| 5 | | ||17/03/11 ||SR ||OpenMP-parallelization does not work for low optimization level (-O0). Program crashes with different error messages, e.g. "{{{integer devide by zero}}}" and/or gives wrong results. The problem seems to lie within routine {{{flow_statistics}}}. If {{{$OMP}}}-directives are removed from {{{flow_statistics}}}, everything seems to work. \\ \\ Another error source in case of 1D-decomposition along {{{y}}} might be in routine {{{poisfft}}}. Here, removing array {{{work_fftx}}} from the {{{PRIVATE}}} clause in lines 1072 and 1183 (in r695) seems to help. \\ \\ Using the Intel-compiler option {{{-override-limits}}} removes the problem completely, but only if the Piacsek-Williams advection scheme is switched on.\\ \\ '''So far, usage of Intel compiler option {{{-openmp}}} is not recommended! '''|| |
