I’m compiling a large F90 application that uses the FFTW libraries for some operations.
I’m using the current trial version of PGI Workstation 5.2 for x864-64 on a dual Opteron system.
Currently the application segfaults when it accesses the FFTW functions. But if I compile a version using native FFT routines instead of FFTW, it works OK (just slower…).
I’m compiling FFTW with pgcc/pgCC and pgf90, using these compiler flags:
-tp k8-64 -O3 -fastsse
Usually the F90 application is linked to shared FFTW libraries. But, i notice that while i am using the “–enable-shared” option to configure FFTW, no shared libs are being built.
So i checked the pgi manpages and tried adding the flag “-shared” or “-shared -fPIC” to my CFLAGS/FFLAGS. With these options, the FFTW configure script aborts immediately, as the sample program it tests the compiler with segfaults on execution. The test program is just a one-line void function.
Any suggestions would be appreciated. I’ve removed the optimization flags, but there is no change. This is my first forray into x86-64 architecture, have compiled the same software before on Xeon with PGI 5.1 compilers and had no problems, although i was mixing gcc and pgf90 if i recall correctly, for compiling FFTW.