Merge remote-tracking branch 'origin/development' into embb453_rwlock

CHANGELOG.md

 Embedded Multicore Building Blocks (EMB²)
 =========================================
 
+Version 0.3.1
+-------------
+
+### Features:
+- None
+
+### Changes and improvements:
+- Removed one function argument from algorithms::Invoke
+- Added "explicit" specifier to base type constructor of Atomic<BaseType*>
+- Added "const" qualifier to dereference operator and member access operator of AtomicPointer<>
+- Changed AtomicBase<>::CompareAndSwap to atomically return expected value
+- Replaced constant in dataflow_cpp_test_simple.cc with corresponding macro
+- Added initialization of atomic variable in hazard_pointer_test.cc to avoid warning with GCC 5.1
+- Changed initial value of allocated_object_from_different_thread
+- Added tests for ID Pool and check for memory leaks
+- Updated unit test for the UniqueLock::Swap
+
+### Bug fixes:
+- Fixed implementation of ID pool (provided fewer elements than specified by capacity)
+- Fixed unsigned overflow bug in timed wait function of condition variables
+- Fixed implementation of UniqueLock::Swap
+
+### Build system:
+- Improved CMake output for automatic initialization option
+- Fixed cpplint and unsigned/signed warnings
+
+### Documentation:
+- Fixed documentation of UniqueLock class
+- Updated README file
+
+
 Version 0.3.0
 -------------
 

CMakeLists.txt

@@ -28,7 +28,7 @@ cmake_minimum_required (VERSION 2.8.9)
 # Version number
 set (EMBB_BASE_VERSION_MAJOR 0)
 set (EMBB_BASE_VERSION_MINOR 3)
-set (EMBB_BASE_VERSION_PATCH 0)
+set (EMBB_BASE_VERSION_PATCH 1)
 
 # Fix compilation for CMake versions >= 3.1
 #
@@ -59,7 +59,9 @@ IF(NOT OpenCL_FOUND)
   MESSAGE( STATUS "OpenCL is not there, will build without MTAPI OpenCL Plugin." )
 ENDIF()
 
-
+# give the user the possibility, to append compiler flags
+set (CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${EXTRA_CMAKE_CXX_FLAGS}")
+set (CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${EXTRA_CMAKE_C_FLAGS}")
 
 if(NOT CMAKE_BUILD_TYPE)
 set(CMAKE_BUILD_TYPE "Release" CACHE STRING
@@ -100,6 +102,13 @@ else()
 endif()
 message("   (set with command line option -DWARNINGS_ARE_ERRORS=ON/OFF)")
 
+if (USE_AUTOMATIC_INITIALIZATION STREQUAL ON)
+  message("-- MTAPI/Tasks automatic initialization enabled (default)")
+else()
+  message("-- MTAPI/Tasks automatic initialization disabled")
+endif()
+message("   (set with command line option -DUSE_AUTOMATIC_INITIALIZATION=ON/OFF)")
+
 include(CMakeCommon/SetCompilerFlags.cmake)
 SetGNUCompilerFlags(compiler_libs compiler_flags)
 SetVisualStudioCompilerFlags(compiler_libs compiler_flags)

README.md

@@ -270,8 +270,8 @@ If you want to use the C++ functionalities of EMB², you have to link the
 following libraries (names will be different on Windows and on Linux) in the
 given order:
 
-    embb_base, embb_base_cpp, embb_mtapi_c, embb_mtapi_cpp, embb_containers_cpp,
-    embb_algorithms_cpp, embb_dataflow_cpp
+    embb_dataflow_cpp, embb_algorithms_cpp, embb_containers_cpp,
+    embb_mtapi_cpp, embb_mtapi_c, embb_base_cpp, embb_base_c
 
 The C++ header files can be included as follows:
 
@@ -284,7 +284,7 @@ The C++ header files can be included as follows:
 
 The following libraries have to be linked in the given order:
 
-    embb_base_c, mtapi_c
+    embb_mtapi_c, embb_base_c
 
 The C header files can be included as follows:
 
@@ -323,6 +323,8 @@ Known Bugs and Limitations
   is bounded by a predefined but modifiable constant (see functions
   embb_thread_get_max_count() / embb_thread_set_max_count() and class
   embb::base::Thread).
+- While MTAPI fully supports heterogeneous systems, the algorithms and
+  dataflow components are currently limited to homogeneous systems.
 
