Change internal audio representation range

...from float [-32768...32767] to float [-1.0...1.0]

Using the range [-1.0..1.0] requires fewer translations of the range for our 
various encoders and decoders.  This also gets rid of a hacky 
translation from float to int to float in decavcodec audio decoding.


git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@3908 b64f7644-9d1e-0410-96f1-a4d463321fa5

13 files changed, 70 insertions, 67 deletions

diff --git a/libhb/deca52.c b/libhb/deca52.c
index 0958a9139..3148bc672 100644
--- a/libhb/deca52.c
+++ b/libhb/deca52.c
@@ -106,7 +106,7 @@ static int deca52Init( hb_work_object_t * w, hb_job_t * job )
     /* will only be actually used if we're not doing AC3 passthru */
     pv->out_discrete_channels = HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(audio->config.out.mixdown);
 
-    pv->level     = 32768.0;
+    pv->level     = 1.0;
     pv->dynamic_range_compression = audio->config.out.dynamic_range_compression;
 
     return 0;
diff --git a/libhb/decavcodec.c b/libhb/decavcodec.c
index 5fe5f8e81..b35d57119 100644
--- a/libhb/decavcodec.c
+++ b/libhb/decavcodec.c
@@ -111,7 +111,6 @@ struct hb_work_private_s
     void*           buffer;
     struct SwsContext *sws_context; // if we have to rescale or convert color space
     hb_downmix_t    *downmix;
-    hb_sample_t     *downmix_buffer;
     int cadence[12];
     hb_chan_map_t   *out_map;
 };
@@ -205,6 +204,7 @@ static int decavcodecInit( hb_work_object_t * w, hb_job_t * job )
     pv->parser = av_parser_init( codec_id );
 
     pv->context = avcodec_alloc_context();
+    hb_ff_set_sample_fmt( pv->context, codec );
     hb_avcodec_open( pv->context, codec );
 
     if ( w->audio != NULL )
@@ -305,11 +305,6 @@ static void closePrivData( hb_work_private_t ** ppv )
         {
             hb_downmix_close( &(pv->downmix) );
         }
-        if ( pv->downmix_buffer )
-        {
-            free( pv->downmix_buffer );
-            pv->downmix_buffer = NULL;
-        }
         free( pv );
     }
     *ppv = NULL;
@@ -528,6 +523,7 @@ static int decavcodecBSInfo( hb_work_object_t *w, const hb_buffer_t *buf,
 
     AVCodecParserContext *parser = av_parser_init( codec->id );
     AVCodecContext *context = avcodec_alloc_context();
+    hb_ff_set_sample_fmt( context, codec );
     hb_avcodec_open( context, codec );
     uint8_t *buffer = av_malloc( AVCODEC_MAX_AUDIO_FRAME_SIZE );
     int out_size = AVCODEC_MAX_AUDIO_FRAME_SIZE;
@@ -1160,6 +1156,7 @@ static int decavcodecvWork( hb_work_object_t * w, hb_buffer_t ** buf_in,
         // There's a mis-feature in ffmpeg that causes the context to be 
         // incorrectly initialized the 1st time avcodec_open is called.
         // If you close it and open a 2nd time, it finishes the job.
+        hb_ff_set_sample_fmt( pv->context, codec );
         hb_avcodec_open( pv->context, codec );
         hb_avcodec_close( pv->context );
         hb_avcodec_open( pv->context, codec );
@@ -1280,6 +1277,7 @@ static void init_ffmpeg_context( hb_work_object_t *w )
     if ( ! pv->context->codec )
     {
         AVCodec *codec = avcodec_find_decoder( pv->context->codec_id );
+        hb_ff_set_sample_fmt( pv->context, codec );
         hb_avcodec_open( pv->context, codec );
     }
     // set up our best guess at the frame duration.
@@ -1491,7 +1489,7 @@ static int decavcodecviInfo( hb_work_object_t *w, hb_work_info_t *info )
 static hb_buffer_t * downmixAudio( 
     hb_audio_t *audio, 
     hb_work_private_t *pv, 
-    int16_t *buffer, 
+    hb_sample_t *buffer, 
     int channels,
     int nsamples )
 {
@@ -1499,20 +1497,12 @@ static hb_buffer_t * downmixAudio(
 
