]> git.sesse.net Git - vlc/blobdiff - modules/packetizer/h264.c
* modules/packetizer/h264.c: fixed massive memory leak.
[vlc] / modules / packetizer / h264.c
index fa4daf40dfb35ea43d584d63321f00d7156c9f5b..08ac7ad60e9aa903428b7b924c5c11d54a7e5bab 100644 (file)
@@ -1,12 +1,12 @@
 /*****************************************************************************
  * h264.c: h264/avc video packetizer
  *****************************************************************************
- * Copyright (C) 2001, 2002 VideoLAN
+ * Copyright (C) 2001, 2002 the VideoLAN team
  * $Id$
  *
  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
  *          Eric Petit <titer@videolan.org>
- *          Gildas Bazin <gbazin@netcourrier.com>
+ *          Gildas Bazin <gbazin@videolan.org>
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
@@ -42,6 +42,8 @@ static int  Open ( vlc_object_t * );
 static void Close( vlc_object_t * );
 
 vlc_module_begin();
+    set_category( CAT_SOUT );
+    set_subcategory( SUBCAT_SOUT_PACKETIZER );
     set_description( _("H264 video packetizer") );
     set_capability( "packetizer", 50 );
     set_callbacks( Open, Close );
@@ -65,14 +67,23 @@ struct decoder_sys_t
     vlc_bool_t b_slice;
     block_t    *p_frame;
 
-    int64_t      i_dts;
-    int64_t      i_pts;
-    unsigned int i_flags;
-
     vlc_bool_t   b_sps;
+    vlc_bool_t   b_pps;
 
     /* avcC data */
     int i_avcC_length_size;
+    block_t *p_sps;
+    block_t *p_pps;
+
+    /* Useful values of the Sequence Parameter Set */
+    int i_log2_max_frame_num;
+    int b_frame_mbs_only;
+
+    /* Useful values of the Slice Header */
+    int i_nal_type;
+    int i_nal_ref_idc;
+    int i_idr_pic_id;
+    int i_frame_num;
 };
 
 enum
@@ -91,7 +102,8 @@ enum nal_unit_type_e
     NAL_SLICE_IDR   = 5,    /* ref_idc != 0 */
     NAL_SEI         = 6,    /* ref_idc == 0 */
     NAL_SPS         = 7,
-    NAL_PPS         = 8
+    NAL_PPS         = 8,
+    NAL_AU_DELIMITER= 9
     /* ref_idc == 0 for 6,9,10,11,12 */
 };
 
@@ -105,7 +117,7 @@ enum nal_priority_e
 
 static block_t *ParseNALBlock( decoder_t *, block_t * );
 
-static block_t *nal_get_encoded( decoder_t *, uint8_t *p, int );
+static block_t *nal_get_annexeb( decoder_t *, uint8_t *p, int );
 
 /*****************************************************************************
  * Open: probe the packetizer and return score
@@ -119,7 +131,8 @@ static int Open( vlc_object_t *p_this )
         p_dec->fmt_in.i_codec != VLC_FOURCC( 'H', '2', '6', '4') &&
         p_dec->fmt_in.i_codec != VLC_FOURCC( 'V', 'S', 'S', 'H') &&
         p_dec->fmt_in.i_codec != VLC_FOURCC( 'v', 's', 's', 'h') &&
-        ( p_dec->fmt_in.i_codec != VLC_FOURCC( 'a', 'v', 'c', '1') || p_dec->fmt_in.i_extra < 7 ) )
+        ( p_dec->fmt_in.i_codec != VLC_FOURCC( 'a', 'v', 'c', '1') ||
+          p_dec->fmt_in.i_extra < 7 ) )
     {
         return VLC_EGENERIC;
     }
@@ -139,14 +152,22 @@ static int Open( vlc_object_t *p_this )
     p_sys->bytestream = block_BytestreamInit( p_dec );
     p_sys->b_slice = VLC_FALSE;
     p_sys->p_frame = NULL;
-    p_sys->i_dts   = 0;
-    p_sys->i_pts   = 0;
-    p_sys->i_flags = 0;
     p_sys->b_sps   = VLC_FALSE;
+    p_sys->b_pps   = VLC_FALSE;
+    p_sys->p_sps   = 0;
+    p_sys->p_pps   = 0;
+
+    p_sys->i_nal_type = -1;
+    p_sys->i_nal_ref_idc = -1;
+    p_sys->i_idr_pic_id = -1;
+    p_sys->i_frame_num = -1;
 
