mpegvideo_enc.c 152 KB
Newer Older
1 2
/*
 * The simplest mpeg encoder (well, it was the simplest!)
3
 * Copyright (c) 2000,2001 Fabrice Bellard
4 5
 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
 *
6 7
 * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
 *
8
 * This file is part of Libav.
9
 *
10
 * Libav is free software; you can redistribute it and/or
11 12 13 14
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
15
 * Libav is distributed in the hope that it will be useful,
16 17 18 19 20
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
21
 * License along with Libav; if not, write to the Free Software
22 23 24 25
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
26
 * @file
27 28 29
 * The simplest mpeg encoder (well, it was the simplest!).
 */

30
#include "libavutil/intmath.h"
31
#include "libavutil/mathematics.h"
32
#include "libavutil/opt.h"
33 34 35
#include "avcodec.h"
#include "dsputil.h"
#include "mpegvideo.h"
36
#include "h263.h"
37
#include "mathops.h"
38 39 40
#include "mjpegenc.h"
#include "msmpeg4.h"
#include "faandct.h"
41
#include "thread.h"
42
#include "aandcttab.h"
43
#include "flv.h"
44
#include "mpeg4video.h"
45
#include "internal.h"
46
#include "bytestream.h"
47 48 49 50 51 52 53 54
#include <limits.h>

//#undef NDEBUG
//#include <assert.h>

static int encode_picture(MpegEncContext *s, int picture_number);
static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
static int sse_mb(MpegEncContext *s);
55 56
static void denoise_dct_c(MpegEncContext *s, DCTELEM *block);
static int dct_quantize_trellis_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
57 58 59 60 61 62

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

//#define DEBUG

63 64
static uint8_t default_mv_penalty[MAX_FCODE + 1][MAX_MV * 2 + 1];
static uint8_t default_fcode_tab[MAX_MV * 2 + 1];
65

66 67 68 69 70
const AVOption ff_mpv_generic_options[] = {
    FF_MPV_COMMON_OPTS
    { NULL },
};

71 72 73 74
void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64],
                       uint16_t (*qmat16)[2][64],
                       const uint16_t *quant_matrix,
                       int bias, int qmin, int qmax, int intra)
75 76
{
    int qscale;
77
    int shift = 0;
78

79
    for (qscale = qmin; qscale <= qmax; qscale++) {
80
        int i;
81
        if (dsp->fdct == ff_jpeg_fdct_islow_8 ||
82 83
            dsp->fdct == ff_jpeg_fdct_islow_10 ||
            dsp->fdct == ff_faandct) {
84 85 86 87 88 89 90
            for (i = 0; i < 64; i++) {
                const int j = dsp->idct_permutation[i];
                /* 16 <= qscale * quant_matrix[i] <= 7905
                 * Assume x = ff_aanscales[i] * qscale * quant_matrix[i]
                 *             19952 <=              x  <= 249205026
                 * (1 << 36) / 19952 >= (1 << 36) / (x) >= (1 << 36) / 249205026
                 *           3444240 >= (1 << 36) / (x) >= 275 */
91 92

                qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
93
                                        (qscale * quant_matrix[j]));
94
            }
95
        } else if (dsp->fdct == ff_fdct_ifast) {
96 97 98 99 100 101 102
            for (i = 0; i < 64; i++) {
                const int j = dsp->idct_permutation[i];
                /* 16 <= qscale * quant_matrix[i] <= 7905
                 * Assume x = ff_aanscales[i] * qscale * quant_matrix[i]
                 *             19952 <=              x  <= 249205026
                 * (1 << 36) / 19952 >= (1 << 36) / (x) >= (1 << 36) / 249205026
                 *           3444240 >= (1 << 36) / (x) >= 275 */
103 104

                qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
105 106
                                        (ff_aanscales[i] * qscale *
                                         quant_matrix[j]));
107 108
            }
        } else {
109 110
            for (i = 0; i < 64; i++) {
                const int j = dsp->idct_permutation[i];
111
                /* We can safely suppose that 16 <= quant_matrix[i] <= 255
112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128
                 * Assume x = qscale * quant_matrix[i]
                 * So             16 <=              x  <= 7905
                 * so (1 << 19) / 16 >= (1 << 19) / (x) >= (1 << 19) / 7905
                 * so          32768 >= (1 << 19) / (x) >= 67 */
                qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
                                        (qscale * quant_matrix[j]));
                //qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) /
                //                    (qscale * quant_matrix[i]);
                qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) /
                                       (qscale * quant_matrix[j]);

                if (qmat16[qscale][0][i] == 0 ||
                    qmat16[qscale][0][i] == 128 * 256)
                    qmat16[qscale][0][i] = 128 * 256 - 1;
                qmat16[qscale][1][i] =
                    ROUNDED_DIV(bias << (16 - QUANT_BIAS_SHIFT),
                                qmat16[qscale][0][i]);
129 130 131
            }
        }

132 133
        for (i = intra; i < 64; i++) {
            int64_t max = 8191;
134
            if (dsp->fdct == ff_fdct_ifast) {
135
                max = (8191LL * ff_aanscales[i]) >> 14;
136
            }
137
            while (((max * qmat[qscale][i]) >> shift) > INT_MAX) {
138 139 140 141
                shift++;
            }
        }
    }
142 143 144 145
    if (shift) {
        av_log(NULL, AV_LOG_INFO,
               "Warning, QMAT_SHIFT is larger than %d, overflows possible\n",
               QMAT_SHIFT - shift);
146 147 148
    }
}

149 150 151 152 153
static inline void update_qscale(MpegEncContext *s)
{
    s->qscale = (s->lambda * 139 + FF_LAMBDA_SCALE * 64) >>
                (FF_LAMBDA_SHIFT + 7);
    s->qscale = av_clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
154

155 156
    s->lambda2 = (s->lambda * s->lambda + FF_LAMBDA_SCALE / 2) >>
                 FF_LAMBDA_SHIFT;
157 158
}

159 160
void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix)
{
161 162
    int i;

163
    if (matrix) {
164
        put_bits(pb, 1, 1);
165 166
        for (i = 0; i < 64; i++) {
            put_bits(pb, 8, matrix[ff_zigzag_direct[i]]);
167
        }
168
    } else
169 170 171
        put_bits(pb, 1, 0);
}

172 173 174
/**
 * init s->current_picture.qscale_table from s->lambda_table
 */
175 176
void ff_init_qscale_tab(MpegEncContext *s)
{
177
    int8_t * const qscale_table = s->current_picture.f.qscale_table;
178 179
    int i;

180 181 182 183 184
    for (i = 0; i < s->mb_num; i++) {
        unsigned int lam = s->lambda_table[s->mb_index2xy[i]];
        int qp = (lam * 139 + FF_LAMBDA_SCALE * 64) >> (FF_LAMBDA_SHIFT + 7);
        qscale_table[s->mb_index2xy[i]] = av_clip(qp, s->avctx->qmin,
                                                  s->avctx->qmax);
185 186 187
    }
}

188 189 190 191
static void copy_picture_attributes(MpegEncContext *s,
                                    AVFrame *dst,
                                    AVFrame *src)
{
192 193 194 195 196 197
    int i;

    dst->pict_type              = src->pict_type;
    dst->quality                = src->quality;
    dst->coded_picture_number   = src->coded_picture_number;
    dst->display_picture_number = src->display_picture_number;
198
    //dst->reference              = src->reference;
199 200 201 202
    dst->pts                    = src->pts;
    dst->interlaced_frame       = src->interlaced_frame;
    dst->top_field_first        = src->top_field_first;

203 204
    if (s->avctx->me_threshold) {
        if (!src->motion_val[0])
205
            av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
206
        if (!src->mb_type)
207
            av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
208
        if (!src->ref_index[0])
209
            av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226
        if (src->motion_subsample_log2 != dst->motion_subsample_log2)
            av_log(s->avctx, AV_LOG_ERROR,
                   "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
                   src->motion_subsample_log2, dst->motion_subsample_log2);

        memcpy(dst->mb_type, src->mb_type,
               s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));

        for (i = 0; i < 2; i++) {
            int stride = ((16 * s->mb_width ) >>
                          src->motion_subsample_log2) + 1;
            int height = ((16 * s->mb_height) >> src->motion_subsample_log2);

            if (src->motion_val[i] &&
                src->motion_val[i] != dst->motion_val[i]) {
                memcpy(dst->motion_val[i], src->motion_val[i],
                       2 * stride * height * sizeof(int16_t));
227
            }
228 229 230
            if (src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]) {
                memcpy(dst->ref_index[i], src->ref_index[i],
                       s->mb_stride * 4 * s->mb_height * sizeof(int8_t));
231 232 233 234 235
            }
        }
    }
}

236 237 238
static void update_duplicate_context_after_me(MpegEncContext *dst,
                                              MpegEncContext *src)
{
239 240 241 242 243 244 245 246 247 248
#define COPY(a) dst->a= src->a
    COPY(pict_type);
    COPY(current_picture);
    COPY(f_code);
    COPY(b_code);
    COPY(qscale);
    COPY(lambda);
    COPY(lambda2);
    COPY(picture_in_gop_number);
    COPY(gop_picture_number);
249 250 251
    COPY(frame_pred_frame_dct); // FIXME don't set in encode_header
    COPY(progressive_frame);    // FIXME don't set in encode_header
    COPY(partitioned_frame);    // FIXME don't set in encode_header
252 253 254 255
#undef COPY
}

/**
256
 * Set the given MpegEncContext to defaults for encoding.
257 258
 * the changed fields will not depend upon the prior state of the MpegEncContext.
 */
259 260
static void MPV_encode_defaults(MpegEncContext *s)
{
261
    int i;
262
    ff_MPV_common_defaults(s);
263

264 265
    for (i = -16; i < 16; i++) {
        default_fcode_tab[i + MAX_MV] = 1;
266
    }
267 268
    s->me.mv_penalty = default_mv_penalty;
    s->fcode_tab     = default_fcode_tab;
269 270 271
}

/* init video encoder */
272
av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
273 274 275 276 277 278 279 280
{
    MpegEncContext *s = avctx->priv_data;
    int i;
    int chroma_h_shift, chroma_v_shift;

    MPV_encode_defaults(s);

    switch (avctx->codec_id) {
281
    case AV_CODEC_ID_MPEG2VIDEO:
282 283
        if (avctx->pix_fmt != AV_PIX_FMT_YUV420P &&
            avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
284 285
            av_log(avctx, AV_LOG_ERROR,
                   "only YUV420 and YUV422 are supported\n");
286 287 288
            return -1;
        }
        break;
289
    case AV_CODEC_ID_LJPEG:
290 291 292 293 294 295 296
        if (avctx->pix_fmt != AV_PIX_FMT_YUVJ420P &&
            avctx->pix_fmt != AV_PIX_FMT_YUVJ422P &&
            avctx->pix_fmt != AV_PIX_FMT_YUVJ444P &&
            avctx->pix_fmt != AV_PIX_FMT_BGRA     &&
            ((avctx->pix_fmt != AV_PIX_FMT_YUV420P &&
              avctx->pix_fmt != AV_PIX_FMT_YUV422P &&
              avctx->pix_fmt != AV_PIX_FMT_YUV444P) ||
297
             avctx->strict_std_compliance > FF_COMPLIANCE_UNOFFICIAL)) {
298 299 300 301
            av_log(avctx, AV_LOG_ERROR, "colorspace not supported in LJPEG\n");
            return -1;
        }
        break;
302
    case AV_CODEC_ID_MJPEG:
303 304 305 306
        if (avctx->pix_fmt != AV_PIX_FMT_YUVJ420P &&
            avctx->pix_fmt != AV_PIX_FMT_YUVJ422P &&
            ((avctx->pix_fmt != AV_PIX_FMT_YUV420P &&
              avctx->pix_fmt != AV_PIX_FMT_YUV422P) ||
307
             avctx->strict_std_compliance > FF_COMPLIANCE_UNOFFICIAL)) {
308 309 310 311 312
            av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
            return -1;
        }
        break;
    default:
313
        if (avctx->pix_fmt != AV_PIX_FMT_YUV420P) {
314 315 316 317 318 319
            av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
            return -1;
        }
    }

    switch (avctx->pix_fmt) {
320 321
    case AV_PIX_FMT_YUVJ422P:
    case AV_PIX_FMT_YUV422P:
322 323
        s->chroma_format = CHROMA_422;
        break;
324 325
    case AV_PIX_FMT_YUVJ420P:
    case AV_PIX_FMT_YUV420P:
326 327 328 329 330 331
    default:
        s->chroma_format = CHROMA_420;
        break;
    }

    s->bit_rate = avctx->bit_rate;
332 333 334 335 336 337 338 339 340 341 342 343 344 345
    s->width    = avctx->width;
    s->height   = avctx->height;
    if (avctx->gop_size > 600 &&
        avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL) {
        av_log(avctx, AV_LOG_ERROR,
               "Warning keyframe interval too large! reducing it ...\n");
        avctx->gop_size = 600;
    }
    s->gop_size     = avctx->gop_size;
    s->avctx        = avctx;
    s->flags        = avctx->flags;
    s->flags2       = avctx->flags2;
    s->max_b_frames = avctx->max_b_frames;
    s->codec_id     = avctx->codec->id;
346 347 348 349 350 351
#if FF_API_MPV_GLOBAL_OPTS
    if (avctx->luma_elim_threshold)
        s->luma_elim_threshold   = avctx->luma_elim_threshold;
    if (avctx->chroma_elim_threshold)
        s->chroma_elim_threshold = avctx->chroma_elim_threshold;
#endif
352 353 354 355 356
    s->strict_std_compliance = avctx->strict_std_compliance;
    s->quarter_sample     = (avctx->flags & CODEC_FLAG_QPEL) != 0;
    s->mpeg_quant         = avctx->mpeg_quant;
    s->rtp_mode           = !!avctx->rtp_payload_size;
    s->intra_dc_precision = avctx->intra_dc_precision;
357 358 359 360
    s->user_specified_pts = AV_NOPTS_VALUE;

    if (s->gop_size <= 1) {
        s->intra_only = 1;
361
        s->gop_size   = 12;
362 363 364 365 366 367 368 369 370
    } else {
        s->intra_only = 0;
    }

    s->me_method = avctx->me_method;

    /* Fixed QSCALE */
    s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);

371 372 373 374 375
#if FF_API_MPV_GLOBAL_OPTS
    if (s->flags & CODEC_FLAG_QP_RD)
        s->mpv_flags |= FF_MPV_FLAG_QP_RD;
#endif

376 377 378 379 380 381
    s->adaptive_quant = (s->avctx->lumi_masking ||
                         s->avctx->dark_masking ||
                         s->avctx->temporal_cplx_masking ||
                         s->avctx->spatial_cplx_masking  ||
                         s->avctx->p_masking      ||
                         s->avctx->border_masking ||
382
                         (s->mpv_flags & FF_MPV_FLAG_QP_RD)) &&
383
                        !s->fixed_qscale;
384

385
    s->loop_filter      = !!(s->flags & CODEC_FLAG_LOOP_FILTER);
386

387 388 389 390
    if (avctx->rc_max_rate && !avctx->rc_buffer_size) {
        av_log(avctx, AV_LOG_ERROR,
               "a vbv buffer size is needed, "
               "for encoding with a maximum bitrate\n");
391 392 393
        return -1;
    }

394 395 396
    if (avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate) {
        av_log(avctx, AV_LOG_INFO,
               "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
397 398
    }

399
    if (avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate) {
400 401 402 403
        av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
        return -1;
    }

404
    if (avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate) {
405 406 407 408
        av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n");
        return -1;
    }

409 410 411 412 413
    if (avctx->rc_max_rate &&
        avctx->rc_max_rate == avctx->bit_rate &&
        avctx->rc_max_rate != avctx->rc_min_rate) {
        av_log(avctx, AV_LOG_INFO,
               "impossible bitrate constraints, this will fail\n");
414 415
    }

416 417 418
    if (avctx->rc_buffer_size &&
        avctx->bit_rate * (int64_t)avctx->time_base.num >
            avctx->rc_buffer_size * (int64_t)avctx->time_base.den) {
419 420 421 422
        av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
        return -1;
    }

423 424 425 426 427
    if (!s->fixed_qscale &&
        avctx->bit_rate * av_q2d(avctx->time_base) >
            avctx->bit_rate_tolerance) {
        av_log(avctx, AV_LOG_ERROR,
               "bitrate tolerance too small for bitrate\n");
428 429 430
        return -1;
    }

431 432
    if (s->avctx->rc_max_rate &&
        s->avctx->rc_min_rate == s->avctx->rc_max_rate &&
433 434
        (s->codec_id == AV_CODEC_ID_MPEG1VIDEO ||
         s->codec_id == AV_CODEC_ID_MPEG2VIDEO) &&
435 436 437 438 439
        90000LL * (avctx->rc_buffer_size - 1) >
            s->avctx->rc_max_rate * 0xFFFFLL) {
        av_log(avctx, AV_LOG_INFO,
               "Warning vbv_delay will be set to 0xFFFF (=VBR) as the "
               "specified vbv buffer is too large for the given bitrate!\n");
440 441
    }

442 443 444
    if ((s->flags & CODEC_FLAG_4MV)  && s->codec_id != AV_CODEC_ID_MPEG4 &&
        s->codec_id != AV_CODEC_ID_H263 && s->codec_id != AV_CODEC_ID_H263P &&
        s->codec_id != AV_CODEC_ID_FLV1) {
445 446 447 448
        av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
        return -1;
    }

449 450 451
    if (s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE) {
        av_log(avctx, AV_LOG_ERROR,
               "OBMC is only supported with simple mb decision\n");
452 453 454
        return -1;
    }

455
    if (s->quarter_sample && s->codec_id != AV_CODEC_ID_MPEG4) {
456 457 458 459
        av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
        return -1;
    }

460
    if (s->max_b_frames                    &&
461 462 463
        s->codec_id != AV_CODEC_ID_MPEG4      &&
        s->codec_id != AV_CODEC_ID_MPEG1VIDEO &&
        s->codec_id != AV_CODEC_ID_MPEG2VIDEO) {
464 465 466 467
        av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
        return -1;
    }

468 469 470
    if ((s->codec_id == AV_CODEC_ID_MPEG4 ||
         s->codec_id == AV_CODEC_ID_H263  ||
         s->codec_id == AV_CODEC_ID_H263P) &&
471 472 473 474
        (avctx->sample_aspect_ratio.num > 255 ||
         avctx->sample_aspect_ratio.den > 255)) {
        av_log(avctx, AV_LOG_ERROR,
               "Invalid pixel aspect ratio %i/%i, limit is 255/255\n",
475 476 477 478
               avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den);
        return -1;
    }

479
    if ((s->flags & (CODEC_FLAG_INTERLACED_DCT | CODEC_FLAG_INTERLACED_ME)) &&
480
        s->codec_id != AV_CODEC_ID_MPEG4 && s->codec_id != AV_CODEC_ID_MPEG2VIDEO) {
481 482 483 484
        av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
        return -1;
    }