     if ( pv->downmix )
     {
-        pv->downmix_buffer = realloc(pv->downmix_buffer, nsamples * sizeof(hb_sample_t));
-        
-        int i;
-        for( i = 0; i < nsamples; ++i )
-        {
-            pv->downmix_buffer[i] = buffer[i];
-        }
-
         int n_ch_samples = nsamples / channels;
         int out_channels = HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(audio->config.out.mixdown);
 
         buf = hb_buffer_init( n_ch_samples * out_channels * sizeof(float) );
         hb_sample_t *samples = (hb_sample_t *)buf->data;
-        hb_downmix(pv->downmix, samples, pv->downmix_buffer, n_ch_samples);
+        hb_downmix(pv->downmix, samples, buffer, n_ch_samples);
     }
     else
     {
@@ -1540,7 +1530,7 @@ static void decodeAudio( hb_audio_t * audio, hb_work_private_t *pv, uint8_t *dat
 
     while ( pos < size )
     {
-        int16_t *buffer = pv->buffer;
+        float *buffer = pv->buffer;
         if ( buffer == NULL )
         {
             pv->buffer = av_malloc( AVCODEC_MAX_AUDIO_FRAME_SIZE );
@@ -1556,7 +1546,7 @@ static void decodeAudio( hb_audio_t * audio, hb_work_private_t *pv, uint8_t *dat
 
         int out_size = AVCODEC_MAX_AUDIO_FRAME_SIZE;
         int nsamples;
-        int len = avcodec_decode_audio3( context, buffer, &out_size, &avp );
+        int len = avcodec_decode_audio3( context, (int16_t*)buffer, &out_size, &avp );
         if ( len < 0 )
         {
             return;
@@ -1574,7 +1564,7 @@ static void decodeAudio( hb_audio_t * audio, hb_work_private_t *pv, uint8_t *dat
         {
             // We require signed 16-bit ints for the output format. If
             // we got something different convert it.
-            if ( context->sample_fmt != SAMPLE_FMT_S16 )
+            if ( context->sample_fmt != AV_SAMPLE_FMT_FLT )
             {
                 // Note: av_audio_convert seems to be a work-in-progress but
                 //       looks like it will eventually handle general audio
@@ -1583,27 +1573,31 @@ static void decodeAudio( hb_audio_t * audio, hb_work_private_t *pv, uint8_t *dat
                 //       anything more complicated than a one-for-one format
                 //       conversion we'd probably want to cache the converter
                 //       context in the pv.
-                int isamp = av_get_bits_per_sample_fmt( context->sample_fmt ) / 8;
-                AVAudioConvert *ctx = av_audio_convert_alloc( SAMPLE_FMT_S16, 1,
-                                                              context->sample_fmt, 1,
-                                                              NULL, 0 );
-                // get output buffer size (in 2-byte samples) then malloc a buffer
+                int isamp;
+                AVAudioConvert *ctx;
+
+                isamp = av_get_bits_per_sample_fmt( context->sample_fmt ) / 8;
+                ctx = av_audio_convert_alloc( AV_SAMPLE_FMT_FLT, 1,
+                                              context->sample_fmt, 1,
+                                              NULL, 0 );
+
+                // get output buffer size then malloc a buffer
                 nsamples = out_size / isamp;
-                buffer = av_malloc( nsamples * 2 );
+                buffer = av_malloc( nsamples * sizeof(hb_sample_t) );
 
-                // we're doing straight sample format conversion which behaves as if
-                // there were only one channel.
+                // we're doing straight sample format conversion which 
+                // behaves as if there were only one channel.
                 const void * const ibuf[6] = { pv->buffer };
                 void * const obuf[6] = { buffer };
                 const int istride[6] = { isamp };
-                const int ostride[6] = { 2 };
+                const int ostride[6] = { sizeof(hb_sample_t) };
 
                 av_audio_convert( ctx, obuf, ostride, ibuf, istride, nsamples );
                 av_audio_convert_free( ctx );
             }
             else
             {
-                nsamples = out_size / 2;
+                nsamples = out_size / sizeof(hb_sample_t);
             }
 
             if ( pts == AV_NOPTS_VALUE )
diff --git a/libhb/decdca.c b/libhb/decdca.c
index 9b85905fd..62f460915 100644
--- a/libhb/decdca.c
+++ b/libhb/decdca.c
@@ -84,7 +84,7 @@ static int decdcaInit( hb_work_object_t * w, hb_job_t * job )
     /* will only be actually used if we're not doing AC3 passthru */
     pv->out_discrete_channels = HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(audio->config.out.mixdown);
 
