(转)x264源码分析（1）：main、parse、encode、x264_encoder_open函数代码分析

转自：http://nkwavelet.blog.163.com/blog/static/2277560382013103010312144/

x264版本：   x264-snapshot-20140226-2245

 1.     首先对主函数进行分析，main函数很简洁，主要有三个步骤，见下图：

2.   接下来分析一下Parse函数中的主要过程：

static int parse( int argc, char **argv, x264_param_t *param, cli_opt_t *opt )

{

        // 设置编码器默认参数，此处并不是真正设置编码器参数，而是设置到变量defaults中，

// 该变量用于命令行help查看的时候，显示默认的参数值而已。
// 该函数的详细分析见：x264编码器默认参数值

x264_param_default( &defaults );

/* Presets are applied before all other options. */
for( optind = 0;; )
{ // 获取preset 和tune
int c = getopt_long( argc, argv, short_options, long_options, NULL );
... ... ... ...
}

       
// 如果preset设置为placebo，此设置获取更大的psnr值，但是速度非常慢，通常不采用此设置

if ( preset && !strcasecmp( preset, "placebo" ) )
b_turbo = 0;

// 首先设置缺省的编码器参数，然后根据给定的preset和tune修改部分编码器参数
// 该函数代码详细分析见：x264_param_xxxx 系列函数代码分析
if ( x264_param_default_preset( param, preset, tune ) < 0 )
return -1;

/*  解释命令行参数 */
for( optind = 0;; )
{      ... ... ... ...     }

/* If first pass mode is used, apply faster settings. */
if ( b_turbo )  
x264_param_apply_fastfirstpass( param );

/* Apply profile restrictions. 主要用于检测设置的profile与其他的编码器参数是否有冲突 */
// 该函数代码详细分析见：x264_param_xxxx 系列函数代码分析
if ( x264_param_apply_profile( param, profile ) < 0 )
return -1;

// 根据输出文件名的后缀名选择相应的文件操作函数集合
if ( select_output( muxer, output_filename, param ) )
return -1;

// 打开输出文件，如果输出文件是flv格式，则此处调用的函数
// 就是 output/flv.c 文件中的open_file 函数
cli_output.open_file( output_filename, &opt->hout, &output_opt );

// 选择输入文件打开方式，有avs, y4m, ffms, lavf, raw等方式，
// 具体方式还依赖编译选项的支持
if( select_input( demuxer, demuxername, input_filename, &opt->hin, &info, &input_opt ) )
return -1;

// 打开输入文件
cli_input.open_file( input_filename, &opt->hin, &info, &input_opt );

// 计算宽高比和帧率，即约分数表示
x264_reduce_fraction( &info.sar_width, &info.sar_height );
x264_reduce_fraction( &info.fps_num, &info.fps_den );

// 初始化视频filter参数，该函数定义在x264.c中
if( init_vid_filters( vid_filters, &opt->hin, &info, param, output_csp ) )
return -1;

/* set param flags from the post-filtered video */
param->b_vfr_input = info.vfr;
param->i_fps_num = info.fps_num;
param->i_fps_den = info.fps_den;
param->i_timebase_num = info.timebase_num;
param->i_timebase_den = info.timebase_den;
param->vui.i_sar_width  = info.sar_width;
param->vui.i_sar_height = info.sar_height;

/* Automatically reduce reference frame count to match the user's target 
 level if the user didn't explicitly set a reference frame count. */
param->i_frame_reference = .......
}

 

3.  编码函数encode分析如下：

static int encode( x264_param_t *param, cli_opt_t *opt )
{
      x264_t *h  =  NULL;         // 结构体x264_t的定义在common/common.h文件中
      x264_picture_t  pic;        // 结构体x264_picture_t的定义在x264.h文件中
      cli_pic_t   cli_pic;            // 结构体cli_pic_t的定义在input/input.h文件中

 

      // 函数x264_encoder_open定义在encoder/encoder.c文件中，这个函数是对不正确的参数进行修改，
      // 并对各结构体参数和cabac编码、预测等需要的参数进行初始化，打开编码器句柄。
      // 通过x264_encoder_parameters得到设置给x264的参数，通过x264_encoder_reconfig更新x264参数
      x264_t *h = x264_encoder_open( param );

 