 
 Development and Contribution

algorithms_cpp/include/embb/algorithms/invoke.h

@@ -49,33 +49,37 @@ typedef embb::base::Function<void> InvokeFunctionType;
 #ifdef DOXYGEN
 
 /**
- * Spawns one to ten function objects at once and runs them in parallel.
+ * Spawns two to ten function objects at once and runs them in parallel.
  *
  * Blocks until all of them are done.
  *
  * \ingroup CPP_ALGORITHMS_INVOKE
  */
-template<typename Function1, ...>
+template<typename Function1, typename Function2, ...>
 void Invoke(
   Function1 func1,
   /**< [in] First function object to invoke */
+  Function2 func2,
+  /**< [in] Second function object to invoke */
   ...);
 
 /**
-* Spawns one to ten function objects at once and runs them in parallel using the
+* Spawns two to ten function objects at once and runs them in parallel using the
 * given embb::mtapi::ExecutionPolicy.
 *
 * Blocks until all of them are done.
 *
 * \ingroup CPP_ALGORITHMS_INVOKE
 */
-template<typename Function1, ...>
+template<typename Function1, typename Function2, ...>
 void Invoke(
   Function1 func1,
   /**< [in] Function object to invoke */
+  Function2 func2,
+  /**< [in] Second function object to invoke */
   ...,
-  const embb::mtapi::ExecutionPolicy & policy
-  /**< [in] embb::mtapi::ExecutionPolicy to use */
+  const embb::tasks::ExecutionPolicy & policy
+  /**< [in] embb::tasks::ExecutionPolicy to use */
   );
 
 #else // DOXYGEN
@@ -118,13 +122,6 @@ class TaskWrapper {
 };
 } // namespace internal
 
-template<typename Function1>
-void Invoke(
-  Function1 func1,
-  const embb::tasks::ExecutionPolicy& policy) {
-  internal::TaskWrapper<Function1> wrap1(func1, policy);
-}
-
 template<typename Function1, typename Function2>
 void Invoke(
   Function1 func1,
@@ -290,12 +287,6 @@ template<typename Function1, typename Function2, typename Function3,
   internal::TaskWrapper<Function10> wrap10(func10, policy);
 }
 
-template<typename Function1>
-void Invoke(
-  Function1 func1) {
-  Invoke(func1, embb::tasks::ExecutionPolicy());
-}
-
 template<typename Function1, typename Function2>
 void Invoke(
   Function1 func1,

algorithms_cpp/test/invoke_test.cc

@@ -44,7 +44,6 @@ static void Invocable10() {}
 
 void InvokeTest::Test() {
   using embb::algorithms::Invoke;
-  Invoke(&Invocable1);
   Invoke(&Invocable1, &Invocable2);
   Invoke(&Invocable1, &Invocable2, &Invocable3);
   Invoke(&Invocable1, &Invocable2, &Invocable3, &Invocable4);
@@ -61,4 +60,24 @@ void InvokeTest::Test() {
          &Invocable6, &Invocable7, &Invocable8, &Invocable9);
   Invoke(&Invocable1, &Invocable2, &Invocable3, &Invocable4, &Invocable5,
          &Invocable6, &Invocable7, &Invocable8, &Invocable9, &Invocable10);
+
+  embb::tasks::ExecutionPolicy policy;
+  Invoke(&Invocable1, &Invocable2, policy);
+  Invoke(&Invocable1, &Invocable2, &Invocable3, policy);
+  Invoke(&Invocable1, &Invocable2, &Invocable3, &Invocable4, policy);
+  Invoke(&Invocable1, &Invocable2, &Invocable3, &Invocable4, &Invocable5,
+         policy);
+  Invoke(&Invocable1, &Invocable2, &Invocable3, &Invocable4, &Invocable5,
+         &Invocable6, policy);
+  Invoke(&Invocable1, &Invocable2, &Invocable3, &Invocable4, &Invocable5,
+         &Invocable6, &Invocable7, policy);
+  Invoke(&Invocable1, &Invocable2, &Invocable3, &Invocable4, &Invocable5,
+         &Invocable6, &Invocable7, &Invocable8, policy);
+  Invoke(&Invocable1, &Invocable2, &Invocable3, &Invocable4, &Invocable5,
+         &Invocable6, &Invocable7, &Invocable8, &Invocable9, policy);
+  Invoke(&Invocable1, &Invocable2, &Invocable3, &Invocable4, &Invocable5,
+         &Invocable6, &Invocable7, &Invocable8, &Invocable9, policy);
+  Invoke(&Invocable1, &Invocable2, &Invocable3, &Invocable4, &Invocable5,
+         &Invocable6, &Invocable7, &Invocable8, &Invocable9, &Invocable10,
+         policy);
 }

base_c/src/condition_variable.c

@@ -83,8 +83,8 @@ int embb_condition_wait_until(embb_condition_t* condition_var,
   embb_time_t now;
   embb_time_now(&now);
   /* Check if absolute timepoint (in milliseconds) still is in the future */
-  if (time->seconds * 1000 + time->nanoseconds / 1000000
-      - now.seconds * 1000 - now.nanoseconds / 1000000 > 0) {
+  if ((time->seconds * 1000 + time->nanoseconds / 1000000)
+      > (now.seconds * 1000 + now.nanoseconds / 1000000)) {
     /* Convert to (unsigned type) milliseconds and round up */
     DWORD time_diff = (DWORD) (
         time->seconds * 1000 + time->nanoseconds / 1000000

base_c/src/internal/thread_index.c

@@ -128,6 +128,20 @@ void embb_internal_thread_index_set_max(unsigned int max) {
   *embb_max_number_thread_indices() = max;
 }
 
+/**
+ * \pre the calling thread is the only active thread
+ *
+ * \post the thread indices count and calling thread index is reset
+ */
 void embb_internal_thread_index_reset() {
+  /** This function is only called in tests, usually when all other threads
+   * except the main thread have terminated. However, the main thread still has
+   * potentially stored its old index value in its thread local storage,
+   * which might be assigned additionally to another thread (as the counter is
+   * reset), which may lead to hard to detect bugs. Therefore, reset the thread
+   * local thread id here.
+   */
+  embb_internal_thread_index_var = UINT_MAX;
+
   embb_counter_init(embb_thread_index_counter());
 }
\ No newline at end of file

base_c/test/condition_var_test.cc

@@ -38,7 +38,7 @@ ConditionVarTest::ConditionVarTest()
   embb_condition_init(&cond_wait_);
   embb_mutex_init(&mutex_cond_wait_, EMBB_MUTEX_PLAIN);
 
-  CreateUnit("Timed wait timouts")
+  CreateUnit("Timed wait timeouts")
       .Add(&ConditionVarTest::TestTimedWaitTimeouts, this);
   if (num_threads_ >= 2) {
     CreateUnit("Condition Notify Test")
@@ -64,10 +64,10 @@ void ConditionVarTest::TestNotify() {
 
     while (embb_counter_get(&counter_)
         < static_cast<unsigned int>(num_threads_-1))
-    {}  // all threads entered critical section
+    {}  // All threads entered critical section
     embb_mutex_lock(&mutex_cond_notify_);
     embb_mutex_unlock(&mutex_cond_notify_);
-    // All threads called wait on the condition (Even last thread)
+    // All threads called wait on the condition (even last thread)
 
     embb_counter_init(&counter_);
 
