]> git.sesse.net Git - vlc/blobdiff - modules/codec/ffmpeg/encoder.c
* modules/codec/ffmpeg: ffmpeg's av_log() messages now go to our messages
[vlc] / modules / codec / ffmpeg / encoder.c
index d97c257088075c8f2b426e3e4ce5780aed4f2794..a341cf936f50e07cf0216640e24b6e8124707279 100644 (file)
@@ -1,7 +1,7 @@
 /*****************************************************************************
  * encoder.c: video and audio encoder using the ffmpeg library
  *****************************************************************************
- * Copyright (C) 1999-2004 VideoLAN
+ * Copyright (C) 1999-2004 the VideoLAN team
  * $Id$
  *
  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
@@ -43,6 +43,9 @@
 #if LIBAVCODEC_BUILD < 4704
 #   define AV_NOPTS_VALUE 0
 #endif
+#if LIBAVCODEC_BUILD < 4684
+#    define FF_QP2LAMBDA 118
+#endif
 
 #include "ffmpeg.h"
 
@@ -51,6 +54,8 @@
 #define HURRY_UP_GUARD2 (300000)
 #define HURRY_UP_GUARD3 (100000)
 
+#define MAX_FRAME_DELAY (FF_MAX_B_FRAMES + 2)
+
 /*****************************************************************************
  * Local prototypes
  *****************************************************************************/
@@ -129,25 +134,43 @@ struct encoder_sys_t
     float      f_rc_buffer_aggressivity;
     vlc_bool_t b_pre_me;
     vlc_bool_t b_hurry_up;
-    vlc_bool_t b_interlace;
+    vlc_bool_t b_interlace, b_interlace_me;
     float      f_i_quant_factor;
     int        i_noise_reduction;
     vlc_bool_t b_mpeg4_matrix;
     vlc_bool_t b_trellis;
+    int        i_quality; /* for VBR */
+    float      f_lumi_masking, f_dark_masking, f_p_masking, f_border_masking;
+    int        i_luma_elim, i_chroma_elim;
+
+    /* Used to work around stupid timestamping behaviour in libavcodec */
+    uint64_t i_framenum;
+    mtime_t  pi_delay_pts[MAX_FRAME_DELAY];
 };
 
 static const char *ppsz_enc_options[] = {
-    "keyint", "bframes", "vt", "qmin", "qmax", "hq", "strict_rc",
+    "keyint", "bframes", "vt", "qmin", "qmax", "hq", "strict-rc",
     "rc-buffer-size", "rc-buffer-aggressivity", "pre-me", "hurry-up",
     "interlace", "i-quant-factor", "noise-reduction", "mpeg4-matrix",
-    "trellis", NULL
+    "trellis", "qscale", "strict", "lumi-masking", "dark-masking",
+    "p-masking", "border-masking", "luma-elim-threshold",
+    "chroma-elim-threshold", NULL
 };
 
+static const uint16_t mpa_bitrate_tab[2][15] =
+{
+    {0, 32, 48, 56, 64, 80, 96, 112, 128, 160, 192, 224, 256, 320, 384},
+    {0, 8, 16, 24, 32, 40, 48, 56, 64, 80, 96, 112, 128, 144, 160}
+};
+
+static const uint16_t mpa_freq_tab[6] =
+{ 44100, 48000, 32000, 22050, 24000, 16000 };
+
 /*****************************************************************************
  * OpenEncoder: probe the encoder
  *****************************************************************************/
-extern int16_t ff_mpeg4_default_intra_matrix[];
-extern int16_t ff_mpeg4_default_non_intra_matrix[];
+extern int16_t IMPORT_SYMBOL ff_mpeg4_default_intra_matrix[];
+extern int16_t IMPORT_SYMBOL ff_mpeg4_default_non_intra_matrix[];
 
 int E_(OpenEncoder)( vlc_object_t *p_this )
 {
@@ -158,6 +181,9 @@ int E_(OpenEncoder)( vlc_object_t *p_this )
     int i_codec_id, i_cat;
     char *psz_namecodec;
     vlc_value_t val;
+    vlc_value_t lockval;
+
+    var_Get( p_enc->p_libvlc, "avcodec", &lockval );
 
     if( !E_(GetFfmpegCodec)( p_enc->fmt_out.i_codec, &i_cat, &i_codec_id,
                              &psz_namecodec ) )
@@ -201,6 +227,7 @@ int E_(OpenEncoder)( vlc_object_t *p_this )
         msg_Err( p_enc, "out of memory" );
         return VLC_EGENERIC;
     }
+    memset( p_sys, 0, sizeof(encoder_sys_t) );
     p_enc->p_sys = p_sys;
     p_sys->p_codec = p_codec;
 
