0,
                        1, pix_op,
                        s->mv[dir][1][0], s->mv[dir][1][1], 8);
        } else {
            

        }
        break;
    }
}


/* put block[] to dest[] */
static inline void put_dct(MpegEncContext *s, 
                           DCTELEM *block, int i, UINT8 *dest, int line_size)
{
    if (!s->mpeg2)
1294
        s->dct_unquantize(s, block, i, s->qscale);
F
Fabrice Bellard 已提交
1295
    ff_idct_put (dest, line_size, block);
F
Fabrice Bellard 已提交
1296 1297 1298 1299 1300 1301
}

/* add block[] to dest[] */
static inline void add_dct(MpegEncContext *s, 
                           DCTELEM *block, int i, UINT8 *dest, int line_size)
{
M
Michael Niedermayer 已提交
1302 1303 1304 1305
    if (s->block_last_index[i] >= 0) {
        ff_idct_add (dest, line_size, block);
    }
}
1306

M
Michael Niedermayer 已提交
1307 1308 1309
static inline void add_dequant_dct(MpegEncContext *s, 
                           DCTELEM *block, int i, UINT8 *dest, int line_size)
{
F
Fabrice Bellard 已提交
1310
    if (s->block_last_index[i] >= 0) {
M
Michael Niedermayer 已提交
1311
        s->dct_unquantize(s, block, i, s->qscale);
1312

F
Fabrice Bellard 已提交
1313
        ff_idct_add (dest, line_size, block);
F
Fabrice Bellard 已提交
1314 1315 1316
    }
}

1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349
/**
 * cleans dc, ac, coded_block for the current non intra MB
 */
void ff_clean_intra_table_entries(MpegEncContext *s)
{
    int wrap = s->block_wrap[0];
    int xy = s->block_index[0];
    
    s->dc_val[0][xy           ] = 
    s->dc_val[0][xy + 1       ] = 
    s->dc_val[0][xy     + wrap] =
    s->dc_val[0][xy + 1 + wrap] = 1024;
    /* ac pred */
    memset(s->ac_val[0][xy       ], 0, 32 * sizeof(INT16));
    memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(INT16));
    if (s->msmpeg4_version>=3) {
        s->coded_block[xy           ] =
        s->coded_block[xy + 1       ] =
        s->coded_block[xy     + wrap] =
        s->coded_block[xy + 1 + wrap] = 0;
    }
    /* chroma */
    wrap = s->block_wrap[4];
    xy = s->mb_x + 1 + (s->mb_y + 1) * wrap;
    s->dc_val[1][xy] =
    s->dc_val[2][xy] = 1024;
    /* ac pred */
    memset(s->ac_val[1][xy], 0, 16 * sizeof(INT16));
    memset(s->ac_val[2][xy], 0, 16 * sizeof(INT16));
    
    s->mbintra_table[s->mb_x + s->mb_y*s->mb_width]= 0;
}

F
Fabrice Bellard 已提交
1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361
/* generic function called after a macroblock has been parsed by the
   decoder or after it has been encoded by the encoder.

   Important variables used:
   s->mb_intra : true if intra macroblock
   s->mv_dir   : motion vector direction
   s->mv_type  : motion vector type
   s->mv       : motion vector
   s->interlaced_dct : true if interlaced dct used (mpeg2)
 */
void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
{
1362
    int mb_x, mb_y;
M
Michael Niedermayer 已提交
1363
    const int mb_xy = s->mb_y * s->mb_width + s->mb_x;
F
Fabrice Bellard 已提交
1364 1365 1366 1367

    mb_x = s->mb_x;
    mb_y = s->mb_y;

1368
#ifdef FF_POSTPROCESS
1369
    /* Obsolete. Exists for compatibility with mplayer only. */
1370 1371
    quant_store[mb_y][mb_x]=s->qscale;
    //printf("[%02d][%02d] %d\n",mb_x,mb_y,s->qscale);
1372
#else
1373
    if(s->avctx->quant_store) s->avctx->quant_store[mb_y*s->avctx->qstride+mb_x] = s->qscale;
1374 1375
#endif

F
Fabrice Bellard 已提交
1376 1377
    /* update DC predictors for P macroblocks */
    if (!s->mb_intra) {
1378
        if (s->h263_pred || s->h263_aic) {
M
Michael Niedermayer 已提交
1379
            if(s->mbintra_table[mb_xy])
1380
                ff_clean_intra_table_entries(s);
F
Fabrice Bellard 已提交
1381
        } else {
1382 1383
            s->last_dc[0] =
            s->last_dc[1] =
F
Fabrice Bellard 已提交
1384 1385 1386
            s->last_dc[2] = 128 << s->intra_dc_precision;
        }
    }
1387
    else if (s->h263_pred || s->h263_aic)
M
Michael Niedermayer 已提交
1388
        s->mbintra_table[mb_xy]=1;
1389

1390
    /* update motion predictor, not for B-frames as they need the motion_val from the last P/S-Frame */
M
Michael Niedermayer 已提交
1391 1392
    if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE) { //FIXME move into h263.c if possible, format specific stuff shouldnt be here
        int motion_x, motion_y;
F
Fabrice Bellard 已提交
1393
        
M
Michael Niedermayer 已提交
1394 1395
        const int wrap = s->block_wrap[0];
        const int xy = s->block_index[0];
F
Fabrice Bellard 已提交
1396 1397 1398 1399 1400 1401 1402 1403 1404
        if (s->mb_intra) {
            motion_x = 0;
            motion_y = 0;
            goto motion_init;
        } else if (s->mv_type == MV_TYPE_16X16) {
            motion_x = s->mv[0][0][0];
            motion_y = s->mv[0][0][1];
        motion_init:
            /* no update if 8X8 because it has been done during parsing */
1405 1406 1407 1408 1409 1410 1411 1412
            s->motion_val[xy][0] = motion_x;
            s->motion_val[xy][1] = motion_y;
            s->motion_val[xy + 1][0] = motion_x;
            s->motion_val[xy + 1][1] = motion_y;
            s->motion_val[xy + wrap][0] = motion_x;
            s->motion_val[xy + wrap][1] = motion_y;
            s->motion_val[xy + 1 + wrap][0] = motion_x;
            s->motion_val[xy + 1 + wrap][1] = motion_y;
F
Fabrice Bellard 已提交
1413 1414 1415
        }
    }
    
1416
    if (!(s->encoding && (s->intra_only || s->pict_type==B_TYPE))) {
F
Fabrice Bellard 已提交
1417
        UINT8 *dest_y, *dest_cb, *dest_cr;
M
Michael Niedermayer 已提交
1418 1419 1420
        int dct_linesize, dct_offset;
        op_pixels_func *op_pix;
        qpel_mc_func *op_qpix;
1421

1422 1423 1424
        /* avoid copy if macroblock skipped in last frame too 
           dont touch it for B-frames as they need the skip info from the next p-frame */
        if (s->pict_type != B_TYPE) {
M
Michael Niedermayer 已提交
1425
            UINT8 *mbskip_ptr = &s->mbskip_table[mb_xy];
1426 1427
            if (s->mb_skiped) {
                s->mb_skiped = 0;
1428 1429 1430
                /* if previous was skipped too, then nothing to do ! 
                   skip only during decoding as we might trash the buffers during encoding a bit */
                if (*mbskip_ptr != 0 && !s->encoding) 
1431 1432 1433 1434 1435 1436
                    goto the_end;
                *mbskip_ptr = 1; /* indicate that this time we skiped it */
            } else {
                *mbskip_ptr = 0; /* not skipped */
            }
        }
F
Fabrice Bellard 已提交
1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451

        dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize) + mb_x * 16;
        dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
        dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;

        if (s->interlaced_dct) {
            dct_linesize = s->linesize * 2;
            dct_offset = s->linesize;
        } else {
            dct_linesize = s->linesize;
            dct_offset = s->linesize * 8;
        }

        if (!s->mb_intra) {
            /* motion handling */
1452
            /* decoding or more than one mb_type (MC was allready done otherwise) */
M
Michael Niedermayer 已提交
1453
            if((!s->encoding) || (s->mb_type[mb_xy]&(s->mb_type[mb_xy]-1))){
1454
                if ((!s->no_rounding) || s->pict_type==B_TYPE){                
1455 1456 1457 1458 1459 1460
                    op_pix = put_pixels_tab;
                    op_qpix= qpel_mc_rnd_tab;
                }else{
                    op_pix = put_no_rnd_pixels_tab;
                    op_qpix= qpel_mc_no_rnd_tab;
                }
F
Fabrice Bellard 已提交
1461