485
    // FIXME mpeg2 uses that too
486
    if (s->mpeg_quant && s->codec_id != AV_CODEC_ID_MPEG4) {
487 488
        av_log(avctx, AV_LOG_ERROR,
               "mpeg2 style quantization not supported by codec\n");
489 490 491
        return -1;
    }

492 493 494 495 496 497
#if FF_API_MPV_GLOBAL_OPTS
    if (s->flags & CODEC_FLAG_CBP_RD)
        s->mpv_flags |= FF_MPV_FLAG_CBP_RD;
#endif

    if ((s->mpv_flags & FF_MPV_FLAG_CBP_RD) && !avctx->trellis) {
498 499 500 501
        av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
        return -1;
    }

502
    if ((s->mpv_flags & FF_MPV_FLAG_QP_RD) &&
503
        s->avctx->mb_decision != FF_MB_DECISION_RD) {
504 505 506 507
        av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
        return -1;
    }

508 509 510 511 512
    if (s->avctx->scenechange_threshold < 1000000000 &&
        (s->flags & CODEC_FLAG_CLOSED_GOP)) {
        av_log(avctx, AV_LOG_ERROR,
               "closed gop with scene change detection are not supported yet, "
               "set threshold to 1000000000\n");
513 514 515
        return -1;
    }

516
    if (s->flags & CODEC_FLAG_LOW_DELAY) {
517
        if (s->codec_id != AV_CODEC_ID_MPEG2VIDEO) {
518 519
            av_log(avctx, AV_LOG_ERROR,
                  "low delay forcing is only available for mpeg2\n");
520 521
            return -1;
        }
522 523 524
        if (s->max_b_frames != 0) {
            av_log(avctx, AV_LOG_ERROR,
                   "b frames cannot be used with low delay\n");
525 526 527 528
            return -1;
        }
    }

529 530 531 532
    if (s->q_scale_type == 1) {
        if (avctx->qmax > 12) {
            av_log(avctx, AV_LOG_ERROR,
                   "non linear quant only supports qmax <= 12 currently\n");
533 534 535 536
            return -1;
        }
    }

537
    if (s->avctx->thread_count > 1         &&
538 539 540 541
        s->codec_id != AV_CODEC_ID_MPEG4      &&
        s->codec_id != AV_CODEC_ID_MPEG1VIDEO &&
        s->codec_id != AV_CODEC_ID_MPEG2VIDEO &&
        (s->codec_id != AV_CODEC_ID_H263P)) {
542 543
        av_log(avctx, AV_LOG_ERROR,
               "multi threaded encoding not supported by codec\n");
544 545 546
        return -1;
    }

547 548 549 550
    if (s->avctx->thread_count < 1) {
        av_log(avctx, AV_LOG_ERROR,
               "automatic thread number detection not supported by codec,"
               "patch welcome\n");
551 552 553
        return -1;
    }

554 555
    if (s->avctx->thread_count > 1)
        s->rtp_mode = 1;
556

557
    if (!avctx->time_base.den || !avctx->time_base.num) {
558 559 560 561
        av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
        return -1;
    }

562 563 564 565
    i = (INT_MAX / 2 + 128) >> 8;
    if (avctx->me_threshold >= i) {
        av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n",
               i - 1);
566 567
        return -1;
    }
568 569 570
    if (avctx->mb_threshold >= i) {
        av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n",
               i - 1);
571 572 573
        return -1;
    }

574 575 576
    if (avctx->b_frame_strategy && (avctx->flags & CODEC_FLAG_PASS2)) {
        av_log(avctx, AV_LOG_INFO,
               "notice: b_frame_strategy only affects the first pass\n");
577 578 579
        avctx->b_frame_strategy = 0;
    }

580 581
    i = av_gcd(avctx->time_base.den, avctx->time_base.num);
    if (i > 1) {
582 583 584
        av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
        avctx->time_base.den /= i;
        avctx->time_base.num /= i;
585
        //return -1;
586 587
    }

588 589
    if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG1VIDEO ||
        s->codec_id == AV_CODEC_ID_MPEG2VIDEO || s->codec_id == AV_CODEC_ID_MJPEG) {
590 591 592 593 594 595 596
        // (a + x * 3 / 8) / x
        s->intra_quant_bias = 3 << (QUANT_BIAS_SHIFT - 3);
        s->inter_quant_bias = 0;
    } else {
        s->intra_quant_bias = 0;
        // (a - x / 4) / x
        s->inter_quant_bias = -(1 << (QUANT_BIAS_SHIFT - 2));
597 598
    }

599 600 601 602 603 604 605 606
    if (avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
        s->intra_quant_bias = avctx->intra_quant_bias;
    if (avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
        s->inter_quant_bias = avctx->inter_quant_bias;

    avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift,
                                  &chroma_v_shift);

607
    if (avctx->codec_id == AV_CODEC_ID_MPEG4 &&
608 609 610 611 612 613
        s->avctx->time_base.den > (1 << 16) - 1) {
        av_log(avctx, AV_LOG_ERROR,
               "timebase %d/%d not supported by MPEG 4 standard, "
               "the maximum admitted value for the timebase denominator "
               "is %d\n", s->avctx->time_base.num, s->avctx->time_base.den,
               (1 << 16) - 1);
614 615 616 617
        return -1;
    }
    s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;

618 619 620
#if FF_API_MPV_GLOBAL_OPTS
    if (avctx->flags2 & CODEC_FLAG2_SKIP_RD)
        s->mpv_flags |= FF_MPV_FLAG_SKIP_RD;
621 622
    if (avctx->flags2 & CODEC_FLAG2_STRICT_GOP)
        s->mpv_flags |= FF_MPV_FLAG_STRICT_GOP;
623 624
    if (avctx->quantizer_noise_shaping)
        s->quantizer_noise_shaping = avctx->quantizer_noise_shaping;
625 626
#endif

627
    switch (avctx->codec->id) {
628
    case AV_CODEC_ID_MPEG1VIDEO:
629
        s->out_format = FMT_MPEG1;
630 631
        s->low_delay  = !!(s->flags & CODEC_FLAG_LOW_DELAY);
        avctx->delay  = s->low_delay ? 0 : (s->max_b_frames + 1);
632
        break;
633
    case AV_CODEC_ID_MPEG2VIDEO:
634
        s->out_format = FMT_MPEG1;
635 636 637
        s->low_delay  = !!(s->flags & CODEC_FLAG_LOW_DELAY);
        avctx->delay  = s->low_delay ? 0 : (s->max_b_frames + 1);
        s->rtp_mode   = 1;
638
        break;
639 640
    case AV_CODEC_ID_LJPEG:
    case AV_CODEC_ID_MJPEG:
641 642
        s->out_format = FMT_MJPEG;
        s->intra_only = 1; /* force intra only for jpeg */
643
        if (avctx->codec->id == AV_CODEC_ID_LJPEG &&
644
            avctx->pix_fmt   == AV_PIX_FMT_BGRA) {
645 646 647
            s->mjpeg_vsample[0] = s->mjpeg_hsample[0] =
            s->mjpeg_vsample[1] = s->mjpeg_hsample[1] =
            s->mjpeg_vsample[2] = s->mjpeg_hsample[2] = 1;
648
        } else {
649
            s->mjpeg_vsample[0] = 2;
650 651
            s->mjpeg_vsample[1] = 2 >> chroma_v_shift;
            s->mjpeg_vsample[2] = 2 >> chroma_v_shift;
652
            s->mjpeg_hsample[0] = 2;
653 654
            s->mjpeg_hsample[1] = 2 >> chroma_h_shift;
            s->mjpeg_hsample[2] = 2 >> chroma_h_shift;
655
        }
656 657
        if (!(CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) ||
            ff_mjpeg_encode_init(s) < 0)
658
            return -1;
659 660
        avctx->delay = 0;
        s->low_delay = 1;
661
        break;
662
    case AV_CODEC_ID_H261:
663 664
        if (!CONFIG_H261_ENCODER)
            return -1;
665
        if (ff_h261_get_picture_format(s->width, s->height) < 0) {
666 667 668 669
            av_log(avctx, AV_LOG_ERROR,
                   "The specified picture size of %dx%d is not valid for the "
                   "H.261 codec.\nValid sizes are 176x144, 352x288\n",
                    s->width, s->height);
670 671 672
            return -1;
        }
        s->out_format = FMT_H261;
673 674
        avctx->delay  = 0;
        s->low_delay  = 1;
675
        break;
676
    case AV_CODEC_ID_H263:
677 678
        if (!CONFIG_H263_ENCODER)
        return -1;
679
        if (ff_match_2uint16(ff_h263_format, FF_ARRAY_ELEMS(ff_h263_format),
680 681 682 683 684 685
                             s->width, s->height) == 8) {
            av_log(avctx, AV_LOG_INFO,
                   "The specified picture size of %dx%d is not valid for "
                   "the H.263 codec.\nValid sizes are 128x96, 176x144, "
                   "352x288, 704x576, and 1408x1152."
                   "Try H.263+.\n", s->width, s->height);
686 687 688
            return -1;
        }
        s->out_format = FMT_H263;
689 690
        avctx->delay  = 0;
        s->low_delay  = 1;
691
        break;
692
    case AV_CODEC_ID_H263P:
693
        s->out_format = FMT_H263;
694
        s->h263_plus  = 1;
695
        /* Fx */
696 697 698 699
        s->h263_aic        = (avctx->flags & CODEC_FLAG_AC_PRED) ? 1 : 0;
        s->modified_quant  = s->h263_aic;
        s->loop_filter     = (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1 : 0;
        s->unrestricted_mv = s->obmc || s->loop_filter || s->umvplus;
700 701 702

        /* /Fx */
        /* These are just to be sure */
703 704
        avctx->delay = 0;
        s->low_delay = 1;
705
        break;
706
    case AV_CODEC_ID_FLV1:
707 708
        s->out_format      = FMT_H263;
        s->h263_flv        = 2; /* format = 1; 11-bit codes */
709
        s->unrestricted_mv = 1;
710 711 712
        s->rtp_mode  = 0; /* don't allow GOB */
        avctx->delay = 0;
        s->low_delay = 1;
713
        break;
714
    case AV_CODEC_ID_RV10:
715
        s->out_format = FMT_H263;
716 717
        avctx->delay  = 0;
        s->low_delay  = 1;
718
        break;
719
    case AV_CODEC_ID_RV20:
720 721 722 723 724 725 726 727
        s->out_format      = FMT_H263;
        avctx->delay       = 0;
        s->low_delay       = 1;
        s->modified_quant  = 1;
        s->h263_aic        = 1;
        s->h263_plus       = 1;
        s->loop_filter     = 1;
        s->unrestricted_mv = 0;
728
        break;
729
    case AV_CODEC_ID_MPEG4:
730 731
        s->out_format      = FMT_H263;
        s->h263_pred       = 1;
732
        s->unrestricted_mv = 1;
733 734
        s->low_delay       = s->max_b_frames ? 0 : 1;
        avctx->delay       = s->low_delay ? 0 : (s->max_b_frames + 1);
735
        break;
736
    case AV_CODEC_ID_MSMPEG4V2:
737 738
        s->out_format      = FMT_H263;
        s->h263_pred       = 1;
739
        s->unrestricted_mv = 1;
740 741 742
        s->msmpeg4_version = 2;
        avctx->delay       = 0;
        s->low_delay       = 1;
743
        break;
744
    case AV_CODEC_ID_MSMPEG4V3:
745 746 747 748 749 750 751
        s->out_format        = FMT_H263;
        s->h263_pred         = 1;
        s->unrestricted_mv   = 1;
        s->msmpeg4_version   = 3;
        s->flipflop_rounding = 1;
        avctx->delay         = 0;
        s->low_delay         = 1;
752
        break;
753
    case AV_CODEC_ID_WMV1:
754 755 756 757 758 759 760
        s->out_format        = FMT_H263;
        s->h263_pred         = 1;
        s->unrestricted_mv   = 1;
        s->msmpeg4_version   = 4;
        s->flipflop_rounding = 1;
        avctx->delay         = 0;
        s->low_delay         = 1;
761
        break;
762
    case AV_CODEC_ID_WMV2:
763 764 765 766 767 768 769
        s->out_format        = FMT_H263;
        s->h263_pred         = 1;
        s->unrestricted_mv   = 1;
        s->msmpeg4_version   = 5;
        s->flipflop_rounding = 1;
        avctx->delay         = 0;
        s->low_delay         = 1;
770 771 772 773 774
        break;
    default:
        return -1;
    }

775
    avctx->has_b_frames = !s->low_delay;
776 777 778

    s->encoding = 1;

779 780
    s->progressive_frame    =
    s->progressive_sequence = !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT |
781 782
                                                CODEC_FLAG_INTERLACED_ME) ||
                                s->alternate_scan);
783

784
    /* init */
785
    if (ff_MPV_common_init(s) < 0)
786 787
        return -1;

788 789 790
    if (ARCH_X86)
        ff_MPV_encode_init_x86(s);

791
    if (!s->dct_quantize)
792
        s->dct_quantize = ff_dct_quantize_c;
793 794
    if (!s->denoise_dct)
        s->denoise_dct  = denoise_dct_c;
795
    s->fast_dct_quantize = s->dct_quantize;
796 797
    if (avctx->trellis)
        s->dct_quantize  = dct_quantize_trellis_c;
798

799 800
    if ((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
        s->chroma_qscale_table = ff_h263_chroma_qscale_table;
801

802
    s->quant_precision = 5;
803 804 805 806

    ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
    ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);

807
    if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
808
        ff_h261_encode_init(s);
809
    if (CONFIG_H263_ENCODER && s->out_format == FMT_H263)
810
        ff_h263_encode_init(s);
811
    if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
812
        ff_msmpeg4_encode_init(s);
813
    if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
814
        && s->out_format == FMT_MPEG1)
815 816 817
        ff_mpeg1_encode_init(s);

    /* init q matrix */
818 819
    for (i = 0; i < 64; i++) {
        int j = s->dsp.idct_permutation[i];
820
        if (CONFIG_MPEG4_ENCODER && s->codec_id == AV_CODEC_ID_MPEG4 &&
821
            s->mpeg_quant) {
822 823
            s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
            s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
824
        } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) {
825 826
            s->intra_matrix[j] =
            s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
827 828
        } else {
            /* mpeg1/2 */
829 830 831
            s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
            s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
        }
832
        if (s->avctx->intra_matrix)
833
            s->intra_matrix[j] = s->avctx->intra_matrix[i];
834
        if (s->avctx->inter_matrix)
835 836 837 838 839 840
            s->inter_matrix[j] = s->avctx->inter_matrix[i];
    }

    /* precompute matrix */
    /* for mjpeg, we do include qscale in the matrix */
    if (s->out_format != FMT_MJPEG) {
841
        ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
842 843
                          s->intra_matrix, s->intra_quant_bias, avctx->qmin,
                          31, 1);
844
        ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
845 846
                          s->inter_matrix, s->inter_quant_bias, avctx->qmin,
                          31, 0);
847 848
    }

849
    if (ff_rate_control_init(s) < 0)
850 851 852 853 854
        return -1;

    return 0;
}

855
av_cold int ff_MPV_encode_end(AVCodecContext *avctx)
856 857 858 859 860
{
    MpegEncContext *s = avctx->priv_data;

    ff_rate_control_uninit(s);

861
    ff_MPV_common_end(s);
862 863
    if ((CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) &&
        s->out_format == FMT_MJPEG)
864 865 866 867 868 869 870
        ff_mjpeg_encode_close(s);

    av_freep(&avctx->extradata);

    return 0;
}

871 872
static int get_sae(uint8_t *src, int ref, int stride)
{
873
    int x,y;
874
    int acc = 0;
875

876 877 878
    for (y = 0; y < 16; y++) {
        for (x = 0; x < 16; x++) {
            acc += FFABS(src[x + y * stride] - ref);
879 880 881 882 883 884
        }
    }

    return acc;
}

885 886 887
static int get_intra_count(MpegEncContext *s, uint8_t *src,
                           uint8_t *ref, int stride)
{
888
    int x, y, w, h;
889
    int acc = 0;
890

891 892
    w = s->width  & ~15;
    h = s->height & ~15;
893

894 895 896 897 898 899 900
    for (y = 0; y < h; y += 16) {
        for (x = 0; x < w; x += 16) {
            int offset = x + y * stride;
            int sad  = s->dsp.sad[0](NULL, src + offset, ref + offset, stride,
                                     16);
            int mean = (s->dsp.pix_sum(src + offset, stride) + 128) >> 8;
            int sae  = get_sae(src + offset, mean, stride);
901

902
            acc += sae + 500 < sad;
903 904 905 906 907 908
        }
    }
    return acc;
}


909 910 911
static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg)
{
    AVFrame *pic = NULL;
912 913
    int64_t pts;
    int i;
914 915
    const int encoding_delay = s->max_b_frames ? s->max_b_frames :
                                                 (s->low_delay ? 0 : 1);
916 917 918 919 920 921 922 923 924 925 926 927 928 929 930
    int direct = 1;

    if (pic_arg) {
        pts = pic_arg->pts;
        pic_arg->display_picture_number = s->input_picture_number++;

        if (pts != AV_NOPTS_VALUE) {
            if (s->user_specified_pts != AV_NOPTS_VALUE) {
                int64_t time = pts;
                int64_t last = s->user_specified_pts;

                if (time <= last) {
                    av_log(s->avctx, AV_LOG_ERROR,
                           "Error, Invalid timestamp=%"PRId64", "
                           "last=%"PRId64"\n", pts, s->user_specified_pts);
931 932
                    return -1;
                }
933 934 935

                if (!s->low_delay && pic_arg->display_picture_number == 1)
                    s->dts_delta = time - last;
936
            }
937 938 939 940 941 942 943 944 945 946
            s->user_specified_pts = pts;
        } else {
            if (s->user_specified_pts != AV_NOPTS_VALUE) {
                s->user_specified_pts =
                pts = s->user_specified_pts + 1;
                av_log(s->avctx, AV_LOG_INFO,
                       "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n",
                       pts);
            } else {
                pts = pic_arg->display_picture_number;
947 948 949 950
            }
        }
    }

951 952 953 954 955 956 957 958 959 960
  if (pic_arg) {
    if (encoding_delay && !(s->flags & CODEC_FLAG_INPUT_PRESERVED))
        direct = 0;
    if (pic_arg->linesize[0] != s->linesize)
        direct = 0;
    if (pic_arg->linesize[1] != s->uvlinesize)
        direct = 0;
    if (pic_arg->linesize[2] != s->uvlinesize)
        direct = 0;

961 962
    av_dlog(s->avctx, "%d %d %d %d\n", pic_arg->linesize[0],
            pic_arg->linesize[1], s->linesize, s->uvlinesize);
963 964 965

    if (direct) {
        i = ff_find_unused_picture(s, 1);
966 967
        if (i < 0)
            return i;
968

