[FFmpeg-devel] [PATCH] Issue 636 : ALAC encoding sometimes fails

[Vitor Sessak]

M?ns Rullg?rd wrote:
> Michael Niedermayer wrote:
>> On Fri, Dec 05, 2008 at 10:09:16AM +0530, Jai Menon wrote:
>>> Hi,
>>>
>>> Attached patch fixes issue 636 and a few other fringe cases where the
>>> alac encoder produces
>>> lossy output. I'm behind a NAT and svn doesn't work, so could someone
>>> please apply. Thanks.
>>>
>>> Regards,
>>>
>>> Jai
>>> Index: libavcodec/alacenc.c
>>> ===================================================================
>>> --- libavcodec/alacenc.c	(revision 15797)
>>> +++ libavcodec/alacenc.c	(working copy)
>>> @@ -253,7 +253,8 @@
>>>
>>>              sum >>= lpc.lpc_quant;
>>>              sum += samples[0];
>>> -            residual[i] = samples[lpc.lpc_order+1] - sum;
>>> +            residual[i] = (samples[lpc.lpc_order+1] - sum) << (32 -
> s->write_sample_size) >>
>>> +                          (32 - s->write_sample_size);
>>>              res_val = residual[i];
>> you are missing a int32_t cast in there, without it this wont work on
>> systems where int is not exactly 32bit
> 
> Alternatively substitute 8*sizeof(int) for 32.  For maximum portability,
> use CHAR_BIT rather than 8.
> 
> I assume the intent here is to sign-extend a smaller value to 32 bits.
> Are compilers clever enough to realise this and use more efficient
> methods where available?  If not, we should add a macro or inline
> function doing it properly.

In alac.c there is an inline function extend_sign32() for this. 
Efficiency issues were discussed at thread "[PATCH] Yet more ALAC 
cleanup (#define removal)" circa september 2007. Note that in alac.c 
case, the function did masking+bit extension...

-Vitor