      // 该函数定义在encoder/encoder.c文件中，该函数只有一句话，将线程中x264_t参数集拷贝到param中
      x264_encoder_parameters( h, param );

 

      /* ticks/frame = ticks/second / frames/second */
      ticks_per_frame = (int64_t)param->i_timebase_den * param->i_fps_den

                                       / param->i_timebase_num  / param->i_fps_num;
      ticks_per_frame = X264_MAX( ticks_per_frame, 1 );

 

      if ( !param->b_repeat_headers )
      {
              // Write SPS/PPS/SEI
             x264_nal_t *headers;
             int i_nal;

 

             // 该函数定义在encoder/encoder.c文件中
             x264_encoder_headers( h, &headers, &i_nal );

 

             // 写文件头，如果是flv文件，则调用output/flv.c 文件中的write_headers函数
             cli_output.write_headers( opt->hout, headers );

      }

 

      /* Encode frames，循环编码每一帧 */

      for( ; !b_ctrl_c && (i_frame < param->i_frame_total || !param->i_frame_total); i_frame++ )

      {

             // 从输入文件中获取一帧，存放在cli_pic中

             if ( filter.get_frame( opt->hin, &cli_pic, i_frame + opt->i_seek ) )
                   break;

             

              // 初始化一帧图片,该函数定义在common/common.c文件中
              x264_picture_init( &pic );

 

              // 将cli_pic 转换成x264_picture_t格式的pic
              convert_cli_to_lib_pic( &pic, &cli_pic );

 

              // 对一帧编码，返回编码后的字节数，该函数定义在x264.c中
              i_frame_size = encode_frame( h, opt->hout, &pic, &last_dts );
              // 此处将函数encode_frame展开来分析
             {
                      x264_picture_t pic_out;
                     x264_nal_t *nal;
                     int i_nal;
                     int i_frame_size = 0;

 

                     // 具体执行编码一帧图片，该函数定义在x264的encoder/encoder.c文件中，执行完之后得到nal 和i_nal的值
                     // 该函数详细分析见：x264_encoder_encode 函数代码分析
                     i_frame_size = x264_encoder_encode( h, &nal, &i_nal, pic, &pic_out );
                    if ( i_frame_size )
                   {          // 将一帧数据写入到输出文件中
                              i_frame_size = cli_output.write_frame( hout, nal[0].p_payload, i_frame_size, &pic_out );
                             *last_dts = pic_out.i_dts;
                   } 
                  return i_frame_size;
            }

 

           if( i_frame_size < 0 )
           {    
                     b_ctrl_c = 1;        /* lie to exit the loop */
                     retval = -1;
           }    
          else if ( i_frame_size )
          {    
                    i_file += i_frame_size;
                    i_frame_output++;
                   if( i_frame_output == 1 )
                           first_dts = prev_dts = last_dts;
           }

 

           // 释放一帧
          if ( filter.release_frame( opt->hin, &cli_pic, i_frame + opt->i_seek ) )
                      break;

      }  // 循环编码一帧结束

 

      /* Flush delayed frames */
      while( !b_ctrl_c && x264_encoder_delayed_frames( h ) )
      {
                  i_frame_size = encode_frame( h, opt->hout, NULL, &last_dts );
                  ... ... ... ...
       }

 

       // 关闭编码器
       x264_encoder_close( h );

 

       // 关闭输出文件
       cli_output.close_file( opt->hout, largest_pts, second_largest_pts );

 

       return retval;

}

4.  编码函数x264_encoder_open分析如下：

x264_t* x264_encoder_open( x264_param_t *param )
{
/* 结构体x264_t的定义在common/common.h中 */
x264_t *h;

/* Create a copy of param, 创建一个param拷贝 */
memcpy( &h->param, param, sizeof(x264_param_t) );

/*  x264线程初始化 */
x264_threading_init();

/* 该函数位于encoder/encoder.c中，主要用于检查h->param中的参数设置是否有冲突  */
if( x264_validate_parameters( h, 1 ) < 0 )
goto fail;

/* 如果提供了外部量化矩阵文件，则读取分析该量化矩阵文件 */

if( h->param.psz_cqm_file )
if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 ) 
goto fail;

