blob: 2500a410c6080501b6caeb0c48921bb93259c7c1 [file] [log] [blame]
/*
* The simplest mpeg encoder (well, it was the simplest!)
* Copyright (c) 2000,2001 Fabrice Bellard
* Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
*
* 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file libavcodec/mpegvideo_enc.c
* The simplest mpeg encoder (well, it was the simplest!).
*/
#include "avcodec.h"
#include "dsputil.h"
#include "mpegvideo.h"
#include "mpegvideo_common.h"
#include "mjpegenc.h"
#include "msmpeg4.h"
#include "h263.h"
#include "faandct.h"
#include "aandcttab.h"
#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);
/* enable all paranoid tests for rounding, overflows, etc... */
//#define PARANOID
//#define DEBUG
static uint8_t default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
static uint8_t default_fcode_tab[MAX_MV*2+1];
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)
{
int qscale;
int shift=0;
for(qscale=qmin; qscale<=qmax; qscale++){
int i;
if (dsp->fdct == ff_jpeg_fdct_islow
#ifdef FAAN_POSTSCALE
|| dsp->fdct == ff_faandct
#endif
) {
for(i=0;i<64;i++) {
const int j= dsp->idct_permutation[i];
/* 16 <= qscale * quant_matrix[i] <= 7905 */
/* 19952 <= ff_aanscales[i] * qscale * quant_matrix[i] <= 249205026 */
/* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1 << 36) / 249205026 */
/* 3444240 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
(qscale * quant_matrix[j]));
}
} else if (dsp->fdct == fdct_ifast
#ifndef FAAN_POSTSCALE
|| dsp->fdct == ff_faandct
#endif
) {
for(i=0;i<64;i++) {
const int j= dsp->idct_permutation[i];
/* 16 <= qscale * quant_matrix[i] <= 7905 */
/* 19952 <= ff_aanscales[i] * qscale * quant_matrix[i] <= 249205026 */
/* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
/* 3444240 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
(ff_aanscales[i] * qscale * quant_matrix[j]));
}
} else {
for(i=0;i<64;i++) {
const int j= dsp->idct_permutation[i];
/* We can safely suppose that 16 <= quant_matrix[i] <= 255
So 16 <= qscale * quant_matrix[i] <= 7905
so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67
*/
qmat[qscale][i] = (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]);
}
}
for(i=intra; i<64; i++){
int64_t max= 8191;
if (dsp->fdct == fdct_ifast
#ifndef FAAN_POSTSCALE
|| dsp->fdct == ff_faandct
#endif
) {
max = (8191LL*ff_aanscales[i]) >> 14;
}
while(((max * qmat[qscale][i]) >> shift) > INT_MAX){
shift++;
}
}
}
if(shift){
av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger than %d, overflows possible\n", QMAT_SHIFT - shift);
}
}
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);
s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
}
void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){
int i;
if(matrix){
put_bits(pb, 1, 1);
for(i=0;i<64;i++) {
put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
}
}else
put_bits(pb, 1, 0);
}
static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
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;
// dst->reference = src->reference;
dst->pts = src->pts;
dst->interlaced_frame = src->interlaced_frame;
dst->top_field_first = src->top_field_first;
if(s->avctx->me_threshold){
if(!src->motion_val[0])
av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
if(!src->mb_type)
av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
if(!src->ref_index[0])
av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
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));
}
if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
memcpy(dst->ref_index[i], src->ref_index[i], s->b8_stride*2*s->mb_height*sizeof(int8_t));
}
}
}
}
static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
#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);
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
#undef COPY
}
/**
* sets the given MpegEncContext to defaults for encoding.
* the changed fields will not depend upon the prior state of the MpegEncContext.