1462 1463
                if (s->mv_dir & MV_DIR_FORWARD) {
                    MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix);
1464
                    if ((!s->no_rounding) || s->pict_type==B_TYPE)
1465 1466 1467 1468 1469 1470 1471
                        op_pix = avg_pixels_tab;
                    else
                        op_pix = avg_no_rnd_pixels_tab;
                }
                if (s->mv_dir & MV_DIR_BACKWARD) {
                    MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix);
                }
F
Fabrice Bellard 已提交
1472 1473
            }

M
Michael Niedermayer 已提交
1474 1475 1476
            /* skip dequant / idct if we are really late ;) */
            if(s->hurry_up>1) goto the_end;

F
Fabrice Bellard 已提交
1477
            /* add dct residue */
M
Michael Niedermayer 已提交
1478
            if(s->encoding || !(s->mpeg2 || s->h263_msmpeg4 || s->codec_id==CODEC_ID_MPEG4)){
M
Michael Niedermayer 已提交
1479 1480 1481 1482 1483
                add_dequant_dct(s, block[0], 0, dest_y, dct_linesize);
                add_dequant_dct(s, block[1], 1, dest_y + 8, dct_linesize);
                add_dequant_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
                add_dequant_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);

M
Michael Niedermayer 已提交
1484 1485 1486 1487
                if(!(s->flags&CODEC_FLAG_GRAY)){
                    add_dequant_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
                    add_dequant_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
                }
M
Michael Niedermayer 已提交
1488 1489 1490 1491 1492
            } else {
                add_dct(s, block[0], 0, dest_y, dct_linesize);
                add_dct(s, block[1], 1, dest_y + 8, dct_linesize);
                add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
                add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);
F
Fabrice Bellard 已提交
1493

M
Michael Niedermayer 已提交
1494 1495 1496 1497
                if(!(s->flags&CODEC_FLAG_GRAY)){
                    add_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
                    add_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
                }
M
Michael Niedermayer 已提交
1498
            }
F
Fabrice Bellard 已提交
1499 1500 1501 1502 1503 1504 1505
        } else {
            /* dct only in intra block */
            put_dct(s, block[0], 0, dest_y, dct_linesize);
            put_dct(s, block[1], 1, dest_y + 8, dct_linesize);
            put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
            put_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);

M
Michael Niedermayer 已提交
1506 1507 1508 1509
            if(!(s->flags&CODEC_FLAG_GRAY)){
                put_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
                put_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
            }
F
Fabrice Bellard 已提交
1510 1511
        }
    }
1512
 the_end:
1513
    emms_c(); //FIXME remove
F
Fabrice Bellard 已提交
1514 1515
}

1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559
static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold, int skip_dc)
{
    static const char tab[64]=
        {3,2,2,1,1,1,1,1,
         1,1,1,1,1,1,1,1,
         1,1,1,1,1,1,1,1,
         0,0,0,0,0,0,0,0,
         0,0,0,0,0,0,0,0,
         0,0,0,0,0,0,0,0,
         0,0,0,0,0,0,0,0,
         0,0,0,0,0,0,0,0};
    int score=0;
    int run=0;
    int i;
    DCTELEM *block= s->block[n];
    const int last_index= s->block_last_index[n];

    if(skip_dc) skip_dc=1;
    
    /* are all which we could set to zero are allready zero? */
    if(last_index<=skip_dc - 1) return;

    for(i=0; i<=last_index; i++){
        const int j = zigzag_direct[i];
        const int level = ABS(block[j]);
        if(level==1){
            if(skip_dc && i==0) continue;
            score+= tab[run];
            run=0;
        }else if(level>1){
            return;
        }else{
            run++;
        }
    }
    if(score >= threshold) return;
    for(i=skip_dc; i<=last_index; i++){
        const int j = zigzag_direct[i];
        block[j]=0;
    }
    if(block[0]) s->block_last_index[n]= 0;
    else         s->block_last_index[n]= -1;
}

1560 1561 1562 1563 1564
static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
{
    int i;
    const int maxlevel= s->max_qcoeff;
    const int minlevel= s->min_qcoeff;
1565 1566
        
    for(i=0;i<=last_index; i++){
1567 1568 1569 1570 1571 1572 1573 1574
        const int j = zigzag_direct[i];
        int level = block[j];
       
        if     (level>maxlevel) level=maxlevel;
        else if(level<minlevel) level=minlevel;
        block[j]= level;
    }
}
1575 1576

static void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
F
Fabrice Bellard 已提交
1577
{
1578 1579 1580
    const int mb_x= s->mb_x;
    const int mb_y= s->mb_y;
    int i;
1581
    int skip_dct[6];
1582 1583 1584 1585 1586 1587 1588 1589 1590
#if 0
        if (s->interlaced_dct) {
            dct_linesize = s->linesize * 2;
            dct_offset = s->linesize;
        } else {
            dct_linesize = s->linesize;
            dct_offset = s->linesize * 8;
        }
#endif
1591
    for(i=0; i<6; i++) skip_dct[i]=0;
1592

1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603
    if (s->mb_intra) {
        UINT8 *ptr;
        int wrap;

        wrap = s->linesize;
        ptr = s->new_picture[0] + (mb_y * 16 * wrap) + mb_x * 16;
        get_pixels(s->block[0], ptr               , wrap);
        get_pixels(s->block[1], ptr            + 8, wrap);
        get_pixels(s->block[2], ptr + 8 * wrap    , wrap);
        get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap);

M
Michael Niedermayer 已提交
1604 1605 1606 1607 1608 1609 1610
        if(s->flags&CODEC_FLAG_GRAY){
            skip_dct[4]= 1;
            skip_dct[5]= 1;
        }else{
            wrap >>=1;
            ptr = s->new_picture[1] + (mb_y * 8 * wrap) + mb_x * 8;
            get_pixels(s->block[4], ptr, wrap);
1611

M
Michael Niedermayer 已提交
1612 1613 1614
            ptr = s->new_picture[2] + (mb_y * 8 * wrap) + mb_x * 8;
            get_pixels(s->block[5], ptr, wrap);
        }
1615 1616 1617 1618
    }else{
        op_pixels_func *op_pix;
        qpel_mc_func *op_qpix;
        UINT8 *dest_y, *dest_cb, *dest_cr;
1619 1620
        UINT8 *ptr_y, *ptr_cb, *ptr_cr;
        int wrap_y, wrap_c;
1621 1622 1623 1624

        dest_y  = s->current_picture[0] + (mb_y * 16 * s->linesize       ) + mb_x * 16;
        dest_cb = s->current_picture[1] + (mb_y * 8  * (s->linesize >> 1)) + mb_x * 8;
        dest_cr = s->current_picture[2] + (mb_y * 8  * (s->linesize >> 1)) + mb_x * 8;
1625 1626 1627 1628 1629
        wrap_y = s->linesize;
        wrap_c = wrap_y>>1;
        ptr_y  = s->new_picture[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
        ptr_cb = s->new_picture[1] + (mb_y * 8 * wrap_c) + mb_x * 8;
        ptr_cr = s->new_picture[2] + (mb_y * 8 * wrap_c) + mb_x * 8;
1630

1631
        if ((!s->no_rounding) || s->pict_type==B_TYPE){
1632 1633
            op_pix = put_pixels_tab;
            op_qpix= qpel_mc_rnd_tab;
M
Michael Niedermayer 已提交
1634
        }else{
1635 1636 1637 1638 1639 1640
            op_pix = put_no_rnd_pixels_tab;
            op_qpix= qpel_mc_no_rnd_tab;
        }

        if (s->mv_dir & MV_DIR_FORWARD) {
            MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix);
1641
           if ((!s->no_rounding) || s->pict_type==B_TYPE)
1642 1643 1644 1645 1646 1647
                op_pix = avg_pixels_tab;
            else
                op_pix = avg_no_rnd_pixels_tab;
        }
        if (s->mv_dir & MV_DIR_BACKWARD) {
            MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix);
1648
        }
1649