@@ -209,9 +236,10 @@ int E_(OpenEncoder)( vlc_object_t *p_this )
 
     p_sys->p_buffer_out = NULL;
     p_sys->p_buffer = NULL;
-    p_sys->b_inited = 0;
 
     p_sys->p_context = p_context = avcodec_alloc_context();
+    p_context->debug = config_GetInt( p_enc, "ffmpeg-debug" );
+    p_context->opaque = (void *)p_this;
 
     /* Set CPU capabilities */
     p_context->dsp_mask = 0;
@@ -233,7 +261,7 @@ int E_(OpenEncoder)( vlc_object_t *p_this )
         p_context->dsp_mask |= FF_MM_SSE2;
     }
 
-    sout_ParseCfg( p_enc, ENC_CFG_PREFIX, ppsz_enc_options, p_enc->p_cfg );
+    sout_CfgParse( p_enc, ENC_CFG_PREFIX, ppsz_enc_options, p_enc->p_cfg );
 
     var_Get( p_enc, ENC_CFG_PREFIX "keyint", &val );
     p_sys->i_key_int = val.i_int;
@@ -247,6 +275,9 @@ int E_(OpenEncoder)( vlc_object_t *p_this )
     var_Get( p_enc, ENC_CFG_PREFIX "interlace", &val );
     p_sys->b_interlace = val.b_bool;
 
+    var_Get( p_enc, ENC_CFG_PREFIX "interlace-me", &val );
+    p_sys->b_interlace_me = val.b_bool;
+
     var_Get( p_enc, ENC_CFG_PREFIX "pre-me", &val );
     p_sys->b_pre_me = val.b_bool;
 
@@ -274,6 +305,10 @@ int E_(OpenEncoder)( vlc_object_t *p_this )
     var_Get( p_enc, ENC_CFG_PREFIX "mpeg4-matrix", &val );
     p_sys->b_mpeg4_matrix = val.b_bool;
 
+    var_Get( p_enc, ENC_CFG_PREFIX "qscale", &val );
+    if( val.f_float < 0.01 || val.f_float > 255.0 ) val.f_float = 0;
+    p_sys->i_quality = (int)(FF_QP2LAMBDA * val.f_float + 0.5);
+
     var_Get( p_enc, ENC_CFG_PREFIX "hq", &val );
     if( val.psz_string && *val.psz_string )
     {
@@ -295,8 +330,27 @@ int E_(OpenEncoder)( vlc_object_t *p_this )
     var_Get( p_enc, ENC_CFG_PREFIX "trellis", &val );
     p_sys->b_trellis = val.b_bool;
 
+    var_Get( p_enc, ENC_CFG_PREFIX "strict", &val );
+    if( val.i_int < - 1 || val.i_int > 1 ) val.i_int = 0;
+    p_context->strict_std_compliance = val.i_int;
+
+    var_Get( p_enc, ENC_CFG_PREFIX "lumi-masking", &val );
+    p_sys->f_lumi_masking = val.f_float;
+    var_Get( p_enc, ENC_CFG_PREFIX "dark-masking", &val );
+    p_sys->f_dark_masking = val.f_float;
+    var_Get( p_enc, ENC_CFG_PREFIX "p-masking", &val );
+    p_sys->f_p_masking = val.f_float;
+    var_Get( p_enc, ENC_CFG_PREFIX "border-masking", &val );
+    p_sys->f_border_masking = val.f_float;
+    var_Get( p_enc, ENC_CFG_PREFIX "luma-elim-threshold", &val );
+    p_sys->i_luma_elim = val.i_int;
+    var_Get( p_enc, ENC_CFG_PREFIX "chroma-elim-threshold", &val );
+    p_sys->i_chroma_elim = val.i_int;
+
     if( p_enc->fmt_in.i_cat == VIDEO_ES )
     {
+        int i_aspect_num, i_aspect_den;
+
         if( !p_enc->fmt_in.video.i_width || !p_enc->fmt_in.video.i_height )
         {
             msg_Warn( p_enc, "invalid size %ix%i", p_enc->fmt_in.video.i_width,
@@ -308,8 +362,13 @@ int E_(OpenEncoder)( vlc_object_t *p_this )
         p_context->width = p_enc->fmt_in.video.i_width;
         p_context->height = p_enc->fmt_in.video.i_height;
 