969
        pic = &s->picture[i].f;
970
        pic->reference = 3;
971

972 973 974
        for (i = 0; i < 4; i++) {
            pic->data[i]     = pic_arg->data[i];
            pic->linesize[i] = pic_arg->linesize[i];
975
        }
976
        if (ff_alloc_picture(s, (Picture *) pic, 1) < 0) {
977 978
            return -1;
        }
979 980
    } else {
        i = ff_find_unused_picture(s, 0);
981 982
        if (i < 0)
            return i;
983

984
        pic = &s->picture[i].f;
985
        pic->reference = 3;
986

987
        if (ff_alloc_picture(s, (Picture *) pic, 0) < 0) {
988 989
            return -1;
        }
990

991 992 993 994 995
        if (pic->data[0] + INPLACE_OFFSET == pic_arg->data[0] &&
            pic->data[1] + INPLACE_OFFSET == pic_arg->data[1] &&
            pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]) {
            // empty
        } else {
996
            int h_chroma_shift, v_chroma_shift;
997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016
            avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift,
                                          &v_chroma_shift);

            for (i = 0; i < 3; i++) {
                int src_stride = pic_arg->linesize[i];
                int dst_stride = i ? s->uvlinesize : s->linesize;
                int h_shift = i ? h_chroma_shift : 0;
                int v_shift = i ? v_chroma_shift : 0;
                int w = s->width  >> h_shift;
                int h = s->height >> v_shift;
                uint8_t *src = pic_arg->data[i];
                uint8_t *dst = pic->data[i];

                if (!s->avctx->rc_buffer_size)
                    dst += INPLACE_OFFSET;

                if (src_stride == dst_stride)
                    memcpy(dst, src, src_stride * h);
                else {
                    while (h--) {
1017 1018 1019 1020 1021 1022 1023 1024 1025
                        memcpy(dst, src, w);
                        dst += dst_stride;
                        src += src_stride;
                    }
                }
            }
        }
    }
    copy_picture_attributes(s, pic, pic_arg);
1026
    pic->pts = pts; // we set this here to avoid modifiying pic_arg
1027 1028 1029
  }

    /* shift buffer entries */
1030 1031
    for (i = 1; i < MAX_PICTURE_COUNT /*s->encoding_delay + 1*/; i++)
        s->input_picture[i - 1] = s->input_picture[i];
1032

1033
    s->input_picture[encoding_delay] = (Picture*) pic;
1034 1035 1036 1037

    return 0;
}

1038 1039
static int skip_check(MpegEncContext *s, Picture *p, Picture *ref)
{
1040
    int x, y, plane;
1041 1042
    int score = 0;
    int64_t score64 = 0;
1043

1044
    for (plane = 0; plane < 3; plane++) {
1045
        const int stride = p->f.linesize[plane];
1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059
        const int bw = plane ? 1 : 2;
        for (y = 0; y < s->mb_height * bw; y++) {
            for (x = 0; x < s->mb_width * bw; x++) {
                int off = p->f.type == FF_BUFFER_TYPE_SHARED ? 0 : 16;
                uint8_t *dptr = p->f.data[plane] + 8 * (x + y * stride) + off;
                uint8_t *rptr = ref->f.data[plane] + 8 * (x + y * stride);
                int v   = s->dsp.frame_skip_cmp[1](s, dptr, rptr, stride, 8);

                switch (s->avctx->frame_skip_exp) {
                case 0: score    =  FFMAX(score, v);          break;
                case 1: score   += FFABS(v);                  break;
                case 2: score   += v * v;                     break;
                case 3: score64 += FFABS(v * v * (int64_t)v); break;
                case 4: score64 += v * v * (int64_t)(v * v);  break;
1060 1061 1062 1063 1064
                }
            }
        }
    }

1065 1066
    if (score)
        score64 = score;
1067

1068
    if (score64 < s->avctx->frame_skip_threshold)
1069
        return 1;
1070
    if (score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda) >> 8))
1071 1072 1073 1074
        return 1;
    return 0;
}

1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089
static int encode_frame(AVCodecContext *c, AVFrame *frame)
{
    AVPacket pkt = { 0 };
    int ret, got_output;

    av_init_packet(&pkt);
    ret = avcodec_encode_video2(c, &pkt, frame, &got_output);
    if (ret < 0)
        return ret;

    ret = pkt.size;
    av_free_packet(&pkt);
    return ret;
}

1090 1091
static int estimate_best_b_count(MpegEncContext *s)
{
1092
    AVCodec *codec    = avcodec_find_encoder(s->avctx->codec_id);
1093
    AVCodecContext *c = avcodec_alloc_context3(NULL);
1094 1095
    AVFrame input[FF_MAX_B_FRAMES + 2];
    const int scale = s->avctx->brd_scale;
1096
    int i, j, out_size, p_lambda, b_lambda, lambda2;
1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120
    int64_t best_rd  = INT64_MAX;
    int best_b_count = -1;

    assert(scale >= 0 && scale <= 3);

    //emms_c();
    //s->next_picture_ptr->quality;
    p_lambda = s->last_lambda_for[AV_PICTURE_TYPE_P];
    //p_lambda * FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
    b_lambda = s->last_lambda_for[AV_PICTURE_TYPE_B];
    if (!b_lambda) // FIXME we should do this somewhere else
        b_lambda = p_lambda;
    lambda2  = (b_lambda * b_lambda + (1 << FF_LAMBDA_SHIFT) / 2) >>
               FF_LAMBDA_SHIFT;

    c->width        = s->width  >> scale;
    c->height       = s->height >> scale;
    c->flags        = CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR |
                      CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
    c->flags       |= s->avctx->flags & CODEC_FLAG_QPEL;
    c->mb_decision  = s->avctx->mb_decision;
    c->me_cmp       = s->avctx->me_cmp;
    c->mb_cmp       = s->avctx->mb_cmp;
    c->me_sub_cmp   = s->avctx->me_sub_cmp;
1121
    c->pix_fmt      = AV_PIX_FMT_YUV420P;
1122 1123
    c->time_base    = s->avctx->time_base;
    c->max_b_frames = s->max_b_frames;
1124

1125
    if (avcodec_open2(c, codec, NULL) < 0)
1126 1127
        return -1;

1128 1129 1130 1131 1132
    for (i = 0; i < s->max_b_frames + 2; i++) {
        int ysize = c->width * c->height;
        int csize = (c->width / 2) * (c->height / 2);
        Picture pre_input, *pre_input_ptr = i ? s->input_picture[i - 1] :
                                                s->next_picture_ptr;
1133 1134

        avcodec_get_frame_defaults(&input[i]);
1135 1136 1137 1138 1139 1140
        input[i].data[0]     = av_malloc(ysize + 2 * csize);
        input[i].data[1]     = input[i].data[0] + ysize;
        input[i].data[2]     = input[i].data[1] + csize;
        input[i].linesize[0] = c->width;
        input[i].linesize[1] =
        input[i].linesize[2] = c->width / 2;
1141

1142 1143
        if (pre_input_ptr && (!i || s->input_picture[i - 1])) {
            pre_input = *pre_input_ptr;
1144

1145 1146 1147 1148
            if (pre_input.f.type != FF_BUFFER_TYPE_SHARED && i) {
                pre_input.f.data[0] += INPLACE_OFFSET;
                pre_input.f.data[1] += INPLACE_OFFSET;
                pre_input.f.data[2] += INPLACE_OFFSET;
1149 1150
            }

1151 1152 1153 1154 1155 1156 1157 1158 1159
            s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0],
                                 pre_input.f.data[0], pre_input.f.linesize[0],
                                 c->width,      c->height);
            s->dsp.shrink[scale](input[i].data[1], input[i].linesize[1],
                                 pre_input.f.data[1], pre_input.f.linesize[1],
                                 c->width >> 1, c->height >> 1);
            s->dsp.shrink[scale](input[i].data[2], input[i].linesize[2],
                                 pre_input.f.data[2], pre_input.f.linesize[2],
                                 c->width >> 1, c->height >> 1);
1160 1161 1162
        }
    }

1163 1164
    for (j = 0; j < s->max_b_frames + 1; j++) {
        int64_t rd = 0;
1165

1166
        if (!s->input_picture[j])
1167 1168
            break;

1169
        c->error[0] = c->error[1] = c->error[2] = 0;
1170

1171 1172
        input[0].pict_type = AV_PICTURE_TYPE_I;
        input[0].quality   = 1 * FF_QP2LAMBDA;
1173 1174 1175

        out_size = encode_frame(c, &input[0]);

1176
        //rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
1177

1178 1179
        for (i = 0; i < s->max_b_frames + 1; i++) {
            int is_p = i % (j + 1) == j || i == s->max_b_frames;
1180

1181 1182 1183
            input[i + 1].pict_type = is_p ?
                                     AV_PICTURE_TYPE_P : AV_PICTURE_TYPE_B;
            input[i + 1].quality   = is_p ? p_lambda : b_lambda;
1184 1185 1186

            out_size = encode_frame(c, &input[i + 1]);

1187 1188 1189 1190
            rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
        }

        /* get the delayed frames */
1191
        while (out_size) {
1192
            out_size = encode_frame(c, NULL);
1193 1194 1195 1196 1197
            rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
        }

        rd += c->error[0] + c->error[1] + c->error[2];

1198 1199 1200
        if (rd < best_rd) {
            best_rd = rd;
            best_b_count = j;
1201 1202 1203 1204 1205 1206
        }
    }

    avcodec_close(c);
    av_freep(&c);

1207
    for (i = 0; i < s->max_b_frames + 2; i++) {
1208 1209 1210 1211 1212 1213
        av_freep(&input[i].data[0]);
    }

    return best_b_count;
}

1214 1215
static int select_input_picture(MpegEncContext *s)
{
1216 1217
    int i;

1218 1219 1220
    for (i = 1; i < MAX_PICTURE_COUNT; i++)
        s->reordered_input_picture[i - 1] = s->reordered_input_picture[i];
    s->reordered_input_picture[MAX_PICTURE_COUNT - 1] = NULL;
1221 1222

    /* set next picture type & ordering */
1223 1224 1225 1226
    if (s->reordered_input_picture[0] == NULL && s->input_picture[0]) {
        if (/*s->picture_in_gop_number >= s->gop_size ||*/
            s->next_picture_ptr == NULL || s->intra_only) {
            s->reordered_input_picture[0] = s->input_picture[0];
1227
            s->reordered_input_picture[0]->f.pict_type = AV_PICTURE_TYPE_I;
1228 1229 1230
            s->reordered_input_picture[0]->f.coded_picture_number =
                s->coded_picture_number++;
        } else {
1231 1232
            int b_frames;

1233 1234 1235 1236
            if (s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor) {
                if (s->picture_in_gop_number < s->gop_size &&
                    skip_check(s, s->input_picture[0], s->next_picture_ptr)) {
                    // FIXME check that te gop check above is +-1 correct
1237
                    if (s->input_picture[0]->f.type == FF_BUFFER_TYPE_SHARED) {
1238
                        for (i = 0; i < 4; i++)
1239 1240
                            s->input_picture[0]->f.data[i] = NULL;
                        s->input_picture[0]->f.type = 0;
1241 1242 1243
                    } else {
                        assert(s->input_picture[0]->f.type == FF_BUFFER_TYPE_USER ||
                               s->input_picture[0]->f.type == FF_BUFFER_TYPE_INTERNAL);
1244

1245
                        s->avctx->release_buffer(s->avctx,
1246
                                                 &s->input_picture[0]->f);
1247 1248 1249 1250 1251 1252 1253 1254 1255
                    }

                    emms_c();
                    ff_vbv_update(s, 0);

                    goto no_output_pic;
                }
            }

1256 1257
            if (s->flags & CODEC_FLAG_PASS2) {
                for (i = 0; i < s->max_b_frames + 1; i++) {
1258
                    int pict_num = s->input_picture[0]->f.display_picture_number + i;
1259

1260
                    if (pict_num >= s->rc_context.num_entries)
1261
                        break;
1262 1263
                    if (!s->input_picture[i]) {
                        s->rc_context.entry[pict_num - 1].new_pict_type = AV_PICTURE_TYPE_P;
1264 1265 1266
                        break;
                    }

1267
                    s->input_picture[i]->f.pict_type =
1268 1269 1270 1271
                        s->rc_context.entry[pict_num].new_pict_type;
                }
            }

1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284
            if (s->avctx->b_frame_strategy == 0) {
                b_frames = s->max_b_frames;
                while (b_frames && !s->input_picture[b_frames])
                    b_frames--;
            } else if (s->avctx->b_frame_strategy == 1) {
                for (i = 1; i < s->max_b_frames + 1; i++) {
                    if (s->input_picture[i] &&
                        s->input_picture[i]->b_frame_score == 0) {
                        s->input_picture[i]->b_frame_score =
                            get_intra_count(s,
                                            s->input_picture[i    ]->f.data[0],
                                            s->input_picture[i - 1]->f.data[0],
                                            s->linesize) + 1;
1285 1286
                    }
                }
1287 1288 1289 1290 1291
                for (i = 0; i < s->max_b_frames + 1; i++) {
                    if (s->input_picture[i] == NULL ||
                        s->input_picture[i]->b_frame_score - 1 >
                            s->mb_num / s->avctx->b_sensitivity)
                        break;
1292 1293
                }

1294
                b_frames = FFMAX(0, i - 1);
1295 1296

                /* reset scores */
1297 1298
                for (i = 0; i < b_frames + 1; i++) {
                    s->input_picture[i]->b_frame_score = 0;
1299
                }
1300 1301 1302
            } else if (s->avctx->b_frame_strategy == 2) {
                b_frames = estimate_best_b_count(s);
            } else {
1303
                av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
1304
                b_frames = 0;
1305 1306 1307 1308
            }

            emms_c();

1309
            for (i = b_frames - 1; i >= 0; i--) {
1310
                int type = s->input_picture[i]->f.pict_type;
1311 1312
                if (type && type != AV_PICTURE_TYPE_B)
                    b_frames = i;
1313
            }
1314 1315 1316 1317
            if (s->input_picture[b_frames]->f.pict_type == AV_PICTURE_TYPE_B &&
                b_frames == s->max_b_frames) {
                av_log(s->avctx, AV_LOG_ERROR,
                       "warning, too many b frames in a row\n");
1318 1319
            }

1320
            if (s->picture_in_gop_number + b_frames >= s->gop_size) {
1321
                if ((s->mpv_flags & FF_MPV_FLAG_STRICT_GOP) &&
1322 1323 1324 1325 1326 1327 1328
                    s->gop_size > s->picture_in_gop_number) {
                    b_frames = s->gop_size - s->picture_in_gop_number - 1;
                } else {
                    if (s->flags & CODEC_FLAG_CLOSED_GOP)
                        b_frames = 0;
                    s->input_picture[b_frames]->f.pict_type = AV_PICTURE_TYPE_I;
                }
1329 1330
            }

1331 1332
            if ((s->flags & CODEC_FLAG_CLOSED_GOP) && b_frames &&
                s->input_picture[b_frames]->f.pict_type == AV_PICTURE_TYPE_I)
1333 1334
                b_frames--;

1335
            s->reordered_input_picture[0] = s->input_picture[b_frames];
1336 1337
            if (s->reordered_input_picture[0]->f.pict_type != AV_PICTURE_TYPE_I)
                s->reordered_input_picture[0]->f.pict_type = AV_PICTURE_TYPE_P;
1338 1339 1340
            s->reordered_input_picture[0]->f.coded_picture_number =
                s->coded_picture_number++;
            for (i = 0; i < b_frames; i++) {
1341
                s->reordered_input_picture[i + 1] = s->input_picture[i];
1342 1343 1344 1345
                s->reordered_input_picture[i + 1]->f.pict_type =
                    AV_PICTURE_TYPE_B;
                s->reordered_input_picture[i + 1]->f.coded_picture_number =
                    s->coded_picture_number++;
1346 1347 1348 1349
            }
        }
    }
no_output_pic:
1350 1351 1352 1353
    if (s->reordered_input_picture[0]) {
        s->reordered_input_picture[0]->f.reference =
           s->reordered_input_picture[0]->f.pict_type !=
               AV_PICTURE_TYPE_B ? 3 : 0;
1354

1355
        ff_copy_picture(&s->new_picture, s->reordered_input_picture[0]);
1356

1357 1358 1359 1360
        if (s->reordered_input_picture[0]->f.type == FF_BUFFER_TYPE_SHARED ||
            s->avctx->rc_buffer_size) {
            // input is a shared pix, so we can't modifiy it -> alloc a new
            // one & ensure that the shared one is reuseable
1361

1362
            Picture *pic;
1363
            int i = ff_find_unused_picture(s, 0);
1364 1365
            if (i < 0)
                return i;
1366
            pic = &s->picture[i];
1367

1368
            pic->f.reference = s->reordered_input_picture[0]->f.reference;
1369
            if (ff_alloc_picture(s, pic, 0) < 0) {
1370 1371
                return -1;
            }
1372 1373

            /* mark us unused / free shared pic */
1374
            if (s->reordered_input_picture[0]->f.type == FF_BUFFER_TYPE_INTERNAL)
1375
                s->avctx->release_buffer(s->avctx,
1376
                                         &s->reordered_input_picture[0]->f);
1377
            for (i = 0; i < 4; i++)
1378 1379
                s->reordered_input_picture[0]->f.data[i] = NULL;
            s->reordered_input_picture[0]->f.type = 0;
1380

1381 1382
            copy_picture_attributes(s, &pic->f,
                                    &s->reordered_input_picture[0]->f);
1383

1384 1385
            s->current_picture_ptr = pic;
        } else {
1386 1387
            // input is not a shared pix -> reuse buffer for current_pix

1388 1389 1390 1391
            assert(s->reordered_input_picture[0]->f.type ==
                       FF_BUFFER_TYPE_USER ||
                   s->reordered_input_picture[0]->f.type ==
                       FF_BUFFER_TYPE_INTERNAL);
1392

1393 1394
            s->current_picture_ptr = s->reordered_input_picture[0];
            for (i = 0; i < 4; i++) {
1395
                s->new_picture.f.data[i] += INPLACE_OFFSET;
1396 1397
            }
        }
1398
        ff_copy_picture(&s->current_picture, s->current_picture_ptr);
1399

1400
        s->picture_number = s->new_picture.f.display_picture_number;
1401 1402
    } else {
        memset(&s->new_picture, 0, sizeof(Picture));
1403
    }
1404
    return 0;
1405 1406
}

1407 1408
int ff_MPV_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
                          const AVFrame *pic_arg, int *got_packet)
1409 1410
{
    MpegEncContext *s = avctx->priv_data;
1411
    int i, stuffing_count, ret;
1412
    int context_count = s->slice_context_count;
1413 1414 1415

    s->picture_in_gop_number++;

1416
    if (load_input_picture(s, pic_arg) < 0)
1417 1418
        return -1;

1419
    if (select_input_picture(s) < 0) {
1420 1421
        return -1;
    }
1422 1423