*/
static void MPV_encode_defaults(MpegEncContext *s){
int i;
MPV_common_defaults(s);
for(i=-16; i<16; i++){
default_fcode_tab[i + MAX_MV]= 1;
}
s->me.mv_penalty= default_mv_penalty;
s->fcode_tab= default_fcode_tab;
}
/* init video encoder */
av_cold int MPV_encode_init(AVCodecContext *avctx)
{
MpegEncContext *s = avctx->priv_data;
int i;
int chroma_h_shift, chroma_v_shift;
MPV_encode_defaults(s);
switch (avctx->codec_id) {
case CODEC_ID_MPEG2VIDEO:
if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){
av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n");
return -1;
}
break;
case CODEC_ID_LJPEG:
case CODEC_ID_MJPEG:
if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P && avctx->pix_fmt != PIX_FMT_RGB32 &&
((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL)){
av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
return -1;
}
break;
default:
if(avctx->pix_fmt != PIX_FMT_YUV420P){
av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
return -1;
}
}
switch (avctx->pix_fmt) {
case PIX_FMT_YUVJ422P:
case PIX_FMT_YUV422P:
s->chroma_format = CHROMA_422;
break;
case PIX_FMT_YUVJ420P:
case PIX_FMT_YUV420P:
default:
s->chroma_format = CHROMA_420;
break;
}
s->bit_rate = avctx->bit_rate;
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;
s->luma_elim_threshold = avctx->luma_elim_threshold;
s->chroma_elim_threshold= avctx->chroma_elim_threshold;
s->strict_std_compliance= avctx->strict_std_compliance;
s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
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;
s->user_specified_pts = AV_NOPTS_VALUE;
if (s->gop_size <= 1) {
s->intra_only = 1;
s->gop_size = 12;
} else {
s->intra_only = 0;
}
s->me_method = avctx->me_method;
/* Fixed QSCALE */
s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
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
|| (s->flags&CODEC_FLAG_QP_RD))
&& !s->fixed_qscale;
s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);
s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC);
s->q_scale_type= !!(s->flags2 & CODEC_FLAG2_NON_LINEAR_QUANT);
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");
return -1;
}
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");
}
if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
return -1;
}
if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n");
return -1;
}
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");
}
if(avctx->rc_buffer_size && avctx->bit_rate*av_q2d(avctx->time_base) > avctx->rc_buffer_size){
av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
return -1;
}
if(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");
return -1;
}
if( s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate
&& (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
&& 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");
}
if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4
&& s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
return -1;
}
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");
return -1;
}
if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
return -1;
}
if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
return -1;
}
if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
return -1;
}
if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
return -1;
}
if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN))
&& s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
return -1;
}
if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
return -1;
}
if((s->flags & CODEC_FLAG_CBP_RD) && !avctx->trellis){
av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
return -1;
}
if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
return -1;
}
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");
return -1;
}
if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){
av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n");
return -1;
}
if(s->flags & CODEC_FLAG_LOW_DELAY){
if (s->codec_id != CODEC_ID_MPEG2VIDEO){
av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg2\n");
return -1;
}
if (s->max_b_frames != 0){
av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n");
return -1;
}
}
if(s->q_scale_type == 1){
if(s->codec_id != CODEC_ID_MPEG2VIDEO){
av_log(avctx, AV_LOG_ERROR, "non linear quant is only available for mpeg2\n");
return -1;
}
if(avctx->qmax > 12){
av_log(avctx, AV_LOG_ERROR, "non linear quant only supports qmax <= 12 currently\n");
return -1;
}
}
if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4
&& s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO
&& (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
return -1;
}
if(s->avctx->thread_count > 1)
s->rtp_mode= 1;
if(!avctx->time_base.den || !avctx->time_base.num){
av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
return -1;
}
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);
return -1;
}
if(avctx->mb_threshold >= i){
av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
return -1;
}
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");
