Closed dhubber closed 8 years ago
I think we have a working solution which I would prefer to leave (since it's simple to understand and should not affect performance), which is essentially each OpenMP thread computes its own min/max and is then combined in a short critical statement (i.e. doing min/max manually). So I'll close this idea now since I don't think there's any need to follow up on it!
Have both C++11 atomic min/max operations for min/max reduction when computing timestep levels and OpenMP locks (for older compilers) in the same code but chosen by compiler flags (which is dependent on the compiler or on selected Makefile options).