[MPlayer-dev-eng] [PATCH] Make mp3lib SIMD optimizations work on AMD64, Part 4

[Zuxy Meng]

Hi,

2007/5/23, Guillaume Poirier <gpoirier at mplayerhq.hu>:
> Hi,
>
> Zuxy Meng wrote:
> > Hi,
> >
> > 2007/5/22, Guillaume POIRIER <poirierg at gmail.com>:
> >> Hi,
> >>
> >> Looks Ok to me.
> >> (I tested that it doesn't break compilation on AMD64, so there's no
> >> obvious regression with regards to bringing AMD64 support ... :-) )
> >>
> >> Commit anytime
> >
> > Committed. Part 4 is big...I didn't bother to decipher the assembly in
> > tabinit_MMX.c; instead I recorded what was produced by
> > make_decode_table_MMX() and used that data to initialize
> > mp3lib_decwins directly. Therefore tabinit_MMX.c itself isn't needed
> > any more.
> >
> > I guess this approach causes least trouble.
>
> Looks Ok to me. Once it's committed, please do explain where these
> magic values come from by putting  doxygen-like comment on top of the
> new array, referring to the commit just just made (please also add the
> revision number), pointing out that the values are essentially always
> the same no matter the input file.

How about this?

/**
  This array of magic numbers were calculated by the pure function
  make_decode_tables_MMX(32768), which had been implemented in (now deleted)
  tabinit_MMX.c, r21060.
  */
static short __attribute__((aligned(8))) mp3lib_decwins[] =
...

-- 
Zuxy
Beauty is truth,
While truth is beauty.
PGP KeyID: E8555ED6