avctx->b_frame_strategy = 0;
}
i= av_gcd(avctx->time_base.den, avctx->time_base.num);
if(i > 1){
av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
avctx->time_base.den /= i;
avctx->time_base.num /= i;
// return -1;
}
if(s->codec_id==CODEC_ID_MJPEG){
s->intra_quant_bias= 1<<(QUANT_BIAS_SHIFT-1); //(a + x/2)/x
s->inter_quant_bias= 0;
}else if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO){
s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
s->inter_quant_bias= 0;
}else{
s->intra_quant_bias=0;
s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x
}
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);
if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
av_log(avctx, AV_LOG_ERROR, "timebase not supported by mpeg 4 standard\n");
return -1;
}
s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
switch(avctx->codec->id) {
case CODEC_ID_MPEG1VIDEO:
s->out_format = FMT_MPEG1;
s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
break;
case CODEC_ID_MPEG2VIDEO:
s->out_format = FMT_MPEG1;
s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
s->rtp_mode= 1;
break;
case CODEC_ID_LJPEG:
case CODEC_ID_MJPEG:
s->out_format = FMT_MJPEG;
s->intra_only = 1; /* force intra only for jpeg */
s->mjpeg_vsample[0] = 2;
s->mjpeg_vsample[1] = 2>>chroma_v_shift;
s->mjpeg_vsample[2] = 2>>chroma_v_shift;
s->mjpeg_hsample[0] = 2;
s->mjpeg_hsample[1] = 2>>chroma_h_shift;
s->mjpeg_hsample[2] = 2>>chroma_h_shift;
if (!(CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER)
|| ff_mjpeg_encode_init(s) < 0)
return -1;
avctx->delay=0;
s->low_delay=1;
break;
case CODEC_ID_H261:
if (!CONFIG_H261_ENCODER) return -1;
if (ff_h261_get_picture_format(s->width, s->height) < 0) {
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);
return -1;
}
s->out_format = FMT_H261;
avctx->delay=0;
s->low_delay=1;
break;
case CODEC_ID_H263:
if (!CONFIG_H263_ENCODER) return -1;
if (h263_get_picture_format(s->width, s->height) == 7) {
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);
return -1;
}
s->out_format = FMT_H263;
s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
avctx->delay=0;
s->low_delay=1;
break;
case CODEC_ID_H263P:
s->out_format = FMT_H263;
s->h263_plus = 1;
/* Fx */
s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
s->h263_aic= (avctx->flags & CODEC_FLAG_AC_PRED) ? 1:0;
s->modified_quant= s->h263_aic;
s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;
/* /Fx */
/* These are just to be sure */
avctx->delay=0;
s->low_delay=1;
break;
case CODEC_ID_FLV1:
s->out_format = FMT_H263;
s->h263_flv = 2; /* format = 1; 11-bit codes */
s->unrestricted_mv = 1;
s->rtp_mode=0; /* don't allow GOB */
avctx->delay=0;
s->low_delay=1;
break;
case CODEC_ID_RV10:
s->out_format = FMT_H263;
avctx->delay=0;
s->low_delay=1;
break;
case CODEC_ID_RV20:
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= s->obmc || s->loop_filter || s->umvplus;
break;
case CODEC_ID_MPEG4:
s->out_format = FMT_H263;
s->h263_pred = 1;
s->unrestricted_mv = 1;
s->low_delay= s->max_b_frames ? 0 : 1;
avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
break;
case CODEC_ID_MSMPEG4V1:
s->out_format = FMT_H263;
s->h263_msmpeg4 = 1;
s->h263_pred = 1;
s->unrestricted_mv = 1;
s->msmpeg4_version= 1;
avctx->delay=0;
s->low_delay=1;
break;
case CODEC_ID_MSMPEG4V2:
s->out_format = FMT_H263;
s->h263_msmpeg4 = 1;
s->h263_pred = 1;
s->unrestricted_mv = 1;
s->msmpeg4_version= 2;
avctx->delay=0;
s->low_delay=1;
break;
case CODEC_ID_MSMPEG4V3:
s->out_format = FMT_H263;
s->h263_msmpeg4 = 1;
s->h263_pred = 1;
s->unrestricted_mv = 1;
s->msmpeg4_version= 3;
s->flipflop_rounding=1;
avctx->delay=0;
s->low_delay=1;
break;
case CODEC_ID_WMV1:
s->out_format = FMT_H263;
s->h263_msmpeg4 = 1;
s->h263_pred = 1;
s->unrestricted_mv = 1;
s->msmpeg4_version= 4;
s->flipflop_rounding=1;
avctx->delay=0;
s->low_delay=1;
break;
case CODEC_ID_WMV2:
s->out_format = FMT_H263;
s->h263_msmpeg4 = 1;
s->h263_pred = 1;
s->unrestricted_mv = 1;
s->msmpeg4_version= 5;
s->flipflop_rounding=1;
avctx->delay=0;
s->low_delay=1;
break;
default:
return -1;
}
avctx->has_b_frames= !s->low_delay;
s->encoding = 1;
/* init */
if (MPV_common_init(s) < 0)
return -1;
if(!s->dct_quantize)
s->dct_quantize = dct_quantize_c;
if(!s->denoise_dct)
s->denoise_dct = denoise_dct_c;
s->fast_dct_quantize = s->dct_quantize;
if(avctx->trellis)
s->dct_quantize = dct_quantize_trellis_c;
if((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
s->chroma_qscale_table= ff_h263_chroma_qscale_table;
s->progressive_frame=
s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
s->quant_precision=5;
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);
if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
ff_h261_encode_init(s);
if (CONFIG_ANY_H263_ENCODER && s->out_format == FMT_H263)
h263_encode_init(s);
if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
ff_msmpeg4_encode_init(s);
if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
&& s->out_format == FMT_MPEG1)
ff_mpeg1_encode_init(s);
/* init q matrix */
