use std::atomic<std::size_t> instead of flag for better performance in arm; cross-platform supporting; add more test cases

include/circ_elem_array.h

@@ -16,10 +16,10 @@ struct alignas(std::max_align_t) elem_array_head {
     using uc_t = std::uint16_t;
     using ac_t = std::atomic<uc_t>;
 
+    std::atomic<std::size_t> lc_ { 0 }; // write spin lock flag
+
     ac_t cc_ { 0 }; // connection counter, using for broadcast
     ac_t wt_ { 0 }; // write index
-
-    std::atomic_flag lc_ ATOMIC_FLAG_INIT; // write spin lock flag
 };
 
 enum : std::size_t {
@@ -93,7 +93,7 @@ public:
     }
 
     void* acquire(void) {
-        while (lc_.test_and_set(std::memory_order_acquire)) {
+        while (lc_.exchange(1, std::memory_order_acquire)) {
             std::this_thread::yield();
         }
         elem_t* el = elem(wt_.load(std::memory_order_relaxed));
@@ -114,7 +114,7 @@ public:
 
     void commit(void* /*ptr*/) {
         wt_.fetch_add(1, std::memory_order_relaxed);
-        lc_.clear(std::memory_order_release);
+        lc_.store(0, std::memory_order_release);
     }
 
     uc_t cursor(void) const {

include/circ_queue.h

@@ -84,8 +84,27 @@ public:
         return old;
     }
 
+    bool push(T const & item) {
+        if (elems_ == nullptr) return false;
+        auto ptr = elems_->acquire();
+        ::new (ptr) T(item);
+        elems_->commit(ptr);
+        return true;
+    }
+
+    template <typename P>
+    auto push(P&& param) // disable this if P is the same as T
+     -> std::enable_if_t<!std::is_same<std::remove_reference_t<P>, T>::value, bool> {
+        if (elems_ == nullptr) return false;
+        auto ptr = elems_->acquire();
+        ::new (ptr) T { std::forward<P>(param) };
+        elems_->commit(ptr);
+        return true;
+    }
+
     template <typename... P>
-    bool push(P&&... params) {
+    auto push(P&&... params) // some old compilers are not support this well
+     -> std::enable_if_t<(sizeof...(P) != 1), bool> {
         if (elems_ == nullptr) return false;
         auto ptr = elems_->acquire();
         ::new (ptr) T { std::forward<P>(params)... };

test/test_circ.cpp

@@ -293,41 +293,42 @@ void Unit::test_prod_cons_3v1(void) {
     test_prod_cons<3, 1>();
 }
 
-template <int B, int E>
-struct test_cons_performance {
+template <int P, int C>
+struct test_performance {
     static void start(void) {
-        test_prod_cons<1, B, false>();
-        test_cons_performance<B + 1, E>::start();
+        test_performance<P - 1, C - 1>::start();
+        test_prod_cons<P, C, false>();
     }
 };
 
-template <int E>
-struct test_cons_performance<E, E> {
+template <int C>
+struct test_performance<1, C> {
     static void start(void) {
-        test_prod_cons<1, E, false>();
+        test_performance<1, C - 1>::start();
+        test_prod_cons<1, C, false>();
     }
 };
 
-template <int B, int E>
-struct test_prod_performance {
+template <int P>
+struct test_performance<P, 1> {
     static void start(void) {
-        test_prod_cons<B, 1, false>();
-        test_prod_performance<B + 1, E>::start();
+        test_performance<P - 1, 1>::start();
+        test_prod_cons<P, 1, false>();
     }
 };
 
-template <int E>
-struct test_prod_performance<E, E> {
+template <>
+struct test_performance<1, 1> {
     static void start(void) {
-        test_prod_cons<E, 1, false>();
+        test_prod_cons<1, 1, false>();
     }
 };
 
 void Unit::test_prod_cons_performance(void) {
-    test_cons_performance<1, 10>::start();
-    test_prod_performance<1, 10>::start();
-    test_prod_cons<3, 3, false>();  // just test
-    test_prod_cons<5, 5>();         // test & verify
+    test_performance<1 , 10>::start();
+    test_performance<10, 1 >::start();
+    test_performance<10, 10>::start();
+    test_prod_cons  <3 , 3 >(); // test & verify
 }
 
 void Unit::test_queue(void) {