+#if LIBAVCODEC_BUILD >= 4754
+        p_context->time_base.num = p_enc->fmt_in.video.i_frame_rate_base;
+        p_context->time_base.den = p_enc->fmt_in.video.i_frame_rate;
+#else
         p_context->frame_rate = p_enc->fmt_in.video.i_frame_rate;
         p_context->frame_rate_base= p_enc->fmt_in.video.i_frame_rate_base;
+#endif
 
         /* Defaults from ffmpeg.c */
         p_context->qblur = 0.5;
@@ -319,18 +378,29 @@ int E_(OpenEncoder)( vlc_object_t *p_this )
         p_context->i_quant_offset = 0.0;
         p_context->i_quant_factor = -0.8;
 
+        p_context->lumi_masking = p_sys->f_lumi_masking;
+        p_context->dark_masking = p_sys->f_dark_masking;
+        p_context->p_masking = p_sys->f_p_masking;
+#if LIBAVCODEC_BUILD >= 4741
+        p_context->border_masking = p_sys->f_border_masking;
+#endif
+        p_context->luma_elim_threshold = p_sys->i_luma_elim;
+        p_context->chroma_elim_threshold = p_sys->i_chroma_elim;
+
         if( p_sys->i_key_int > 0 )
             p_context->gop_size = p_sys->i_key_int;
         p_context->max_b_frames =
-            __MIN( p_sys->i_b_frames, FF_MAX_B_FRAMES );
+            __MAX( __MIN( p_sys->i_b_frames, FF_MAX_B_FRAMES ), 0 );
         p_context->b_frame_strategy = 0;
 
 #if LIBAVCODEC_BUILD >= 4687
+        av_reduce( &i_aspect_num, &i_aspect_den,
+                   p_enc->fmt_in.video.i_aspect,
+                   VOUT_ASPECT_FACTOR, 1 << 30 /* something big */ );
         av_reduce( &p_context->sample_aspect_ratio.num,
-                  &p_context->sample_aspect_ratio.den,
-                  p_enc->fmt_in.video.i_aspect *
-                  (int64_t)p_context->height / p_context->width,
-                  VOUT_ASPECT_FACTOR, 1 << 30 /* something big */ );
+                   &p_context->sample_aspect_ratio.den,
+                   i_aspect_num * (int64_t)p_context->height,
+                   i_aspect_den * (int64_t)p_context->width, 1 << 30 );
 #else
         p_context->aspect_ratio = ((float)p_enc->fmt_in.video.i_aspect) /
             VOUT_ASPECT_FACTOR;
@@ -339,11 +409,29 @@ int E_(OpenEncoder)( vlc_object_t *p_this )
         p_sys->p_buffer_out = malloc( AVCODEC_MAX_VIDEO_FRAME_SIZE );
 
         p_enc->fmt_in.i_codec = VLC_FOURCC('I','4','2','0');
+        p_context->pix_fmt = E_(GetFfmpegChroma)( p_enc->fmt_in.i_codec );
+#if LIBAVCODEC_BUILD >= 4714
+        if( p_codec->pix_fmts )
+        {
+            const enum PixelFormat *p = p_codec->pix_fmts;
+            for( ; *p != -1; p++ )
+            {
+                if( *p == p_context->pix_fmt ) break;
+            }
+            if( *p == -1 ) p_context->pix_fmt = p_codec->pix_fmts[0];
+            p_enc->fmt_in.i_codec = E_(GetVlcChroma)( p_context->pix_fmt );
+        }
+#else
+        p_enc->fmt_in.i_codec = E_(GetVlcChroma)( p_context->pix_fmt );
+#endif
 
         if ( p_sys->b_strict_rc )
         {
             p_context->rc_max_rate = p_enc->fmt_out.i_bitrate;
             p_context->rc_buffer_size = p_sys->i_rc_buffer_size;
+            /* This is from ffmpeg's ffmpeg.c : */
+            p_context->rc_initial_buffer_occupancy
+                = p_sys->i_rc_buffer_size * 3/4;
             p_context->rc_buffer_aggressivity = p_sys->f_rc_buffer_aggressivity;
         }
 