     /* Setup properties */
     es_format_Copy( &p_dec->fmt_out, &p_dec->fmt_in );
     p_dec->fmt_out.i_codec = VLC_FOURCC( 'h', '2', '6', '4' );
+    /* FIXME: FFMPEG isn't happy at all if you leave this */
+    if( p_dec->fmt_out.i_extra ) free( p_dec->fmt_out.p_extra );
+    p_dec->fmt_out.i_extra = 0; p_dec->fmt_out.p_extra = 0;
 
     if( p_dec->fmt_in.i_codec == VLC_FOURCC( 'a', 'v', 'c', '1' ) )
     {
@@ -163,8 +184,10 @@ static int Open( vlc_object_t *p_this )
         for( i = 0; i < i_sps; i++ )
         {
             int i_length = GetWBE( p );
-            block_t *p_sps = nal_get_encoded( p_dec, p+2, i_length );
+            block_t *p_sps = nal_get_annexeb( p_dec, p + 2, i_length );
 
+            p_sys->p_sps = block_Duplicate( p_sps );
+            p_sps->i_pts = p_sps->i_dts = mdate();
             ParseNALBlock( p_dec, p_sps );
             p += 2 + i_length;
         }
@@ -173,8 +196,10 @@ static int Open( vlc_object_t *p_this )
         for( i = 0; i < i_pps; i++ )
         {
             int i_length = GetWBE( p );
-            block_t *p_pps = nal_get_encoded( p_dec, p+2, i_length );
+            block_t *p_pps = nal_get_annexeb( p_dec, p + 2, i_length );
 
+            p_sys->p_pps = block_Duplicate( p_pps );
+            p_pps->i_pts = p_pps->i_dts = mdate();
             ParseNALBlock( p_dec, p_pps );
             p += 2 + i_length;
         }
@@ -188,6 +213,22 @@ static int Open( vlc_object_t *p_this )
     {
         /* Set callback */
         p_dec->pf_packetize = Packetize;
+
+        /* */
+        if( p_dec->fmt_in.i_extra > 0 )
+        {
+            block_t *p_init = block_New( p_dec, p_dec->fmt_in.i_extra );
+            block_t *p_pic;
+
+            memcpy( p_init->p_buffer, p_dec->fmt_in.p_extra,
+                    p_dec->fmt_in.i_extra );
+
+            while( ( p_pic = Packetize( p_dec, &p_init ) ) )
+            {
+                /* Should not occur because we should only receive SPS/PPS */
+                block_Release( p_pic );
+            }
+        }
     }
 
     return VLC_SUCCESS;
@@ -201,6 +242,9 @@ static void Close( vlc_object_t *p_this )
     decoder_t *p_dec = (decoder_t*)p_this;
     decoder_sys_t *p_sys = p_dec->p_sys;
 
+    if( p_sys->p_frame ) block_ChainRelease( p_sys->p_frame );
+    if( p_sys->p_sps ) block_Release( p_sys->p_sps );
+    if( p_sys->p_pps ) block_Release( p_sys->p_pps );
     block_BytestreamRelease( &p_sys->bytestream );
     free( p_sys );
 }
@@ -223,7 +267,7 @@ static block_t *Packetize( decoder_t *p_dec, block_t **pp_block )
         {
             case STATE_NOSYNC:
                 if( block_FindStartcodeFromOffset( &p_sys->bytestream,
-                        &p_sys->i_offset, p_sys->startcode, 4 ) == VLC_SUCCESS )
+                      &p_sys->i_offset, p_sys->startcode+1, 3 ) == VLC_SUCCESS)
                 {
                     p_sys->i_state = STATE_NEXT_SYNC;
                 }
@@ -246,7 +290,7 @@ static block_t *Packetize( decoder_t *p_dec, block_t **pp_block )
             case STATE_NEXT_SYNC:
                 /* Find the next startcode */
                 if( block_FindStartcodeFromOffset( &p_sys->bytestream,
-                        &p_sys->i_offset, p_sys->startcode, 4 ) != VLC_SUCCESS )
+                      &p_sys->i_offset, p_sys->startcode+1, 3 ) != VLC_SUCCESS)
                 {
                     /* Need more data */
                     return NULL;
@@ -260,6 +304,7 @@ static block_t *Packetize( decoder_t *p_dec, block_t **pp_block )
                 block_GetBytes( &p_sys->bytestream, p_pic->p_buffer,
                                 p_pic->i_buffer );
 
