[vorbis-dev] EGCS broken; tested 2.95 through CVS current

[Monty]

Just so folks know, all the modern variants of EGCS (now named gcc) out there
appear to be producing occasionally bogus x86 FPU code at -O1 or higher.  Guess
what... Vorbis trips the bug whatever it is.

I just grabbed and built the latest CVS code, and it's also broken.  I'm constructing a small failure case example for a bug report now.  It looks like FPU arithmetic into a register isn't being stored back into memory.  I don't know for sure yet; I don't read Pentium FPU code well :-P

The results have been confirmed on multiple machines running different distributions.  The bug is entirely deterministic.

GCC 2.7.2.2 tests OK.  I'm looking for a copy of 2.8 to try as well.

So, for the immediate moment (until EGCS is fixed or I find a workaround), do not use EGCS with optimization to compile Vorbis.   Yes, I know this sucks.

Monty

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/