[FFmpeg-devel] [PATCH] LGPL AC3 decoding

[Justin Ruggles]

Jason Garrett-Glaser wrote:
> This patch contains a clean-room-written
> ac3_decode_transform_coeffs_ch, written solely from the spec and
> Justin's copious advice and descriptions (no knowledge of the original
> function).  Oddly enough, it seems to actually be closer to the
> original liba52 code, despite the fact that I never saw that either.
> According to Diego, this should be sufficient to finally LGPL ffac3;
> I'll let someone else handle the header/configure changes for that.
> 
> Bonus: mine's faster, too.  With GCC 3.4 on 32-bit Core 2 Conroe
> (First 1234901 bytes of Broadway-5.1-48khz-448kbit.ac3):
> 
> Before: 27908.6 dezicycles (+/- 9.7)
> After: 24820.6 dezicycles (+/- 12.4)
> Total: 12.4% faster ac3_decode_transform_coeffs_ch
> 
> PSNR-wise it matches the original exactly; this was achieved through
> Justin supplying me with the single line of code for the bap=0 noise
> from the original code.  Note that I have not tested this on anything
> other than that sample of Broadway, so I cannot guarantee it works for
> all files.

It looks good to me.  I tested it out on a few other files and also got
identical results.

In my opinion, this is all that is needed in order to relicense ac3dec.c
to LGPL.  I really wish it hadn't come to this, but Walken stopped
responding to my and Diego's emails on the subject.  I am glad we got a
faster function in the end though.  Thank you Jason!

-Justin