@@ -75,7 +75,7 @@ void ConditionVarTest::TestNotify() {
     embb_mutex_lock(&mutex_cond_wait_);
     embb_condition_wait_for(&cond_wait_, &mutex_cond_wait_, &duration);
     while (embb_counter_get(&counter_) == 0)
-    {} //if hangs here signal has not succeded
+    {} // If test hangs here, signalling has not succeeded
     PT_ASSERT_EQ_MSG(embb_counter_get(&counter_), static_cast<unsigned int>(1),
         "Only one thread notified");
 
@@ -85,7 +85,7 @@ void ConditionVarTest::TestNotify() {
 
     while (embb_counter_get(&counter_) !=
         static_cast<unsigned int>(num_threads_-1))
-    {} // If this hangs then not all threads were notified.
+    {} // If test hangs here, not all threads were notified
 
     embb_mutex_unlock(&mutex_cond_wait_);
     embb_mutex_destroy(&mutex_cond_wait_);
@@ -105,13 +105,13 @@ void ConditionVarTest::TestTimedWaitTimeouts() {
   embb_time_t time;
   embb_duration_t duration = EMBB_DURATION_INIT;
 
-  // Wait for now tests already passed time point
+  // Wait for "now" tests already passed time point
   embb_time_now(&time);
   embb_mutex_lock(&mutex);
   int status = embb_condition_wait_until(&cond, &mutex, &time);
   PT_EXPECT_EQ(status, EMBB_TIMEDOUT);
 
-  // Wait for a future timepoint
+  // Wait for a future time point
   status = embb_duration_set_milliseconds(&duration, 1);
   PT_EXPECT_EQ(status, EMBB_SUCCESS);
   status = embb_time_in(&time, &duration); // Time now

base_c/test/time_test.cc

@@ -36,6 +36,9 @@ namespace test {
 
 TimeTest::TimeTest() {
   CreateUnit("Time in duration").Add(&TimeTest::TestTimeInDuration, this);
+  CreateUnit("Monotonicity").Add(
+      &TimeTest::TestMonotonicity, this,
+      1, partest::TestSuite::GetDefaultNumIterations() * 10);
 }
 
 void TimeTest::TestTimeInDuration() {
@@ -48,6 +51,20 @@ void TimeTest::TestTimeInDuration() {
   PT_EXPECT_EQ(status, EMBB_SUCCESS);
 }
 
+void TimeTest::TestMonotonicity() {
+  embb_time_t first;
+  embb_time_t second;
+  int status1 = embb_time_in(&first, embb_duration_zero());
+  int status2 = embb_time_in(&second, embb_duration_zero());
+  PT_EXPECT_EQ(status1, EMBB_SUCCESS);
+  PT_EXPECT_EQ(status2, EMBB_SUCCESS);
+  unsigned long long first_abs = first.seconds * 1000 +
+                                 first.nanoseconds / 1000000;
+  unsigned long long second_abs = second.seconds * 1000 +
+                                  second.nanoseconds / 1000000;
+  PT_EXPECT_GE(second_abs, first_abs);
+}
+
 } // namespace test
 } // namespace base
 } // namespace embb

base_c/test/time_test.h

@@ -42,9 +42,14 @@ class TimeTest : public partest::TestCase {
 
  private:
   /**
-   * Tests time in duration method.
+   * Tests time-in-duration method.
    */
   void TestTimeInDuration();
+
+  /**
+   * Tests that succeedingly taken times are monotonously increasing.
+   */
+  void TestMonotonicity();
 };
 
 } // namespace test

base_cpp/include/embb/base/atomic.h

@@ -478,7 +478,7 @@ class Atomic<BaseType*> : public embb::base::internal::atomic::
  public:
   Atomic() : embb::base::internal::atomic::
     AtomicPointer<BaseType, ptrdiff_t, sizeof(BaseType*)>() {}
-  Atomic(BaseType* p) : embb::base::internal::atomic::
+  explicit Atomic(BaseType* p) : embb::base::internal::atomic::
     AtomicPointer<BaseType, ptrdiff_t, sizeof(BaseType*)>(p) {}
 