1650 1651 1652 1653
        diff_pixels(s->block[0], ptr_y                 , dest_y                 , wrap_y);
        diff_pixels(s->block[1], ptr_y              + 8, dest_y              + 8, wrap_y);
        diff_pixels(s->block[2], ptr_y + 8 * wrap_y    , dest_y + 8 * wrap_y    , wrap_y);
        diff_pixels(s->block[3], ptr_y + 8 * wrap_y + 8, dest_y + 8 * wrap_y + 8, wrap_y);
M
Michael Niedermayer 已提交
1654 1655 1656 1657 1658 1659 1660 1661 1662
        
        if(s->flags&CODEC_FLAG_GRAY){
            skip_dct[4]= 1;
            skip_dct[5]= 1;
        }else{
            diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
            diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
        }

1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686
        /* pre quantization */         
        if(s->mc_mb_var[s->mb_width*mb_y+ mb_x]<2*s->qscale*s->qscale){
            if(pix_abs8x8(ptr_y               , dest_y               , wrap_y) < 20*s->qscale) skip_dct[0]= 1;
            if(pix_abs8x8(ptr_y            + 8, dest_y            + 8, wrap_y) < 20*s->qscale) skip_dct[1]= 1;
            if(pix_abs8x8(ptr_y + 8*wrap_y    , dest_y + 8*wrap_y    , wrap_y) < 20*s->qscale) skip_dct[2]= 1;
            if(pix_abs8x8(ptr_y + 8*wrap_y + 8, dest_y + 8*wrap_y + 8, wrap_y) < 20*s->qscale) skip_dct[3]= 1;
            if(pix_abs8x8(ptr_cb              , dest_cb              , wrap_y) < 20*s->qscale) skip_dct[4]= 1;
            if(pix_abs8x8(ptr_cr              , dest_cr              , wrap_y) < 20*s->qscale) skip_dct[5]= 1;
#if 0
{
 static int stat[7];
 int num=0;
 for(i=0; i<6; i++)
  if(skip_dct[i]) num++;
 stat[num]++;
 
 if(s->mb_x==0 && s->mb_y==0){
  for(i=0; i<7; i++){
   printf("%6d %1d\n", stat[i], i);
  }
 }
}
#endif
        }
1687

1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703
    }
            
#if 0
            {
                float adap_parm;
                
                adap_parm = ((s->avg_mb_var << 1) + s->mb_var[s->mb_width*mb_y+mb_x] + 1.0) /
                            ((s->mb_var[s->mb_width*mb_y+mb_x] << 1) + s->avg_mb_var + 1.0);
            
                printf("\ntype=%c qscale=%2d adap=%0.2f dquant=%4.2f var=%4d avgvar=%4d", 
                        (s->mb_type[s->mb_width*mb_y+mb_x] > 0) ? 'I' : 'P', 
                        s->qscale, adap_parm, s->qscale*adap_parm,
                        s->mb_var[s->mb_width*mb_y+mb_x], s->avg_mb_var);
            }
#endif
    /* DCT & quantize */
1704 1705 1706 1707
    if(s->out_format==FMT_MJPEG){
        for(i=0;i<6;i++) {
            int overflow;
            s->block_last_index[i] = dct_quantize(s, s->block[i], i, 8, &overflow);
1708
            if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1709 1710 1711
        }
    }else{
        for(i=0;i<6;i++) {
1712 1713 1714
            if(!skip_dct[i]){
                int overflow;
                s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1715
            // FIXME we could decide to change to quantizer instead of clipping
1716 1717
            // JS: I don't think that would be a good idea it could lower quality instead
            //     of improve it. Just INTRADC clipping deserves changes in quantizer
1718 1719 1720
                if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
            }else
                s->block_last_index[i]= -1;
1721
        }
1722 1723 1724 1725 1726 1727
        if(s->luma_elim_threshold && !s->mb_intra)
            for(i=0; i<4; i++)
                dct_single_coeff_elimination(s, i, s->luma_elim_threshold, 0);
        if(s->chroma_elim_threshold && !s->mb_intra)
            for(i=4; i<6; i++)
                dct_single_coeff_elimination(s, i, s->chroma_elim_threshold, 1);
1728 1729
    }

M
Michael Niedermayer 已提交
1730 1731 1732 1733 1734 1735 1736
    if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
        s->block_last_index[4]=
        s->block_last_index[5]= 0;
        s->block[4][0]=
        s->block[5][0]= 128;
    }

1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755
    /* huffman encode */
    switch(s->out_format) {
    case FMT_MPEG1:
        mpeg1_encode_mb(s, s->block, motion_x, motion_y);
        break;
    case FMT_H263:
        if (s->h263_msmpeg4)
            msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
        else if(s->h263_pred)
            mpeg4_encode_mb(s, s->block, motion_x, motion_y);
        else
            h263_encode_mb(s, s->block, motion_x, motion_y);
        break;
    case FMT_MJPEG:
        mjpeg_encode_mb(s, s->block);
        break;
    }
}

1756
void ff_copy_bits(PutBitContext *pb, UINT8 *src, int length)
1757
{
1758 1759 1760 1761
    int bytes= length>>4;
    int bits= length&15;
    int i;

1762 1763
    if(length==0) return;

1764 1765
    for(i=0; i<bytes; i++) put_bits(pb, 16, be2me_16(((uint16_t*)src)[i]));
    put_bits(pb, bits, be2me_16(((uint16_t*)src)[i])>>(16-bits));
1766 1767
}

1768
static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785
    int i;

    memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?

    /* mpeg1 */
    d->mb_incr= s->mb_incr;
    for(i=0; i<3; i++)
        d->last_dc[i]= s->last_dc[i];
    
    /* statistics */
    d->mv_bits= s->mv_bits;
    d->i_tex_bits= s->i_tex_bits;
    d->p_tex_bits= s->p_tex_bits;
    d->i_count= s->i_count;
    d->p_count= s->p_count;
    d->skip_count= s->skip_count;
    d->misc_bits= s->misc_bits;
1786
    d->last_bits= 0;
1787 1788

    d->mb_skiped= s->mb_skiped;
1789 1790
}

1791
static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811
    int i;

    memcpy(d->mv, s->mv, 2*4*2*sizeof(int)); 
    memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
    
    /* mpeg1 */
    d->mb_incr= s->mb_incr;
    for(i=0; i<3; i++)
        d->last_dc[i]= s->last_dc[i];
    
    /* statistics */
    d->mv_bits= s->mv_bits;
    d->i_tex_bits= s->i_tex_bits;
    d->p_tex_bits= s->p_tex_bits;
    d->i_count= s->i_count;
    d->p_count= s->p_count;
    d->skip_count= s->skip_count;
    d->misc_bits= s->misc_bits;

    d->mb_intra= s->mb_intra;
1812
    d->mb_skiped= s->mb_skiped;
1813 1814 1815
    d->mv_type= s->mv_type;
    d->mv_dir= s->mv_dir;
    d->pb= s->pb;
1816 1817 1818 1819
    if(s->data_partitioning){
        d->pb2= s->pb2;
        d->tex_pb= s->tex_pb;
    }
1820 1821 1822 1823 1824
    d->block= s->block;
    for(i=0; i<6; i++)
        d->block_last_index[i]= s->block_last_index[i];
}

