- previous method did not work with kernels generated for walberla where
  block size changes are made at runtime
- device query does not always work, since the compile system may have
  no GPU or not the same GPU
-> max block size is passed as parameter and only optionally determined
   by a device query