    /* output? */
1424
    if (s->new_picture.f.data[0]) {
1425 1426 1427
        if (!pkt->data &&
            (ret = ff_alloc_packet(pkt, s->mb_width*s->mb_height*MAX_MB_BYTES)) < 0)
            return ret;
1428 1429 1430 1431 1432 1433
        if (s->mb_info) {
            s->mb_info_ptr = av_packet_new_side_data(pkt,
                                 AV_PKT_DATA_H263_MB_INFO,
                                 s->mb_width*s->mb_height*12);
            s->prev_mb_info = s->last_mb_info = s->mb_info_size = 0;
        }
1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444

        for (i = 0; i < context_count; i++) {
            int start_y = s->thread_context[i]->start_mb_y;
            int   end_y = s->thread_context[i]->  end_mb_y;
            int h       = s->mb_height;
            uint8_t *start = pkt->data + (size_t)(((int64_t) pkt->size) * start_y / h);
            uint8_t *end   = pkt->data + (size_t)(((int64_t) pkt->size) *   end_y / h);

            init_put_bits(&s->thread_context[i]->pb, start, end - start);
        }

1445
        s->pict_type = s->new_picture.f.pict_type;
1446
        //emms_c();
1447
        ff_MPV_frame_start(s, avctx);
1448 1449 1450 1451 1452 1453 1454 1455 1456 1457
vbv_retry:
        if (encode_picture(s, s->picture_number) < 0)
            return -1;

        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;
1458 1459
        // FIXME f/b_count in avctx
        avctx->p_count     = s->mb_num - s->i_count - s->skip_count;
1460 1461
        avctx->skip_count  = s->skip_count;

1462
        ff_MPV_frame_end(s);
1463

1464
        if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1465 1466
            ff_mjpeg_encode_picture_trailer(s);

1467 1468 1469
        if (avctx->rc_buffer_size) {
            RateControlContext *rcc = &s->rc_context;
            int max_size = rcc->buffer_index * avctx->rc_max_available_vbv_use;
1470

1471 1472 1473 1474 1475
            if (put_bits_count(&s->pb) > max_size &&
                s->lambda < s->avctx->lmax) {
                s->next_lambda = FFMAX(s->lambda + 1, s->lambda *
                                       (s->qscale + 1) / s->qscale);
                if (s->adaptive_quant) {
1476
                    int i;
1477 1478 1479 1480 1481
                    for (i = 0; i < s->mb_height * s->mb_stride; i++)
                        s->lambda_table[i] =
                            FFMAX(s->lambda_table[i] + 1,
                                  s->lambda_table[i] * (s->qscale + 1) /
                                  s->qscale);
1482
                }
1483 1484 1485 1486
                s->mb_skipped = 0;        // done in MPV_frame_start()
                // done in encode_picture() so we must undo it
                if (s->pict_type == AV_PICTURE_TYPE_P) {
                    if (s->flipflop_rounding          ||
1487 1488
                        s->codec_id == AV_CODEC_ID_H263P ||
                        s->codec_id == AV_CODEC_ID_MPEG4)
1489 1490
                        s->no_rounding ^= 1;
                }
1491 1492 1493
                if (s->pict_type != AV_PICTURE_TYPE_B) {
                    s->time_base       = s->last_time_base;
                    s->last_non_b_time = s->time - s->pp_time;
1494
                }
1495 1496
                for (i = 0; i < context_count; i++) {
                    PutBitContext *pb = &s->thread_context[i]->pb;
1497 1498 1499 1500 1501 1502 1503 1504
                    init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
                }
                goto vbv_retry;
            }

            assert(s->avctx->rc_max_rate);
        }

1505
        if (s->flags & CODEC_FLAG_PASS1)
1506 1507
            ff_write_pass1_stats(s);

1508 1509 1510
        for (i = 0; i < 4; i++) {
            s->current_picture_ptr->f.error[i] = s->current_picture.f.error[i];
            avctx->error[i] += s->current_picture_ptr->f.error[i];
1511 1512
        }

1513 1514 1515 1516
        if (s->flags & CODEC_FLAG_PASS1)
            assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits +
                   avctx->i_tex_bits + avctx->p_tex_bits ==
                       put_bits_count(&s->pb));
1517 1518 1519
        flush_put_bits(&s->pb);
        s->frame_bits  = put_bits_count(&s->pb);

1520 1521 1522 1523
        stuffing_count = ff_vbv_update(s, s->frame_bits);
        if (stuffing_count) {
            if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) <
                    stuffing_count + 50) {
1524 1525 1526 1527
                av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
                return -1;
            }

1528
            switch (s->codec_id) {
1529 1530
            case AV_CODEC_ID_MPEG1VIDEO:
            case AV_CODEC_ID_MPEG2VIDEO:
1531
                while (stuffing_count--) {
1532 1533 1534
                    put_bits(&s->pb, 8, 0);
                }
            break;
1535
            case AV_CODEC_ID_MPEG4:
1536 1537 1538
                put_bits(&s->pb, 16, 0);
                put_bits(&s->pb, 16, 0x1C3);
                stuffing_count -= 4;
1539
                while (stuffing_count--) {
1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550
                    put_bits(&s->pb, 8, 0xFF);
                }
            break;
            default:
                av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
            }
            flush_put_bits(&s->pb);
            s->frame_bits  = put_bits_count(&s->pb);
        }

        /* update mpeg1/2 vbv_delay for CBR */
1551 1552 1553 1554 1555
        if (s->avctx->rc_max_rate                          &&
            s->avctx->rc_min_rate == s->avctx->rc_max_rate &&
            s->out_format == FMT_MPEG1                     &&
            90000LL * (avctx->rc_buffer_size - 1) <=
                s->avctx->rc_max_rate * 0xFFFFLL) {
1556
            int vbv_delay, min_delay;
1557 1558 1559 1560 1561
            double inbits  = s->avctx->rc_max_rate *
                             av_q2d(s->avctx->time_base);
            int    minbits = s->frame_bits - 8 *
                             (s->vbv_delay_ptr - s->pb.buf - 1);
            double bits    = s->rc_context.buffer_index + minbits - inbits;
1562

1563 1564 1565
            if (bits < 0)
                av_log(s->avctx, AV_LOG_ERROR,
                       "Internal error, negative bits\n");
1566

1567
            assert(s->repeat_first_field == 0);
1568

1569 1570 1571
            vbv_delay = bits * 90000 / s->avctx->rc_max_rate;
            min_delay = (minbits * 90000LL + s->avctx->rc_max_rate - 1) /
                        s->avctx->rc_max_rate;
1572

1573
            vbv_delay = FFMAX(vbv_delay, min_delay);
1574

1575 1576 1577
            assert(vbv_delay < 0xFFFF);

            s->vbv_delay_ptr[0] &= 0xF8;
1578 1579
            s->vbv_delay_ptr[0] |= vbv_delay >> 13;
            s->vbv_delay_ptr[1]  = vbv_delay >> 5;
1580
            s->vbv_delay_ptr[2] &= 0x07;
1581 1582
            s->vbv_delay_ptr[2] |= vbv_delay << 3;
            avctx->vbv_delay     = vbv_delay * 300;
1583
        }
1584
        s->total_bits     += s->frame_bits;
1585
        avctx->frame_bits  = s->frame_bits;
1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597

        pkt->pts = s->current_picture.f.pts;
        if (!s->low_delay) {
            if (!s->current_picture.f.coded_picture_number)
                pkt->dts = pkt->pts - s->dts_delta;
            else
                pkt->dts = s->reordered_pts;
            s->reordered_pts = s->input_picture[0]->f.pts;
        } else
            pkt->dts = pkt->pts;
        if (s->current_picture.f.key_frame)
            pkt->flags |= AV_PKT_FLAG_KEY;
1598 1599
        if (s->mb_info)
            av_packet_shrink_side_data(pkt, AV_PKT_DATA_H263_MB_INFO, s->mb_info_size);
1600 1601
    } else {
        s->frame_bits = 0;
1602
    }
1603
    assert((s->frame_bits & 7) == 0);
1604

1605 1606 1607
    pkt->size = s->frame_bits / 8;
    *got_packet = !!pkt->size;
    return 0;
1608 1609
}

1610 1611
static inline void dct_single_coeff_elimination(MpegEncContext *s,
                                                int n, int threshold)
1612
{
1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624
    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;
1625
    int i;
1626 1627
    DCTELEM *block = s->block[n];
    const int last_index = s->block_last_index[n];
1628 1629
    int skip_dc;

1630 1631 1632 1633 1634
    if (threshold < 0) {
        skip_dc = 0;
        threshold = -threshold;
    } else
        skip_dc = 1;
1635

Diego Biurrun's avatar
Diego Biurrun committed
1636
    /* Are all we could set to zero already zero? */
1637 1638
    if (last_index <= skip_dc - 1)
        return;
1639

1640
    for (i = 0; i <= last_index; i++) {
1641 1642
        const int j = s->intra_scantable.permutated[i];
        const int level = FFABS(block[j]);
1643 1644 1645 1646 1647 1648
        if (level == 1) {
            if (skip_dc && i == 0)
                continue;
            score += tab[run];
            run = 0;
        } else if (level > 1) {
1649
            return;
1650
        } else {
1651 1652 1653
            run++;
        }
    }
1654 1655 1656
    if (score >= threshold)
        return;
    for (i = skip_dc; i <= last_index; i++) {
1657
        const int j = s->intra_scantable.permutated[i];
1658
        block[j] = 0;
1659
    }
1660 1661 1662 1663
    if (block[0])
        s->block_last_index[n] = 0;
    else
        s->block_last_index[n] = -1;
1664 1665
}

1666 1667
static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block,
                               int last_index)
1668 1669
{
    int i;
1670 1671 1672
    const int maxlevel = s->max_qcoeff;
    const int minlevel = s->min_qcoeff;
    int overflow = 0;
1673

1674 1675 1676 1677
    if (s->mb_intra) {
        i = 1; // skip clipping of intra dc
    } else
        i = 0;
1678

1679 1680
    for (; i <= last_index; i++) {
        const int j = s->intra_scantable.permutated[i];
1681 1682
        int level = block[j];

1683 1684
        if (level > maxlevel) {
            level = maxlevel;
1685
            overflow++;
1686 1687
        } else if (level < minlevel) {
            level = minlevel;
1688 1689 1690
            overflow++;
        }

1691
        block[j] = level;
1692 1693
    }

1694 1695 1696 1697
    if (overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
        av_log(s->avctx, AV_LOG_INFO,
               "warning, clipping %d dct coefficients to %d..%d\n",
               overflow, minlevel, maxlevel);
1698 1699
}

1700 1701
static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride)
{
1702
    int x, y;
1703 1704 1705
    // FIXME optimize
    for (y = 0; y < 8; y++) {
        for (x = 0; x < 8; x++) {
1706
            int x2, y2;
1707 1708 1709
            int sum = 0;
            int sqr = 0;
            int count = 0;
1710

1711 1712 1713
            for (y2 = FFMAX(y - 1, 0); y2 < FFMIN(8, y + 2); y2++) {
                for (x2= FFMAX(x - 1, 0); x2 < FFMIN(8, x + 2); x2++) {
                    int v = ptr[x2 + y2 * stride];
1714
                    sum += v;
1715
                    sqr += v * v;
1716 1717 1718
                    count++;
                }
            }
1719
            weight[x + 8 * y]= (36 * ff_sqrt(count * sqr - sum * sum)) / count;
1720 1721 1722 1723
        }
    }
}

1724 1725 1726 1727
static av_always_inline void encode_mb_internal(MpegEncContext *s,
                                                int motion_x, int motion_y,
                                                int mb_block_height,
                                                int mb_block_count)
1728 1729 1730
{
    int16_t weight[8][64];
    DCTELEM orig[8][64];
1731 1732
    const int mb_x = s->mb_x;
    const int mb_y = s->mb_y;
1733 1734
    int i;
    int skip_dct[8];
1735
    int dct_offset = s->linesize * 8; // default for progressive frames
1736 1737 1738
    uint8_t *ptr_y, *ptr_cb, *ptr_cr;
    int wrap_y, wrap_c;

1739 1740
    for (i = 0; i < mb_block_count; i++)
        skip_dct[i] = s->skipdct;
1741

1742 1743 1744
    if (s->adaptive_quant) {
        const int last_qp = s->qscale;
        const int mb_xy = mb_x + mb_y * s->mb_stride;
1745

1746
        s->lambda = s->lambda_table[mb_xy];
1747 1748
        update_qscale(s);

1749
        if (!(s->mpv_flags & FF_MPV_FLAG_QP_RD)) {
1750
            s->qscale = s->current_picture_ptr->f.qscale_table[mb_xy];
1751
            s->dquant = s->qscale - last_qp;
1752

1753 1754
            if (s->out_format == FMT_H263) {
                s->dquant = av_clip(s->dquant, -2, 2);
1755

1756
                if (s->codec_id == AV_CODEC_ID_MPEG4) {
1757 1758 1759 1760
                    if (!s->mb_intra) {
                        if (s->pict_type == AV_PICTURE_TYPE_B) {
                            if (s->dquant & 1 || s->mv_dir & MV_DIRECT)
                                s->dquant = 0;
1761
                        }
1762 1763
                        if (s->mv_type == MV_TYPE_8X8)
                            s->dquant = 0;
1764 1765 1766 1767 1768
                    }
                }
            }
        }
        ff_set_qscale(s, last_qp + s->dquant);
1769
    } else if (s->mpv_flags & FF_MPV_FLAG_QP_RD)
1770 1771 1772 1773
        ff_set_qscale(s, s->qscale + s->dquant);

    wrap_y = s->linesize;
    wrap_c = s->uvlinesize;
1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793
    ptr_y  = s->new_picture.f.data[0] +
             (mb_y * 16 * wrap_y)              + mb_x * 16;
    ptr_cb = s->new_picture.f.data[1] +
             (mb_y * mb_block_height * wrap_c) + mb_x * 8;
    ptr_cr = s->new_picture.f.data[2] +
             (mb_y * mb_block_height * wrap_c) + mb_x * 8;

    if (mb_x * 16 + 16 > s->width || mb_y * 16 + 16 > s->height) {
        uint8_t *ebuf = s->edge_emu_buffer + 32;
        s->dsp.emulated_edge_mc(ebuf, ptr_y, wrap_y, 16, 16, mb_x * 16,
                                mb_y * 16, s->width, s->height);
        ptr_y = ebuf;
        s->dsp.emulated_edge_mc(ebuf + 18 * wrap_y, ptr_cb, wrap_c, 8,
                                mb_block_height, mb_x * 8, mb_y * 8,
                                s->width >> 1, s->height >> 1);
        ptr_cb = ebuf + 18 * wrap_y;
        s->dsp.emulated_edge_mc(ebuf + 18 * wrap_y + 8, ptr_cr, wrap_c, 8,
                                mb_block_height, mb_x * 8, mb_y * 8,
                                s->width >> 1, s->height >> 1);
        ptr_cr = ebuf + 18 * wrap_y + 8;
1794 1795 1796
    }

    if (s->mb_intra) {
1797
        if (s->flags & CODEC_FLAG_INTERLACED_DCT) {
1798 1799
            int progressive_score, interlaced_score;

1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815
            s->interlaced_dct = 0;
            progressive_score = s->dsp.ildct_cmp[4](s, ptr_y,
                                                    NULL, wrap_y, 8) +
                                s->dsp.ildct_cmp[4](s, ptr_y + wrap_y * 8,
                                                    NULL, wrap_y, 8) - 400;

            if (progressive_score > 0) {
                interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y,
                                                       NULL, wrap_y * 2, 8) +
                                   s->dsp.ildct_cmp[4](s, ptr_y + wrap_y,
                                                       NULL, wrap_y * 2, 8);
                if (progressive_score > interlaced_score) {
                    s->interlaced_dct = 1;

                    dct_offset = wrap_y;
                    wrap_y <<= 1;
1816
                    if (s->chroma_format == CHROMA_422)
1817
                        wrap_c <<= 1;
1818 1819 1820 1821
                }
            }
        }

1822 1823 1824 1825
        s->dsp.get_pixels(s->block[0], ptr_y                  , wrap_y);
        s->dsp.get_pixels(s->block[1], ptr_y              + 8 , wrap_y);
        s->dsp.get_pixels(s->block[2], ptr_y + dct_offset     , wrap_y);
        s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8 , wrap_y);
1826

1827 1828 1829 1830
        if (s->flags & CODEC_FLAG_GRAY) {
            skip_dct[4] = 1;
            skip_dct[5] = 1;
        } else {
1831 1832
            s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
            s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
1833 1834 1835 1836 1837
            if (!s->chroma_y_shift) { /* 422 */
                s->dsp.get_pixels(s->block[6],
                                  ptr_cb + (dct_offset >> 1), wrap_c);
                s->dsp.get_pixels(s->block[7],
                                  ptr_cr + (dct_offset >> 1), wrap_c);
1838 1839
            }
        }
1840
    } else {
1841 1842 1843 1844 1845 1846 1847 1848
        op_pixels_func (*op_pix)[4];
        qpel_mc_func (*op_qpix)[16];
        uint8_t *dest_y, *dest_cb, *dest_cr;

        dest_y  = s->dest[0];
        dest_cb = s->dest[1];
        dest_cr = s->dest[2];

1849 1850 1851 1852 1853 1854
        if ((!s->no_rounding) || s->pict_type == AV_PICTURE_TYPE_B) {
            op_pix  = s->dsp.put_pixels_tab;
            op_qpix = s->dsp.put_qpel_pixels_tab;
        } else {
            op_pix  = s->dsp.put_no_rnd_pixels_tab;
            op_qpix = s->dsp.put_no_rnd_qpel_pixels_tab;
1855 1856 1857
        }

        if (s->mv_dir & MV_DIR_FORWARD) {
1858 1859 1860
            ff_MPV_motion(s, dest_y, dest_cb, dest_cr, 0,
                          s->last_picture.f.data,
                          op_pix, op_qpix);
1861 1862
            op_pix  = s->dsp.avg_pixels_tab;
            op_qpix = s->dsp.avg_qpel_pixels_tab;
1863 1864
        }
        if (s->mv_dir & MV_DIR_BACKWARD) {
1865 1866 1867
            ff_MPV_motion(s, dest_y, dest_cb, dest_cr, 1,
                          s->next_picture.f.data,
                          op_pix, op_qpix);
1868 1869
        }

1870
        if (s->flags & CODEC_FLAG_INTERLACED_DCT) {
1871 1872
            int progressive_score, interlaced_score;

1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896
            s->interlaced_dct = 0;
            progressive_score = s->dsp.ildct_cmp[0](s, dest_y,
                                                    ptr_y,              wrap_y,
                                                    8) +
                                s->dsp.ildct_cmp[0](s, dest_y + wrap_y * 8,
                                                    ptr_y + wrap_y * 8, wrap_y,
                                                    8) - 400;

            if (s->avctx->ildct_cmp == FF_CMP_VSSE)
                progressive_score -= 400;

            if (progressive_score > 0) {
                interlaced_score = s->dsp.ildct_cmp[0](s, dest_y,
                                                       ptr_y,
                                                       wrap_y * 2, 8) +
                                   s->dsp.ildct_cmp[0](s, dest_y + wrap_y,
                                                       ptr_y + wrap_y,
                                                       wrap_y * 2, 8);

                if (progressive_score > interlaced_score) {
                    s->interlaced_dct = 1;

                    dct_offset = wrap_y;
                    wrap_y <<= 1;
1897
                    if (s->chroma_format == CHROMA_422)
1898
                        wrap_c <<= 1;
1899 1900 1901 1902
                }
            }
        }