@@ -368,15 +456,29 @@ int E_(OpenEncoder)( vlc_object_t *p_this )
 
         if ( p_sys->b_interlace )
         {
-            p_context->flags |= CODEC_FLAG_INTERLACED_DCT;
+            if ( p_context->height <= 280 )
+            {
+                if ( p_context->height != 16 || p_context->width != 16 )
+                    msg_Warn( p_enc,
+                        "disabling interlaced video because height=%d <= 280",
+                        p_context->height );
+            }
+            else
+            {
+                p_context->flags |= CODEC_FLAG_INTERLACED_DCT;
 #if LIBAVCODEC_BUILD >= 4698
-            p_context->flags |= CODEC_FLAG_INTERLACED_ME;
+                if ( p_sys->b_interlace_me )
+                    p_context->flags |= CODEC_FLAG_INTERLACED_ME;
+            }
 #endif
         }
 
         if ( p_sys->b_trellis )
             p_context->flags |= CODEC_FLAG_TRELLIS_QUANT;
 
+        if ( p_sys->i_qmin > 0 && p_sys->i_qmin == p_sys->i_qmax )
+            p_context->flags |= CODEC_FLAG_QSCALE;
+
 #if LIBAVCODEC_BUILD >= 4702
         if ( p_enc->i_threads >= 1 )
             p_context->thread_count = p_enc->i_threads;
@@ -392,9 +494,21 @@ int E_(OpenEncoder)( vlc_object_t *p_this )
         p_context->max_qdiff = 3;
 
         p_context->mb_decision = p_sys->i_hq;
+
+        if( p_sys->i_quality )
+        {
+            p_context->flags |= CODEC_FLAG_QSCALE;
+#if LIBAVCODEC_BUILD >= 4668
+            p_context->global_quality = p_sys->i_quality;
+#endif
+        }
     }
     else if( p_enc->fmt_in.i_cat == AUDIO_ES )
     {
+        /* work around bug in libmp3lame encoding */
+        if( i_codec_id == CODEC_ID_MP3 && p_enc->fmt_in.audio.i_channels > 2 )
+            p_enc->fmt_in.audio.i_channels = 2;
+
         p_enc->fmt_in.i_codec  = AOUT_FMT_S16_NE;
         p_context->sample_rate = p_enc->fmt_in.audio.i_rate;
         p_context->channels    = p_enc->fmt_in.audio.i_channels;
@@ -415,19 +529,67 @@ int E_(OpenEncoder)( vlc_object_t *p_this )
     p_context->extradata = NULL;
     p_context->flags |= CODEC_FLAG_GLOBAL_HEADER;
 
