[vorbis-dev] libvorbis on Classic MacOS

[Segher Boessenkool]

Timothy Wayper wrote:

> The only warnings look harmless. MrC doesn't recognize the double 
> paranethesis convention around intentional "side-effect" assignments, so 
> there's a few warnings about those. Plus a few unused parameters/vars, 
> including those caused by it using the slow "no FPU control" mode. 
> Nothing horrible. There are no changes to the code required, except for 
> the inclusion of fp.h in vorbisfile.c for rint(). Alternatively, the use 
> of floor((x) + 0.5) for rint gives the same results.

That makes me think...  did you set all -DBLABLA's to sane
values?

> Hadn't even thought about Altivec, but I'll look into that immediately. 

There were some problems with wrongly aligned stack frames
on MrC with Altivec, I believe.  I've never done any much
Altivec in C on MacOS myself, though, so treat this as
hearsay.

> I haven't enabled it, but I don't think MrC automatically SIMD's 
> things... and I presume the reference implementation doesn't have a G4 
> accelerated backend ;-)

Correct on both accounts, I think.  But maybe some (system)
library uses Altivec internally and you get bitten by the
stack problem.

> I was concerned that maybe I was running out of stack space (since it's 
> manually adjusted on classic MacOS), but I turned that up to a 2Mb's, 
> and it made no difference.

Vorbis doesn't use nearly that much, fortunately ;)

> Thanks for your comments, that's actually very helpful! Now I know it 
> can work, I'll try a little harder :-)

Glad to be of help.  MacOS "Classic" is kind of dear to
me... sniff...

> Since the problem happens relatively early on, I'll run the same data 
> through a machine running Vorbis 1.0 on another (working) platform, and 
> compare it step by step with the problem case...

Insert lots and lots of fprintf()'s.

<p>Happy hunting,

Segher

p.s.  If you suspect miscompilation of a certain routine, feel
free to send me asm dumps (+ original code).

<p>--- >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.