1903 1904 1905 1906 1907 1908
        s->dsp.diff_pixels(s->block[0], ptr_y, dest_y, wrap_y);
        s->dsp.diff_pixels(s->block[1], ptr_y + 8, dest_y + 8, wrap_y);
        s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset,
                           dest_y + dct_offset, wrap_y);
        s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8,
                           dest_y + dct_offset + 8, wrap_y);
1909

1910 1911 1912 1913
        if (s->flags & CODEC_FLAG_GRAY) {
            skip_dct[4] = 1;
            skip_dct[5] = 1;
        } else {
1914 1915
            s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
            s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1916 1917 1918 1919 1920
            if (!s->chroma_y_shift) { /* 422 */
                s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset >> 1),
                                   dest_cb + (dct_offset >> 1), wrap_c);
                s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset >> 1),
                                   dest_cr + (dct_offset >> 1), wrap_c);
1921 1922 1923
            }
        }
        /* pre quantization */
1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954
        if (s->current_picture.mc_mb_var[s->mb_stride * mb_y + mb_x] <
                2 * s->qscale * s->qscale) {
            // FIXME optimize
            if (s->dsp.sad[1](NULL, ptr_y , dest_y,
                              wrap_y, 8) < 20 * s->qscale)
                skip_dct[0] = 1;
            if (s->dsp.sad[1](NULL, ptr_y + 8,
                              dest_y + 8, wrap_y, 8) < 20 * s->qscale)
                skip_dct[1] = 1;
            if (s->dsp.sad[1](NULL, ptr_y + dct_offset,
                              dest_y + dct_offset, wrap_y, 8) < 20 * s->qscale)
                skip_dct[2] = 1;
            if (s->dsp.sad[1](NULL, ptr_y + dct_offset + 8,
                              dest_y + dct_offset + 8,
                              wrap_y, 8) < 20 * s->qscale)
                skip_dct[3] = 1;
            if (s->dsp.sad[1](NULL, ptr_cb, dest_cb,
                              wrap_c, 8) < 20 * s->qscale)
                skip_dct[4] = 1;
            if (s->dsp.sad[1](NULL, ptr_cr, dest_cr,
                              wrap_c, 8) < 20 * s->qscale)
                skip_dct[5] = 1;
            if (!s->chroma_y_shift) { /* 422 */
                if (s->dsp.sad[1](NULL, ptr_cb + (dct_offset >> 1),
                                  dest_cb + (dct_offset >> 1),
                                  wrap_c, 8) < 20 * s->qscale)
                    skip_dct[6] = 1;
                if (s->dsp.sad[1](NULL, ptr_cr + (dct_offset >> 1),
                                  dest_cr + (dct_offset >> 1),
                                  wrap_c, 8) < 20 * s->qscale)
                    skip_dct[7] = 1;
1955 1956 1957 1958
            }
        }
    }

1959
    if (s->quantizer_noise_shaping) {
1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978
        if (!skip_dct[0])
            get_visual_weight(weight[0], ptr_y                 , wrap_y);
        if (!skip_dct[1])
            get_visual_weight(weight[1], ptr_y              + 8, wrap_y);
        if (!skip_dct[2])
            get_visual_weight(weight[2], ptr_y + dct_offset    , wrap_y);
        if (!skip_dct[3])
            get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
        if (!skip_dct[4])
            get_visual_weight(weight[4], ptr_cb                , wrap_c);
        if (!skip_dct[5])
            get_visual_weight(weight[5], ptr_cr                , wrap_c);
        if (!s->chroma_y_shift) { /* 422 */
            if (!skip_dct[6])
                get_visual_weight(weight[6], ptr_cb + (dct_offset >> 1),
                                  wrap_c);
            if (!skip_dct[7])
                get_visual_weight(weight[7], ptr_cr + (dct_offset >> 1),
                                  wrap_c);
1979
        }
1980
        memcpy(orig[0], s->block[0], sizeof(DCTELEM) * 64 * mb_block_count);
1981 1982 1983
    }

    /* DCT & quantize */
1984
    assert(s->out_format != FMT_MJPEG || s->qscale == 8);
1985
    {
1986 1987
        for (i = 0; i < mb_block_count; i++) {
            if (!skip_dct[i]) {
1988 1989
                int overflow;
                s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1990 1991 1992 1993 1994 1995 1996 1997 1998
                // FIXME we could decide to change to quantizer instead of
                // clipping
                // 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
                if (overflow)
                    clip_coeffs(s, s->block[i], s->block_last_index[i]);
            } else
                s->block_last_index[i] = -1;
1999
        }
2000
        if (s->quantizer_noise_shaping) {
2001 2002 2003 2004 2005
            for (i = 0; i < mb_block_count; i++) {
                if (!skip_dct[i]) {
                    s->block_last_index[i] =
                        dct_quantize_refine(s, s->block[i], weight[i],
                                            orig[i], i, s->qscale);
2006 2007 2008 2009
                }
            }
        }

2010 2011
        if (s->luma_elim_threshold && !s->mb_intra)
            for (i = 0; i < 4; i++)
2012
                dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
2013 2014
        if (s->chroma_elim_threshold && !s->mb_intra)
            for (i = 4; i < mb_block_count; i++)
2015 2016
                dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);

2017
        if (s->mpv_flags & FF_MPV_FLAG_CBP_RD) {
2018 2019 2020
            for (i = 0; i < mb_block_count; i++) {
                if (s->block_last_index[i] == -1)
                    s->coded_score[i] = INT_MAX / 256;
2021 2022 2023 2024
            }
        }
    }

2025 2026 2027 2028 2029
    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] = (1024 + s->c_dc_scale / 2) / s->c_dc_scale;
2030 2031
    }

2032
    // non c quantize code returns incorrect block_last_index FIXME
2033
    if (s->alternate_scan && s->dct_quantize != ff_dct_quantize_c) {
2034
        for (i = 0; i < mb_block_count; i++) {
2035
            int j;
2036 2037 2038 2039
            if (s->block_last_index[i] > 0) {
                for (j = 63; j > 0; j--) {
                    if (s->block[i][s->intra_scantable.permutated[j]])
                        break;
2040
                }
2041
                s->block_last_index[i] = j;
2042 2043 2044 2045 2046 2047
            }
        }
    }

    /* huffman encode */
    switch(s->codec_id){ //FIXME funct ptr could be slightly faster
2048 2049
    case AV_CODEC_ID_MPEG1VIDEO:
    case AV_CODEC_ID_MPEG2VIDEO:
2050
        if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
2051
            ff_mpeg1_encode_mb(s, s->block, motion_x, motion_y);
Aurelien Jacobs's avatar
Aurelien Jacobs committed
2052
        break;
2053
    case AV_CODEC_ID_MPEG4:
2054
        if (CONFIG_MPEG4_ENCODER)
2055
            ff_mpeg4_encode_mb(s, s->block, motion_x, motion_y);
Aurelien Jacobs's avatar
Aurelien Jacobs committed
2056
        break;
2057 2058 2059
    case AV_CODEC_ID_MSMPEG4V2:
    case AV_CODEC_ID_MSMPEG4V3:
    case AV_CODEC_ID_WMV1:
2060
        if (CONFIG_MSMPEG4_ENCODER)
2061
            ff_msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
2062
        break;
2063
    case AV_CODEC_ID_WMV2:
2064
        if (CONFIG_WMV2_ENCODER)
2065 2066
            ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
        break;
2067
    case AV_CODEC_ID_H261:
2068
        if (CONFIG_H261_ENCODER)
2069 2070
            ff_h261_encode_mb(s, s->block, motion_x, motion_y);
        break;
2071 2072 2073 2074 2075
    case AV_CODEC_ID_H263:
    case AV_CODEC_ID_H263P:
    case AV_CODEC_ID_FLV1:
    case AV_CODEC_ID_RV10:
    case AV_CODEC_ID_RV20:
2076
        if (CONFIG_H263_ENCODER)
2077
            ff_h263_encode_mb(s, s->block, motion_x, motion_y);
Aurelien Jacobs's avatar
Aurelien Jacobs committed
2078
        break;
2079
    case AV_CODEC_ID_MJPEG:
2080
        if (CONFIG_MJPEG_ENCODER)
2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096
            ff_mjpeg_encode_mb(s, s->block);
        break;
    default:
        assert(0);
    }
}

static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
{
    if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y,  8, 6);
    else                                encode_mb_internal(s, motion_x, motion_y, 16, 8);
}

static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
    int i;

2097
    memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster than a loop?
2098 2099 2100 2101

    /* mpeg1 */
    d->mb_skip_run= s->mb_skip_run;
    for(i=0; i<3; i++)
2102
        d->last_dc[i] = s->last_dc[i];
2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117

    /* 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->f_count= s->f_count;
    d->b_count= s->b_count;
    d->skip_count= s->skip_count;
    d->misc_bits= s->misc_bits;
    d->last_bits= 0;

    d->mb_skipped= 0;
    d->qscale= s->qscale;
    d->dquant= s->dquant;
2118 2119

    d->esc3_level_length= s->esc3_level_length;
2120 2121 2122 2123 2124 2125
}

static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
    int i;

    memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
2126
    memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster than a loop?
2127 2128 2129 2130

    /* mpeg1 */
    d->mb_skip_run= s->mb_skip_run;
    for(i=0; i<3; i++)
2131
        d->last_dc[i] = s->last_dc[i];
2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156

    /* 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->f_count= s->f_count;
    d->b_count= s->b_count;
    d->skip_count= s->skip_count;
    d->misc_bits= s->misc_bits;

    d->mb_intra= s->mb_intra;
    d->mb_skipped= s->mb_skipped;
    d->mv_type= s->mv_type;
    d->mv_dir= s->mv_dir;
    d->pb= s->pb;
    if(s->data_partitioning){
        d->pb2= s->pb2;
        d->tex_pb= s->tex_pb;
    }
    d->block= s->block;
    for(i=0; i<8; i++)
        d->block_last_index[i]= s->block_last_index[i];
    d->interlaced_dct= s->interlaced_dct;
    d->qscale= s->qscale;
2157 2158

    d->esc3_level_length= s->esc3_level_length;
2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193
}

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 score;
    uint8_t *dest_backup[3];

    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];
    }

    if(*next_block){
        memcpy(dest_backup, s->dest, sizeof(s->dest));
        s->dest[0] = s->rd_scratchpad;
        s->dest[1] = s->rd_scratchpad + 16*s->linesize;
        s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
        assert(s->linesize >= 32); //FIXME
    }

    encode_mb(s, motion_x, motion_y);

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

    if(s->avctx->mb_decision == FF_MB_DECISION_RD){
2194
        ff_MPV_decode_mb(s, s->block);
2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241

        score *= s->lambda2;
        score += sse_mb(s) << FF_LAMBDA_SHIFT;
    }

    if(*next_block){
        memcpy(s->dest, dest_backup, sizeof(s->dest));
    }

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

        copy_context_after_encode(best, s, type);
    }
}

static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
    uint32_t *sq = ff_squareTbl + 256;
    int acc=0;
    int x,y;

    if(w==16 && h==16)
        return s->dsp.sse[0](NULL, src1, src2, stride, 16);
    else if(w==8 && h==8)
        return s->dsp.sse[1](NULL, src1, src2, stride, 8);

    for(y=0; y<h; y++){
        for(x=0; x<w; x++){
            acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
        }
    }

    assert(acc>=0);

    return acc;
}

static int sse_mb(MpegEncContext *s){
    int w= 16;
    int h= 16;

    if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
    if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;

    if(w==16 && h==16)
      if(s->avctx->mb_cmp == FF_CMP_NSSE){
2242 2243 2244
        return  s->dsp.nsse[0](s, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
               +s->dsp.nsse[1](s, s->new_picture.f.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
               +s->dsp.nsse[1](s, s->new_picture.f.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
2245
      }else{
2246 2247 2248
        return  s->dsp.sse[0](NULL, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
               +s->dsp.sse[1](NULL, s->new_picture.f.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
               +s->dsp.sse[1](NULL, s->new_picture.f.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
2249 2250
      }
    else
2251 2252 2253
        return  sse(s, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], w, h, s->linesize)
               +sse(s, s->new_picture.f.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], w>>1, h>>1, s->uvlinesize)
               +sse(s, s->new_picture.f.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], w>>1, h>>1, s->uvlinesize);
2254 2255 2256
}

static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
2257
    MpegEncContext *s= *(void**)arg;
2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275


    s->me.pre_pass=1;
    s->me.dia_size= s->avctx->pre_dia_size;
    s->first_slice_line=1;
    for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
        for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
            ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
        }
        s->first_slice_line=0;
    }

    s->me.pre_pass=0;

    return 0;
}

static int estimate_motion_thread(AVCodecContext *c, void *arg){
2276
    MpegEncContext *s= *(void**)arg;
2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291

    ff_check_alignment();

    s->me.dia_size= s->avctx->dia_size;
    s->first_slice_line=1;
    for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
        s->mb_x=0; //for block init below
        ff_init_block_index(s);
        for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
            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 */
2292
            if(s->pict_type==AV_PICTURE_TYPE_B)
2293 2294 2295 2296 2297 2298 2299 2300 2301 2302
                ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
            else
                ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
        }
        s->first_slice_line=0;
    }
    return 0;
}

static int mb_var_thread(AVCodecContext *c, void *arg){
2303
    MpegEncContext *s= *(void**)arg;
2304 2305 2306 2307 2308 2309 2310 2311
    int mb_x, mb_y;

    ff_check_alignment();

    for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
        for(mb_x=0; mb_x < s->mb_width; mb_x++) {
            int xx = mb_x * 16;
            int yy = mb_y * 16;
2312
            uint8_t *pix = s->new_picture.f.data[0] + (yy * s->linesize) + xx;
2313 2314 2315
            int varc;
            int sum = s->dsp.pix_sum(pix, s->linesize);

2316
            varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)sum*sum)>>8) + 500 + 128)>>8;
2317 2318 2319 2320 2321 2322 2323 2324 2325 2326

            s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
            s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
            s->me.mb_var_sum_temp    += varc;
        }
    }
    return 0;
}

static void write_slice_end(MpegEncContext *s){
2327
    if(CONFIG_MPEG4_ENCODER && s->codec_id==AV_CODEC_ID_MPEG4){
2328 2329 2330 2331 2332
        if(s->partitioned_frame){
            ff_mpeg4_merge_partitions(s);
        }

        ff_mpeg4_stuffing(&s->pb);
2333
    }else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
2334 2335 2336
        ff_mjpeg_encode_stuffing(&s->pb);
    }

2337
    avpriv_align_put_bits(&s->pb);
2338 2339 2340 2341 2342 2343
    flush_put_bits(&s->pb);

    if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
        s->misc_bits+= get_bits_diff(s);
}

2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386
static void write_mb_info(MpegEncContext *s)
{
    uint8_t *ptr = s->mb_info_ptr + s->mb_info_size - 12;
    int offset = put_bits_count(&s->pb);
    int mba  = s->mb_x + s->mb_width * (s->mb_y % s->gob_index);
    int gobn = s->mb_y / s->gob_index;
    int pred_x, pred_y;
    if (CONFIG_H263_ENCODER)
        ff_h263_pred_motion(s, 0, 0, &pred_x, &pred_y);
    bytestream_put_le32(&ptr, offset);
    bytestream_put_byte(&ptr, s->qscale);
    bytestream_put_byte(&ptr, gobn);
    bytestream_put_le16(&ptr, mba);
    bytestream_put_byte(&ptr, pred_x); /* hmv1 */
    bytestream_put_byte(&ptr, pred_y); /* vmv1 */
    /* 4MV not implemented */
    bytestream_put_byte(&ptr, 0); /* hmv2 */
    bytestream_put_byte(&ptr, 0); /* vmv2 */
}

static void update_mb_info(MpegEncContext *s, int startcode)
{
    if (!s->mb_info)
        return;
    if (put_bits_count(&s->pb) - s->prev_mb_info*8 >= s->mb_info*8) {
        s->mb_info_size += 12;
        s->prev_mb_info = s->last_mb_info;
    }
    if (startcode) {
        s->prev_mb_info = put_bits_count(&s->pb)/8;
        /* This might have incremented mb_info_size above, and we return without
         * actually writing any info into that slot yet. But in that case,
         * this will be called again at the start of the after writing the
         * start code, actually writing the mb info. */
        return;
    }

    s->last_mb_info = put_bits_count(&s->pb)/8;
    if (!s->mb_info_size)
        s->mb_info_size += 12;
    write_mb_info(s);
}

2387
static int encode_thread(AVCodecContext *c, void *arg){
2388
    MpegEncContext *s= *(void**)arg;
2389
    int mb_x, mb_y, pdif = 0;
Baptiste Coudurier's avatar
Baptiste Coudurier committed
2390
    int chr_h= 16>>s->chroma_y_shift;
2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420
    int i, j;
    MpegEncContext best_s, backup_s;
    uint8_t bit_buf[2][MAX_MB_BYTES];
    uint8_t bit_buf2[2][MAX_MB_BYTES];
    uint8_t bit_buf_tex[2][MAX_MB_BYTES];
    PutBitContext pb[2], pb2[2], tex_pb[2];

    ff_check_alignment();

    for(i=0; i<2; i++){
        init_put_bits(&pb    [i], bit_buf    [i], MAX_MB_BYTES);
        init_put_bits(&pb2   [i], bit_buf2   [i], MAX_MB_BYTES);
        init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
    }

    s->last_bits= put_bits_count(&s->pb);
    s->mv_bits=0;
    s->misc_bits=0;
    s->i_tex_bits=0;
    s->p_tex_bits=0;
    s->i_count=0;
    s->f_count=0;
    s->b_count=0;
    s->skip_count=0;

    for(i=0; i<3; i++){
        /* init last dc values */
        /* note: quant matrix value (8) is implied here */
        s->last_dc[i] = 128 << s->intra_dc_precision;