/* 用于2pass模式，保存编码的统计数据 */
if( h->param.rc.psz_stat_out ) 
h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out ); 
if( h->param.rc.psz_stat_in ) 
h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );

/* 将fps和timebase化为既约分数 */
x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den ); 
x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );

/* Init x264_t */

h->i_frame = -1;

h->i_frame_num = 0;

h->i_idr_pic_id = 0;

·/* 设置vui的宽高比，也就是将vui.i_sar_width和vui.i_sar_height按比例缩放到65535以内并化为即约分数，通常可以忽略 */

x264_set_aspect_ratio( h, &h->param, 1 );

/* 初始化sps和pps，这两个函数位于encoder/set.c中 */
x264_sps_init( h->sps, h->param.i_sps_id, &h->param ); 
x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );

/* 检查设定的level是否有效，该函数及数组x264_levels[] 定义定义在encoder/set.c中  */

/* 有关各种level的极限值，可参考h264官方协议：Table A-1 – Level limits */

x264_validate_levels( h, 1 );

h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;

/* 初始化量化矩阵  */
if( x264_cqm_init( h ) < 0 )
goto fail;

/* 每行每列及一帧总的宏块数目  */
h->mb.i_mb_width  = h->sps->i_mb_width; 
h->mb.i_mb_height = h->sps->i_mb_height; 
h->mb.i_mb_count  = h->mb.i_mb_width * h->mb.i_mb_height; 

// 对于csp取值X264_CSP_I420而言，以下两个值都是1
h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422; 
h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;

/* Adaptive MBAFF and subme 0 are not supported as we require halving motion 
 * vectors during prediction, resulting in hpel mvs. 
 * The chosen solution is to make MBAFF non-adaptive in this case. */
// 忽略隔行，因此下面变量值通常为0
h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;

/* Init frames. 初始化一些帧设置，包括延迟帧数、参考帧数、关键帧间隔等 */ 
// i_bframe_adaptive的默认值为X264_B_ADAPT_FAST，rc.b_stat_read通常取值0

// i_bframe_adaptive可以通过命令行参数--b-adapt设定，可以取值0、1、2，默认值为1，即X264_B_ADAPT_FAST

// param.i_bframe的值可以通过命令行参数--bframes设定
if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read ) 
h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4; 
else 
h->frames.i_delay = h->param.i_bframe;

// mb_tree默认是开启的，i_vbv_buffer_size默认值为0，
// rc.i_lookahead默认值为40，但是对于不同的preset，rc.i_lookahead取不同的值
// rc.i_lookahead的值还可以通过命令行参数--rc-lookahead设定
if ( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size ) 
h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead ); 

i_slicetype_length = h->frames.i_delay; 
h->frames.i_delay += h->i_thread_frames - 1;  // h->i_thread_frames通常取值1
h->frames.i_delay += h->param.i_sync_lookahead;  // h->param.i_sync_lookahead通常取值0
h->frames.i_delay += h->param.b_vfr_input;  // h->param.b_vfr_input默认值为1

// param.i_bframe_pyramid的值可以通过命令行参数--b-pyramid设定，可以取值0、1、2
// --b-pyramid表示是否允许B帧作为参考帧，默认值为normal=2，none=0表示不允许B帧作为参考帧
h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0; 

// 获取前向、后向最大参考帧数，解码缓冲区最大帧数
h->frames.i_max_ref0 = h->param.i_frame_reference; 
h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference ); 
h->frames.i_max_dpb = h->sps->vui.i_max_dec_frame_buffering; 
h->frames.b_have_lowres = !h->param.rc.b_stat_read 
 && ( h->param.rc.i_rc_method == X264_RC_ABR 
 || h->param.rc.i_rc_method == X264_RC_CRF 
 || h->param.i_bframe_adaptive 
 || h->param.i_scenecut_threshold 
 || h->param.rc.b_mb_tree 
 || h->param.analyse.i_weighted_pred );  // h->frames.b_have_lowres通常取值1
h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0; 
h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);

// i_keyint_max默认值250，上一个idr帧和上一个关键帧的序号，初始化为负值
h->frames.i_last_idr = 
h->frames.i_last_keyframe = - h->param.i_keyint_max;
h->frames.i_input = 0; 
h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;  // 初始化帧的最大pts和次大pts值为-1
h->frames.i_poc_last_open_gop = -1;

