[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: compiler-options?

"Christopher C. Chimelis" wrote:
> Just use -mcpu.  It should do the right thing in all cases.

Well, I remember bad experiences using -mcpu alone in the past,
so I advised to use the linker option, too. But I must admit
that this was years ago (most likely egcs 1.0.x on RH 5.1/AXP).

> > What are the "symptoms" that a special code needs -mieee? Most packages
> > which I compile, are used for molecular phylogeny. MrBayes, as the name
> > says, works with Bayesian methods. I guess, that fpu-operations are
> > heavily used in these kind of calculations.
> Yes, it sounds like the stuff that you want to compile is definitely in
> the "needs -mieee" category :-)

Kerstin, be careful! Using -mieee is a big burden if you
don't really need it. This option definitely isn't
necessary because of floating point calculations per se.
The need for -mieee usually can be avoided in source code
(if it boils down to denormals and one has the option to
patch the sources).

> [...] I would recommend compiling them
> with Compaq's compilers since they schedule code for EV5 and EV6 much
> better than gcc probably ever will. [...]

Yes, using Compaq's compilers is worth a try. But this
decision is independent of the other (using IEEE software
emulation on Alphas or not).

Stefan Schroepfer

Reply to: