[Vorbis-dev] Vorbis encoding bug on amd64

Erik de Castro Lopo mle+la at mega-nerd.com
Wed Aug 29 04:51:59 PDT 2007


xiphmont at xiph.org wrote:

> It does, it just manifests differently.  In any case, fix committed,
> verify it's correct please.

These two? :

    r13651 | xiphmont | 2007-08-29 21:25:58 +1000 (Wed, 29 Aug 2007) | 3 lines
    One more patch to bug #300
    ------------------------------------------------------------------------
    r13644 | xiphmont | 2007-08-29 04:25:49 +1000 (Wed, 29 Aug 2007) | 4 lines

    Bug #300 was not completely killed; a setting for uncoupled modes in
    setup_22.h got missed.

Sorry, no, that didn't fix it.

Just to make sure, I even added the bare bones of a test suite to SVN.
To run it do "make check". The test suite generates source data which 
is a little different for the previous test data and I'm now getting
slightly different results.

On x86 (ie 32 bit Intel) I'm getting:

    vorbis_44100.oga     : ok
    vorbis_48000.oga     : ok
    vorbis_32000.oga     : ok
    vorbis_22050.oga     : ok
    vorbis_16000.oga     : ok
    vorbis_96000.oga     : ok

all tests passing, while in x86-64 the (very simple) test fails at 
all sample rates tested:

    vorbis_44100.oga     : Error : max_abs (628.911743) too big.
    vorbis_48000.oga     : Error : max_abs (672.001526) too big.
    vorbis_32000.oga     : Error : max_abs (616.815552) too big.
    vorbis_22050.oga     : Error : max_abs (57612.203125) too big.
    vorbis_16000.oga     : Error : max_abs (65091.523438) too big.
    vorbis_96000.oga     : Error : max_abs (1026.016479) too big.

Erik
-- 
-----------------------------------------------------------------
Erik de Castro Lopo
-----------------------------------------------------------------
A fanatic is one who can't change his mind and won't change the subject.
         -- Winston Churchill


More information about the Vorbis-dev mailing list