   BaseType* operator=(BaseType* p) {

base_cpp/include/embb/base/internal/atomic/atomic_base.h

@@ -177,8 +177,7 @@ CompareAndSwap(BaseType& expected, BaseType desired) {
     compare_and_swap(&AtomicValue, &native_expected, native_desired)) !=0
     ? true : false;
 
-  if (!return_val)
-    expected = Load();
+  memcpy(&expected, &native_expected, sizeof(expected));
 
   return return_val;
 }

base_cpp/include/embb/base/internal/atomic/atomic_pointer.h

@@ -65,8 +65,8 @@ class AtomicPointer : public AtomicArithmetic<BaseType*, DifferenceType, S> {
   bool IsPointer() const;
 
   // The methods below are documented in atomic.h
-  BaseType* operator->();
-  BaseType& operator*();
+  BaseType* operator->() const;
+  BaseType& operator*() const;
 };
 
 template<typename BaseType, typename DifferenceType, size_t S>
@@ -93,13 +93,13 @@ inline bool AtomicPointer<BaseType, DifferenceType, S>::
 
 template<typename BaseType, typename DifferenceType, size_t S>
 inline BaseType* AtomicPointer<BaseType, DifferenceType, S>::
-  operator->() {
+  operator->() const {
   return this->Load();
 }
 
 template<typename BaseType, typename DifferenceType, size_t S>
 inline BaseType& AtomicPointer<BaseType, DifferenceType, S>::
-  operator*() {
+  operator*() const {
   return *(this->Load());
 }
 

base_cpp/include/embb/base/internal/mutex-inl.h

@@ -28,6 +28,7 @@
 #define EMBB_BASE_INTERNAL_MUTEX_INL_H_
 
 #include <cassert>
+#include <algorithm>
 
 namespace embb {
 namespace base {
@@ -95,8 +96,8 @@ void UniqueLock<Mutex>::Unlock() {
 
 template<typename Mutex>
 void UniqueLock<Mutex>::Swap(UniqueLock<Mutex>& other) {
-  locked_ = other.locked_;
-  mutex_ = other.Release();
+  std::swap(mutex_, other.mutex_);
+  std::swap(locked_, other.locked_);
 }
 
 template<typename Mutex>

base_cpp/include/embb/base/mutex.h

@@ -439,11 +439,11 @@ class UniqueLock {
   void Unlock();
 
   /**
-   * Transfers ownership of a mutex to this lock.
+   * Exchanges ownership of the wrapped mutex with another lock.
    */
   void Swap(
     UniqueLock<Mutex>& other
-    /**< [IN/OUT] Lock from which ownership shall be transferred */
+    /**< [IN/OUT] The lock to exchange ownership with */
     );
 
   /**

base_cpp/test/mutex_test.cc

@@ -191,13 +191,21 @@ void MutexTest::TestUniqueLock() {
   }
 
   { // Test lock swapping
-    UniqueLock<> lock1;
-    UniqueLock<> lock2(mutex_);
-    PT_EXPECT_EQ(lock1.OwnsLock(), false);
-    PT_EXPECT_EQ(lock2.OwnsLock(), true);
-    lock1.Swap(lock2);
+    UniqueLock<> lock1(mutex_);
     PT_EXPECT_EQ(lock1.OwnsLock(), true);
+
+    {
+      UniqueLock<> lock2;
       PT_EXPECT_EQ(lock2.OwnsLock(), false);
+
+      lock1.Swap(lock2);
+      PT_EXPECT_EQ(lock1.OwnsLock(), false);
+      PT_EXPECT_EQ(lock2.OwnsLock(), true);
+    }
+
+    // At this point, "lock2" was destroyed and "mutex_" must be unlocked.
+    UniqueLock<> lock3(mutex_, embb::base::try_lock);
+    PT_EXPECT_EQ(lock3.OwnsLock(), true);
   }
 }
 