2421
        s->current_picture.f.error[i] = 0;
2422 2423 2424 2425 2426 2427 2428
    }
    s->mb_skip_run = 0;
    memset(s->last_mv, 0, sizeof(s->last_mv));

    s->last_mv_dir = 0;

    switch(s->codec_id){
2429 2430 2431
    case AV_CODEC_ID_H263:
    case AV_CODEC_ID_H263P:
    case AV_CODEC_ID_FLV1:
2432
        if (CONFIG_H263_ENCODER)
Aurelien Jacobs's avatar
Aurelien Jacobs committed
2433
            s->gob_index = ff_h263_get_gob_height(s);
2434
        break;
2435
    case AV_CODEC_ID_MPEG4:
2436
        if(CONFIG_MPEG4_ENCODER && s->partitioned_frame)
2437 2438 2439 2440 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
            ff_mpeg4_init_partitions(s);
        break;
    }

    s->resync_mb_x=0;
    s->resync_mb_y=0;
    s->first_slice_line = 1;
    s->ptr_lastgob = s->pb.buf;
    for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
        s->mb_x=0;
        s->mb_y= mb_y;

        ff_set_qscale(s, s->qscale);
        ff_init_block_index(s);

        for(mb_x=0; mb_x < s->mb_width; mb_x++) {
            int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
            int mb_type= s->mb_type[xy];
//            int d;
            int dmin= INT_MAX;
            int dir;

            if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
                av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
                return -1;
            }
            if(s->data_partitioning){
                if(   s->pb2   .buf_end - s->pb2   .buf - (put_bits_count(&s->    pb2)>>3) < MAX_MB_BYTES
                   || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
                    av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
                    return -1;
                }
            }

            s->mb_x = mb_x;
            s->mb_y = mb_y;  // moved into loop, can get changed by H.261
            ff_update_block_index(s);

2475
            if(CONFIG_H261_ENCODER && s->codec_id == AV_CODEC_ID_H261){
2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491
                ff_h261_reorder_mb_index(s);
                xy= s->mb_y*s->mb_stride + s->mb_x;
                mb_type= s->mb_type[xy];
            }

            /* write gob / video packet header  */
            if(s->rtp_mode){
                int current_packet_size, is_gob_start;

                current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);

                is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;

                if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;

                switch(s->codec_id){
2492 2493
                case AV_CODEC_ID_H263:
                case AV_CODEC_ID_H263P:
2494 2495 2496
                    if(!s->h263_slice_structured)
                        if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
                    break;
2497
                case AV_CODEC_ID_MPEG2VIDEO:
2498
                    if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
2499
                case AV_CODEC_ID_MPEG1VIDEO:
2500 2501 2502 2503 2504 2505 2506 2507
                    if(s->mb_skip_run) is_gob_start=0;
                    break;
                }

                if(is_gob_start){
                    if(s->start_mb_y != mb_y || mb_x!=0){
                        write_slice_end(s);

2508
                        if(CONFIG_MPEG4_ENCODER && s->codec_id==AV_CODEC_ID_MPEG4 && s->partitioned_frame){
2509 2510 2511 2512 2513
                            ff_mpeg4_init_partitions(s);
                        }
                    }

                    assert((put_bits_count(&s->pb)&7) == 0);
2514
                    current_packet_size= put_bits_ptr(&s->pb) - s->ptr_lastgob;
2515 2516 2517 2518 2519 2520 2521

                    if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
                        int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
                        int d= 100 / s->avctx->error_rate;
                        if(r % d == 0){
                            current_packet_size=0;
                            s->pb.buf_ptr= s->ptr_lastgob;
2522
                            assert(put_bits_ptr(&s->pb) == s->ptr_lastgob);
2523 2524 2525 2526 2527 2528 2529
                        }
                    }

                    if (s->avctx->rtp_callback){
                        int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
                        s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
                    }
2530
                    update_mb_info(s, 1);
2531 2532

                    switch(s->codec_id){
2533
                    case AV_CODEC_ID_MPEG4:
2534
                        if (CONFIG_MPEG4_ENCODER) {
Aurelien Jacobs's avatar
Aurelien Jacobs committed
2535 2536
                            ff_mpeg4_encode_video_packet_header(s);
                            ff_mpeg4_clean_buffers(s);
2537
                        }
2538
                    break;
2539 2540
                    case AV_CODEC_ID_MPEG1VIDEO:
                    case AV_CODEC_ID_MPEG2VIDEO:
2541
                        if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) {
Aurelien Jacobs's avatar
Aurelien Jacobs committed
2542 2543
                            ff_mpeg1_encode_slice_header(s);
                            ff_mpeg1_clean_buffers(s);
2544
                        }
2545
                    break;
2546 2547
                    case AV_CODEC_ID_H263:
                    case AV_CODEC_ID_H263P:
2548
                        if (CONFIG_H263_ENCODER)
2549
                            ff_h263_encode_gob_header(s, mb_y);
2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573
                    break;
                    }

                    if(s->flags&CODEC_FLAG_PASS1){
                        int bits= put_bits_count(&s->pb);
                        s->misc_bits+= bits - s->last_bits;
                        s->last_bits= bits;
                    }

                    s->ptr_lastgob += current_packet_size;
                    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;
            }

            s->mb_skipped=0;
            s->dquant=0; //only for QP_RD

2574 2575
            update_mb_info(s, 0);

2576
            if (mb_type & (mb_type-1) || (s->mpv_flags & FF_MPV_FLAG_QP_RD)) { // more than 1 MB type possible or FF_MPV_FLAG_QP_RD
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
                int next_block=0;
                int pb_bits_count, pb2_bits_count, tex_pb_bits_count;

                copy_context_before_encode(&backup_s, s, -1);
                backup_s.pb= s->pb;
                best_s.data_partitioning= s->data_partitioning;
                best_s.partitioned_frame= s->partitioned_frame;
                if(s->data_partitioning){
                    backup_s.pb2= s->pb2;
                    backup_s.tex_pb= s->tex_pb;
                }

                if(mb_type&CANDIDATE_MB_TYPE_INTER){
                    s->mv_dir = MV_DIR_FORWARD;
                    s->mv_type = MV_TYPE_16X16;
                    s->mb_intra= 0;
                    s->mv[0][0][0] = s->p_mv_table[xy][0];
                    s->mv[0][0][1] = s->p_mv_table[xy][1];
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
                                 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
                }
                if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
                    s->mv_dir = MV_DIR_FORWARD;
                    s->mv_type = MV_TYPE_FIELD;
                    s->mb_intra= 0;
                    for(i=0; i<2; i++){
                        j= s->field_select[0][i] = s->p_field_select_table[i][xy];
                        s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
                        s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
                    }
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
                                 &dmin, &next_block, 0, 0);
                }
                if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
                    s->mv_dir = MV_DIR_FORWARD;
                    s->mv_type = MV_TYPE_16X16;
                    s->mb_intra= 0;
                    s->mv[0][0][0] = 0;
                    s->mv[0][0][1] = 0;
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
                                 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
                }
                if(mb_type&CANDIDATE_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++){
2624 2625
                        s->mv[0][i][0] = s->current_picture.f.motion_val[0][s->block_index[i]][0];
                        s->mv[0][i][1] = s->current_picture.f.motion_val[0][s->block_index[i]][1];
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 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
                    }
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
                                 &dmin, &next_block, 0, 0);
                }
                if(mb_type&CANDIDATE_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];
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
                                 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
                }
                if(mb_type&CANDIDATE_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];
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
                                 &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
                }
                if(mb_type&CANDIDATE_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];
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
                                 &dmin, &next_block, 0, 0);
                }
                if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
                    s->mv_dir = MV_DIR_FORWARD;
                    s->mv_type = MV_TYPE_FIELD;
                    s->mb_intra= 0;
                    for(i=0; i<2; i++){
                        j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
                        s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
                        s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
                    }
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
                                 &dmin, &next_block, 0, 0);
                }
                if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
                    s->mv_dir = MV_DIR_BACKWARD;
                    s->mv_type = MV_TYPE_FIELD;
                    s->mb_intra= 0;
                    for(i=0; i<2; i++){
                        j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
                        s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
                        s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
                    }
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
                                 &dmin, &next_block, 0, 0);
                }
                if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
                    s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
                    s->mv_type = MV_TYPE_FIELD;
                    s->mb_intra= 0;
                    for(dir=0; dir<2; dir++){
                        for(i=0; i<2; i++){
                            j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
                            s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
                            s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
                        }
                    }
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
                                 &dmin, &next_block, 0, 0);
                }
                if(mb_type&CANDIDATE_MB_TYPE_INTRA){
                    s->mv_dir = 0;
                    s->mv_type = MV_TYPE_16X16;
                    s->mb_intra= 1;
                    s->mv[0][0][0] = 0;
                    s->mv[0][0][1] = 0;
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
                                 &dmin, &next_block, 0, 0);
                    if(s->h263_pred || s->h263_aic){
                        if(best_s.mb_intra)
                            s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
                        else
                            ff_clean_intra_table_entries(s); //old mode?
                    }
                }

2713
                if ((s->mpv_flags & FF_MPV_FLAG_QP_RD) && dmin < INT_MAX) {
2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731
                    if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
                        const int last_qp= backup_s.qscale;
                        int qpi, qp, dc[6];
                        DCTELEM ac[6][16];
                        const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
                        static const int dquant_tab[4]={-1,1,-2,2};

                        assert(backup_s.dquant == 0);

                        //FIXME intra
                        s->mv_dir= best_s.mv_dir;
                        s->mv_type = MV_TYPE_16X16;
                        s->mb_intra= best_s.mb_intra;
                        s->mv[0][0][0] = best_s.mv[0][0][0];
                        s->mv[0][0][1] = best_s.mv[0][0][1];
                        s->mv[1][0][0] = best_s.mv[1][0][0];
                        s->mv[1][0][1] = best_s.mv[1][0][1];

2732
                        qpi = s->pict_type == AV_PICTURE_TYPE_B ? 2 : 0;
2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758
                        for(; qpi<4; qpi++){
                            int dquant= dquant_tab[qpi];
                            qp= last_qp + dquant;
                            if(qp < s->avctx->qmin || qp > s->avctx->qmax)
                                continue;
                            backup_s.dquant= dquant;
                            if(s->mb_intra && s->dc_val[0]){
                                for(i=0; i<6; i++){
                                    dc[i]= s->dc_val[0][ s->block_index[i] ];
                                    memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
                                }
                            }

                            encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
                                         &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
                            if(best_s.qscale != qp){
                                if(s->mb_intra && s->dc_val[0]){
                                    for(i=0; i<6; i++){
                                        s->dc_val[0][ s->block_index[i] ]= dc[i];
                                        memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
                                    }
                                }
                            }
                        }
                    }
                }
2759
                if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
2760 2761 2762 2763 2764 2765 2766 2767 2768 2769
                    int mx= s->b_direct_mv_table[xy][0];
                    int my= s->b_direct_mv_table[xy][1];

                    backup_s.dquant = 0;
                    s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
                    s->mb_intra= 0;
                    ff_mpeg4_set_direct_mv(s, mx, my);
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
                                 &dmin, &next_block, mx, my);
                }
2770
                if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
2771 2772 2773 2774 2775 2776 2777
                    backup_s.dquant = 0;
                    s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
                    s->mb_intra= 0;
                    ff_mpeg4_set_direct_mv(s, 0, 0);
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
                                 &dmin, &next_block, 0, 0);
                }
2778
                if (!best_s.mb_intra && s->mpv_flags & FF_MPV_FLAG_SKIP_RD) {
2779 2780 2781 2782 2783 2784
                    int coded=0;
                    for(i=0; i<6; i++)
                        coded |= s->block_last_index[i];
                    if(coded){
                        int mx,my;
                        memcpy(s->mv, best_s.mv, sizeof(s->mv));
2785
                        if(CONFIG_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810
                            mx=my=0; //FIXME find the one we actually used
                            ff_mpeg4_set_direct_mv(s, mx, my);
                        }else if(best_s.mv_dir&MV_DIR_BACKWARD){
                            mx= s->mv[1][0][0];
                            my= s->mv[1][0][1];
                        }else{
                            mx= s->mv[0][0][0];
                            my= s->mv[0][0][1];
                        }

                        s->mv_dir= best_s.mv_dir;
                        s->mv_type = best_s.mv_type;
                        s->mb_intra= 0;
/*                        s->mv[0][0][0] = best_s.mv[0][0][0];
                        s->mv[0][0][1] = best_s.mv[0][0][1];
                        s->mv[1][0][0] = best_s.mv[1][0][0];
                        s->mv[1][0][1] = best_s.mv[1][0][1];*/
                        backup_s.dquant= 0;
                        s->skipdct=1;
                        encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
                                        &dmin, &next_block, mx, my);
                        s->skipdct=0;
                    }
                }

2811
                s->current_picture.f.qscale_table[xy] = best_s.qscale;
2812 2813 2814 2815 2816

                copy_context_after_encode(s, &best_s, -1);

                pb_bits_count= put_bits_count(&s->pb);
                flush_put_bits(&s->pb);
2817
                avpriv_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2818 2819 2820 2821 2822
                s->pb= backup_s.pb;

                if(s->data_partitioning){
                    pb2_bits_count= put_bits_count(&s->pb2);
                    flush_put_bits(&s->pb2);
2823
                    avpriv_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2824 2825 2826 2827
                    s->pb2= backup_s.pb2;

                    tex_pb_bits_count= put_bits_count(&s->tex_pb);
                    flush_put_bits(&s->tex_pb);
2828
                    avpriv_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2829 2830 2831 2832
                    s->tex_pb= backup_s.tex_pb;
                }
                s->last_bits= put_bits_count(&s->pb);

2833
                if (CONFIG_H263_ENCODER &&
2834
                    s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
2835 2836 2837 2838 2839 2840 2841 2842 2843
                    ff_h263_update_motion_val(s);

                if(next_block==0){ //FIXME 16 vs linesize16
                    s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
                    s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
                    s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
                }

                if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
2844
                    ff_MPV_decode_mb(s, s->block);
2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877
            } else {
                int motion_x = 0, motion_y = 0;
                s->mv_type=MV_TYPE_16X16;
                // only one MB-Type possible

                switch(mb_type){
                case CANDIDATE_MB_TYPE_INTRA:
                    s->mv_dir = 0;
                    s->mb_intra= 1;
                    motion_x= s->mv[0][0][0] = 0;
                    motion_y= s->mv[0][0][1] = 0;
                    break;
                case CANDIDATE_MB_TYPE_INTER:
                    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];
                    break;
                case CANDIDATE_MB_TYPE_INTER_I:
                    s->mv_dir = MV_DIR_FORWARD;
                    s->mv_type = MV_TYPE_FIELD;
                    s->mb_intra= 0;
                    for(i=0; i<2; i++){
                        j= s->field_select[0][i] = s->p_field_select_table[i][xy];
                        s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
                        s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
                    }
                    break;
                case CANDIDATE_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++){
2878 2879
                        s->mv[0][i][0] = s->current_picture.f.motion_val[0][s->block_index[i]][0];
                        s->mv[0][i][1] = s->current_picture.f.motion_val[0][s->block_index[i]][1];
2880 2881 2882
                    }
                    break;
                case CANDIDATE_MB_TYPE_DIRECT:
2883
                    if (CONFIG_MPEG4_ENCODER) {
Aurelien Jacobs's avatar
Aurelien Jacobs committed
2884 2885 2886 2887 2888
                        s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
                        s->mb_intra= 0;
                        motion_x=s->b_direct_mv_table[xy][0];
                        motion_y=s->b_direct_mv_table[xy][1];
                        ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
2889
                    }
2890 2891
                    break;
                case CANDIDATE_MB_TYPE_DIRECT0:
2892
                    if (CONFIG_MPEG4_ENCODER) {
Aurelien Jacobs's avatar
Aurelien Jacobs committed
2893 2894 2895
                        s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
                        s->mb_intra= 0;
                        ff_mpeg4_set_direct_mv(s, 0, 0);
2896
                    }
2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958
                    break;
                case CANDIDATE_MB_TYPE_BIDIR:
                    s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
                    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];
                    break;
                case CANDIDATE_MB_TYPE_BACKWARD:
                    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];
                    break;
                case CANDIDATE_MB_TYPE_FORWARD:
                    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];
                    break;
                case CANDIDATE_MB_TYPE_FORWARD_I:
                    s->mv_dir = MV_DIR_FORWARD;
                    s->mv_type = MV_TYPE_FIELD;
                    s->mb_intra= 0;
                    for(i=0; i<2; i++){
                        j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
                        s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
                        s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
                    }
                    break;
                case CANDIDATE_MB_TYPE_BACKWARD_I:
                    s->mv_dir = MV_DIR_BACKWARD;
                    s->mv_type = MV_TYPE_FIELD;
                    s->mb_intra= 0;
                    for(i=0; i<2; i++){
                        j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
                        s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
                        s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
                    }
                    break;
                case CANDIDATE_MB_TYPE_BIDIR_I:
                    s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
                    s->mv_type = MV_TYPE_FIELD;
                    s->mb_intra= 0;
                    for(dir=0; dir<2; dir++){
                        for(i=0; i<2; i++){
                            j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
                            s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
                            s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
                        }
                    }
                    break;
                default:
                    av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
                }

                encode_mb(s, motion_x, motion_y);

                // RAL: Update last macroblock type
                s->last_mv_dir = s->mv_dir;

2959
                if (CONFIG_H263_ENCODER &&
2960
                    s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
2961 2962
                    ff_h263_update_motion_val(s);

2963
                ff_MPV_decode_mb(s, s->block);
2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978
            }

            /* 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;
            }

            if(s->flags&CODEC_FLAG_PSNR){
                int w= 16;
                int h= 16;

                if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
                if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;

2979 2980
                s->current_picture.f.error[0] += sse(
                    s, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
2981
                    s->dest[0], w, h, s->linesize);
2982 2983
                s->current_picture.f.error[1] += sse(
                    s, s->new_picture.f.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
Baptiste Coudurier's avatar
Baptiste Coudurier committed
2984
                    s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2985 2986
                s->current_picture.f.error[2] += sse(
                    s, s->new_picture.f.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
Baptiste Coudurier's avatar
Baptiste Coudurier committed
2987
                    s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2988 2989
            }
            if(s->loop_filter){
2990
                if(CONFIG_H263_ENCODER && s->out_format == FMT_H263)
2991 2992
                    ff_h263_loop_filter(s);
            }
2993 2994
            av_dlog(s->avctx, "MB %d %d bits\n",
                    s->mb_x + s->mb_y * s->mb_stride, put_bits_count(&s->pb));
2995 2996 2997 2998
        }
    }

    //not beautiful here but we must write it before flushing so it has to be here
2999
    if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == AV_PICTURE_TYPE_I)
3000
        ff_msmpeg4_encode_ext_header(s);
3001 3002 3003 3004 3005 3006

    write_slice_end(s);

    /* Send the last GOB if RTP */
    if (s->avctx->rtp_callback) {
        int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
3007
        pdif = put_bits_ptr(&s->pb) - s->ptr_lastgob;
3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037
        /* Call the RTP callback to send the last GOB */
        emms_c();
        s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
    }

    return 0;
}

#define MERGE(field) dst->field += src->field; src->field=0
static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
    MERGE(me.scene_change_score);
    MERGE(me.mc_mb_var_sum_temp);
    MERGE(me.mb_var_sum_temp);
}

static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
    int i;