for(i=0;i<64;i++) {
int j= s->dsp.idct_permutation[i];
if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
}else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
s->intra_matrix[j] =
s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
}else
{ /* mpeg1/2 */
s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
}
if(s->avctx->intra_matrix)
s->intra_matrix[j] = s->avctx->intra_matrix[i];
if(s->avctx->inter_matrix)
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) {
ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
}
if(ff_rate_control_init(s) < 0)
return -1;
return 0;
}
av_cold int MPV_encode_end(AVCodecContext *avctx)
{
MpegEncContext *s = avctx->priv_data;
ff_rate_control_uninit(s);
MPV_common_end(s);
if ((CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
ff_mjpeg_encode_close(s);
av_freep(&avctx->extradata);
return 0;
}
static int get_sae(uint8_t *src, int ref, int stride){
int x,y;
int acc=0;
for(y=0; y<16; y++){
for(x=0; x<16; x++){
acc+= FFABS(src[x+y*stride] - ref);
}
}
return acc;
}
static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
int x, y, w, h;
int acc=0;
w= s->width &~15;
h= s->height&~15;
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);
acc+= sae + 500 < sad;
}
}
return acc;
}
static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
AVFrame *pic=NULL;
int64_t pts;
int i;
const int encoding_delay= s->max_b_frames;
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);
return -1;
}
}
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;
}
}
}
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;
// av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
if(direct){
i= ff_find_unused_picture(s, 1);
pic= (AVFrame*)&s->picture[i];
pic->reference= 3;
for(i=0; i<4; i++){
pic->data[i]= pic_arg->data[i];
pic->linesize[i]= pic_arg->linesize[i];
}
alloc_picture(s, (Picture*)pic, 1);
}else{
i= ff_find_unused_picture(s, 0);
pic= (AVFrame*)&s->picture[i];
pic->reference= 3;
alloc_picture(s, (Picture*)pic, 0);
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{
int h_chroma_shift, v_chroma_shift;
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--){
memcpy(dst, src, w);
dst += dst_stride;
src += src_stride;
}
}
}
}
}
copy_picture_attributes(s, pic, pic_arg);
pic->pts= pts; //we set this here to avoid modifiying pic_arg
}
/* shift buffer entries */
for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
s->input_picture[i-1]= s->input_picture[i];
s->input_picture[encoding_delay]= (Picture*)pic;
return 0;
}
static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
int x, y, plane;
int score=0;
int64_t score64=0;
for(plane=0; plane<3; plane++){
const int stride= p->linesize[plane];
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->type == FF_BUFFER_TYPE_SHARED ? 0: 16;
int v= s->dsp.frame_skip_cmp[1](s, p->data[plane] + 8*(x + y*stride)+off, ref->data[plane] + 8*(x + y*stride), 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;
}
}
}
}
if(score) score64= score;
if(score64 < s->avctx->frame_skip_threshold)
return 1;
if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
return 1;
return 0;
}
static int estimate_best_b_count(MpegEncContext *s){
AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
AVCodecContext *c= avcodec_alloc_context();
AVFrame input[FF_MAX_B_FRAMES+2];
const int scale= s->avctx->brd_scale;
int i, j, out_size, p_lambda, b_lambda, lambda2;
int outbuf_size= s->width * s->height; //FIXME
uint8_t *outbuf= av_malloc(outbuf_size);
int64_t best_rd= INT64_MAX;
int best_b_count= -1;
assert(scale>=0 && scale <=3);
// emms_c();
p_lambda= s->last_lambda_for[FF_P_TYPE]; //s->next_picture_ptr->quality;
b_lambda= s->last_lambda_for[FF_B_TYPE]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else
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;
c->pix_fmt = PIX_FMT_YUV420P;
c->time_base= s->avctx->time_base;
c->max_b_frames= s->max_b_frames;
if (avcodec_open(c, codec) < 0)
return -1;
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;
avcodec_get_frame_defaults(&input[i]);
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;
if(pre_input_ptr && (!i || s->input_picture[i-1])) {
pre_input= *pre_input_ptr;
if(pre_input.type != FF_BUFFER_TYPE_SHARED && i) {
pre_input.data[0]+=INPLACE_OFFSET;
pre_input.data[1]+=INPLACE_OFFSET;
pre_input.data[2]+=INPLACE_OFFSET;
}
s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height);
s->dsp.shrink[scale](input[i].data[1], input[i].linesize[1], pre_input.data[1], pre_input.linesize[1], c->width>>1, c->height>>1);
s->dsp.shrink[scale](input[i].data[2], input[i].linesize[2], pre_input.data[2], pre_input.linesize[2], c->width>>1, c->height>>1);
}
}
for(j=0; j<s->max_b_frames+1; j++){
int64_t rd=0;
if(!s->input_picture[j])
break;
c->error[0]= c->error[1]= c->error[2]= 0;
input[0].pict_type= FF_I_TYPE;
input[0].quality= 1 * FF_QP2LAMBDA;
out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
// rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
for(i=0; i<s->max_b_frames+1; i++){