1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854
static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type, 
                           PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
                           int *dmin, int *next_block, int motion_x, int motion_y)
{
    int bits_count;
    
    copy_context_before_encode(s, backup, type);

    s->block= s->blocks[*next_block];
    s->pb= pb[*next_block];
    if(s->data_partitioning){
        s->pb2   = pb2   [*next_block];
        s->tex_pb= tex_pb[*next_block];
    }

    encode_mb(s, motion_x, motion_y);

    bits_count= get_bit_count(&s->pb);
    if(s->data_partitioning){
        bits_count+= get_bit_count(&s->pb2);
        bits_count+= get_bit_count(&s->tex_pb);
    }

    if(bits_count<*dmin){
        *dmin= bits_count;
        *next_block^=1;

        copy_context_after_encode(best, s, type);
    }
}
1855

1856 1857 1858 1859
static void encode_picture(MpegEncContext *s, int picture_number)
{
    int mb_x, mb_y, last_gob, pdif = 0;
    int i;
1860
    int bits;
1861
    MpegEncContext best_s, backup_s;
1862 1863 1864 1865 1866 1867 1868 1869 1870 1871
    UINT8 bit_buf[2][3000];
    UINT8 bit_buf2[2][3000];
    UINT8 bit_buf_tex[2][3000];
    PutBitContext pb[2], pb2[2], tex_pb[2];

    for(i=0; i<2; i++){
        init_put_bits(&pb    [i], bit_buf    [i], 3000, NULL, NULL);
        init_put_bits(&pb2   [i], bit_buf2   [i], 3000, NULL, NULL);
        init_put_bits(&tex_pb[i], bit_buf_tex[i], 3000, NULL, NULL);
    }
F
Fabrice Bellard 已提交
1872 1873

    s->picture_number = picture_number;
1874

1875 1876 1877 1878 1879 1880 1881
    s->block_wrap[0]=
    s->block_wrap[1]=
    s->block_wrap[2]=
    s->block_wrap[3]= s->mb_width*2 + 2;
    s->block_wrap[4]=
    s->block_wrap[5]= s->mb_width + 2;
    
1882
    /* Reset the average MB variance */
1883 1884
    s->mb_var_sum = 0;
    s->mc_mb_var_sum = 0;
1885 1886 1887 1888 1889

    /* we need to initialize some time vars before we can encode b-frames */
    if (s->h263_pred && !s->h263_msmpeg4)
        ff_set_mpeg4_time(s, s->picture_number); 

1890
    /* Estimate motion for every MB */
1891
    if(s->pict_type != I_TYPE){
1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905
        for(mb_y=0; mb_y < s->mb_height; mb_y++) {
            s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1;
            s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1);
            s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1;
            s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2);
            for(mb_x=0; mb_x < s->mb_width; mb_x++) {
                s->mb_x = mb_x;
                s->mb_y = mb_y;
                s->block_index[0]+=2;
                s->block_index[1]+=2;
                s->block_index[2]+=2;
                s->block_index[3]+=2;

                /* compute motion vector & mb_type and store in context */
1906 1907 1908 1909
                if(s->pict_type==B_TYPE)
                    ff_estimate_b_frame_motion(s, mb_x, mb_y);
                else
                    ff_estimate_p_frame_motion(s, mb_x, mb_y);
1910
//                s->mb_type[mb_y*s->mb_width + mb_x]=MB_TYPE_INTER;
1911 1912
            }
        }
1913
        emms_c();
1914
    }else /* if(s->pict_type == I_TYPE) */{
1915 1916 1917
        /* I-Frame */
        //FIXME do we need to zero them?
        memset(s->motion_val[0], 0, sizeof(INT16)*(s->mb_width*2 + 2)*(s->mb_height*2 + 2)*2);
1918
        memset(s->p_mv_table   , 0, sizeof(INT16)*(s->mb_width+2)*(s->mb_height+2)*2);
1919
        memset(s->mb_type      , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height);
1920 1921
    }

1922
    if(s->mb_var_sum < s->mc_mb_var_sum && s->pict_type == P_TYPE){ //FIXME subtract MV bits
M
Michael Niedermayer 已提交
1923
        s->pict_type= I_TYPE;
1924
        memset(s->mb_type   , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height);
1925 1926 1927 1928
        if(s->max_b_frames==0){
            s->input_pict_type= I_TYPE;
            s->input_picture_in_gop_number=0;
        }
1929
//printf("Scene change detected, encoding as I Frame\n");
M
Michael Niedermayer 已提交
1930
    }
1931 1932 1933 1934 1935 1936 1937
    
    if(s->pict_type==P_TYPE || s->pict_type==S_TYPE) 
        s->f_code= ff_get_best_fcode(s, s->p_mv_table, MB_TYPE_INTER);
        ff_fix_long_p_mvs(s);
    if(s->pict_type==B_TYPE){
        s->f_code= ff_get_best_fcode(s, s->b_forw_mv_table, MB_TYPE_FORWARD);
        s->b_code= ff_get_best_fcode(s, s->b_back_mv_table, MB_TYPE_BACKWARD);
1938 1939 1940 1941 1942

        ff_fix_long_b_mvs(s, s->b_forw_mv_table, s->f_code, MB_TYPE_FORWARD);
        ff_fix_long_b_mvs(s, s->b_back_mv_table, s->b_code, MB_TYPE_BACKWARD);
        ff_fix_long_b_mvs(s, s->b_bidir_forw_mv_table, s->f_code, MB_TYPE_BIDIR);
        ff_fix_long_b_mvs(s, s->b_bidir_back_mv_table, s->b_code, MB_TYPE_BIDIR);
1943
    }
1944
    
1945 1946
//printf("f_code %d ///\n", s->f_code);

M
Michael Niedermayer 已提交
1947 1948
//    printf("%d %d\n", s->avg_mb_var, s->mc_mb_var);

1949 1950 1951 1952 1953
    if(s->flags&CODEC_FLAG_PASS2)
        s->qscale = ff_rate_estimate_qscale_pass2(s);
    else if (!s->fixed_qscale) 
        s->qscale = ff_rate_estimate_qscale(s);

F
Fabrice Bellard 已提交
1954 1955
    if (s->out_format == FMT_MJPEG) {
        /* for mjpeg, we do include qscale in the matrix */
1956
        s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0];
F
Fabrice Bellard 已提交
1957
        for(i=1;i<64;i++)
1958
            s->intra_matrix[i] = CLAMP_TO_8BIT((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
1959 1960
        convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, 
                       s->q_intra_matrix16_bias, s->intra_matrix, s->intra_quant_bias);
F
Fabrice Bellard 已提交
1961 1962
    }

1963
    s->last_bits= get_bit_count(&s->pb);
F
Fabrice Bellard 已提交
1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981
    switch(s->out_format) {
    case FMT_MJPEG:
        mjpeg_picture_header(s);
        break;
    case FMT_H263:
        if (s->h263_msmpeg4) 
            msmpeg4_encode_picture_header(s, picture_number);
        else if (s->h263_pred)
            mpeg4_encode_picture_header(s, picture_number);
        else if (s->h263_rv10) 
            rv10_encode_picture_header(s, picture_number);
        else
            h263_encode_picture_header(s, picture_number);
        break;
    case FMT_MPEG1:
        mpeg1_encode_picture_header(s, picture_number);
        break;
    }
1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992
    bits= get_bit_count(&s->pb);
    s->header_bits= bits - s->last_bits;
    s->last_bits= bits;
    s->mv_bits=0;
    s->misc_bits=0;
    s->i_tex_bits=0;
    s->p_tex_bits=0;
    s->i_count=0;
    s->p_count=0;
    s->skip_count=0;

F
Fabrice Bellard 已提交
1993 1994 1995 1996 1997 1998 1999 2000 2001
    /* init last dc values */
    /* note: quant matrix value (8) is implied here */
    s->last_dc[0] = 128;
    s->last_dc[1] = 128;
    s->last_dc[2] = 128;
    s->mb_incr = 1;
    s->last_mv[0][0][0] = 0;
    s->last_mv[0][0][1] = 0;

