davinci: save one memcpy per audio encoding pass

Write directly into the special DSP buffer and only copy the remainder to a temporarily buffer.

modules/codec/davinci/audio1_encoder.c

@@ -79,6 +79,8 @@ struct xdm_sys_t
 
     /* remainder buffer management */
     uint8_t  *p_tmp;
+    size_t   i_tmp;
+
     size_t   i_nb_bytes;
     uint32_t i_nb_samples;
     mtime_t  start_date;
@@ -558,8 +560,6 @@ static block_t *EncodeAudio1( encoder_t *p_enc, aout_buffer_t *p_buffer )
     }
     else
     {
-        // FIXME: optimize by using xdm->in.descs[0].buf for filling and xdm->p_tmp as overflow
-        // This will save several memcpy's
         uint32_t i_nb_delayed_samples = 0;
         size_t i_size = 0;
 
@@ -567,9 +567,16 @@ static block_t *EncodeAudio1( encoder_t *p_enc, aout_buffer_t *p_buffer )
         if( xdm->start_date == 0 )
             xdm->start_date = p_buffer->start_date;
 
+        if( xdm->i_tmp > 0 )
+        {
+            memcpy( (uint8_t*)xdm->in.descs[0].buf + xdm->i_nb_bytes, xdm->p_tmp, xdm->i_tmp );
+            xdm->i_nb_bytes += xdm->i_tmp;
+            xdm->i_tmp = 0;
+        }
+
         if( xdm->i_nb_bytes + p_buffer->i_nb_bytes <= (size_t)xdm->in.descs[0].bufSize )
         {
-            memcpy( xdm->p_tmp + xdm->i_nb_bytes, p_buffer->p_buffer, p_buffer->i_nb_bytes );
+            memcpy( (uint8_t*)xdm->in.descs[0].buf + xdm->i_nb_bytes, p_buffer->p_buffer, p_buffer->i_nb_bytes );
 
             xdm->i_nb_bytes += p_buffer->i_nb_bytes;
             xdm->i_nb_samples += p_buffer->i_nb_samples;
@@ -586,7 +593,7 @@ static block_t *EncodeAudio1( encoder_t *p_enc, aout_buffer_t *p_buffer )
         {
             i_size = xdm->in.descs[0].bufSize - xdm->i_nb_bytes;
             i_size = __MIN(i_size, p_buffer->i_nb_bytes);
-            memcpy( xdm->p_tmp + xdm->i_nb_bytes, p_buffer->p_buffer, i_size );
+            memcpy( (uint8_t*)xdm->in.descs[0].buf + xdm->i_nb_bytes, p_buffer->p_buffer, i_size );
 
             i_nb_delayed_samples = p_buffer->i_nb_samples - (xdm->i_frames - xdm->i_nb_samples);
             xdm->i_nb_samples += (p_buffer->i_nb_samples - i_nb_delayed_samples);
@@ -601,7 +608,6 @@ static block_t *EncodeAudio1( encoder_t *p_enc, aout_buffer_t *p_buffer )
 
         /* always fill entire encoder bufer */
         assert( (uint32_t)xdm->in.descs[0].bufSize == xdm->i_nb_bytes );
-        memcpy( (uint8_t*)xdm->in.descs[0].buf, xdm->p_tmp, xdm->i_nb_bytes );
 
         /* samples in DSP buffer */
         i_nb_samples = xdm->i_nb_samples;
@@ -614,8 +620,9 @@ static block_t *EncodeAudio1( encoder_t *p_enc, aout_buffer_t *p_buffer )
         {
             memset( xdm->p_tmp, 0, (uint32_t)xdm->in.descs[0].bufSize );
             memcpy( xdm->p_tmp, p_buffer->p_buffer + i_size, p_buffer->i_nb_bytes - i_size );
+            xdm->i_tmp = p_buffer->i_nb_bytes - i_size;
 