+                if( !p_pic->p_buffer[p_pic->i_buffer-1] ) p_pic->i_buffer--;
                 p_sys->i_offset = 0;
 
                 /* Parse the NAL */
@@ -268,6 +313,10 @@ static block_t *Packetize( decoder_t *p_dec, block_t **pp_block )
                     p_sys->i_state = STATE_NOSYNC;
                     break;
                 }
+#if 0
+                msg_Dbg( p_dec, "pts="I64Fd" dts="I64Fd,
+                         p_pic->i_pts, p_pic->i_dts );
+#endif
 
                 /* So p_block doesn't get re-added several times */
                 *pp_block = block_BytestreamPop( &p_sys->bytestream );
@@ -294,6 +343,22 @@ static block_t *PacketizeAVC1( decoder_t *p_dec, block_t **pp_block )
     p_block = *pp_block;
     *pp_block = NULL;
 
+#if 0
+    if( //(p_block->i_flags & BLOCK_FLAG_TYPE_I) &&
+        p_sys->p_sps && p_sys->p_pps )
+    {
+        block_t *p_pic;
+        block_t *p_sps = block_Duplicate( p_sys->p_sps );
+        block_t *p_pps = block_Duplicate( p_sys->p_pps );
+        p_sps->i_dts = p_pps->i_dts = p_block->i_dts;
+        p_sps->i_pts = p_pps->i_pts = p_block->i_pts;
+        p_pic = ParseNALBlock( p_dec, p_sps );
+        if( p_pic ) block_ChainAppend( &p_ret, p_pic );
+        p_pic = ParseNALBlock( p_dec, p_pps );
+        if( p_pic ) block_ChainAppend( &p_ret, p_pic );
+    }
+#endif
+
     for( p = p_block->p_buffer; p < &p_block->p_buffer[p_block->i_buffer]; )
     {
         block_t *p_pic;
@@ -307,10 +372,11 @@ static block_t *PacketizeAVC1( decoder_t *p_dec, block_t **pp_block )
 
         if( i_size > 0 )
         {
-            block_t *p_part = nal_get_encoded( p_dec, p, i_size );
+            block_t *p_part = nal_get_annexeb( p_dec, p, i_size );
 
             p_part->i_dts = p_block->i_dts;
             p_part->i_pts = p_block->i_pts;
+
             /* Parse the NAL */
             if( ( p_pic = ParseNALBlock( p_dec, p_part ) ) )
             {
@@ -319,76 +385,30 @@ static block_t *PacketizeAVC1( decoder_t *p_dec, block_t **pp_block )
         }
         p += i_size;
     }
+    block_Release( p_block );
 
     return p_ret;
 }
 
-static block_t *nal_get_encoded( decoder_t *p_dec, uint8_t *p, int i_size )
+static block_t *nal_get_annexeb( decoder_t *p_dec, uint8_t *p, int i_size )
 {
     block_t *p_nal;
-    int     i_nal_size = 5;
-    uint8_t *src = &p[1];
-    uint8_t *end = &p[i_size];
-    uint8_t *dst;
-    int     i_count = 0;
 
-    /* 1: compute real size */
-    while( src < end )
-    {
-        if( i_count == 2 && *src <= 0x03 )
-        {
-            i_nal_size++;
-            i_count = 0;
-        }
-        if( *src == 0 )
-        {
-            i_count++;
-        }
-        else
-        {
-            i_count = 0;
-        }
-        i_nal_size++;
-        src++;
-    }
-
-    /* 2: encode it */
-    p_nal = block_New( p_dec, i_nal_size );
-    i_count = 0;
-    src = p;
-    dst = p_nal->p_buffer;
+    p_nal = block_New( p_dec, 3 + i_size );
 