-    pv->level     = 32768.0;
+    pv->level     = 1.0;
 
     return 0;
 }
@@ -268,7 +268,7 @@ static hb_buffer_t * Decode( hb_work_object_t * w )
         {
             for ( k = 0; k < pv->out_discrete_channels; k++ )
             {
-                samples_out[(pv->out_discrete_channels*j)+k]   = samples_in[(256*k)+j] * 32767;
+                samples_out[(pv->out_discrete_channels*j)+k]   = samples_in[(256*k)+j];
             }
         }
 
diff --git a/libhb/declpcm.c b/libhb/declpcm.c
index f47fe87b9..fc120a333 100644
--- a/libhb/declpcm.c
+++ b/libhb/declpcm.c
@@ -198,7 +198,7 @@ static hb_buffer_t *Decode( hb_work_object_t *w )
         case 16: // 2 byte, big endian, signed (the right shift sign extends)
             while ( --count >= 0 )
             {
-                *odat++ = ( (int)( frm[0] << 24 ) >> 16 ) | frm[1];
+                *odat++ = (float)(( (int)( frm[0] << 24 ) >> 16 ) | frm[1]) / 32768.0;
                 frm += 2;
             }
             break;
@@ -209,9 +209,9 @@ static hb_buffer_t *Decode( hb_work_object_t *w )
             while ( --count >= 0 )
             {
                 *odat++ = (float)( ( (int)( frm[0] << 24 ) >> 12 ) |
-                                   ( frm[1] << 4 ) | ( frm[2] >> 4 ) ) / 16.;
+                                   ( frm[1] << 4 ) | ( frm[2] >> 4 ) ) / (16. * 32768.0);
                 *odat++ = (float)( ( (int)( frm[2] << 28 ) >> 16 ) |
-                                   ( frm[3] << 8 ) | frm[4] ) / 16.;
+                                   ( frm[3] << 8 ) | frm[4] ) / (16. * 32768.0);
                 frm += 5;
             }
             break;
@@ -227,13 +227,13 @@ static hb_buffer_t *Decode( hb_work_object_t *w )
             while ( --count >= 0 )
             {
                 *odat++ = (float)( ( (int)( frm[0] << 24 ) >> 8 ) |
-                            ( frm[1] << 8 ) | frm[8] ) / 256.;
+                            ( frm[1] << 8 ) | frm[8] ) / (256. * 32768.0);
                 *odat++ = (float)( ( (int)( frm[2] << 24 ) >> 8 ) |
-                            ( frm[3] << 8 ) | frm[9] ) / 256.;
+                            ( frm[3] << 8 ) | frm[9] ) / (256. * 32768.0);
                 *odat++ = (float)( ( (int)( frm[4] << 24 ) >> 8 ) |
-                            ( frm[5] << 8 ) | frm[10] ) / 256.;
+                            ( frm[5] << 8 ) | frm[10] ) / (256. * 32768.0);
                 *odat++ = (float)( ( (int)( frm[6] << 24 ) >> 8 ) |
-                            ( frm[7] << 8 ) | frm[11] ) / 256.;
+                            ( frm[7] << 8 ) | frm[11] ) / (256. * 32768.0);
                 frm += 12;
             }
             break;
diff --git a/libhb/encac3.c b/libhb/encac3.c
index 179086607..693e8b02e 100644
--- a/libhb/encac3.c
+++ b/libhb/encac3.c
@@ -18,7 +18,6 @@ struct hb_work_private_s
     unsigned long    output_bytes;
     hb_list_t      * list;
     uint8_t        * buf;
-    float          * samples;
 };
 
 int  encac3Init( hb_work_object_t *, hb_job_t * );
@@ -53,7 +52,6 @@ int encac3Init( hb_work_object_t * w, hb_job_t * job )
     pv->output_bytes = AC3_MAX_CODED_FRAME_SIZE;
 
     pv->buf = malloc( pv->input_samples * sizeof( float ) );
-    pv->samples = malloc( pv->input_samples * sizeof( float ) );
 
     codec = avcodec_find_encoder( CODEC_ID_AC3 );
     if( !codec )
@@ -127,12 +125,6 @@ void encac3Close( hb_work_object_t * w )
             pv->buf = NULL;
         }
 