    MERGE(dct_count[0]); //note, the other dct vars are not part of the context
    MERGE(dct_count[1]);
    MERGE(mv_bits);
    MERGE(i_tex_bits);
    MERGE(p_tex_bits);
    MERGE(i_count);
    MERGE(f_count);
    MERGE(b_count);
    MERGE(skip_count);
    MERGE(misc_bits);
    MERGE(error_count);
    MERGE(padding_bug_score);
3038 3039 3040
    MERGE(current_picture.f.error[0]);
    MERGE(current_picture.f.error[1]);
    MERGE(current_picture.f.error[2]);
3041 3042 3043 3044 3045 3046 3047 3048 3049 3050

    if(dst->avctx->noise_reduction){
        for(i=0; i<64; i++){
            MERGE(dct_error_sum[0][i]);
            MERGE(dct_error_sum[1][i]);
        }
    }

    assert(put_bits_count(&src->pb) % 8 ==0);
    assert(put_bits_count(&dst->pb) % 8 ==0);
3051
    avpriv_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
3052 3053 3054 3055 3056
    flush_put_bits(&dst->pb);
}

static int estimate_qp(MpegEncContext *s, int dry_run){
    if (s->next_lambda){
3057 3058
        s->current_picture_ptr->f.quality =
        s->current_picture.f.quality = s->next_lambda;
3059 3060
        if(!dry_run) s->next_lambda= 0;
    } else if (!s->fixed_qscale) {
3061 3062 3063
        s->current_picture_ptr->f.quality =
        s->current_picture.f.quality = ff_rate_estimate_qscale(s, dry_run);
        if (s->current_picture.f.quality < 0)
3064 3065 3066 3067 3068
            return -1;
    }

    if(s->adaptive_quant){
        switch(s->codec_id){
3069
        case AV_CODEC_ID_MPEG4:
3070
            if (CONFIG_MPEG4_ENCODER)
Aurelien Jacobs's avatar
Aurelien Jacobs committed
3071
                ff_clean_mpeg4_qscales(s);
3072
            break;
3073 3074 3075
        case AV_CODEC_ID_H263:
        case AV_CODEC_ID_H263P:
        case AV_CODEC_ID_FLV1:
3076
            if (CONFIG_H263_ENCODER)
Aurelien Jacobs's avatar
Aurelien Jacobs committed
3077
                ff_clean_h263_qscales(s);
3078
            break;
3079 3080
        default:
            ff_init_qscale_tab(s);
3081 3082 3083 3084 3085
        }

        s->lambda= s->lambda_table[0];
        //FIXME broken
    }else
3086
        s->lambda = s->current_picture.f.quality;
3087 3088 3089 3090
    update_qscale(s);
    return 0;
}

3091 3092
/* must be called before writing the header */
static void set_frame_distances(MpegEncContext * s){
3093
    assert(s->current_picture_ptr->f.pts != AV_NOPTS_VALUE);
3094
    s->time = s->current_picture_ptr->f.pts * s->avctx->time_base.num;
3095

3096
    if(s->pict_type==AV_PICTURE_TYPE_B){
3097 3098 3099 3100 3101 3102 3103 3104 3105
        s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
        assert(s->pb_time > 0 && s->pb_time < s->pp_time);
    }else{
        s->pp_time= s->time - s->last_non_b_time;
        s->last_non_b_time= s->time;
        assert(s->picture_number==0 || s->pp_time > 0);
    }
}

3106 3107 3108 3109
static int encode_picture(MpegEncContext *s, int picture_number)
{
    int i;
    int bits;
3110
    int context_count = s->slice_context_count;
3111 3112 3113 3114 3115 3116 3117 3118 3119

    s->picture_number = picture_number;

    /* Reset the average MB variance */
    s->me.mb_var_sum_temp    =
    s->me.mc_mb_var_sum_temp = 0;

    /* we need to initialize some time vars before we can encode b-frames */
    // RAL: Condition added for MPEG1VIDEO
3120
    if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO || s->codec_id == AV_CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->msmpeg4_version))
3121
        set_frame_distances(s);
3122
    if(CONFIG_MPEG4_ENCODER && s->codec_id == AV_CODEC_ID_MPEG4)
3123
        ff_set_mpeg4_time(s);
3124 3125 3126

    s->me.scene_change_score=0;

3127
//    s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion
3128

3129
    if(s->pict_type==AV_PICTURE_TYPE_I){
3130 3131
        if(s->msmpeg4_version >= 3) s->no_rounding=1;
        else                        s->no_rounding=0;
3132
    }else if(s->pict_type!=AV_PICTURE_TYPE_B){
3133
        if(s->flipflop_rounding || s->codec_id == AV_CODEC_ID_H263P || s->codec_id == AV_CODEC_ID_MPEG4)
3134 3135 3136 3137 3138 3139 3140 3141
            s->no_rounding ^= 1;
    }

    if(s->flags & CODEC_FLAG_PASS2){
        if (estimate_qp(s,1) < 0)
            return -1;
        ff_get_2pass_fcode(s);
    }else if(!(s->flags & CODEC_FLAG_QSCALE)){
3142
        if(s->pict_type==AV_PICTURE_TYPE_B)
3143 3144 3145 3146 3147 3148 3149
            s->lambda= s->last_lambda_for[s->pict_type];
        else
            s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
        update_qscale(s);
    }

    s->mb_intra=0; //for the rate distortion & bit compare functions
3150
    for(i=1; i<context_count; i++){
3151 3152 3153
        ff_update_duplicate_context(s->thread_context[i], s);
    }

3154 3155
    if(ff_init_me(s)<0)
        return -1;
3156 3157

    /* Estimate motion for every MB */
3158
    if(s->pict_type != AV_PICTURE_TYPE_I){
3159 3160
        s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
        s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
3161 3162
        if(s->pict_type != AV_PICTURE_TYPE_B && s->avctx->me_threshold==0){
            if((s->avctx->pre_me && s->last_non_b_pict_type==AV_PICTURE_TYPE_I) || s->avctx->pre_me==2){
3163
                s->avctx->execute(s->avctx, pre_estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
3164 3165 3166
            }
        }

3167
        s->avctx->execute(s->avctx, estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
3168
    }else /* if(s->pict_type == AV_PICTURE_TYPE_I) */{
3169 3170 3171 3172 3173 3174
        /* I-Frame */
        for(i=0; i<s->mb_stride*s->mb_height; i++)
            s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;

        if(!s->fixed_qscale){
            /* finding spatial complexity for I-frame rate control */
3175
            s->avctx->execute(s->avctx, mb_var_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
3176 3177
        }
    }
3178
    for(i=1; i<context_count; i++){
3179 3180 3181 3182 3183 3184
        merge_context_after_me(s, s->thread_context[i]);
    }
    s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
    s->current_picture.   mb_var_sum= s->current_picture_ptr->   mb_var_sum= s->me.   mb_var_sum_temp;
    emms_c();

3185 3186
    if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == AV_PICTURE_TYPE_P){
        s->pict_type= AV_PICTURE_TYPE_I;
3187 3188
        for(i=0; i<s->mb_stride*s->mb_height; i++)
            s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
3189 3190
        av_dlog(s, "Scene change detected, encoding as I Frame %d %d\n",
                s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
3191 3192 3193
    }

    if(!s->umvplus){
3194
        if(s->pict_type==AV_PICTURE_TYPE_P || s->pict_type==AV_PICTURE_TYPE_S) {
3195 3196 3197 3198 3199 3200
            s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);

            if(s->flags & CODEC_FLAG_INTERLACED_ME){
                int a,b;
                a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
                b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
3201
                s->f_code= FFMAX3(s->f_code, a, b);
3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215
            }

            ff_fix_long_p_mvs(s);
            ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
            if(s->flags & CODEC_FLAG_INTERLACED_ME){
                int j;
                for(i=0; i<2; i++){
                    for(j=0; j<2; j++)
                        ff_fix_long_mvs(s, s->p_field_select_table[i], j,
                                        s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
                }
            }
        }

3216
        if(s->pict_type==AV_PICTURE_TYPE_B){
3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249
            int a, b;

            a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
            b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
            s->f_code = FFMAX(a, b);

            a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
            b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
            s->b_code = FFMAX(a, b);

            ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
            ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
            ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
            ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
            if(s->flags & CODEC_FLAG_INTERLACED_ME){
                int dir, j;
                for(dir=0; dir<2; dir++){
                    for(i=0; i<2; i++){
                        for(j=0; j<2; j++){
                            int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
                                          : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
                            ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
                                            s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
                        }
                    }
                }
            }
        }
    }

    if (estimate_qp(s, 0) < 0)
        return -1;

3250
    if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==AV_PICTURE_TYPE_I && !(s->flags & CODEC_FLAG_QSCALE))
3251 3252 3253 3254 3255 3256 3257 3258 3259
        s->qscale= 3; //reduce clipping problems

    if (s->out_format == FMT_MJPEG) {
        /* for mjpeg, we do include qscale in the matrix */
        for(i=1;i<64;i++){
            int j= s->dsp.idct_permutation[i];

            s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
        }
3260
        s->y_dc_scale_table=
3261 3262
        s->c_dc_scale_table= ff_mpeg2_dc_scale_table[s->intra_dc_precision];
        s->intra_matrix[0] = ff_mpeg2_dc_scale_table[s->intra_dc_precision][8];
3263
        ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
3264 3265 3266 3267 3268
                       s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
        s->qscale= 8;
    }

    //FIXME var duplication
3269 3270 3271 3272
    s->current_picture_ptr->f.key_frame =
    s->current_picture.f.key_frame = s->pict_type == AV_PICTURE_TYPE_I; //FIXME pic_ptr
    s->current_picture_ptr->f.pict_type =
    s->current_picture.f.pict_type = s->pict_type;
3273

3274
    if (s->current_picture.f.key_frame)
3275 3276 3277 3278 3279
        s->picture_in_gop_number=0;

    s->last_bits= put_bits_count(&s->pb);
    switch(s->out_format) {
    case FMT_MJPEG:
3280
        if (CONFIG_MJPEG_ENCODER)
3281 3282 3283
            ff_mjpeg_encode_picture_header(s);
        break;
    case FMT_H261:
3284
        if (CONFIG_H261_ENCODER)
3285 3286 3287
            ff_h261_encode_picture_header(s, picture_number);
        break;
    case FMT_H263:
3288
        if (CONFIG_WMV2_ENCODER && s->codec_id == AV_CODEC_ID_WMV2)
3289
            ff_wmv2_encode_picture_header(s, picture_number);
3290
        else if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
3291
            ff_msmpeg4_encode_picture_header(s, picture_number);
3292
        else if (CONFIG_MPEG4_ENCODER && s->h263_pred)
3293
            ff_mpeg4_encode_picture_header(s, picture_number);
3294
        else if (CONFIG_RV10_ENCODER && s->codec_id == AV_CODEC_ID_RV10)
3295
            ff_rv10_encode_picture_header(s, picture_number);
3296
        else if (CONFIG_RV20_ENCODER && s->codec_id == AV_CODEC_ID_RV20)
3297
            ff_rv20_encode_picture_header(s, picture_number);
3298
        else if (CONFIG_FLV_ENCODER && s->codec_id == AV_CODEC_ID_FLV1)
3299
            ff_flv_encode_picture_header(s, picture_number);
3300
        else if (CONFIG_H263_ENCODER)
3301
            ff_h263_encode_picture_header(s, picture_number);
3302 3303
        break;
    case FMT_MPEG1:
3304
        if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
3305
            ff_mpeg1_encode_picture_header(s, picture_number);
3306 3307 3308 3309 3310 3311 3312 3313 3314
        break;
    case FMT_H264:
        break;
    default:
        assert(0);
    }
    bits= put_bits_count(&s->pb);
    s->header_bits= bits - s->last_bits;

3315
    for(i=1; i<context_count; i++){
3316 3317
        update_duplicate_context_after_me(s->thread_context[i], s);
    }
3318 3319
    s->avctx->execute(s->avctx, encode_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
    for(i=1; i<context_count; i++){
3320 3321 3322 3323 3324 3325
        merge_context_after_encode(s, s->thread_context[i]);
    }
    emms_c();
    return 0;
}

3326
static void denoise_dct_c(MpegEncContext *s, DCTELEM *block){
3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349
    const int intra= s->mb_intra;
    int i;

    s->dct_count[intra]++;

    for(i=0; i<64; i++){
        int level= block[i];

        if(level){
            if(level>0){
                s->dct_error_sum[intra][i] += level;
                level -= s->dct_offset[intra][i];
                if(level<0) level=0;
            }else{
                s->dct_error_sum[intra][i] -= level;
                level += s->dct_offset[intra][i];
                if(level>0) level=0;
            }
            block[i]= level;
        }
    }
}

3350 3351 3352
static int dct_quantize_trellis_c(MpegEncContext *s,
                                  DCTELEM *block, int n,
                                  int qscale, int *overflow){
3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466
    const int *qmat;
    const uint8_t *scantable= s->intra_scantable.scantable;
    const uint8_t *perm_scantable= s->intra_scantable.permutated;
    int max=0;
    unsigned int threshold1, threshold2;
    int bias=0;
    int run_tab[65];
    int level_tab[65];
    int score_tab[65];
    int survivor[65];
    int survivor_count;
    int last_run=0;
    int last_level=0;
    int last_score= 0;
    int last_i;
    int coeff[2][64];
    int coeff_count[64];
    int qmul, qadd, start_i, last_non_zero, i, dc;
    const int esc_length= s->ac_esc_length;
    uint8_t * length;
    uint8_t * last_length;
    const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);

    s->dsp.fdct (block);

    if(s->dct_error_sum)
        s->denoise_dct(s, block);
    qmul= qscale*16;
    qadd= ((qscale-1)|1)*8;

    if (s->mb_intra) {
        int q;
        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;
            qadd=0;
        }

        /* note: block[0] is assumed to be positive */
        block[0] = (block[0] + (q >> 1)) / q;
        start_i = 1;
        last_non_zero = 0;
        qmat = s->q_intra_matrix[qscale];
        if(s->mpeg_quant || s->out_format == FMT_MPEG1)
            bias= 1<<(QMAT_SHIFT-1);
        length     = s->intra_ac_vlc_length;
        last_length= s->intra_ac_vlc_last_length;
    } else {
        start_i = 0;
        last_non_zero = -1;
        qmat = s->q_inter_matrix[qscale];
        length     = s->inter_ac_vlc_length;
        last_length= s->inter_ac_vlc_last_length;
    }
    last_i= start_i;

    threshold1= (1<<QMAT_SHIFT) - bias - 1;
    threshold2= (threshold1<<1);

    for(i=63; i>=start_i; i--) {
        const int j = scantable[i];
        int level = block[j] * qmat[j];

        if(((unsigned)(level+threshold1))>threshold2){
            last_non_zero = i;
            break;
        }
    }

    for(i=start_i; i<=last_non_zero; i++) {
        const int j = scantable[i];
        int level = block[j] * qmat[j];

//        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;
                coeff[0][i]= level;
                coeff[1][i]= level-1;
//                coeff[2][k]= level-2;
            }else{
                level= (bias - level)>>QMAT_SHIFT;
                coeff[0][i]= -level;
                coeff[1][i]= -level+1;
//                coeff[2][k]= -level+2;
            }
            coeff_count[i]= FFMIN(level, 2);
            assert(coeff_count[i]);
            max |=level;
        }else{
            coeff[0][i]= (level>>31)|1;
            coeff_count[i]= 1;
        }
    }

    *overflow= s->max_qcoeff < max; //overflow might have happened

    if(last_non_zero < start_i){
        memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
        return last_non_zero;
    }

    score_tab[start_i]= 0;
    survivor[0]= start_i;
    survivor_count= 1;

    for(i=start_i; i<=last_non_zero; i++){
3467
        int level_index, j, zero_distortion;
3468
        int dct_coeff= FFABS(block[ scantable[i] ]);
3469
        int best_score=256*256*256*120;
3470

3471
        if (s->dsp.fdct == ff_fdct_ifast)
3472
            dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12;
3473
        zero_distortion= dct_coeff*dct_coeff;
3474

3475
        for(level_index=0; level_index < coeff_count[i]; level_index++){
3476
            int distortion;
3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496
            int level= coeff[level_index][i];
            const int alevel= FFABS(level);
            int unquant_coeff;

            assert(level);

            if(s->out_format == FMT_H263){
                unquant_coeff= alevel*qmul + qadd;
            }else{ //MPEG1
                j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
                if(s->mb_intra){
                        unquant_coeff = (int)(  alevel  * qscale * s->intra_matrix[j]) >> 3;
                        unquant_coeff =   (unquant_coeff - 1) | 1;
                }else{
                        unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
                        unquant_coeff =   (unquant_coeff - 1) | 1;
                }
                unquant_coeff<<= 3;
            }

3497
            distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion;
3498 3499 3500 3501
            level+=64;
            if((level&(~127)) == 0){
                for(j=survivor_count-1; j>=0; j--){
                    int run= i - survivor[j];
3502
                    int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514
                    score += score_tab[i-run];

                    if(score < best_score){
                        best_score= score;
                        run_tab[i+1]= run;
                        level_tab[i+1]= level-64;
                    }
                }

                if(s->out_format == FMT_H263){
                    for(j=survivor_count-1; j>=0; j--){
                        int run= i - survivor[j];
3515
                        int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3516 3517 3518 3519 3520 3521 3522 3523 3524 3525
                        score += score_tab[i-run];
                        if(score < last_score){
                            last_score= score;
                            last_run= run;
                            last_level= level-64;
                            last_i= i+1;
                        }
                    }
                }
            }else{
3526
                distortion += esc_length*lambda;
3527 3528
                for(j=survivor_count-1; j>=0; j--){
                    int run= i - survivor[j];
3529
                    int score= distortion + score_tab[i-run];
3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540

                    if(score < best_score){
                        best_score= score;
                        run_tab[i+1]= run;
                        level_tab[i+1]= level-64;
                    }
                }

                if(s->out_format == FMT_H263){
                  for(j=survivor_count-1; j>=0; j--){
                        int run= i - survivor[j];
3541
                        int score= distortion + score_tab[i-run];
3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667
                        if(score < last_score){
                            last_score= score;
                            last_run= run;
                            last_level= level-64;
                            last_i= i+1;
                        }
                    }
                }
            }
        }

        score_tab[i+1]= best_score;

        //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
        if(last_non_zero <= 27){
            for(; survivor_count; survivor_count--){
                if(score_tab[ survivor[survivor_count-1] ] <= best_score)
                    break;
            }
        }else{
            for(; survivor_count; survivor_count--){
                if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
                    break;
            }
        }

        survivor[ survivor_count++ ]= i+1;
    }

    if(s->out_format != FMT_H263){
        last_score= 256*256*256*120;
        for(i= survivor[0]; i<=last_non_zero + 1; i++){
            int score= score_tab[i];
            if(i) score += lambda*2; //FIXME exacter?