containers_cpp/include/embb/containers/internal/hazard_pointer-inl.h

@@ -30,386 +30,360 @@
 namespace embb {
 namespace containers {
 namespace internal {
-template< typename ElementT >
-FixedSizeList<ElementT>::FixedSizeList(size_t max_size) :
-  max_size(max_size),
-  size(0) {
-  elementsArray = static_cast<ElementT*>(
-    embb::base::Allocation::Allocate(sizeof(ElementT) *
-    max_size));
-}
-
-template< typename ElementT >
-inline size_t FixedSizeList<ElementT>::GetSize() const {
-  return size;
-}
-
-template< typename ElementT >
-inline size_t FixedSizeList<ElementT>::GetMaxSize() const {
-  return max_size;
-}
-
-template< typename ElementT >
-inline void FixedSizeList<ElementT>::clear() {
-  size = 0;
-}
-
-template< typename ElementT >
-typename FixedSizeList<ElementT>::iterator
-FixedSizeList<ElementT>::begin() const {
-  return &elementsArray[0];
-}
-
-template< typename ElementT >
-typename FixedSizeList<ElementT>::iterator
-FixedSizeList<ElementT>::end() const {
-  return &elementsArray[size];
-}
-
-template< typename ElementT >
-FixedSizeList< ElementT > &
-FixedSizeList<ElementT>::operator= (const FixedSizeList & other) {
-  size = 0;
-
-  if (max_size < other.size) {
-    EMBB_THROW(embb::base::ErrorException, "Copy target to small");
-  }
-
-  for (const_iterator it = other.begin(); it != other.end(); ++it) {
-    PushBack(*it);
-  }
-  return *this;
-}
-
-template< typename ElementT >
-bool FixedSizeList<ElementT>::PushBack(ElementT const el) {
-  if (size + 1 > max_size) {
-    return false;
-  }
-  elementsArray[size] = el;
-  size++;
-  return true;
-}
-
-template< typename ElementT >
-FixedSizeList<ElementT>::~FixedSizeList() {
-  embb::base::Allocation::Free(elementsArray);
-}
-
-template< typename GuardType >
-bool HazardPointerThreadEntry<GuardType>::IsActive() {
-  return is_active;
-}
-
-template< typename GuardType >
-bool HazardPointerThreadEntry<GuardType>::TryReserve() {
-  bool expected = false;
-  return is_active.CompareAndSwap(expected, true);
-}
-
-template< typename GuardType >
-void HazardPointerThreadEntry<GuardType>::Deactivate() {
-  is_active = false;
-}
-
-template< typename GuardType >
-size_t HazardPointerThreadEntry<GuardType>::GetRetiredCounter() {
-  return retired_list.GetSize();
-}
-
-template< typename GuardType >
-FixedSizeList< GuardType >& HazardPointerThreadEntry<GuardType>::
-GetRetired() {
-  return retired_list;
-}
-
-template< typename GuardType >
-FixedSizeList< GuardType >& HazardPointerThreadEntry<GuardType>::
-GetRetiredTemp() {
-  return retired_list_temp;
-}
-
-template< typename GuardType >
-FixedSizeList< GuardType >& HazardPointerThreadEntry<GuardType>::
-GetHazardTemp() {
-  return hazard_pointer_list_temp;
-}
-
-template< typename GuardType >
-void HazardPointerThreadEntry<GuardType>::
-SetRetired(internal::FixedSizeList< GuardType > const & retired_list) {
-  this->retired_list = retired_list;
-}
-
-template< typename GuardType >
-HazardPointerThreadEntry<GuardType>::
-HazardPointerThreadEntry(GuardType undefined_guard, int guards_per_thread,
-  size_t max_size_retired_list) :
-#ifdef EMBB_DEBUG
-  who_is_scanning(-1),
+// Visual Studio is complaining, that the return in the last line of this
+// function is not reachable. This is true, as long as exceptions are enabled.
+// Otherwise, the exception becomes an assertion and with disabling assertions,
+// the code becomes reachable. So, disabling this warning.