-        if ( pv->samples )
-        {
-            free( pv->samples );
-            pv->samples = NULL;
-        }
-
         if ( pv->list )
             hb_list_empty( &pv->list );
 
@@ -147,7 +139,6 @@ static hb_buffer_t * Encode( hb_work_object_t * w )
     uint64_t pts, pos;
     hb_audio_t * audio = w->audio;
     hb_buffer_t * buf;
-    int ii;
 
     if( hb_list_bytes( pv->list ) < pv->input_samples * sizeof( float ) )
     {
@@ -188,15 +179,9 @@ static hb_buffer_t * Encode( hb_work_object_t * w )
                         (float*)pv->buf, AC3_SAMPLES_PER_FRAME);
     }
     
-    for (ii = 0; ii < pv->input_samples; ii++)
-    {
-        // ffmpeg float samples are -1.0 to 1.0
-        pv->samples[ii] = ((float*)pv->buf)[ii] / 32768.0;
-    }
-
     buf = hb_buffer_init( pv->output_bytes );
     buf->size = avcodec_encode_audio( pv->context, buf->data, buf->alloc,
-                                      (short*)pv->samples );
+                                      (short*)pv->buf );
 
     buf->start = pts + 90000 * pos / pv->out_discrete_channels / sizeof( float ) / audio->config.out.samplerate;
     buf->stop  = buf->start + 90000 * AC3_SAMPLES_PER_FRAME / audio->config.out.samplerate;
diff --git a/libhb/encfaac.c b/libhb/encfaac.c
index c4a57100f..b8ff8347c 100644
--- a/libhb/encfaac.c
+++ b/libhb/encfaac.c
@@ -207,6 +207,12 @@ static hb_buffer_t * Encode( hb_work_object_t * w )
     uint64_t pts, pos;
     hb_list_getbytes( pv->list, pv->buf, pv->input_samples * sizeof( float ),
                       &pts, &pos );
+
+    int i;
+    float *fltBuf = (float*)pv->buf;
+    for ( i = 0; i < pv->input_samples; i++ )
+        fltBuf[i] *= 32768.0;
+        
     int size = faacEncEncode( pv->faac, (int32_t *)pv->buf, pv->input_samples,
                               pv->obuf, pv->output_bytes );
 
diff --git a/libhb/enclame.c b/libhb/enclame.c
index 955aca626..ba05e1fc0 100644
--- a/libhb/enclame.c
+++ b/libhb/enclame.c
@@ -50,6 +50,7 @@ int enclameInit( hb_work_object_t * w, hb_job_t * job )
 
     pv->lame = lame_init();
     // use ABR
+    lame_set_scale( pv->lame, 32768.0 );
     lame_set_VBR( pv->lame, vbr_abr );
     lame_set_VBR_mean_bitrate_kbps( pv->lame, audio->config.out.bitrate );
     lame_set_in_samplerate( pv->lame, audio->config.out.samplerate );
@@ -120,7 +121,6 @@ static hb_buffer_t * Encode( hb_work_object_t * w )
     hb_list_getbytes( pv->list, pv->buf, pv->input_samples * sizeof( float ),
                       &pts, &pos);
 
-    memset(samples, 0, 2*1152*sizeof(float));
     for( i = 0; i < 1152; i++ )
     {
         for( j = 0; j < pv->out_discrete_channels; j++ )
diff --git a/libhb/encvorbis.c b/libhb/encvorbis.c
index 609e45ffd..1c7e323ba 100644
--- a/libhb/encvorbis.c
+++ b/libhb/encvorbis.c
@@ -242,9 +242,10 @@ static hb_buffer_t * Encode( hb_work_object_t * w )
     {
         for( j = 0; j < pv->out_discrete_channels; j++)
         {
-            buffer[j][i] = ((float *) pv->buf)[(pv->out_discrete_channels * i + pv->channel_map[j])] / 32768.f;
+            buffer[j][i] = ((float *) pv->buf)[(pv->out_discrete_channels * i + pv->channel_map[j])];
         }
     }
+
     vorbis_analysis_wrote( &pv->vd, OGGVORBIS_FRAME_SIZE );
 