            if(score < last_score){
                last_score= score;
                last_i= i;
                last_level= level_tab[i];
                last_run= run_tab[i];
            }
        }
    }

    s->coded_score[n] = last_score;

    dc= FFABS(block[0]);
    last_non_zero= last_i - 1;
    memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));

    if(last_non_zero < start_i)
        return last_non_zero;

    if(last_non_zero == 0 && start_i == 0){
        int best_level= 0;
        int best_score= dc * dc;

        for(i=0; i<coeff_count[0]; i++){
            int level= coeff[i][0];
            int alevel= FFABS(level);
            int unquant_coeff, score, distortion;

            if(s->out_format == FMT_H263){
                    unquant_coeff= (alevel*qmul + qadd)>>3;
            }else{ //MPEG1
                    unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
                    unquant_coeff =   (unquant_coeff - 1) | 1;
            }
            unquant_coeff = (unquant_coeff + 4) >> 3;
            unquant_coeff<<= 3 + 3;

            distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
            level+=64;
            if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
            else                    score= distortion + esc_length*lambda;

            if(score < best_score){
                best_score= score;
                best_level= level - 64;
            }
        }
        block[0]= best_level;
        s->coded_score[n] = best_score - dc*dc;
        if(best_level == 0) return -1;
        else                return last_non_zero;
    }

    i= last_i;
    assert(last_level);

    block[ perm_scantable[last_non_zero] ]= last_level;
    i -= last_run + 1;

    for(; i>start_i; i -= run_tab[i] + 1){
        block[ perm_scantable[i-1] ]= level_tab[i];
    }

    return last_non_zero;
}

//#define REFINE_STATS 1
static int16_t basis[64][64];

static void build_basis(uint8_t *perm){
    int i, j, x, y;
    emms_c();
    for(i=0; i<8; i++){
        for(j=0; j<8; j++){
            for(y=0; y<8; y++){
                for(x=0; x<8; x++){
                    double s= 0.25*(1<<BASIS_SHIFT);
                    int index= 8*i + j;
                    int perm_index= perm[index];
                    if(i==0) s*= sqrt(0.5);
                    if(j==0) s*= sqrt(0.5);
                    basis[perm_index][8*x + y]= lrintf(s * cos((M_PI/8.0)*i*(x+0.5)) * cos((M_PI/8.0)*j*(y+0.5)));
                }
            }
        }
    }
}

static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
                        DCTELEM *block, int16_t *weight, DCTELEM *orig,
                        int n, int qscale){
    int16_t rem[64];
3668
    LOCAL_ALIGNED_16(DCTELEM, d1, [64]);
3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787
    const uint8_t *scantable= s->intra_scantable.scantable;
    const uint8_t *perm_scantable= s->intra_scantable.permutated;
//    unsigned int threshold1, threshold2;
//    int bias=0;
    int run_tab[65];
    int prev_run=0;
    int prev_level=0;
    int qmul, qadd, start_i, last_non_zero, i, dc;
    uint8_t * length;
    uint8_t * last_length;
    int lambda;
    int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
#ifdef REFINE_STATS
static int count=0;
static int after_last=0;
static int to_zero=0;
static int from_zero=0;
static int raise=0;
static int lower=0;
static int messed_sign=0;
#endif

    if(basis[0][0] == 0)
        build_basis(s->dsp.idct_permutation);

    qmul= qscale*2;
    qadd= (qscale-1)|1;
    if (s->mb_intra) {
        if (!s->h263_aic) {
            if (n < 4)
                q = s->y_dc_scale;
            else
                q = s->c_dc_scale;
        } else{
            /* For AIC we skip quant/dequant of INTRADC */
            q = 1;
            qadd=0;
        }
        q <<= RECON_SHIFT-3;
        /* note: block[0] is assumed to be positive */
        dc= block[0]*q;
//        block[0] = (block[0] + (q >> 1)) / q;
        start_i = 1;
//        if(s->mpeg_quant || s->out_format == FMT_MPEG1)
//            bias= 1<<(QMAT_SHIFT-1);
        length     = s->intra_ac_vlc_length;
        last_length= s->intra_ac_vlc_last_length;
    } else {
        dc= 0;
        start_i = 0;
        length     = s->inter_ac_vlc_length;
        last_length= s->inter_ac_vlc_last_length;
    }
    last_non_zero = s->block_last_index[n];

#ifdef REFINE_STATS
{START_TIMER
#endif
    dc += (1<<(RECON_SHIFT-1));
    for(i=0; i<64; i++){
        rem[i]= dc - (orig[i]<<RECON_SHIFT); //FIXME  use orig dirrectly instead of copying to rem[]
    }
#ifdef REFINE_STATS
STOP_TIMER("memset rem[]")}
#endif
    sum=0;
    for(i=0; i<64; i++){
        int one= 36;
        int qns=4;
        int w;

        w= FFABS(weight[i]) + qns*one;
        w= 15 + (48*qns*one + w/2)/w; // 16 .. 63

        weight[i] = w;
//        w=weight[i] = (63*qns + (w/2)) / w;

        assert(w>0);
        assert(w<(1<<6));
        sum += w*w;
    }
    lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
#ifdef REFINE_STATS
{START_TIMER
#endif
    run=0;
    rle_index=0;
    for(i=start_i; i<=last_non_zero; i++){
        int j= perm_scantable[i];
        const int level= block[j];
        int coeff;

        if(level){
            if(level<0) coeff= qmul*level - qadd;
            else        coeff= qmul*level + qadd;
            run_tab[rle_index++]=run;
            run=0;

            s->dsp.add_8x8basis(rem, basis[j], coeff);
        }else{
            run++;
        }
    }
#ifdef REFINE_STATS
if(last_non_zero>0){
STOP_TIMER("init rem[]")
}
}

{START_TIMER
#endif
    for(;;){
        int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0);
        int best_coeff=0;
        int best_change=0;
        int run2, best_unquant_change=0, analyze_gradient;
#ifdef REFINE_STATS
{START_TIMER
#endif
3788
        analyze_gradient = last_non_zero > 2 || s->quantizer_noise_shaping >= 3;
3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845

        if(analyze_gradient){
#ifdef REFINE_STATS
{START_TIMER
#endif
            for(i=0; i<64; i++){
                int w= weight[i];

                d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12);
            }
#ifdef REFINE_STATS
STOP_TIMER("rem*w*w")}
{START_TIMER
#endif
            s->dsp.fdct(d1);
#ifdef REFINE_STATS
STOP_TIMER("dct")}
#endif
        }

        if(start_i){
            const int level= block[0];
            int change, old_coeff;

            assert(s->mb_intra);

            old_coeff= q*level;

            for(change=-1; change<=1; change+=2){
                int new_level= level + change;
                int score, new_coeff;

                new_coeff= q*new_level;
                if(new_coeff >= 2048 || new_coeff < 0)
                    continue;

                score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff);
                if(score<best_score){
                    best_score= score;
                    best_coeff= 0;
                    best_change= change;
                    best_unquant_change= new_coeff - old_coeff;
                }
            }
        }

        run=0;
        rle_index=0;
        run2= run_tab[rle_index++];
        prev_level=0;
        prev_run=0;

        for(i=start_i; i<64; i++){
            int j= perm_scantable[i];
            const int level= block[j];
            int change, old_coeff;

3846
            if(s->quantizer_noise_shaping < 3 && i > last_non_zero + 1)
3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863
                break;

            if(level){
                if(level<0) old_coeff= qmul*level - qadd;
                else        old_coeff= qmul*level + qadd;
                run2= run_tab[rle_index++]; //FIXME ! maybe after last
            }else{
                old_coeff=0;
                run2--;
                assert(run2>=0 || i >= last_non_zero );
            }

            for(change=-1; change<=1; change+=2){
                int new_level= level + change;
                int score, new_coeff, unquant_change;

                score=0;
3864
                if(s->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level))
3865 3866 3867 3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036
                   continue;

                if(new_level){
                    if(new_level<0) new_coeff= qmul*new_level - qadd;
                    else            new_coeff= qmul*new_level + qadd;
                    if(new_coeff >= 2048 || new_coeff <= -2048)
                        continue;
                    //FIXME check for overflow

                    if(level){
                        if(level < 63 && level > -63){
                            if(i < last_non_zero)
                                score +=   length[UNI_AC_ENC_INDEX(run, new_level+64)]
                                         - length[UNI_AC_ENC_INDEX(run, level+64)];
                            else
                                score +=   last_length[UNI_AC_ENC_INDEX(run, new_level+64)]
                                         - last_length[UNI_AC_ENC_INDEX(run, level+64)];
                        }
                    }else{
                        assert(FFABS(new_level)==1);

                        if(analyze_gradient){
                            int g= d1[ scantable[i] ];
                            if(g && (g^new_level) >= 0)
                                continue;
                        }

                        if(i < last_non_zero){
                            int next_i= i + run2 + 1;
                            int next_level= block[ perm_scantable[next_i] ] + 64;

                            if(next_level&(~127))
                                next_level= 0;

                            if(next_i < last_non_zero)
                                score +=   length[UNI_AC_ENC_INDEX(run, 65)]
                                         + length[UNI_AC_ENC_INDEX(run2, next_level)]
                                         - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
                            else
                                score +=  length[UNI_AC_ENC_INDEX(run, 65)]
                                        + last_length[UNI_AC_ENC_INDEX(run2, next_level)]
                                        - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
                        }else{
                            score += last_length[UNI_AC_ENC_INDEX(run, 65)];
                            if(prev_level){
                                score +=  length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
                                        - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
                            }
                        }
                    }
                }else{
                    new_coeff=0;
                    assert(FFABS(level)==1);

                    if(i < last_non_zero){
                        int next_i= i + run2 + 1;
                        int next_level= block[ perm_scantable[next_i] ] + 64;

                        if(next_level&(~127))
                            next_level= 0;

                        if(next_i < last_non_zero)
                            score +=   length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
                                     - length[UNI_AC_ENC_INDEX(run2, next_level)]
                                     - length[UNI_AC_ENC_INDEX(run, 65)];
                        else
                            score +=   last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
                                     - last_length[UNI_AC_ENC_INDEX(run2, next_level)]
                                     - length[UNI_AC_ENC_INDEX(run, 65)];
                    }else{
                        score += -last_length[UNI_AC_ENC_INDEX(run, 65)];
                        if(prev_level){
                            score +=  last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
                                    - length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
                        }
                    }
                }

                score *= lambda;

                unquant_change= new_coeff - old_coeff;
                assert((score < 100*lambda && score > -100*lambda) || lambda==0);

                score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change);
                if(score<best_score){
                    best_score= score;
                    best_coeff= i;
                    best_change= change;
                    best_unquant_change= unquant_change;
                }
            }
            if(level){
                prev_level= level + 64;
                if(prev_level&(~127))
                    prev_level= 0;
                prev_run= run;
                run=0;
            }else{
                run++;
            }
        }
#ifdef REFINE_STATS
STOP_TIMER("iterative step")}
#endif

        if(best_change){
            int j= perm_scantable[ best_coeff ];

            block[j] += best_change;

            if(best_coeff > last_non_zero){
                last_non_zero= best_coeff;
                assert(block[j]);
#ifdef REFINE_STATS
after_last++;
#endif
            }else{
#ifdef REFINE_STATS
if(block[j]){
    if(block[j] - best_change){
        if(FFABS(block[j]) > FFABS(block[j] - best_change)){
            raise++;
        }else{
            lower++;
        }
    }else{
        from_zero++;
    }
}else{
    to_zero++;
}
#endif
                for(; last_non_zero>=start_i; last_non_zero--){
                    if(block[perm_scantable[last_non_zero]])
                        break;
                }
            }
#ifdef REFINE_STATS
count++;
if(256*256*256*64 % count == 0){
    printf("after_last:%d to_zero:%d from_zero:%d raise:%d lower:%d sign:%d xyp:%d/%d/%d\n", after_last, to_zero, from_zero, raise, lower, messed_sign, s->mb_x, s->mb_y, s->picture_number);
}
#endif
            run=0;
            rle_index=0;
            for(i=start_i; i<=last_non_zero; i++){
                int j= perm_scantable[i];
                const int level= block[j];

                 if(level){
                     run_tab[rle_index++]=run;
                     run=0;
                 }else{
                     run++;
                 }
            }

            s->dsp.add_8x8basis(rem, basis[j], best_unquant_change);
        }else{
            break;
        }
    }
#ifdef REFINE_STATS
if(last_non_zero>0){
STOP_TIMER("iterative search")
}
}
#endif

    return last_non_zero;
}

4037
int ff_dct_quantize_c(MpegEncContext *s,
4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113 4114 4115 4116
                        DCTELEM *block, int n,
                        int qscale, int *overflow)
{
    int i, j, level, last_non_zero, q, start_i;
    const int *qmat;
    const uint8_t *scantable= s->intra_scantable.scantable;
    int bias;
    int max=0;
    unsigned int threshold1, threshold2;

    s->dsp.fdct (block);

    if(s->dct_error_sum)
        s->denoise_dct(s, block);

    if (s->mb_intra) {
        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;

        /* note: block[0] is assumed to be positive */
        block[0] = (block[0] + (q >> 1)) / q;
        start_i = 1;
        last_non_zero = 0;
        qmat = s->q_intra_matrix[qscale];
        bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
    } else {
        start_i = 0;
        last_non_zero = -1;
        qmat = s->q_inter_matrix[qscale];
        bias= s->inter_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
    }
    threshold1= (1<<QMAT_SHIFT) - bias - 1;
    threshold2= (threshold1<<1);
    for(i=63;i>=start_i;i--) {
        j = scantable[i];
        level = block[j] * qmat[j];

        if(((unsigned)(level+threshold1))>threshold2){
            last_non_zero = i;
            break;
        }else{
            block[j]=0;
        }
    }
    for(i=start_i; i<=last_non_zero; i++) {
        j = scantable[i];
        level = block[j] * qmat[j];

//        if(   bias+level >= (1<<QMAT_SHIFT)
//           || bias-level >= (1<<QMAT_SHIFT)){
        if(((unsigned)(level+threshold1))>threshold2){
            if(level>0){
                level= (bias + level)>>QMAT_SHIFT;
                block[j]= level;
            }else{
                level= (bias - level)>>QMAT_SHIFT;
                block[j]= -level;
            }
            max |=level;
        }else{
            block[j]=0;
        }
    }
    *overflow= s->max_qcoeff < max; //overflow might have happened

    /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */
    if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)
        ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero);

    return last_non_zero;
}

4117 4118 4119
#define OFFSET(x) offsetof(MpegEncContext, x)
#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
static const AVOption h263_options[] = {
4120 4121 4122
    { "obmc",         "use overlapped block motion compensation.", OFFSET(obmc), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
    { "structured_slices","Write slice start position at every GOB header instead of just GOB number.", OFFSET(h263_slice_structured), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE},
    { "mb_info",      "emit macroblock info for RFC 2190 packetization, the parameter value is the maximum payload size", OFFSET(mb_info), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, INT_MAX, VE },
4123
    FF_MPV_COMMON_OPTS
4124 4125 4126 4127 4128 4129 4130 4131 4132 4133
    { NULL },
};

static const AVClass h263_class = {
    .class_name = "H.263 encoder",
    .item_name  = av_default_item_name,
    .option     = h263_options,
    .version    = LIBAVUTIL_VERSION_INT,
};

4134
AVCodec ff_h263_encoder = {
4135 4136
    .name           = "h263",
    .type           = AVMEDIA_TYPE_VIDEO,
4137
    .id             = AV_CODEC_ID_H263,
4138
    .priv_data_size = sizeof(MpegEncContext),
4139
    .init           = ff_MPV_encode_init,
4140
    .encode2        = ff_MPV_encode_picture,
4141
    .close          = ff_MPV_encode_end,
4142
    .pix_fmts= (const enum AVPixelFormat[]){AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE},
4143
    .long_name= NULL_IF_CONFIG_SMALL("H.263 / H.263-1996"),
4144
    .priv_class     = &h263_class,
4145 4146
};

4147
static const AVOption h263p_options[] = {
4148 4149 4150 4151
    { "umv",        "Use unlimited motion vectors.",    OFFSET(umvplus), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
    { "aiv",        "Use alternative inter VLC.",       OFFSET(alt_inter_vlc), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
    { "obmc",       "use overlapped block motion compensation.", OFFSET(obmc), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
    { "structured_slices", "Write slice start position at every GOB header instead of just GOB number.", OFFSET(h263_slice_structured), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE},
4152
    FF_MPV_COMMON_OPTS
4153 4154 4155 4156 4157
    { NULL },
};
static const AVClass h263p_class = {
    .class_name = "H.263p encoder",
    .item_name  = av_default_item_name,
4158
    .option     = h263p_options,
4159 4160 4161
    .version    = LIBAVUTIL_VERSION_INT,
};

4162
AVCodec ff_h263p_encoder = {
4163 4164
    .name           = "h263p",
    .type           = AVMEDIA_TYPE_VIDEO,
4165
    .id             = AV_CODEC_ID_H263P,
4166
    .priv_data_size = sizeof(MpegEncContext),
4167
    .init           = ff_MPV_encode_init,
4168
    .encode2        = ff_MPV_encode_picture,
4169
    .close          = ff_MPV_encode_end,
4170
    .capabilities   = CODEC_CAP_SLICE_THREADS,
4171
    .pix_fmts       = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE },
4172
    .long_name      = NULL_IF_CONFIG_SMALL("H.263+ / H.263-1998 / H.263 version 2"),
4173
    .priv_class     = &h263p_class,
4174 4175
};

4176 4177
FF_MPV_GENERIC_CLASS(msmpeg4v2)

4178
AVCodec ff_msmpeg4v2_encoder = {
4179 4180
    .name           = "msmpeg4v2",
    .type           = AVMEDIA_TYPE_VIDEO,
4181
    .id             = AV_CODEC_ID_MSMPEG4V2,
4182
    .priv_data_size = sizeof(MpegEncContext),
4183
    .init           = ff_MPV_encode_init,
4184
    .encode2        = ff_MPV_encode_picture,
4185
    .close          = ff_MPV_encode_end,
4186
    .pix_fmts       = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE },
4187
    .long_name      = NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 2"),
4188
    .priv_class     = &msmpeg4v2_class,
4189 4190
};

4191 4192
FF_MPV_GENERIC_CLASS(msmpeg4v3)

4193
AVCodec ff_msmpeg4v3_encoder = {
4194 4195
    .name           = "msmpeg4",
    .type           = AVMEDIA_TYPE_VIDEO,
4196
    .id             = AV_CODEC_ID_MSMPEG4V3,
4197
    .priv_data_size = sizeof(MpegEncContext),
4198
    .init           = ff_MPV_encode_init,
4199
    .encode2        = ff_MPV_encode_picture,
4200
    .close          = ff_MPV_encode_end,
4201
    .pix_fmts       = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE },
4202
    .long_name      = NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 3"),
4203
    .priv_class     = &msmpeg4v3_class,
4204 4205
};

4206 4207
FF_MPV_GENERIC_CLASS(wmv1)

4208
AVCodec ff_wmv1_encoder = {
4209 4210
    .name           = "wmv1",
    .type           = AVMEDIA_TYPE_VIDEO,
4211
    .id             = AV_CODEC_ID_WMV1,
4212
    .priv_data_size = sizeof(MpegEncContext),
4213
    .init           = ff_MPV_encode_init,
4214
    .encode2        = ff_MPV_encode_picture,
4215
    .close          = ff_MPV_encode_end,
4216
    .pix_fmts       = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE },
4217
    .long_name      = NULL_IF_CONFIG_SMALL("Windows Media Video 7"),
4218
    .priv_class     = &wmv1_class,
4219
};