-    /* add start code */
-    *dst++ = 0x00;
-    *dst++ = 0x00;
-    *dst++ = 0x00;
-    *dst++ = 0x01;
+    /* Add start code */
+    p_nal->p_buffer[0] = 0x00;
+    p_nal->p_buffer[1] = 0x00;
+    p_nal->p_buffer[2] = 0x01;
 
-    /* nal type */
-    *dst++ = *src++;
-
-    while( src < end )
-    {
-        if( i_count == 2 && *src <= 0x03 )
-        {
-            *dst++ = 0x03;
-            i_count = 0;
-        }
-        if( *src == 0 )
-        {
-            i_count++;
-        }
-        else
-        {
-            i_count = 0;
-        }
-        *dst++ = *src++;
-    }
+    /* Copy nalu */
+    memcpy( &p_nal->p_buffer[3], p, i_size );
 
     return p_nal;
 }
 
-static void nal_get_decoded( uint8_t **pp_ret, int *pi_ret, uint8_t *src, int i_src )
+static void nal_get_decoded( uint8_t **pp_ret, int *pi_ret,
+                             uint8_t *src, int i_src )
 {
     uint8_t *end = &src[i_src];
     uint8_t *dst = malloc( i_src );
@@ -397,7 +417,8 @@ static void nal_get_decoded( uint8_t **pp_ret, int *pi_ret, uint8_t *src, int i_
 
     while( src < end )
     {
-        if( src < end - 3 && src[0] == 0x00 && src[1] == 0x00  && src[2] == 0x03 )
+        if( src < end - 3 && src[0] == 0x00 && src[1] == 0x00 &&
+            src[2] == 0x03 )
         {
             *dst++ = 0x00;
             *dst++ = 0x00;
@@ -421,6 +442,7 @@ static inline int bs_read_ue( bs_t *s )
     }
     return( ( 1 << i) - 1 + bs_read( s, i ) );
 }
+
 static inline int bs_read_se( bs_t *s )
 {
     int val = bs_read_ue( s );
@@ -434,69 +456,110 @@ static block_t *ParseNALBlock( decoder_t *p_dec, block_t *p_frag )
     decoder_sys_t *p_sys = p_dec->p_sys;
     block_t *p_pic = NULL;
 
-    const int i_ref_idc = (p_frag->p_buffer[4] >> 5)&0x03;
-    const int i_nal_type= p_frag->p_buffer[4]&0x1f;
+    const int i_nal_ref_idc = (p_frag->p_buffer[3] >> 5)&0x03;
+    const int i_nal_type = p_frag->p_buffer[3]&0x1f;
 
-    if( p_sys->b_slice &&
-        ( i_nal_type == NAL_SLICE || i_nal_type == NAL_SLICE_IDR ||
-          i_nal_type == NAL_SLICE_DPC || i_nal_type == NAL_SPS || i_nal_type == NAL_PPS ) )
+#define OUTPUT \
+    do {                                                \
+        p_pic = block_ChainGather( p_sys->p_frame );    \
+        p_pic->i_length = 0;    /* FIXME */             \
+                                                        \
+        p_sys->p_frame = NULL;                          \
+        p_sys->b_slice = VLC_FALSE;                     \
+    } while(0)
+
+
+    if( p_sys->b_slice && !p_sys->b_sps )
     {
-        if( p_sys->b_sps )
-        {
-            p_pic = block_ChainGather( p_sys->p_frame );
-            p_pic->i_dts = p_sys->i_dts;
-            p_pic->i_pts = p_sys->i_pts;
-            p_pic->i_length = 0;    /* FIXME */
-            p_pic->i_flags = p_sys->i_flags;
-        }
-        else
-        {
-            block_ChainRelease( p_sys->p_frame );
-            msg_Warn( p_dec, "waiting SPS" );
-        }
+        block_ChainRelease( p_sys->p_frame );
+        msg_Warn( p_dec, "waiting for SPS" );
 
-        /* reset context */
+        /* Reset context */
         p_sys->p_frame = NULL;
         p_sys->b_slice = VLC_FALSE;
-        //p_sys->i_dts += 40000;
     }
 
