[FFmpeg-devel] [PATCH] Bugfix: decoding 8- and 24-bit FLAC files

[Michael Donaghy]

Attached is a patch to fix two problems relating to decoding of FLAC files 
with 8 and 24 (and I believe 20, though am unable to test this) bit samples; 
also attached is a 5-second example of each for verification (test by "ffplay 
file.flac" for ffplay built with and without this patch).

The original problem is the macro "DECORRELATE", lines 683-692 after patching 
of libavcodec/flac.c, for dealing with the various types of "joint stereo" 
frames:
*(samples++) = (left ) >> (16 - s->bps);\
*(samples++) = (right) >> (16 - s->bps);\
This is supposed to bitshift two s->bps bit decoded samples into 16-bit format 
for returning; however, it fails when s->bps is > 16 as in this case the 
values need to be bitshifted to the right. The equivalent code for dealing 
with non-joint stereo frames is encapsulated by the shift_to_16_bits 
function:
 *(samples++) = shift_to_16_bits(s->decoded[i][j], s->bps);
Rather than maintaining two separate code paths I felt it best to reuse this 
function, hence the two lines above become:
*(samples++) = shift_to_16_bits(left , s->bps);\
*(samples++) = shift_to_16_bits(right, s->bps);\
Having changed this a second, related bug became apparent: the aforementioned 
shift_to_16_bits (lines 571-582 after patching, of the same file) does not 
handle 8-bit samples correctly (wheras the original joint stereo code did); 
in keeping with the surrounding code I added the following code to handle 
this case.
    } else if (bps == 8) {
        return (data << 8);
The user-visible result of these changes is that 24-bit (and, without 
verification, 20-bit) FLAC files making use of joint stereo frames (i.e. 
encoded with flac -m or flac -M), and 8-bit FLAC files making use of 
non-joint stereo frames (i.e. all 8-bit FLAC files) can now be played 
correctly, where they previously could not. FLAC joint stereo frames with 
12-bit samples, if such exist, can no longer be played correctly, but since 
all stereo FLAC files include non-joint stereo frames, such files are 
unusable with or without this patch; if playback of 12-bit FLAC files is 
desired this could (post-patch) be easily accomplished by adding 
another "else if" clause to shift_to_16_bits().

My suggested commit message would be "Fix problem with decoding of 24- and 
8-bit FLAC files"

Michael Donaghy
-------------- next part --------------
A non-text attachment was scrubbed...
Name: flac_non_16bit.patch
Type: text/x-diff
Size: 785 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20070610/afb37979/attachment.patch>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 8bit.flac
Type: audio/x-flac
Size: 126548 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20070610/afb37979/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 24bit.flac
Type: audio/x-flac
Size: 959883 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20070610/afb37979/attachment-0001.bin>