problems with libfftw3

Message

hwang4 · #1 Post by **hwang4** » Thu Jun 15, 2006 9:18 pm

Hi, my serial vasp is running ok until I tried to compile it with using libfftw3.a instead of the fft libriary that comes with vasp itself. (libfftw3.a is said to be faster, right?)

the only thing I changed is comkment the following first line and uncomment the second line. And I have checked that the path for libfftw3.a i scrooect.

#FFT3D = fft3dfurth.o fft3dlib.o
FFT3D = fftw3d.o fft3dlib.o /usr/local/lib/libfftw3.a

I got no error complaints during compiling. However, the calculation will soon die. The lines from the screen is as follows:

vasp.4.6.26 15Dec04 complex
POSCAR found : 2 types and 34 ions
LDA part: xc-table for Ceperly-Alder, standard interpolation
POSCAR, INCAR and KPOINTS ok, starting setup
FFT: planning ... 12
Segmentation fault

Is this the problem of handling memory? My vasp version is 4.6.26. Anyone can help me out? Thanks a lot.

tjf · #2 Post by **tjf** » Fri Jun 16, 2006 8:51 am

Often segfaults are caused by incompatible data structures in argument lists to callsd subprograms, such as a different calling conventions or incompatible data sizes. Did you compile FFTW yourself? Have you linked against it for other codes? What's you arch/OS/compiler etc?

#3 Post by **admin** » Wed Jun 21, 2006 12:46 pm

1) does vasp run with Juergen Furthmuellers fft3d? If so, there is probably an error in your libfftw3 installation
if it also crashes with the alternative fft:
2) please upgrade to vasp.4.6.28, the .26 release had some problems with memory allocation.
3) the second reason may be that you really need too much memory for your job. In this case it may help to run vasp parallel.