P3M + Heffte tracking ticket

[RudolfWeeber]

The prototype is in my Github in the branch heffte

• Basic implementation of Hefte-based 3d FFt in electrostatic P3M
• Get rid of segfaults for 2 core parallel simulationsand for non-cubic simulation box (primary suspect are NaN in influence functions)
• Do a more detailed performance analysis of the K-space kernel.
  • figure out if all std::complex slowness is avoided, or if we need a different complex implementation
  • annotate different regoins and functions with Caliper and measure performance vs number of cores
    • charge assignment
    • halo communication for real space charge density
    • halo removal
    • 3d FFT
    • k-space force calc
    • Backward 3d fft
    • Introduciotn of space for halos
    • halo comm of E real-space fields
    • force assignment
• instance the Heffte fft3d with a sub-communicaotr, such that the reshaping is done on a single core
• enable OpenMP for heffte
• clean-up Heffte CMake integration

GPU and electrostatic pressure can probably wait until we have the CPU forces right.