[FFmpeg-devel] [PATCH][RFC] Lagarith Decoder.

[Loren Merritt]

On Fri, 18 Sep 2009, Vitor Sessak wrote:
> Nathan Caldwell wrote:
>
>> Ok, so the values that cause this are:
>> rac->prob[i] = 0x6eb
>> scale_factor = 16
>> cumul_prob = 0xfd00
>
> So I suppose the attached file reproduces your problem, no? Does using 
> func2() in the decoder makes it bitexact (no, using it is not ok, but it 
> helps debugging)?

No. The Lagarith reference source code says "double", and doesn't even 
attempt portability, so when compiled on x86_32 (which the only official 
binary is) it uses 80bit long double.

--Loren Merritt