Fritz Jost writes: > > I have come across an issue using GCC 3.4.4 which I am in need of > help. The problem manifests itself as extremely slow floating point > performance for specific arguments passed to the dsin function in > FORTRAN using g77. The following statement : > > y = dsin(dble(1.25169754E-06)) > > when compiled with g77 under GCC 3.4.4 will take 400usec to execute > on my 2+ GHz Opteron platform. I have tested several machines and > they all show the same problem. The same code executes at sub > microsecond speed using GCC 4.0.2 on a Turion-64 notebook as well > as a XEON based machine using gcc 3.3.3. Unfortunately changing the > GCC version on my project is not an option. Can anybody shed some > light on what is going on here ? Please see http://gcc.gnu.org/ml/gcc-help/2006-05/msg00189.html That question is about pow() rather than sin(). but the answer is the same. See also __dubsin() in glibc/sysdeps/ieee754/dbl-64/dsincos.c. Andrew.
