std::execution::sequenced_policy, std::execution::parallel_policy, std::execution::parallel_unsequenced_policy, std::execution::unsequenced_policy

class sequenced_policy { /* unspecified */ };

class parallel_policy { /* unspecified */ };

class parallel_unsequenced_policy { /* unspecified */ };

class unsequenced_policy { /* unspecified */ };

During the execution of a parallel algorithm with any of these three execution policies, if the invocation of an element access function exits via an uncaught exception, std::terminate is called, but the implementations may define additional execution policies that handle exceptions differently.

Notes

When using parallel execution policy, it is the programmer's responsibility to avoid data races and deadlocks:

int a[] = {0,1};
std::vector<int> v;
std::for_each(std::execution::par, std::begin(a), std::end(a), [&](int i) {
  v.push_back(i*2+1); // Error: data race
});

std::atomic<int> x{0};
int a[] = {1,2};
std::for_each(std::execution::par, std::begin(a), std::end(a), [&](int) {
  x.fetch_add(1, std::memory_order_relaxed);
  while (x.load(std::memory_order_relaxed) == 1) { } // Error: assumes execution order
});

int x = 0;
std::mutex m;
int a[] = {1,2};
std::for_each(std::execution::par, std::begin(a), std::end(a), [&](int) {
  std::lock_guard<std::mutex> guard(m);
  ++x; // correct
});

Unsequenced execution policies are the only case where function calls are unsequenced with respect to each other, meaning they can be interleaved. In all other situations in C++, they are indeterminately-sequenced (cannot interleave). Because of that, users are not allowed to allocate or deallocate memory, acquire mutexes, use non-lockfree std::atomic specializations, or, in general, perform any vectorization-unsafe operations when using these policies (vectorization-unsafe functions are the ones that synchronize-with another function, e.g. std::mutex::unlock synchronizes-with the next std::mutex::lock).

int x = 0;
std::mutex m;
int a[] = {1,2};
std::for_each(std::execution::par_unseq, std::begin(a), std::end(a), [&](int) {
  std::lock_guard<std::mutex> guard(m); // Error: lock_guard constructor calls m.lock()
  ++x;
});

If the implementation cannot parallelize or vectorize (e.g. due to lack of resources), all standard execution policies can fall back to sequential execution.

See also