int is_p= i % (j+1) == j || i==s->max_b_frames;
input[i+1].pict_type= is_p ? FF_P_TYPE : FF_B_TYPE;
input[i+1].quality= is_p ? p_lambda : b_lambda;
out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
}
/* get the delayed frames */
while(out_size){
out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
}
rd += c->error[0] + c->error[1] + c->error[2];
if(rd < best_rd){
best_rd= rd;
best_b_count= j;
}
}
av_freep(&outbuf);
avcodec_close(c);
av_freep(&c);
for(i=0; i<s->max_b_frames+2; i++){
av_freep(&input[i].data[0]);
}
return best_b_count;
}
static void select_input_picture(MpegEncContext *s){
int i;
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;
/* set next picture type & ordering */
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];
s->reordered_input_picture[0]->pict_type= FF_I_TYPE;
s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
}else{
int b_frames;
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
//av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n", s->input_picture[0]->data[0], s->input_picture[0]->pts);
if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
for(i=0; i<4; i++)
s->input_picture[0]->data[i]= NULL;
s->input_picture[0]->type= 0;
}else{
assert( s->input_picture[0]->type==FF_BUFFER_TYPE_USER
|| s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
}
emms_c();
ff_vbv_update(s, 0);
goto no_output_pic;
}
}
if(s->flags&CODEC_FLAG_PASS2){
for(i=0; i<s->max_b_frames+1; i++){
int pict_num= s->input_picture[0]->display_picture_number + i;
if(pict_num >= s->rc_context.num_entries)
break;
if(!s->input_picture[i]){
s->rc_context.entry[pict_num-1].new_pict_type = FF_P_TYPE;
break;
}
s->input_picture[i]->pict_type=
s->rc_context.entry[pict_num].new_pict_type;
}
}
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 ]->data[0],
s->input_picture[i-1]->data[0], s->linesize) + 1;
}
}
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;
}
b_frames= FFMAX(0, i-1);
/* reset scores */
for(i=0; i<b_frames+1; i++){
s->input_picture[i]->b_frame_score=0;
}
}else if(s->avctx->b_frame_strategy==2){
b_frames= estimate_best_b_count(s);
}else{
av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
b_frames=0;
}
emms_c();
//static int b_count=0;
//b_count+= b_frames;
//av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
for(i= b_frames - 1; i>=0; i--){
int type= s->input_picture[i]->pict_type;
if(type && type != FF_B_TYPE)
b_frames= i;
}
if(s->input_picture[b_frames]->pict_type == FF_B_TYPE && b_frames == s->max_b_frames){
av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
}
if(s->picture_in_gop_number + b_frames >= s->gop_size){
if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && 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]->pict_type= FF_I_TYPE;
}
}
if( (s->flags & CODEC_FLAG_CLOSED_GOP)
&& b_frames
&& s->input_picture[b_frames]->pict_type== FF_I_TYPE)
b_frames--;
s->reordered_input_picture[0]= s->input_picture[b_frames];
if(s->reordered_input_picture[0]->pict_type != FF_I_TYPE)
s->reordered_input_picture[0]->pict_type= FF_P_TYPE;
s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
for(i=0; i<b_frames; i++){
s->reordered_input_picture[i+1]= s->input_picture[i];
s->reordered_input_picture[i+1]->pict_type= FF_B_TYPE;
s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++;
}
}
}
no_output_pic:
if(s->reordered_input_picture[0]){
s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=FF_B_TYPE ? 3 : 0;
ff_copy_picture(&s->new_picture, s->reordered_input_picture[0]);
if(s->reordered_input_picture[0]->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
int i= ff_find_unused_picture(s, 0);
Picture *pic= &s->picture[i];
pic->reference = s->reordered_input_picture[0]->reference;
alloc_picture(s, pic, 0);
/* mark us unused / free shared pic */
if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL)
s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
for(i=0; i<4; i++)
s->reordered_input_picture[0]->data[i]= NULL;
s->reordered_input_picture[0]->type= 0;
copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
s->current_picture_ptr= pic;
}else{
// input is not a shared pix -> reuse buffer for current_pix
assert( s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
|| s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
s->current_picture_ptr= s->reordered_input_picture[0];
for(i=0; i<4; i++){
s->new_picture.data[i]+= INPLACE_OFFSET;
}
}
ff_copy_picture(&s->current_picture, s->current_picture_ptr);
s->picture_number= s->new_picture.display_picture_number;
//printf("dpn:%d\n", s->picture_number);
}else{
memset(&s->new_picture, 0, sizeof(Picture));
}
}
int MPV_encode_picture(AVCodecContext *avctx,
unsigned char *buf, int buf_size, void *data)
{
MpegEncContext *s = avctx->priv_data;
AVFrame *pic_arg = data;
int i, stuffing_count;
for(i=0; i<avctx->thread_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= buf + (size_t)(((int64_t) buf_size)*start_y/h);
uint8_t *end = buf + (size_t)(((int64_t) buf_size)* end_y/h);
init_put_bits(&s->thread_context[i]->pb, start, end - start);