2002
    /* Get the GOB height based on picture height */
2003
    if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4) {
2004 2005 2006 2007 2008 2009
        if (s->height <= 400)
            s->gob_index = 1;
        else if (s->height <= 800)
            s->gob_index = 2;
        else
            s->gob_index = 4;
2010 2011
    }else if(s->codec_id==CODEC_ID_MPEG4){
        s->gob_index = 1;
2012
    }
2013 2014 2015 2016 2017 2018

    if(s->codec_id==CODEC_ID_MPEG4 && s->data_partitioning && s->pict_type!=B_TYPE)
        ff_mpeg4_init_partitions(s);

    s->resync_mb_x=0;
    s->resync_mb_y=0;
2019
    for(mb_y=0; mb_y < s->mb_height; mb_y++) {
2020
        /* Put GOB header based on RTP MTU for formats which support it per line (H263*)*/
2021 2022 2023 2024 2025 2026
        /* TODO: Put all this stuff in a separate generic function */
        if (s->rtp_mode) {
            if (!mb_y) {
                s->ptr_lastgob = s->pb.buf;
                s->ptr_last_mb_line = s->pb.buf;
            } else if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4 && !(mb_y % s->gob_index)) {
2027
                // MN: we could move the space check from h263 -> here, as its not h263 specific
2028 2029
                last_gob = h263_encode_gob_header(s, mb_y);
                if (last_gob) {
2030 2031 2032 2033 2034
                    s->first_slice_line = 1;
                }else{
                    /*MN: we reset it here instead at the end of each line cuz mpeg4 can have 
                          slice lines starting & ending in the middle*/
                    s->first_slice_line = 0;
2035 2036 2037
                }
            }
        }
M
Michael Niedermayer 已提交
2038 2039 2040

        s->y_dc_scale= s->y_dc_scale_table[ s->qscale ];
        s->c_dc_scale= s->c_dc_scale_table[ s->qscale ];
2041
        
M
Michael Niedermayer 已提交
2042 2043 2044 2045 2046 2047
        s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1;
        s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1);
        s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1;
        s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2);
        s->block_index[4]= s->block_wrap[4]*(mb_y + 1)                    + s->block_wrap[0]*(s->mb_height*2 + 2);
        s->block_index[5]= s->block_wrap[4]*(mb_y + 1 + s->mb_height + 2) + s->block_wrap[0]*(s->mb_height*2 + 2);
2048
        for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2049 2050
            const int mb_type= s->mb_type[mb_y * s->mb_width + mb_x];
            const int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1;
2051
//            int d;
2052
            int dmin=10000000;
2053 2054 2055

            s->mb_x = mb_x;
            s->mb_y = mb_y;
M
Michael Niedermayer 已提交
2056 2057 2058 2059 2060 2061
            s->block_index[0]+=2;
            s->block_index[1]+=2;
            s->block_index[2]+=2;
            s->block_index[3]+=2;
            s->block_index[4]++;
            s->block_index[5]++;
2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094
            
            /* write gob / video packet header for formats which support it at any MB (MPEG4) */
            if(s->rtp_mode && s->mb_y>0 && s->codec_id==CODEC_ID_MPEG4){
                int pdif= pbBufPtr(&s->pb) - s->ptr_lastgob;

                //the *2 is there so we stay below the requested size
                if(pdif + s->mb_line_avgsize/s->mb_width >= s->rtp_payload_size){ 
                    if(s->codec_id==CODEC_ID_MPEG4){
                        if(s->data_partitioning && s->pict_type!=B_TYPE){
                            ff_mpeg4_merge_partitions(s);
                            ff_mpeg4_init_partitions(s);
                        }
                        ff_mpeg4_encode_video_packet_header(s);

                        if(s->flags&CODEC_FLAG_PASS1){
                            int bits= get_bit_count(&s->pb);
                            s->misc_bits+= bits - s->last_bits;
                            s->last_bits= bits;
                        }
                        ff_mpeg4_clean_buffers(s);
                    }
                    s->ptr_lastgob = pbBufPtr(&s->pb);
                    s->first_slice_line=1;
                    s->resync_mb_x=mb_x;
                    s->resync_mb_y=mb_y;
                }

                if(  (s->resync_mb_x   == s->mb_x)
                   && s->resync_mb_y+1 == s->mb_y){
                    s->first_slice_line=0; 
                }
            }

2095
            if(mb_type & (mb_type-1)){ // more than 1 MB type possible
2096
                int next_block=0;
2097
                int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2098 2099

                copy_context_before_encode(&backup_s, s, -1);
2100 2101 2102 2103 2104 2105
                backup_s.pb= s->pb;
                best_s.data_partitioning= s->data_partitioning;
                if(s->data_partitioning){
                    backup_s.pb2= s->pb2;
                    backup_s.tex_pb= s->tex_pb;
                }
2106

2107
                if(mb_type&MB_TYPE_INTER){
2108
                    s->mv_dir = MV_DIR_FORWARD;
M
Michael Niedermayer 已提交
2109
                    s->mv_type = MV_TYPE_16X16;
2110
                    s->mb_intra= 0;
2111 2112
                    s->mv[0][0][0] = s->p_mv_table[xy][0];
                    s->mv[0][0][1] = s->p_mv_table[xy][1];
2113 2114
                    encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTER, pb, pb2, tex_pb, 
                                 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
F
Fabrice Bellard 已提交
2115
                }
2116
                if(mb_type&MB_TYPE_INTER4V){                 
2117
                    s->mv_dir = MV_DIR_FORWARD;
M
Michael Niedermayer 已提交
2118 2119 2120 2121 2122 2123
                    s->mv_type = MV_TYPE_8X8;
                    s->mb_intra= 0;
                    for(i=0; i<4; i++){
                        s->mv[0][i][0] = s->motion_val[s->block_index[i]][0];
                        s->mv[0][i][1] = s->motion_val[s->block_index[i]][1];
                    }
2124 2125
                    encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTER4V, pb, pb2, tex_pb, 
                                 &dmin, &next_block, 0, 0);
2126 2127 2128 2129 2130 2131 2132
                }
                if(mb_type&MB_TYPE_FORWARD){
                    s->mv_dir = MV_DIR_FORWARD;
                    s->mv_type = MV_TYPE_16X16;
                    s->mb_intra= 0;
                    s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
                    s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2133 2134
                    encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_FORWARD, pb, pb2, tex_pb, 
                                 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2135 2136 2137 2138 2139 2140 2141
                }
                if(mb_type&MB_TYPE_BACKWARD){
                    s->mv_dir = MV_DIR_BACKWARD;
                    s->mv_type = MV_TYPE_16X16;
                    s->mb_intra= 0;
                    s->mv[1][0][0] = s->b_back_mv_table[xy][0];
                    s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2142 2143
                    encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_BACKWARD, pb, pb2, tex_pb, 
                                 &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2144 2145 2146 2147 2148 2149 2150 2151 2152
                }
                if(mb_type&MB_TYPE_BIDIR){
                    s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
                    s->mv_type = MV_TYPE_16X16;
                    s->mb_intra= 0;
                    s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
                    s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
                    s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
                    s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2153 2154
                    encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_BIDIR, pb, pb2, tex_pb, 
                                 &dmin, &next_block, 0, 0);
2155 2156 2157 2158 2159 2160 2161 2162 2163
                }
                if(mb_type&MB_TYPE_DIRECT){
                    s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
                    s->mv_type = MV_TYPE_16X16; //FIXME
                    s->mb_intra= 0;
                    s->mv[0][0][0] = s->b_direct_forw_mv_table[xy][0];
                    s->mv[0][0][1] = s->b_direct_forw_mv_table[xy][1];
                    s->mv[1][0][0] = s->b_direct_back_mv_table[xy][0];
                    s->mv[1][0][1] = s->b_direct_back_mv_table[xy][1];
2164 2165
                    encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_DIRECT, pb, pb2, tex_pb, 
                                 &dmin, &next_block, s->b_direct_mv_table[xy][0], s->b_direct_mv_table[xy][1]);
M
Michael Niedermayer 已提交
2166
                }
2167
                if(mb_type&MB_TYPE_INTRA){
2168
                    s->mv_dir = MV_DIR_FORWARD;
M
Michael Niedermayer 已提交
2169
                    s->mv_type = MV_TYPE_16X16;
2170 2171 2172
                    s->mb_intra= 1;
                    s->mv[0][0][0] = 0;
                    s->mv[0][0][1] = 0;
2173 2174
                    encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTRA, pb, pb2, tex_pb, 
                                 &dmin, &next_block, 0, 0);
2175 2176 2177
                    /* force cleaning of ac/dc pred stuff if needed ... */
                    if(s->h263_pred || s->h263_aic)
                        s->mbintra_table[mb_x + mb_y*s->mb_width]=1;
M
Michael Niedermayer 已提交
2178
                }
2179
                copy_context_after_encode(s, &best_s, -1);
2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196
                