+    vlc_mutex_lock( lockval.p_address );
     if( avcodec_open( p_context, p_codec ) )
     {
-        if( p_enc->fmt_in.i_cat == AUDIO_ES && p_context->channels > 2 )
+        vlc_mutex_unlock( lockval.p_address );
+        if( p_enc->fmt_in.i_cat == AUDIO_ES &&
+             (p_context->channels > 2 || i_codec_id == CODEC_ID_MP2
+               || i_codec_id == CODEC_ID_MP3) )
         {
-            p_context->channels = 2;
-            p_enc->fmt_in.audio.i_channels = 2; // FIXME
+            if( p_context->channels > 2 )
+            {
+                p_context->channels = 2;
+                p_enc->fmt_in.audio.i_channels = 2; // FIXME
+                msg_Warn( p_enc, "stereo mode selected (codec limitation)" );
+            }
+
+            if( i_codec_id == CODEC_ID_MP2 || i_codec_id == CODEC_ID_MP3 )
+            {
+                int i_frequency, i;
+
+                for ( i_frequency = 0; i_frequency < 6; i_frequency++ )
+                {
+                    if ( p_enc->fmt_out.audio.i_rate
+                            == mpa_freq_tab[i_frequency] )
+                        break;
+                }
+                if ( i_frequency == 6 )
+                {
+                    msg_Err( p_enc, "MPEG audio doesn't support frequency=%d",
+                             p_enc->fmt_out.audio.i_rate );
+                    free( p_sys );
+                    return VLC_EGENERIC;
+                }
+
+                for ( i = 1; i < 14; i++ )
+                {
+                    if ( p_enc->fmt_out.i_bitrate / 1000
+                          <= mpa_bitrate_tab[i_frequency / 3][i] )
+                        break;
+                }
+                if ( p_enc->fmt_out.i_bitrate / 1000
+                      != mpa_bitrate_tab[i_frequency / 3][i] )
+                {
+                    msg_Warn( p_enc,
+                              "MPEG audio doesn't support bitrate=%d, using %d",
+                              p_enc->fmt_out.i_bitrate,
+                              mpa_bitrate_tab[i_frequency / 3][i] * 1000 );
+                    p_enc->fmt_out.i_bitrate =
+                        mpa_bitrate_tab[i_frequency / 3][i] * 1000;
+                    p_context->bit_rate = p_enc->fmt_out.i_bitrate;
+                }
+            }
+
+            p_context->codec = NULL;
+            vlc_mutex_lock( lockval.p_address );
             if( avcodec_open( p_context, p_codec ) )
             {
+                vlc_mutex_unlock( lockval.p_address );
                 msg_Err( p_enc, "cannot open encoder" );
                 free( p_sys );
                 return VLC_EGENERIC;
             }
-            msg_Warn( p_enc, "stereo mode selected (codec limitation)" );
         }
         else
         {
@@ -436,6 +598,7 @@ int E_(OpenEncoder)( vlc_object_t *p_this )
             return VLC_EGENERIC;
         }
     }
+    vlc_mutex_unlock( lockval.p_address );
 
     p_enc->fmt_out.i_extra = p_context->extradata_size;
     p_enc->fmt_out.p_extra = p_context->extradata;
@@ -448,12 +611,6 @@ int E_(OpenEncoder)( vlc_object_t *p_this )
         p_sys->p_buffer = malloc( p_sys->i_frame_size );
     }
 
-    p_sys->i_last_ref_pts = 0;
-    p_sys->i_buggy_pts_detect = 0;
-    p_sys->i_samples_delay = 0;
-    p_sys->i_pts = 0;
-    p_sys->i_last_pts = 0;
-
     msg_Dbg( p_enc, "found encoder %s", psz_namecodec );
 
     return VLC_SUCCESS;
@@ -584,11 +741,11 @@ static block_t *EncodeVideo( encoder_t *p_enc, picture_t *p_pict )
     /* Let ffmpeg select the frame type */
     frame.pict_type = 0;
 
-    frame.repeat_pict = p_pict->i_nb_fields;
+    frame.repeat_pict = p_pict->i_nb_fields - 2;
 
 #if LIBAVCODEC_BUILD >= 4685
     frame.interlaced_frame = !p_pict->b_progressive;
-    frame.top_field_first = p_pict->b_top_field_first;
+    frame.top_field_first = !!p_pict->b_top_field_first;
 #endif
 
 #if LIBAVCODEC_BUILD < 4702
@@ -656,12 +813,32 @@ static block_t *EncodeVideo( encoder_t *p_enc, picture_t *p_pict )
                       "same PTS (" I64Fd ")", frame.pts );
             return NULL;
         }
+        else if ( p_sys->i_last_pts > frame.pts )
+        {
+            msg_Warn( p_enc, "almost fed libavcodec with a frame in the "
+                      "past (current: " I64Fd ", last: "I64Fd")",
+                      frame.pts, p_sys->i_last_pts );
+            return NULL;
+        }
         else
         {
             p_sys->i_last_pts = frame.pts;
         }
     }
 
+    frame.quality = p_sys->i_quality;
+
+    /* Ugly work-around for stupid libavcodec behaviour */
+#if LIBAVCODEC_BUILD >= 4722
+    p_sys->i_framenum++;
+    p_sys->pi_delay_pts[p_sys->i_framenum % MAX_FRAME_DELAY] = frame.pts;
+    frame.pts = p_sys->i_framenum * AV_TIME_BASE *
+        p_enc->fmt_in.video.i_frame_rate_base;
+    frame.pts += p_enc->fmt_in.video.i_frame_rate - 1;
+    frame.pts /= p_enc->fmt_in.video.i_frame_rate;
+#endif
+    /* End work-around */
+
     i_out = avcodec_encode_video( p_sys->p_context, p_sys->p_buffer_out,
                                   AVCODEC_MAX_VIDEO_FRAME_SIZE, &frame );
 