}
s->picture_in_gop_number++;
if(load_input_picture(s, pic_arg) < 0)
return -1;
select_input_picture(s);
/* output? */
if(s->new_picture.data[0]){
s->pict_type= s->new_picture.pict_type;
//emms_c();
//printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
MPV_frame_start(s, avctx);
vbv_retry:
if (encode_picture(s, s->picture_number) < 0)
return -1;
avctx->real_pict_num = s->picture_number;
avctx->header_bits = s->header_bits;
avctx->mv_bits = s->mv_bits;
avctx->misc_bits = s->misc_bits;
avctx->i_tex_bits = s->i_tex_bits;
avctx->p_tex_bits = s->p_tex_bits;
avctx->i_count = s->i_count;
avctx->p_count = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
avctx->skip_count = s->skip_count;
MPV_frame_end(s);
if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
ff_mjpeg_encode_picture_trailer(s);
if(avctx->rc_buffer_size){
RateControlContext *rcc= &s->rc_context;
int max_size= rcc->buffer_index * avctx->rc_max_available_vbv_use;
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){
int i;
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);
}
s->mb_skipped = 0; //done in MPV_frame_start()
if(s->pict_type==FF_P_TYPE){ //done in encode_picture() so we must undo it
if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
s->no_rounding ^= 1;
}
if(s->pict_type!=FF_B_TYPE){
s->time_base= s->last_time_base;
s->last_non_b_time= s->time - s->pp_time;
}
// av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
for(i=0; i<avctx->thread_count; i++){
PutBitContext *pb= &s->thread_context[i]->pb;
init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
}
goto vbv_retry;
}
assert(s->avctx->rc_max_rate);
}
if(s->flags&CODEC_FLAG_PASS1)
ff_write_pass1_stats(s);
for(i=0; i<4; i++){
s->current_picture_ptr->error[i]= s->current_picture.error[i];
avctx->error[i] += s->current_picture_ptr->error[i];
}
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));
flush_put_bits(&s->pb);
s->frame_bits = put_bits_count(&s->pb);
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){
av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
return -1;
}
switch(s->codec_id){
case CODEC_ID_MPEG1VIDEO:
case CODEC_ID_MPEG2VIDEO:
while(stuffing_count--){
put_bits(&s->pb, 8, 0);
}
break;
case CODEC_ID_MPEG4:
put_bits(&s->pb, 16, 0);
put_bits(&s->pb, 16, 0x1C3);
stuffing_count -= 4;
while(stuffing_count--){
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 */
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){
int vbv_delay;
assert(s->repeat_first_field==0);
vbv_delay= lrintf(90000 * s->rc_context.buffer_index / s->avctx->rc_max_rate);
assert(vbv_delay < 0xFFFF);
s->vbv_delay_ptr[0] &= 0xF8;
s->vbv_delay_ptr[0] |= vbv_delay>>13;
s->vbv_delay_ptr[1] = vbv_delay>>5;
s->vbv_delay_ptr[2] &= 0x07;
s->vbv_delay_ptr[2] |= vbv_delay<<3;
}
s->total_bits += s->frame_bits;
avctx->frame_bits = s->frame_bits;
}else{
assert((pbBufPtr(&s->pb) == s->pb.buf));
s->frame_bits=0;
}
assert((s->frame_bits&7)==0);
return s->frame_bits/8;
}
static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
{
static const char tab[64]=
{3,2,2,1,1,1,1,1,
1,1,1,1,1,1,1,1,
1,1,1,1,1,1,1,1,
0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0};
int score=0;
int run=0;
int i;
DCTELEM *block= s->block[n];
const int last_index= s->block_last_index[n];
int skip_dc;
if(threshold<0){
skip_dc=0;
threshold= -threshold;
}else
skip_dc=1;
/* Are all we could set to zero already zero? */
if(last_index<=skip_dc - 1) return;
for(i=0; i<=last_index; i++){
const int j = s->intra_scantable.permutated[i];
const int level = FFABS(block[j]);
if(level==1){
if(skip_dc && i==0) continue;
score+= tab[run];
run=0;
}else if(level>1){
return;
}else{
run++;
}
}
if(score >= threshold) return;
for(i=skip_dc; i<=last_index; i++){
const int j = s->intra_scantable.permutated[i];
block[j]=0;
}
if(block[0]) s->block_last_index[n]= 0;
else s->block_last_index[n]= -1;
}
static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
{
int i;
const int maxlevel= s->max_qcoeff;
const int minlevel= s->min_qcoeff;
int overflow=0;
if(s->mb_intra){
i=1; //skip clipping of intra dc
}else
i=0;
for(;i<=last_index; i++){
const int j= s->intra_scantable.permutated[i];
int level = block[j];
if (level>maxlevel){
level=maxlevel;
overflow++;
}else if(level<minlevel){
level=minlevel;
overflow++;
}
block[j]= level;
}
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);
}
static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride){
int x, y;
//FIXME optimize
for(y=0; y<8; y++){
for(x=0; x<8; x++){
int x2, y2;
int sum=0;
int sqr=0;
int count=0;
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];
sum += v;
sqr += v*v;
count++;
}
}
weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
}
}
}
static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
{
int16_t weight[8][64];
DCTELEM orig[8][64];
const int mb_x= s->mb_x;
const int mb_y= s->mb_y;
int i;
int skip_dct[8];
int dct_offset = s->linesize*8; //default for progressive frames
uint8_t *ptr_y, *ptr_cb, *ptr_cr;
int wrap_y, wrap_c;