                pb_bits_count= get_bit_count(&s->pb);
                flush_put_bits(&s->pb);
                ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
                s->pb= backup_s.pb;
                
                if(s->data_partitioning){
                    pb2_bits_count= get_bit_count(&s->pb2);
                    flush_put_bits(&s->pb2);
                    ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
                    s->pb2= backup_s.pb2;
                    
                    tex_pb_bits_count= get_bit_count(&s->tex_pb);
                    flush_put_bits(&s->tex_pb);
                    ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
                    s->tex_pb= backup_s.tex_pb;
                }
2197
                s->last_bits= get_bit_count(&s->pb);
F
Fabrice Bellard 已提交
2198
            } else {
2199 2200
                int motion_x, motion_y;
                s->mv_type=MV_TYPE_16X16;
2201
                // only one MB-Type possible
2202 2203
                switch(mb_type){
                case MB_TYPE_INTRA:
2204
                    s->mv_dir = MV_DIR_FORWARD;
2205
                    s->mb_intra= 1;
2206 2207
                    motion_x= s->mv[0][0][0] = 0;
                    motion_y= s->mv[0][0][1] = 0;
2208 2209
                    break;
                case MB_TYPE_INTER:
2210 2211 2212 2213
                    s->mv_dir = MV_DIR_FORWARD;
                    s->mb_intra= 0;
                    motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
                    motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2214
                    break;
2215 2216 2217 2218 2219 2220 2221 2222 2223 2224
                case MB_TYPE_INTER4V:
                    s->mv_dir = MV_DIR_FORWARD;
                    s->mv_type = MV_TYPE_8X8;
                    s->mb_intra= 0;
                    for(i=0; i<4; i++){
                        s->mv[0][i][0] = s->motion_val[s->block_index[i]][0];
                        s->mv[0][i][1] = s->motion_val[s->block_index[i]][1];
                    }
                    motion_x= motion_y= 0;
                    break;
2225
                case MB_TYPE_DIRECT:
2226 2227
                    s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
                    s->mb_intra= 0;
2228 2229 2230 2231 2232 2233 2234 2235
                    motion_x=s->b_direct_mv_table[xy][0];
                    motion_y=s->b_direct_mv_table[xy][1];
                    s->mv[0][0][0] = s->b_direct_forw_mv_table[xy][0];
                    s->mv[0][0][1] = s->b_direct_forw_mv_table[xy][1];
                    s->mv[1][0][0] = s->b_direct_back_mv_table[xy][0];
                    s->mv[1][0][1] = s->b_direct_back_mv_table[xy][1];
                    break;
                case MB_TYPE_BIDIR:
2236
                    s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2237
                    s->mb_intra= 0;
2238 2239 2240 2241 2242 2243
                    motion_x=0;
                    motion_y=0;
                    s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
                    s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
                    s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
                    s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2244 2245
                    break;
                case MB_TYPE_BACKWARD:
2246 2247 2248 2249
                    s->mv_dir = MV_DIR_BACKWARD;
                    s->mb_intra= 0;
                    motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
                    motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2250 2251
                    break;
                case MB_TYPE_FORWARD:
2252 2253 2254 2255 2256
                    s->mv_dir = MV_DIR_FORWARD;
                    s->mb_intra= 0;
                    motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
                    motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
//                    printf(" %d %d ", motion_x, motion_y);
2257 2258
                    break;
                default:
2259 2260
                    motion_x=motion_y=0; //gcc warning fix
                    printf("illegal MB type\n");
2261
                }
2262
                encode_mb(s, motion_x, motion_y);
F
Fabrice Bellard 已提交
2263
            }
2264 2265 2266 2267 2268
            /* clean the MV table in IPS frames for direct mode in B frames */
            if(s->mb_intra /* && I,P,S_TYPE */){
                s->p_mv_table[xy][0]=0;
                s->p_mv_table[xy][1]=0;
            }
F
Fabrice Bellard 已提交
2269

2270
            MPV_decode_mb(s, s->block);
2271
//printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_width, get_bit_count(&s->pb));
F
Fabrice Bellard 已提交
2272
        }
2273 2274


2275 2276 2277
        /* Obtain average GOB size for RTP */
        if (s->rtp_mode) {
            if (!mb_y)
2278
                s->mb_line_avgsize = pbBufPtr(&s->pb) - s->ptr_last_mb_line;
2279
            else if (!(mb_y % s->gob_index)) {    
2280 2281
                s->mb_line_avgsize = (s->mb_line_avgsize + pbBufPtr(&s->pb) - s->ptr_last_mb_line) >> 1;
                s->ptr_last_mb_line = pbBufPtr(&s->pb);
2282 2283 2284
            }
            //fprintf(stderr, "\nMB line: %d\tSize: %u\tAvg. Size: %u", s->mb_y, 
            //                    (s->pb.buf_ptr - s->ptr_last_mb_line), s->mb_line_avgsize);
2285
            if(s->codec_id!=CODEC_ID_MPEG4) s->first_slice_line = 0; //FIXME clean
2286
        }
F
Fabrice Bellard 已提交
2287
    }
2288
    emms_c();
2289

2290 2291 2292 2293
    if(s->codec_id==CODEC_ID_MPEG4 && s->data_partitioning && s->pict_type!=B_TYPE)
        ff_mpeg4_merge_partitions(s);

    if (s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == I_TYPE)
M
Michael Niedermayer 已提交
2294 2295
        msmpeg4_encode_ext_header(s);

2296 2297 2298
    if(s->codec_id==CODEC_ID_MPEG4) 
        ff_mpeg4_stuffing(&s->pb);

2299 2300
    //if (s->gob_number)
    //    fprintf(stderr,"\nNumber of GOB: %d", s->gob_number);
2301 2302 2303 2304
    
    /* Send the last GOB if RTP */    
    if (s->rtp_mode) {
        flush_put_bits(&s->pb);
2305
        pdif = pbBufPtr(&s->pb) - s->ptr_lastgob;
2306 2307 2308
        /* Call the RTP callback to send the last GOB */
        if (s->rtp_callback)
            s->rtp_callback(s->ptr_lastgob, pdif, s->gob_number);
2309
        s->ptr_lastgob = pbBufPtr(&s->pb);
2310 2311
        //fprintf(stderr,"\nGOB: %2d size: %d (last)", s->gob_number, pdif);
    }
F
Fabrice Bellard 已提交
2312 2313
}

M
Michael Niedermayer 已提交
2314
static int dct_quantize_c(MpegEncContext *s, 
F
Fabrice Bellard 已提交
2315
                        DCTELEM *block, int n,
2316
                        int qscale, int *overflow)
F
Fabrice Bellard 已提交
2317 2318 2319
{
    int i, j, level, last_non_zero, q;
    const int *qmat;
2320 2321 2322
    int bias;
    int max=0;
    unsigned int threshold1, threshold2;
2323
    
F
Fabrice Bellard 已提交
2324 2325
    av_fdct (block);

F
Fabrice Bellard 已提交
2326 2327 2328 2329
    /* we need this permutation so that we correct the IDCT
       permutation. will be moved into DCT code */
    block_permute(block);

F
Fabrice Bellard 已提交
2330
    if (s->mb_intra) {
2331 2332 2333 2334 2335 2336 2337 2338 2339 2340
        if (!s->h263_aic) {
            if (n < 4)
                q = s->y_dc_scale;
            else
                q = s->c_dc_scale;
            q = q << 3;
        } else
            /* For AIC we skip quant/dequant of INTRADC */
            q = 1 << 3;
            
