On Mon, Mar 23, 2009 at 12:18 PM, Andy Dougherty <[email protected]> wrote: > >> On Mon, Mar 23, 2009 at 8:26 AM, Reini Urban <[email protected]> wrote: >> > 2009/3/22 Andy Dougherty via RT <[email protected]>: >> >> Well, it's not really completely resolved, but I couldn't think of a >> >> better status for it. The *BSD versions are resolved, and Solaris is >> >> half-resolved (works with Sun's cc, fails with gcc). >> >> >> >> This issue is now listed in Trac as TT #205, so we can close the RT >> >> version of the ticket and keep the Trac version open. >> > >> > IMHO it should have be solved by this config/gen/platform/solaris/math.c >> > entry: >> > >> > #include <math.h> >> > #if defined(__GNUC__) && defined(_LIB_VERSION) >> > _LIB_VERSION_TYPE _LIB_VERSION = _IEEE_; >> > #endif >> > >> > Maybe _LIB_VERSION is not defined? > > It is not solved by that bit of code. At least on Solaris 8, _LIB_VERSION > is not defined. Nor is _LIB_VERSION_TYPE. That bit of code is just > irrelevant. It has never been useful there. It was worth trying, but it > doesn't do anything. > > On Mon, 23 Mar 2009, Steve Peters wrote: > >> It is fixed for Solaris CC, but not for gcc. > > Correct. > >> My guess is inlining >> might be the cause here with gcc. [ . . . ] >> away. Right now, I don't have a Solaris to test this theory, so if >> someone (I'm looking at you Andy ;-) ) could compile Parrot on Solaris >> with gcc without optimizations, I'd be interested in seeing if this is >> fixed or not. > > No, that doesn't fix it. I suspect it's a math library issue, and I just > haven't figured out how to tell gcc to do whatever cc does when you pass > it the -xlibmieee option.
The answer might be here <http://people.redhat.com/drepper/sol-porting.pdf>. It sounds like -lieee -ffast-math might be the replacement to try. Steve Peters [email protected] _______________________________________________ http://lists.parrot.org/mailman/listinfo/parrot-dev