-            xdm->i_nb_bytes = p_buffer->i_nb_bytes - i_size;
+            xdm->i_nb_bytes = 0;
             xdm->i_nb_samples = i_nb_delayed_samples;
 
             xdm->start_date = i_pts_end;

modules/codec/davinci/audio_encoder.c

@@ -74,6 +74,8 @@ struct xdm_sys_t
 
     /* remainder buffer management */
     uint8_t  *p_tmp;
+    size_t   i_tmp;
+
     size_t   i_nb_bytes;
     uint32_t i_nb_samples;
     mtime_t  start_date;
@@ -316,8 +318,6 @@ static block_t *EncodeAudio( encoder_t *p_enc, aout_buffer_t *p_buffer )
     }
     else
     {
-        // FIXME: optimize by using xdm->in.bufs[0] for filling and xdm->p_tmp as overflow
-        // This will save several memcpy's
         uint32_t i_nb_delayed_samples = 0;
         size_t i_size = 0;
 
@@ -325,9 +325,16 @@ static block_t *EncodeAudio( encoder_t *p_enc, aout_buffer_t *p_buffer )
         if( xdm->start_date == 0 )
             xdm->start_date = p_buffer->start_date;
 
+        if( xdm->i_tmp > 0 )
+        {
+            memcpy( (uint8_t*)xdm->in.bufs[0] + xdm->i_nb_bytes, xdm->p_tmp, xdm->i_tmp );
+            xdm->i_nb_bytes += xdm->i_tmp;
+            xdm->i_tmp = 0;
+        }
+
         if( xdm->i_nb_bytes + p_buffer->i_nb_bytes <= (size_t)xdm->in.bufSizes[0] )
         {
-            memcpy( xdm->p_tmp + xdm->i_nb_bytes, p_buffer->p_buffer, p_buffer->i_nb_bytes );
+            memcpy( (uint8_t*)xdm->in.bufs[0] + xdm->i_nb_bytes, p_buffer->p_buffer, p_buffer->i_nb_bytes );
 
             xdm->i_nb_bytes += p_buffer->i_nb_bytes;
             xdm->i_nb_samples += p_buffer->i_nb_samples;
@@ -344,7 +351,7 @@ static block_t *EncodeAudio( encoder_t *p_enc, aout_buffer_t *p_buffer )
         {
             i_size = xdm->in.bufSizes[0] - xdm->i_nb_bytes;
             i_size = __MIN(i_size, p_buffer->i_nb_bytes);
-            memcpy( xdm->p_tmp + xdm->i_nb_bytes, p_buffer->p_buffer, i_size );
+            memcpy( (uint8_t*)xdm->in.bufs[0] + xdm->i_nb_bytes, p_buffer->p_buffer, i_size );
 
             i_nb_delayed_samples = p_buffer->i_nb_samples - (xdm->i_frames - xdm->i_nb_samples);
             xdm->i_nb_samples += (p_buffer->i_nb_samples - i_nb_delayed_samples);
@@ -359,7 +366,6 @@ static block_t *EncodeAudio( encoder_t *p_enc, aout_buffer_t *p_buffer )
 
         /* always fill entire encoder bufer */
         assert( (uint32_t)xdm->in.bufSizes[0] == xdm->i_nb_bytes );
-        memcpy( (uint8_t*)xdm->in.bufs[0], xdm->p_tmp, xdm->i_nb_bytes );
 
         /* samples in DSP buffer */
         i_nb_samples = xdm->i_nb_samples;
@@ -372,8 +378,9 @@ static block_t *EncodeAudio( encoder_t *p_enc, aout_buffer_t *p_buffer )
         {
             memset( xdm->p_tmp, 0, (uint32_t)xdm->in.bufSizes[0] );
             memcpy( xdm->p_tmp, p_buffer->p_buffer + i_size, p_buffer->i_nb_bytes - i_size );
+            xdm->i_tmp = p_buffer->i_nb_bytes - i_size;
 
-            xdm->i_nb_bytes = p_buffer->i_nb_bytes - i_size;
+            xdm->i_nb_bytes = 0;
             xdm->i_nb_samples = i_nb_delayed_samples;
 
             xdm->start_date = i_pts_end;