Re: atlas on HPPA (PA RISC)
In reference to a message from Camm Maguire, dated Jan 06:
> Greetings! Anyone out there using these machines wanting an atlas
> port? I've run into some very odd compiler problems, and am in need
> of an expert to go any further. Basically, single precision floating
> point fails accross the board, double passes all tests. The compiler
> dies with an error if the code tries to tune for too large a cache.
> Cross compilation with -march=1.0 fails.
first of all, you probably don't want -march=1.0 .... try 1.1
also there are some floating point issues on hppa-linux right now. willy
mentioned that the 'long double' type is broken (gcc and glibc don't
agree on its size) but i don't know if that's the problem in atlas.
python2 also has a sqrt on floating point number problem, but i don't
know if it's related to this or not.
if you can provide a simple test case that is failing, it'll help us
track down the problem.
Debian Developer <firstname.lastname@example.org>