aout: always use S16N (rather than FI32) in absence of FPU

None of the real audio outputs support FI32; they already typically
converted down to S16 internally. Moreover, mixing FI32 requires
64-bits (signed) multiplication, which non-FPU processors typically
do not support natively.

src/audio_output/output.c

@@ -234,11 +234,6 @@ int aout_OutputNew( audio_output_t *p_aout,
     if( HAVE_FPU )
         owner->mixer_format.i_format = VLC_CODEC_FL32;
     else
-    /* Otherwise, audio filters will not work. Use fixed-point if the input has
-     * more than 16-bits depth. */
-    if( p_format->i_bitspersample > 16 || !AOUT_FMT_LINEAR(p_format))
-        owner->mixer_format.i_format = VLC_CODEC_FI32;
-    else
     /* Fallback to 16-bits. This avoids pointless conversion to and from
      * 32-bits samples for the sole purpose of software mixing. */
         owner->mixer_format.i_format = VLC_CODEC_S16N;