FFTW r2r in 2017.0

[4f0drlp7eyj3]

Hi,

   Before trying to debug a third-party code with FFTW calls, I thought I'd pick the brains of the experts.

   I build the code either by linking to MKL (via "-mkl=sequential"), or to FFTW 3.3.3. The MKL-linked binary generates incorrect results (zeros in an array where there shouldn't be), whereas the FFTW-linked binary generates the expected results.

   The basic question I have, is if there are limitations or traps to watch out for with these 2 consecutive calls (like not implemented in MKL)?

dfftw_plan_r2r_1d( (INTEGER*8) A, (INTEGER) B, (REAL*8) C[32768], (REAL*8) D[32768], (INTEGER) 0, (INTEGER) 64 )
dfftw_execute_r2r( (INTEGER*8) A, C, D )

Thanks; Chris

[Gennady_F_Intel]

if you pass INTEGER*8, could you please try to explicitly link with ILP64 libs? here is the link to the MKL Linker Adviser - https://software.intel.com/en-us/articles/intel-mkl-link-line-advisor

[4f0drlp7eyj3]

Hi Gennady,

   Thanks for the suggestion. No luck, though, either with building -i8 and linking to ilp64, or just linking.

[Gennady_F_Intel]

Could you please try the latest version 2018? We had issue with FFTW API and the fix available in MKL 2018.