@@ -670,21 +847,37 @@ static block_t *EncodeVideo( encoder_t *p_enc, picture_t *p_pict )
         block_t *p_block = block_New( p_enc, i_out );
         memcpy( p_block->p_buffer, p_sys->p_buffer_out, i_out );
 
-        if( p_sys->p_context->coded_frame->pts != AV_NOPTS_VALUE &&
+        /* FIXME, 3-2 pulldown is not handled correctly */
+        p_block->i_length = I64C(1000000) *
+            p_enc->fmt_in.video.i_frame_rate_base /
+                p_enc->fmt_in.video.i_frame_rate;
+
+        if( !p_sys->p_context->max_b_frames || !p_sys->p_context->delay )
+        {
+            /* No delay -> output pts == input pts */
+            p_block->i_pts = p_block->i_dts = p_pict->date;
+        }
+        else if( p_sys->p_context->coded_frame->pts != AV_NOPTS_VALUE &&
             p_sys->p_context->coded_frame->pts != 0 &&
             p_sys->i_buggy_pts_detect != p_sys->p_context->coded_frame->pts )
         {
             p_sys->i_buggy_pts_detect = p_sys->p_context->coded_frame->pts;
+            p_block->i_pts = p_sys->p_context->coded_frame->pts;
 
-            /* FIXME, 3-2 pulldown is not handled correctly */
-            p_block->i_length = I64C(1000000) *
-                p_enc->fmt_in.video.i_frame_rate_base /
-                p_enc->fmt_in.video.i_frame_rate;
-            p_block->i_pts    = p_sys->p_context->coded_frame->pts;
+            /* Ugly work-around for stupid libavcodec behaviour */
+#if LIBAVCODEC_BUILD >= 4722
+            {
+            int64_t i_framenum = p_block->i_pts *
+                p_enc->fmt_in.video.i_frame_rate /
+                p_enc->fmt_in.video.i_frame_rate_base / AV_TIME_BASE;
+
+            p_block->i_pts = p_sys->pi_delay_pts[i_framenum % MAX_FRAME_DELAY];
+            }
+#endif
+            /* End work-around */
 
-            if( !p_sys->p_context->delay ||
-                ( p_sys->p_context->coded_frame->pict_type != FF_I_TYPE &&
-                  p_sys->p_context->coded_frame->pict_type != FF_P_TYPE ) )
+            if( p_sys->p_context->coded_frame->pict_type != FF_I_TYPE &&
+                p_sys->p_context->coded_frame->pict_type != FF_P_TYPE )
             {
                 p_block->i_dts = p_block->i_pts;
             }
@@ -707,9 +900,6 @@ static block_t *EncodeVideo( encoder_t *p_enc, picture_t *p_pict )
         {
             /* Buggy libavcodec which doesn't update coded_frame->pts
              * correctly */
-            p_block->i_length = I64C(1000000) *
-                p_enc->fmt_in.video.i_frame_rate_base /
-                p_enc->fmt_in.video.i_frame_rate;
             p_block->i_dts = p_block->i_pts = p_pict->date;
         }
 
@@ -820,6 +1010,9 @@ void E_(CloseEncoder)( vlc_object_t *p_this )
 {
     encoder_t *p_enc = (encoder_t *)p_this;
     encoder_sys_t *p_sys = p_enc->p_sys;
+    vlc_value_t lockval;
+
+    var_Get( p_enc->p_libvlc, "avcodec", &lockval );
 
 #if LIBAVCODEC_BUILD >= 4702
     if ( p_sys->b_inited && p_enc->i_threads >= 1 )
@@ -841,7 +1034,9 @@ void E_(CloseEncoder)( vlc_object_t *p_this )
     }
 #endif
 
+    vlc_mutex_lock( lockval.p_address );
     avcodec_close( p_sys->p_context );
+    vlc_mutex_unlock( lockval.p_address );
     av_free( p_sys->p_context );
 
     if( p_sys->p_buffer ) free( p_sys->p_buffer );