c++ - Difference between execution policies and when to use them

Question

I noticed that a majority (if not all) functions in <algorithm> are getting one or more extra overloads. All of these extra overloads add a specific new parameter, for example, std::for_each goes from:

template< class InputIt, class UnaryFunction >
UnaryFunction for_each( InputIt first, InputIt last, UnaryFunction f );

to:

template< class ExecutionPolicy, class InputIt, class UnaryFunction2 >
void for_each( ExecutionPolicy&& policy, InputIt first, InputIt last, UnaryFunction2 f );

What effect does this extra ExecutionPolicy have on these functions?

What are the differences between:

• std::execution::seq
• std::execution::par
• std::execution::par_unseq

And when to use one or the other?

question from:https://stackoverflow.com/questions/39954678/difference-between-execution-policies-and-when-to-use-them

Answer 1

seq means "execute sequentially" and is the exact same thing as the version without an execution policy.

par means "execute in parallel", which permits the implementation to execute on multiple threads in parallel. You are responsible for making sure that no data races happen within f.

par_unseq means that in addition to being allowed to execute in multiple threads, the implementation is also allowed to interleave individual loop iterations within a single thread, i.e. load multiple elements and execute f on all of them only afterwards. This is required to permit a vectorized implementation.