for(i=0; i<mb_block_count; i++) skip_dct[i]=s->skipdct;
if(s->adaptive_quant){
const int last_qp= s->qscale;
const int mb_xy= mb_x + mb_y*s->mb_stride;
s->lambda= s->lambda_table[mb_xy];
update_qscale(s);
if(!(s->flags&CODEC_FLAG_QP_RD)){
s->qscale= s->current_picture_ptr->qscale_table[mb_xy];
s->dquant= s->qscale - last_qp;
if(s->out_format==FMT_H263){
s->dquant= av_clip(s->dquant, -2, 2);
if(s->codec_id==CODEC_ID_MPEG4){
if(!s->mb_intra){
if(s->pict_type == FF_B_TYPE){
if(s->dquant&1 || s->mv_dir&MV_DIRECT)
s->dquant= 0;
}
if(s->mv_type==MV_TYPE_8X8)
s->dquant=0;
}
}
}
}
ff_set_qscale(s, last_qp + s->dquant);
}else if(s->flags&CODEC_FLAG_QP_RD)
ff_set_qscale(s, s->qscale + s->dquant);
wrap_y = s->linesize;
wrap_c = s->uvlinesize;
ptr_y = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
ptr_cb = s->new_picture.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
ptr_cr = s->new_picture.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;
ff_emulated_edge_mc(ebuf , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width , s->height);
ptr_y= ebuf;
ff_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;
ff_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;
}
if (s->mb_intra) {
if(s->flags&CODEC_FLAG_INTERLACED_DCT){
int progressive_score, interlaced_score;
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;
if (s->chroma_format == CHROMA_422)
wrap_c<<=1;
}
}
}
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);
if(s->flags&CODEC_FLAG_GRAY){
skip_dct[4]= 1;
skip_dct[5]= 1;
}else{
s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
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);
}
}
}else{
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];
if ((!s->no_rounding) || s->pict_type==FF_B_TYPE){
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;
}
if (s->mv_dir & MV_DIR_FORWARD) {
MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
op_pix = s->dsp.avg_pixels_tab;
op_qpix= s->dsp.avg_qpel_pixels_tab;
}
if (s->mv_dir & MV_DIR_BACKWARD) {
MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
}
if(s->flags&CODEC_FLAG_INTERLACED_DCT){
int progressive_score, interlaced_score;
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;
if (s->chroma_format == CHROMA_422)
wrap_c<<=1;
}
}
}
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);
if(s->flags&CODEC_FLAG_GRAY){
skip_dct[4]= 1;
skip_dct[5]= 1;
}else{
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);
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);
}
}
/* pre quantization */
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;
}
}
}
if(s->avctx->quantizer_noise_shaping){
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);
}
memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
}
/* DCT & quantize */
assert(s->out_format!=FMT_MJPEG || s->qscale==8);
{
for(i=0;i<mb_block_count;i++) {
if(!skip_dct[i]){
int overflow;
s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
// 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;
}
if(s->avctx->quantizer_noise_shaping){
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);
}
}
}
if(s->luma_elim_threshold && !s->mb_intra)
for(i=0; i<4; i++)
dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
if(s->chroma_elim_threshold && !s->mb_intra)
for(i=4; i<mb_block_count; i++)
dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
if(s->flags & CODEC_FLAG_CBP_RD){
for(i=0;i<mb_block_count;i++) {
if(s->block_last_index[i] == -1)
s->coded_score[i]= INT_MAX/256;
}
}
}
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;
}
//non c quantize code returns incorrect block_last_index FIXME
if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
for(i=0; i<mb_block_count; i++){
int j;
if(s->block_last_index[i]>0){
for(j=63; j>0; j--){
if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
}
s->block_last_index[i]= j;
}
}
}
/* huffman encode */
switch(s->codec_id){ //FIXME funct ptr could be slightly faster
case CODEC_ID_MPEG1VIDEO:
case CODEC_ID_MPEG2VIDEO:
if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
mpeg1_encode_mb(s, s->block, motion_x, motion_y);
break;
case CODEC_ID_MPEG4:
if (CONFIG_MPEG4_ENCODER)
mpeg4_encode_mb(s, s->block, motion_x, motion_y);
break;
case CODEC_ID_MSMPEG4V2:
case CODEC_ID_MSMPEG4V3:
case CODEC_ID_WMV1:
if (CONFIG_MSMPEG4_ENCODER)
msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
break;
case CODEC_ID_WMV2:
if (CONFIG_WMV2_ENCODER)
ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
break;
case CODEC_ID_H261:
if (CONFIG_H261_ENCODER)
ff_h261_encode_mb(s, s->block, motion_x, motion_y);
break;
case CODEC_ID_H263:
case CODEC_ID_H263P:
case CODEC_ID_FLV1:
case CODEC_ID_RV10:
case CODEC_ID_RV20:
if (CONFIG_H263_ENCODER || CONFIG_H263P_ENCODER ||
CONFIG_FLV_ENCODER || CONFIG_RV10_ENCODER || CONFIG_RV20_ENCODER)
h263_encode_mb(s, s->block, motion_x, motion_y);
break;
case CODEC_ID_MJPEG:
if (CONFIG_MJPEG_ENCODER)
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;
memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
/* mpeg1 */