F
Fabrice Bellard 已提交
2341 2342 2343 2344
        /* note: block[0] is assumed to be positive */
        block[0] = (block[0] + (q >> 1)) / q;
        i = 1;
        last_non_zero = 0;
2345 2346
        qmat = s->q_intra_matrix[qscale];
        bias= s->intra_quant_bias<<(QMAT_SHIFT - 3 - QUANT_BIAS_SHIFT);
F
Fabrice Bellard 已提交
2347 2348 2349
    } else {
        i = 0;
        last_non_zero = -1;
2350 2351
        qmat = s->q_inter_matrix[qscale];
        bias= s->inter_quant_bias<<(QMAT_SHIFT - 3 - QUANT_BIAS_SHIFT);
F
Fabrice Bellard 已提交
2352
    }
2353 2354
    threshold1= (1<<(QMAT_SHIFT - 3)) - bias - 1;
    threshold2= threshold1<<1;
F
Fabrice Bellard 已提交
2355 2356 2357 2358 2359 2360

    for(;i<64;i++) {
        j = zigzag_direct[i];
        level = block[j];
        level = level * qmat[j];

2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
        if(((unsigned)(level+threshold1))>threshold2){
            if(level>0){
                level= (bias + level)>>(QMAT_SHIFT - 3);
                block[j]= level;
            }else{
                level= (bias - level)>>(QMAT_SHIFT - 3);
                block[j]= -level;
            }
            max |=level;
F
Fabrice Bellard 已提交
2372
            last_non_zero = i;
2373 2374
        }else{
            block[j]=0;
F
Fabrice Bellard 已提交
2375 2376
        }
    }
2377 2378
    *overflow= s->max_qcoeff < max; //overflow might have happend
    
F
Fabrice Bellard 已提交
2379 2380 2381
    return last_non_zero;
}

2382 2383
static void dct_unquantize_mpeg1_c(MpegEncContext *s, 
                                   DCTELEM *block, int n, int qscale)
F
Fabrice Bellard 已提交
2384
{
M
Michael Niedermayer 已提交
2385
    int i, level, nCoeffs;
F
Fabrice Bellard 已提交
2386 2387
    const UINT16 *quant_matrix;

M
Michael Niedermayer 已提交
2388 2389 2390
    if(s->alternate_scan) nCoeffs= 64;
    else nCoeffs= s->block_last_index[n]+1;
    
F
Fabrice Bellard 已提交
2391 2392 2393 2394 2395 2396 2397
    if (s->mb_intra) {
        if (n < 4) 
            block[0] = block[0] * s->y_dc_scale;
        else
            block[0] = block[0] * s->c_dc_scale;
        /* XXX: only mpeg1 */
        quant_matrix = s->intra_matrix;
M
Michael Niedermayer 已提交
2398 2399 2400
        for(i=1;i<nCoeffs;i++) {
            int j= zigzag_direct[i];
            level = block[j];
F
Fabrice Bellard 已提交
2401 2402 2403
            if (level) {
                if (level < 0) {
                    level = -level;
M
Michael Niedermayer 已提交
2404
                    level = (int)(level * qscale * quant_matrix[j]) >> 3;
F
Fabrice Bellard 已提交
2405 2406 2407
                    level = (level - 1) | 1;
                    level = -level;
                } else {
M
Michael Niedermayer 已提交
2408
                    level = (int)(level * qscale * quant_matrix[j]) >> 3;
F
Fabrice Bellard 已提交
2409 2410 2411 2412 2413 2414
                    level = (level - 1) | 1;
                }
#ifdef PARANOID
                if (level < -2048 || level > 2047)
                    fprintf(stderr, "unquant error %d %d\n", i, level);
#endif
M
Michael Niedermayer 已提交
2415
                block[j] = level;
F
Fabrice Bellard 已提交
2416 2417 2418 2419
            }
        }
    } else {
        i = 0;
2420
        quant_matrix = s->inter_matrix;
M
Michael Niedermayer 已提交
2421
        for(;i<nCoeffs;i++) {
M
Michael Niedermayer 已提交
2422 2423
            int j= zigzag_direct[i];
            level = block[j];
F
Fabrice Bellard 已提交
2424 2425 2426 2427
            if (level) {
                if (level < 0) {
                    level = -level;
                    level = (((level << 1) + 1) * qscale *
M
Michael Niedermayer 已提交
2428
                             ((int) (quant_matrix[j]))) >> 4;
F
Fabrice Bellard 已提交
2429 2430 2431 2432
                    level = (level - 1) | 1;
                    level = -level;
                } else {
                    level = (((level << 1) + 1) * qscale *
M
Michael Niedermayer 已提交
2433
                             ((int) (quant_matrix[j]))) >> 4;
F
Fabrice Bellard 已提交
2434 2435 2436 2437 2438 2439
                    level = (level - 1) | 1;
                }
#ifdef PARANOID
                if (level < -2048 || level > 2047)
                    fprintf(stderr, "unquant error %d %d\n", i, level);
#endif
M
Michael Niedermayer 已提交
2440
                block[j] = level;
F
Fabrice Bellard 已提交
2441 2442 2443 2444
            }
        }
    }
}
2445

2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481
static void dct_unquantize_mpeg2_c(MpegEncContext *s, 
                                   DCTELEM *block, int n, int qscale)
{
    int i, level, nCoeffs;
    const UINT16 *quant_matrix;

    if(s->alternate_scan) nCoeffs= 64;
    else nCoeffs= s->block_last_index[n]+1;
    
    if (s->mb_intra) {
        if (n < 4) 
            block[0] = block[0] * s->y_dc_scale;
        else
            block[0] = block[0] * s->c_dc_scale;
        quant_matrix = s->intra_matrix;
        for(i=1;i<nCoeffs;i++) {
            int j= zigzag_direct[i];
            level = block[j];
            if (level) {
                if (level < 0) {
                    level = -level;
                    level = (int)(level * qscale * quant_matrix[j]) >> 3;
                    level = -level;
                } else {
                    level = (int)(level * qscale * quant_matrix[j]) >> 3;
                }
#ifdef PARANOID
                if (level < -2048 || level > 2047)
                    fprintf(stderr, "unquant error %d %d\n", i, level);
#endif
                block[j] = level;
            }
        }
    } else {
        int sum=-1;
        i = 0;
2482
        quant_matrix = s->inter_matrix;
2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508
        for(;i<nCoeffs;i++) {
            int j= zigzag_direct[i];
            level = block[j];
            if (level) {
                if (level < 0) {
                    level = -level;
                    level = (((level << 1) + 1) * qscale *
                             ((int) (quant_matrix[j]))) >> 4;
                    level = -level;
                } else {
                    level = (((level << 1) + 1) * qscale *
                             ((int) (quant_matrix[j]))) >> 4;
                }
#ifdef PARANOID
                if (level < -2048 || level > 2047)
                    fprintf(stderr, "unquant error %d %d\n", i, level);
#endif
                block[j] = level;
                sum+=level;
            }
        }
        block[63]^=sum&1;
    }
}


2509 2510 2511 2512
static void dct_unquantize_h263_c(MpegEncContext *s, 
                                  DCTELEM *block, int n, int qscale)
{
    int i, level, qmul, qadd;
M
Michael Niedermayer 已提交
2513
    int nCoeffs;
2514
    
2515
    if (s->mb_intra) {
2516 2517 2518 2519 2520 2521
        if (!s->h263_aic) {
            if (n < 4) 
                block[0] = block[0] * s->y_dc_scale;
            else
                block[0] = block[0] * s->c_dc_scale;
        }
2522
        i = 1;
M
Michael Niedermayer 已提交
2523
        nCoeffs= 64; //does not allways use zigzag table 
2524 2525
    } else {
        i = 0;
M
Michael Niedermayer 已提交
2526
        nCoeffs= zigzag_end[ s->block_last_index[n] ];
2527 2528 2529
    }

    qmul = s->qscale << 1;
2530 2531 2532 2533
    if (s->h263_aic && s->mb_intra)
        qadd = 0;
    else
        qadd = (s->qscale - 1) | 1;
2534

M
Michael Niedermayer 已提交
2535
    for(;i<nCoeffs;i++) {
2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550
        level = block[i];
        if (level) {
            if (level < 0) {
                level = level * qmul - qadd;
            } else {
                level = level * qmul + qadd;
            }
#ifdef PARANOID
                if (level < -2048 || level > 2047)
                    fprintf(stderr, "unquant error %d %d\n", i, level);
#endif
            block[i] = level;
        }
    }
}
F
Fabrice Bellard 已提交
2551

