I am just getting acquainted with parallelization, so my question is probably dumb. But here is a brief description of the problem: I want to solve a finite horizon dynamic programming problem by backward induction. I start in the last period (T), and for all possible elements of the state space (state space is discrete) I solve for the optimal values of the choice variables. This gives me the value function, which allows me to iterate one step backward and use the Bellman equation to solve for the period T-1 value function.
Now it is clear that the program has to do the same thing for a lot of times, namely solve for the optimal choices for a given state. This is done inside a DO loop, where the index variable indexes the states. The solution for one state is totally independent of the solution for another state, so I thought there is room for multi-threading. When I found the DO CONCURRENT construct, I thought this is exactly what I need. Although there is a lot of stuff inside this loop (including other loops), but the order of execution in the main loop is irrelevant.
I followed the requirements for DO CONCURRENT (only pure procedures and so on), and I compiled my code with the /Qparallel option. There was no error during compilation. But if anything, my code got slower, instead of faster. Did I completely misunderstand the DO CONCURRENT construct? How can I know if the compiler was able to paralellize at all?
Thanks for any hint,
Andras
