Has anyone noticed that their IDL code runs about 20% slower on linux (using fft for example) than win32? Is this due to the lack of vectorization in the gcc compiler? Can i speed up my fft by calling the LAPACK routines? If so do these use the optimized Linux LAPACK library?