[vorbis-dev] BCB build successful?
Lourens Veen
jsr at dds.nl
Wed Jan 31 01:11:58 PST 2001
Well, I'm no expert but this is on the BladeEnc homepage:
WARNING 1 - Audio Quality
Old versions of BladeEnc were very sensitive for compiler optimisations.
Even well known compilers as Visual C++ and GNU C generated mutated
versions of BladeEnc, producing lower quality output, when optimisations
were enabled.
However, the cause of this problem has later been identified and solved
(it was a compiler error, but it could easily and safely be avoided by
marking a certain variable as 'volatile') and is no longer of any
greater concern.
All compilers we know of generates correctly working BladeEnc binaries
even when using the highest
optimization level, but just to be sure we recommend you the following
test:
First compile a non-optimized version of BladeEnc and encode an mp3 file
(doesn't have to be so long, 20 seconds should be more than enough).
Then compile a fully optimized BladeEnc and use it to encode the same
file, then compare the outputs. They should be binary identical and not
differ by a single byte. If they differ, something has happened and I
recommend you to not trust the optimized binary.
(quote from http://bladeenc.mp3.no, in the Sourcecode section).
So apparently it's solved now, and it was indeed a compiler bug.
Must be source differences then.
Lourens
"David K. Gasaway" wrote:
>
> > Not necessarily I think. If the compiler optimizes the float math and
> > loses some precision in the process, then the vector quantisation and
> > Huffman compression will compress a bit better (and quality will drop
> > a bit).
>
> Optimization *shouldn't* affect precision. Lossly optimization?
> Hmm.
>
> > Lourens
>
--- >8 ----
List archives: http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'vorbis-dev-request at xiph.org'
containing only the word 'unsubscribe' in the body. No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.
More information about the Vorbis-dev
mailing list