     /* Try to extract again */
diff --git a/libhb/hb.c b/libhb/hb.c
index 92e743cc6..4c16156c5 100644
--- a/libhb/hb.c
+++ b/libhb/hb.c
@@ -241,6 +241,29 @@ int hb_ff_layout_xlat(int64_t ff_channel_layout, int channels)
     return hb_layout;
 }
 
+// Set sample format to AV_SAMPLE_FMT_FLT if supported.
+// If it is not supported, we will have to translate using
+// av_audio_convert.
+void hb_ff_set_sample_fmt(AVCodecContext *context, AVCodec *codec)
+{
+    if ( codec && codec->sample_fmts )
+    {
+        if ( codec->type != AVMEDIA_TYPE_AUDIO )
+            return; // Not audio
+
+        const enum AVSampleFormat *fmt;
+
+        for ( fmt = codec->sample_fmts; *fmt != -1; fmt++ )
+        {
+            if ( *fmt == AV_SAMPLE_FMT_FLT )
+            {
+                context->sample_fmt = AV_SAMPLE_FMT_FLT;
+                break;
+            }
+        }
+    }
+}
+
 /**
  * Registers work objects, by adding the work object to a liked list.
  * @param w Handle to hb_work_object_t to register.
diff --git a/libhb/hbffmpeg.h b/libhb/hbffmpeg.h
index 205bd0f64..3c7d1ead6 100644
--- a/libhb/hbffmpeg.h
+++ b/libhb/hbffmpeg.h
@@ -15,3 +15,4 @@ struct SwsContext*
 hb_sws_get_context(int srcW, int srcH, enum PixelFormat srcFormat,
                    int dstW, int dstH, enum PixelFormat dstFormat,
                    int flags);
+void hb_ff_set_sample_fmt(AVCodecContext *context, AVCodec *codec);
diff --git a/libhb/platform/macosx/encca_aac.c b/libhb/platform/macosx/encca_aac.c
index 762559cae..21b458960 100644
--- a/libhb/platform/macosx/encca_aac.c
+++ b/libhb/platform/macosx/encca_aac.c
@@ -280,14 +280,6 @@ static OSStatus inInputDataProc( AudioConverterRef converter, UInt32 *npackets,
 
     *npackets = buffers->mBuffers[0].mDataByteSize / pv->isamplesiz;
 
-    /* transform data from [-32768,32767] to [-1.0,1.0] */
-    float *fdata = buffers->mBuffers[0].mData;
-    int i;
-
-    for( i = 0; i < *npackets * pv->nchannels; i++ ) {
-        fdata[i] = fdata[i] / 32768.f;
-    }
-
     pv->ibytes -= buffers->mBuffers[0].mDataByteSize;
 
     return noErr;
diff --git a/libhb/stream.c b/libhb/stream.c
index ccc10f746..f48edbe4c 100644
--- a/libhb/stream.c
+++ b/libhb/stream.c
@@ -2837,6 +2837,7 @@ static void ffmpeg_add_codec( hb_stream_t *stream, int stream_index )
     context->error_recognition = 1;
     context->error_concealment = FF_EC_GUESS_MVS|FF_EC_DEBLOCK;
     AVCodec *codec = avcodec_find_decoder( context->codec_id );
+    hb_ff_set_sample_fmt( context, codec );
     hb_avcodec_open( context, codec );
 }
 
diff --git a/libhb/sync.c b/libhb/sync.c
index 19ee605f0..bceeb10b5 100644
--- a/libhb/sync.c
+++ b/libhb/sync.c
@@ -1172,9 +1172,9 @@ static hb_buffer_t * OutputAudioFrame( hb_audio_t *audio, hb_buffer_t *buf,
                 sample = (double)*(((float*)buf->data)+ii);
                 sample *= sync->gain_factor;
                 if (sample > 0)
-                    sample = MIN(sample, 32767.0);
+                    sample = MIN(sample, 1.0);
                 else
-                    sample = MAX(sample, -32768.0);
+                    sample = MAX(sample, -1.0);
                 *(((float*)buf->data)+ii) = sample;
             }
         }