d->mb_skip_run= s->mb_skip_run;
for(i=0; i<3; i++)
d->last_dc[i]= s->last_dc[i];
/* statistics */
d->mv_bits= s->mv_bits;
d->i_tex_bits= s->i_tex_bits;
d->p_tex_bits= s->p_tex_bits;
d->i_count= s->i_count;
d->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;
d->esc3_level_length= s->esc3_level_length;
}
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));
memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
/* mpeg1 */
d->mb_skip_run= s->mb_skip_run;
for(i=0; i<3; i++)
d->last_dc[i]= s->last_dc[i];
/* statistics */
d->mv_bits= s->mv_bits;
d->i_tex_bits= s->i_tex_bits;
d->p_tex_bits= s->p_tex_bits;
d->i_count= s->i_count;
d->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;
d->esc3_level_length= s->esc3_level_length;
}
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){
MPV_decode_mb(s, s->block);
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){
return s->dsp.nsse[0](s, s->new_picture.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.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.data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
}else{
return s->dsp.sse[0](NULL, s->new_picture.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.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.data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
}
else
return sse(s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], w, h, s->linesize)
+sse(s, s->new_picture.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.data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[2], w>>1, h>>1, s->uvlinesize);
}
static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
MpegEncContext *s= *(void**)arg;
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){
MpegEncContext *s= *(void**)arg;
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 */
if(s->pict_type==FF_B_TYPE)
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){
MpegEncContext *s= *(void**)arg;
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;
uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
int varc;
int sum = s->dsp.pix_sum(pix, s->linesize);
varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
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){
if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4){
if(s->partitioned_frame){
ff_mpeg4_merge_partitions(s);
}
ff_mpeg4_stuffing(&s->pb);
}else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
ff_mjpeg_encode_stuffing(&s->pb);
}
align_put_bits(&s->pb);
flush_put_bits(&s->pb);
if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
s->misc_bits+= get_bits_diff(s);
}
static int encode_thread(AVCodecContext *c, void *arg){
MpegEncContext *s= *(void**)arg;
int mb_x, mb_y, pdif = 0;
int chr_h= 16>>s->chroma_y_shift;
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];
//printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
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;
s->current_picture.error[i] = 0;
}
s->mb_skip_run = 0;
memset(s->last_mv, 0, sizeof(s->last_mv));
s->last_mv_dir = 0;
switch(s->codec_id){
case CODEC_ID_H263:
case CODEC_ID_H263P:
case CODEC_ID_FLV1:
if (CONFIG_H263_ENCODER || CONFIG_H263P_ENCODER || CONFIG_FLV_ENCODER)
s->gob_index = ff_h263_get_gob_height(s);
break;
case CODEC_ID_MPEG4:
if(CONFIG_MPEG4_ENCODER && s->partitioned_frame)
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++) {
// printf("row %d at %X\n", s->mb_y, (int)s);
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);
if(CONFIG_H261_ENCODER && s->codec_id == CODEC_ID_H261){
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){
case CODEC_ID_H263:
case CODEC_ID_H263P:
if(!s->h263_slice_structured)
if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
break;
case CODEC_ID_MPEG2VIDEO:
if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
case CODEC_ID_MPEG1VIDEO:
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);
if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
ff_mpeg4_init_partitions(s);
}
}
assert((put_bits_count(&s->pb)&7) == 0);
current_packet_size= pbBufPtr(&s->pb) - s->ptr_lastgob;
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;
#ifndef ALT_BITSTREAM_WRITER
s->pb.buf_ptr= s->ptr_lastgob;
#endif
assert(pbBufPtr(&s->pb) == s->ptr_lastgob);
}
}
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);
}
switch(s->codec_id){
case CODEC_ID_MPEG4:
if (CONFIG_MPEG4_ENCODER) {
ff_mpeg4_encode_video_packet_header(s);
ff_mpeg4_clean_buffers(s);
}
break;
case CODEC_ID_MPEG1VIDEO:
case CODEC_ID_MPEG2VIDEO:
if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) {
ff_mpeg1_encode_slice_header(s);
ff_mpeg1_clean_buffers(s);
}
break;
case CODEC_ID_H263:
case CODEC_ID_H263P:
if (CONFIG_H263_ENCODER || CONFIG_H263P_ENCODER)
h263_encode_gob_header(s, mb_y);
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
if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
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++){
s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
}
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