2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664
static void remove_ac(MpegEncContext *s, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, int mb_x, int mb_y)
{
    int dc, dcb, dcr, y, i;
    for(i=0; i<4; i++){
        dc= s->dc_val[0][mb_x*2+1 + (i&1) + (mb_y*2+1 + (i>>1))*(s->mb_width*2+2)];
        for(y=0; y<8; y++){
            int x;
            for(x=0; x<8; x++){
                dest_y[x + (i&1)*8 + (y + (i>>1)*8)*s->linesize]= dc/8;
            }
        }
    }
    dcb = s->dc_val[1][mb_x+1 + (mb_y+1)*(s->mb_width+2)];
    dcr= s->dc_val[2][mb_x+1 + (mb_y+1)*(s->mb_width+2)];
    for(y=0; y<8; y++){
        int x;
        for(x=0; x<8; x++){
            dest_cb[x + y*(s->linesize>>1)]= dcb/8;
            dest_cr[x + y*(s->linesize>>1)]= dcr/8;
        }
    }
}

/**
 * will conceal past errors, and allso drop b frames if needed
 *
 */
void ff_conceal_past_errors(MpegEncContext *s, int unknown_pos)
{
    int mb_x= s->mb_x;
    int mb_y= s->mb_y;
    int mb_dist=0;
    int i, intra_count=0, inter_count=0;
    int intra_conceal= s->msmpeg4_version ? 50 : 50; //FIXME finetune
    int inter_conceal= s->msmpeg4_version ? 50 : 50;
    
    // for last block
    if(mb_x>=s->mb_width)  mb_x= s->mb_width -1;
    if(mb_y>=s->mb_height) mb_y= s->mb_height-1;

    if(s->decoding_error==0 && unknown_pos){
        if(s->data_partitioning && s->pict_type!=B_TYPE)
                s->decoding_error= DECODING_AC_LOST;
        else
                s->decoding_error= DECODING_DESYNC;
    }

    if(s->decoding_error==DECODING_DESYNC && s->pict_type!=B_TYPE) s->next_p_frame_damaged=1;

    for(i=mb_x + mb_y*s->mb_width; i>=0; i--){
        if(s->mbintra_table[i]) intra_count++;
        else                    inter_count++;
    }
    
    if(s->decoding_error==DECODING_AC_LOST){
        intra_conceal*=2;
        inter_conceal*=2;
    }else if(s->decoding_error==DECODING_ACDC_LOST){
        intra_conceal*=2;
        inter_conceal*=2;
    }

    if(unknown_pos && (intra_count<inter_count)){
        intra_conceal= inter_conceal= s->mb_num; 
//        printf("%d %d\n",intra_count, inter_count);
    }

    fprintf(stderr, "concealing errors\n");

    /* for all MBs from the current one back until the last resync marker */
    for(; mb_y>=0 && mb_y>=s->resync_mb_y; mb_y--){
        for(; mb_x>=0; mb_x--){
            uint8_t *dest_y  = s->current_picture[0] + (mb_y * 16*  s->linesize      ) + mb_x * 16;
            uint8_t *dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
            uint8_t *dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
            int mb_x_backup= s->mb_x; //FIXME pass xy to mpeg_motion
            int mb_y_backup= s->mb_y;
            s->mb_x=mb_x;
            s->mb_y=mb_y;
            if(s->mbintra_table[mb_y*s->mb_width + mb_x] && mb_dist<intra_conceal){
                if(s->decoding_error==DECODING_AC_LOST){
                    remove_ac(s, dest_y, dest_cb, dest_cr, mb_x, mb_y);
//                    printf("remove ac to %d %d\n", mb_x, mb_y);
                }else{
                    mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, 
                                s->last_picture, 0, 0, put_pixels_tab,
                                0/*mx*/, 0/*my*/, 16);
                }
            }
            else if(!s->mbintra_table[mb_y*s->mb_width + mb_x] && mb_dist<inter_conceal){
                int mx=0;
                int my=0;

                if(s->decoding_error!=DECODING_DESYNC){
                    int xy= mb_x*2+1 + (mb_y*2+1)*(s->mb_width*2+2);
                    mx= s->motion_val[ xy ][0];
                    my= s->motion_val[ xy ][1];
                }

                mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, 
                            s->last_picture, 0, 0, put_pixels_tab,
                            mx, my, 16);
            }
            s->mb_x= mb_x_backup;
            s->mb_y= mb_y_backup;

            if(mb_x== s->resync_mb_x && mb_y== s->resync_mb_y) return;
            if(!s->mbskip_table[mb_x + mb_y*s->mb_width]) mb_dist++;
        }
        mb_x=s->mb_width-1;
    }
}

F
Fabrice Bellard 已提交
2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714
AVCodec mpeg1video_encoder = {
    "mpeg1video",
    CODEC_TYPE_VIDEO,
    CODEC_ID_MPEG1VIDEO,
    sizeof(MpegEncContext),
    MPV_encode_init,
    MPV_encode_picture,
    MPV_encode_end,
};

AVCodec h263_encoder = {
    "h263",
    CODEC_TYPE_VIDEO,
    CODEC_ID_H263,
    sizeof(MpegEncContext),
    MPV_encode_init,
    MPV_encode_picture,
    MPV_encode_end,
};

AVCodec h263p_encoder = {
    "h263p",
    CODEC_TYPE_VIDEO,
    CODEC_ID_H263P,
    sizeof(MpegEncContext),
    MPV_encode_init,
    MPV_encode_picture,
    MPV_encode_end,
};

AVCodec rv10_encoder = {
    "rv10",
    CODEC_TYPE_VIDEO,
    CODEC_ID_RV10,
    sizeof(MpegEncContext),
    MPV_encode_init,
    MPV_encode_picture,
    MPV_encode_end,
};

AVCodec mjpeg_encoder = {
    "mjpeg",
    CODEC_TYPE_VIDEO,
    CODEC_ID_MJPEG,
    sizeof(MpegEncContext),
    MPV_encode_init,
    MPV_encode_picture,
    MPV_encode_end,
};

F
Fabrice Bellard 已提交
2715 2716
AVCodec mpeg4_encoder = {
    "mpeg4",
F
Fabrice Bellard 已提交
2717
    CODEC_TYPE_VIDEO,
F
Fabrice Bellard 已提交
2718
    CODEC_ID_MPEG4,
F
Fabrice Bellard 已提交
2719 2720 2721 2722 2723 2724
    sizeof(MpegEncContext),
    MPV_encode_init,
    MPV_encode_picture,
    MPV_encode_end,
};

2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745
AVCodec msmpeg4v1_encoder = {
    "msmpeg4v1",
    CODEC_TYPE_VIDEO,
    CODEC_ID_MSMPEG4V1,
    sizeof(MpegEncContext),
    MPV_encode_init,
    MPV_encode_picture,
    MPV_encode_end,
};

AVCodec msmpeg4v2_encoder = {
    "msmpeg4v2",
    CODEC_TYPE_VIDEO,
    CODEC_ID_MSMPEG4V2,
    sizeof(MpegEncContext),
    MPV_encode_init,
    MPV_encode_picture,
    MPV_encode_end,
};

AVCodec msmpeg4v3_encoder = {
F
Fabrice Bellard 已提交
2746 2747
    "msmpeg4",
    CODEC_TYPE_VIDEO,
2748
    CODEC_ID_MSMPEG4V3,
F
Fabrice Bellard 已提交
2749 2750 2751 2752 2753
    sizeof(MpegEncContext),
    MPV_encode_init,
    MPV_encode_picture,
    MPV_encode_end,
};
M
Michael Niedermayer 已提交
2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773

AVCodec wmv1_encoder = {
    "wmv1",
    CODEC_TYPE_VIDEO,
    CODEC_ID_WMV1,
    sizeof(MpegEncContext),
    MPV_encode_init,
    MPV_encode_picture,
    MPV_encode_end,
};

AVCodec wmv2_encoder = {
    "wmv2",
    CODEC_TYPE_VIDEO,
    CODEC_ID_WMV2,
    sizeof(MpegEncContext),
    MPV_encode_init,
    MPV_encode_picture,
    MPV_encode_end,
};