OpenMP benchmarks

The following benchmarks compare pyPDAF configurations across ensemble sizes and observation-grid densities. They are representative measurements from the openmp_test benchmark rather than performance guarantees; results depend on the processor, MPI implementation, OpenMP runtime, and thread placement.

OpenMP speedup

Each cell reports the serial pyPDAF runtime divided by the runtime of the corresponding OpenMP build. Values above 1x therefore indicate a speedup, while values below 1x indicate that the OpenMP build was slower. The two panels show standard GIL-enabled Python and free-threaded Python.

Python and Fortran comparison

Each cell reports the pyPDAF runtime divided by the runtime of the matching Fortran PDAF benchmark. A value of 1x indicates equal runtime, and a larger value indicates that pyPDAF was slower. The panels compare OpenMP-enabled and serial runs separately.