-    if( i_nal_type >= NAL_SLICE && i_nal_type <= NAL_SLICE_IDR )
+    if( !p_sys->b_sps &&
+        i_nal_type >= NAL_SLICE && i_nal_type <= NAL_SLICE_IDR )
+    {
+        p_sys->b_slice = VLC_TRUE;
+        /* Fragment will be discarded later on */
+    }
+    else if( i_nal_type >= NAL_SLICE && i_nal_type <= NAL_SLICE_IDR )
     {
         uint8_t *dec;
-        int     i_dec;
+        int i_dec, i_first_mb, i_slice_type, i_frame_num, i_pic_flags = 0;
+        vlc_bool_t b_pic = VLC_FALSE;
         bs_t s;
 
-        p_sys->b_slice = VLC_TRUE;
-        p_sys->i_dts   = p_frag->i_dts;
-        p_sys->i_pts   = p_frag->i_pts;
-
         /* do not convert the whole frame */
-        nal_get_decoded( &dec, &i_dec, &p_frag->p_buffer[5], __MIN( p_frag->i_buffer - 5, 60 ) );
+        nal_get_decoded( &dec, &i_dec, &p_frag->p_buffer[4],
+                         __MIN( p_frag->i_buffer - 4, 60 ) );
         bs_init( &s, dec, i_dec );
 
-        /* i_first_mb */
-        bs_read_ue( &s );
-        /* picture type */
-        switch( bs_read_ue( &s ) )
+        /* first_mb_in_slice */
+        i_first_mb = bs_read_ue( &s );
+
+        /* slice_type */
+        switch( (i_slice_type = bs_read_ue( &s )) )
         {
             case 0: case 5:
-                p_sys->i_flags = BLOCK_FLAG_TYPE_P;
+                i_pic_flags = BLOCK_FLAG_TYPE_P;
                 break;
             case 1: case 6:
-                p_sys->i_flags =BLOCK_FLAG_TYPE_B;
+                i_pic_flags = BLOCK_FLAG_TYPE_B;
                 break;
             case 2: case 7:
-                p_sys->i_flags = BLOCK_FLAG_TYPE_I;
+                i_pic_flags = BLOCK_FLAG_TYPE_I;
                 break;
             case 3: case 8: /* SP */
-                p_sys->i_flags = BLOCK_FLAG_TYPE_P;
+                i_pic_flags = BLOCK_FLAG_TYPE_P;
                 break;
             case 4: case 9:
-                p_sys->i_flags = BLOCK_FLAG_TYPE_I;
+                i_pic_flags = BLOCK_FLAG_TYPE_I;
                 break;
         }
 
+        /* pic_parameter_set_id */
+        bs_read_ue( &s );
+        /* frame_num */
+        i_frame_num = bs_read( &s, p_sys->i_log2_max_frame_num + 4 );
+
+        /* Detection of the first VCL NAL unit of a primary coded picture
+         * (cf. 7.4.1.2.4) */
+        if( i_frame_num != p_sys->i_frame_num ||
+            ( (i_nal_ref_idc != p_sys->i_nal_ref_idc) &&
+              (!i_nal_ref_idc || !p_sys->i_nal_ref_idc) ) )
+        {
+            b_pic = VLC_TRUE;
+        }
+        p_sys->i_frame_num = i_frame_num;
+        p_sys->i_nal_ref_idc = i_nal_ref_idc;
+
+        if( !p_sys->b_frame_mbs_only )
+        {
+            /* field_pic_flag */
+            if( bs_read( &s, 1 ) )
+            {
+                /* bottom_field_flag */
+                bs_read( &s, 1 );
+            }
+        }
+
+        if( i_nal_type == NAL_SLICE_IDR )
+        {
+            /* id_pic_id */
+            int i_idr_pic_id = bs_read_ue( &s );
+            if( p_sys->i_nal_type != i_nal_type ) b_pic = VLC_TRUE;
+            if( p_sys->i_idr_pic_id != i_idr_pic_id ) b_pic = VLC_TRUE;
+            p_sys->i_idr_pic_id = i_idr_pic_id;
+        }
+        p_sys->i_nal_type = i_nal_type;
+
+        if( b_pic && p_sys->b_slice ) OUTPUT;
+
+        p_sys->b_slice = VLC_TRUE;
+
         free( dec );
     }
     else if( i_nal_type == NAL_SPS )
@@ -506,9 +569,12 @@ static block_t *ParseNALBlock( decoder_t *p_dec, block_t *p_frag )
         bs_t s;
         int i_tmp;
 