/* Allocate room for max refs plus a few extra just in case. */

CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe 
 + h->i_thread_frames + 3) * sizeof(x264_frame_t *) ); 
if ( h->param.analyse.i_weighted_pred > 0 )  // 如果开启了P帧加权预测，则还需要申请额外的帧空间；默认是开启的
CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );

h->i_ref[0] = h->i_ref[1] = 0;  // 初始化前向和后向参考帧数为0
h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0; 
h->i_prev_duration = ((uint64_t)h->param.i_fps_den * h->sps->vui.i_time_scale) 
/ ((uint64_t)h->param.i_fps_num * h->sps->vui.i_num_units_in_tick); 
h->i_disp_fields_last_frame = -1;

/* precalculate the cost of coding various combinations of bits in a single context，该函数定义在encoder/rdo.c中 */
x264_rdo_init();

/* init CPU functions */

x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );  // 该函数定义在common/predict.c中 
x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );  // 该函数定义在common/predict.c中
x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );  // 该函数定义在common/predict.c中
x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );  // 该函数定义在common/predict.c中
x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );  // 该函数定义在common/predict.c中
x264_pixel_init( h->param.cpu, &h->pixf );  // 该函数定义在common/pixel.c中 

x264_dct_init( h->param.cpu, &h->dctf );  // 该函数定义在common/dct.c中 
x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced ); // 该函数定义在common/dct.c中 
memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) ); 
x264_mc_init( h->param.cpu, &h->mc, h->param.b_cpu_independent );  // 该函数定义在common/mc.c中 
x264_quant_init( h, h->param.cpu, &h->quantf );  // 该函数定义在common/quant.c中
x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED ); // 该函数定义在common/deblock.c中

x264_bitstream_init( h->param.cpu, &h->bsf ); // 该函数定义在common/bitstream.c中
if( h->param.b_cabac ) 
x264_cabac_init( h );  // 该函数定义在common/cabac.c中 
else  // 函数x264_cavlc_init定义在common/vlc.c中 
x264_stack_align( x264_cavlc_init, h ); // 该函数定义在common/cpu.h中

mbcmp_init( h );  // 该函数定义在encoder/encoder.c中 
  chroma_dsp_init( h ); // 该函数定义在encoder/encoder.c中

/*  输出cpu可以利用的多媒体指令集，例如mmx、sse2、sse3等 */
p = buf + sprintf( buf, "using cpu capabilities:" );
for( int i = 0; x264_cpu_names[i].flags; i++ ) // 数组x264_cpu_names定义在common/cpu.c中
{ ... ... }
if( !h->param.cpu ) 
p += sprintf( p, " none!" );
x264_log( h, X264_LOG_INFO, "%s\n", buf );

/* 生成数组logs，其中logs[0] = 0.718、 logs[i] = 2 * log2(i + 1) + 1.718，
   其中log2表示以2为底的对数，该函数定义在encoder/analyse.c中  */

float *logs = x264_analyse_prepare_costs( h );

/* 初始化不同qp值所对应的代价，该函数定义在encoder/analyse.c中  */

for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
if ( x264_analyse_init_costs( h, logs, qp ) ) 
goto fail;
if ( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )

goto fail;
x264_free( logs );

static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
/* Checks for known miscompilation issues. */
if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
goto fail;

/* Must be volatile or else GCC will optimize it out. */

volatile int temp = 392;
if( x264_clz( temp ) != 23 )
goto fail;

h->out.i_nal = 0;
h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
 * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
: pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4 + 64;  /* +4 for startcode, +64 for nal_escape assembly padding */

// 有关线程池、线程锁定等代码
... ...

if ( x264_lookahead_init( h, i_slicetype_length ) ) // 该函数定义在encoder/lookahead.c中
goto fail;

for( int i = 0; i < h->param.i_threads; i++ )
if ( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 ) // 该函数定义在common/macroblock.c中
goto fail;

/* 初始化码率控制的一些参数，该函数详细分析见：x264_ratecontrol_new 函数代码分析（码率控制） */

if ( x264_ratecontrol_new( h ) < 0 ) // 该函数定义在encoder/ratecontrol.c中
goto fail;

/* 输出profile、level等一些日志信息 */
... ...

return h;

}