+        if( !p_sys->b_sps ) msg_Dbg( p_dec, "found NAL_SPS" );
+
         p_sys->b_sps = VLC_TRUE;
 
-        nal_get_decoded( &dec, &i_dec, &p_frag->p_buffer[5], p_frag->i_buffer - 5 );
+        nal_get_decoded( &dec, &i_dec, &p_frag->p_buffer[4],
+                         p_frag->i_buffer - 4 );
 
         bs_init( &s, dec, i_dec );
         /* Skip profile(8), constraint_set012, reserver(5), level(8) */
@@ -516,7 +582,7 @@ static block_t *ParseNALBlock( decoder_t *p_dec, block_t *p_frag )
         /* sps id */
         bs_read_ue( &s );
         /* Skip i_log2_max_frame_num */
-        bs_read_ue( &s );
+        p_sys->i_log2_max_frame_num = bs_read_ue( &s );
         /* Read poc_type */
         i_tmp = bs_read_ue( &s );
         if( i_tmp == 0 )
@@ -552,26 +618,26 @@ static block_t *ParseNALBlock( decoder_t *p_dec, block_t *p_frag )
         p_dec->fmt_out.video.i_height = 16 * ( bs_read_ue( &s ) + 1 );
 
         /* b_frame_mbs_only */
-        i_tmp = bs_read( &s, 1 );
-        if( i_tmp == 0 )
+        p_sys->b_frame_mbs_only = bs_read( &s, 1 );
+        if( p_sys->b_frame_mbs_only == 0 )
         {
             bs_skip( &s, 1 );
         }
         /* b_direct8x8_inference */
         bs_skip( &s, 1 );
 
-        /* crop */
+        /* crop */
         i_tmp = bs_read( &s, 1 );
         if( i_tmp )
         {
             /* left */
-            p_dec->fmt_out.video.i_width -= 2 * bs_read_ue( &s );
+            bs_read_ue( &s );
             /* right */
-            p_dec->fmt_out.video.i_width -= 2 * bs_read_ue( &s );
+            bs_read_ue( &s );
             /* top */
-            p_dec->fmt_out.video.i_height -= 2 * bs_read_ue( &s );
+            bs_read_ue( &s );
             /* bottom */
-            p_dec->fmt_out.video.i_height -= 2 * bs_read_ue( &s );
+            bs_read_ue( &s );
         }
 
         /* vui */
@@ -602,27 +668,43 @@ static block_t *ParseNALBlock( decoder_t *p_dec, block_t *p_frag )
                     w = bs_read( &s, 16 );
                     h = bs_read( &s, 16 );
                 }
-                p_dec->fmt_out.video.i_aspect =
-                    VOUT_ASPECT_FACTOR *
-                    w / h *
-                    p_dec->fmt_out.video.i_width / p_dec->fmt_out.video.i_height;
+                if( h != 0 )
+                    p_dec->fmt_out.video.i_aspect = VOUT_ASPECT_FACTOR * w /
+                        h * p_dec->fmt_out.video.i_width /
+                        p_dec->fmt_out.video.i_height;
+                else
+                    p_dec->fmt_out.video.i_aspect = VOUT_ASPECT_FACTOR;
             }
         }
 
         free( dec );
+
+
+        if( p_sys->b_slice ) OUTPUT;
     }
     else if( i_nal_type == NAL_PPS )
     {
         bs_t s;
-        bs_init( &s, &p_frag->p_buffer[5], p_frag->i_buffer - 5 );
+        bs_init( &s, &p_frag->p_buffer[4], p_frag->i_buffer - 4 );
+
+        if( !p_sys->b_pps ) msg_Dbg( p_dec, "found NAL_PPS" );
+        p_sys->b_pps = VLC_TRUE;
 
         /* TODO */
+
+        if( p_sys->b_slice ) OUTPUT;
+    }
+    else if( i_nal_type == NAL_AU_DELIMITER ||
+             i_nal_type == NAL_SEI ||
+             ( i_nal_type >= 13 && i_nal_type <= 18 ) )
+    {
+        if( p_sys->b_slice ) OUTPUT;
     }
 
+#undef OUTPUT
 
     /* Append the block */
     block_ChainAppend( &p_sys->p_frame, p_frag );
 
     return p_pic;
 }
-