+#ifdef EMBB_PLATFORM_COMPILER_MSVC
+#pragma warning(push)
+#pragma warning(disable:4702)
 #endif
-  undefined_guard(undefined_guard),
-  guards_per_thread(guards_per_thread),
-  max_size_retired_list(max_size_retired_list),
-  // initially, each potential thread is active... if that is not the case
-  // another thread could call "HelpScan", and block this thread in making
-  // progress.
-  // Still, threads can be leave the hazard pointer processing (deactivation),
-  // but this can only be done once, i.e., this is not revertable...
-  is_active(1),
-  retired_list(max_size_retired_list),
-  retired_list_temp(max_size_retired_list),
-  hazard_pointer_list_temp(embb::base::Thread::GetThreadsMaxCount() *
-    guards_per_thread) {
-  // Initialize guarded pointer list
-  guarded_pointers = static_cast<embb::base::Atomic<GuardType>*>
-    (embb::base::Allocation::Allocate(
-    sizeof(embb::base::Atomic<GuardType>)*guards_per_thread));
-
-  for (int i = 0; i != guards_per_thread; ++i) {
-    new (static_cast<void*>(&guarded_pointers[i]))
-      embb::base::Atomic<GuardType>(undefined_guard);
-  }
-}
-
-template< typename GuardType >
-HazardPointerThreadEntry<GuardType>::~HazardPointerThreadEntry() {
-  for (int i = 0; i != guards_per_thread; ++i) {
-    guarded_pointers[i].~Atomic();
-  }
-
-  embb::base::Allocation::Free(guarded_pointers);
-}
-
-template< typename GuardType >
-GuardType HazardPointerThreadEntry<GuardType>::GetGuard(int pos) const {
-  return guarded_pointers[pos];
-}
-
-template< typename GuardType >
-void HazardPointerThreadEntry<GuardType>::AddRetired(GuardType pointerToGuard) {
-  retired_list.PushBack(pointerToGuard);
-}
-
-template< typename GuardType >
-void HazardPointerThreadEntry<GuardType>::
-GuardPointer(int guardNumber, GuardType pointerToGuard) {
-  guarded_pointers[guardNumber] = pointerToGuard;
-}
-
-template< typename GuardType >
-void HazardPointerThreadEntry<GuardType>::SetActive(bool active) {
-  is_active = active;
-}
-
-template< typename GuardType >
-unsigned int HazardPointer< GuardType >::GetCurrentThreadIndex() {
-  unsigned int thread_index;
-  int return_val = embb_internal_thread_index(&thread_index);
-
-  if (return_val != EMBB_SUCCESS)
-    EMBB_THROW(embb::base::ErrorException, "Could not get thread id!");
-
-  return thread_index;
-}
-template< typename GuardType >
-bool HazardPointer< GuardType >::IsThresholdExceeded() {
-  double retiredCounterLocThread =
-    static_cast<double>(GetHazardPointerElementForCurrentThread().
-    GetRetiredCounter());
-
-  return (retiredCounterLocThread >=
-    RETIRE_THRESHOLD *
-    static_cast<double>(active_hazard_pointer)*
-    static_cast<double>(guards_per_thread));
-}
-
-template< typename GuardType >
-size_t HazardPointer< GuardType >::GetActiveHazardPointers() {
-  return active_hazard_pointer;
-}
-template< typename GuardType >
-typename HazardPointer< GuardType >::HazardPointerThreadEntry_t &
-HazardPointer< GuardType >::GetHazardPointerElementForCurrentThread() {
-  // For each thread, there is a slot in the hazard pointer array.
-  // Initially, the active flag of a hazard pointer entry is false.
-  // Only the respective thread changes the flag from true to false.
-  // This means that the current thread tells that he is about to
-  // stop operating, and the others are responsible for his retired
-  // list.
-
-  return hazard_pointer_thread_entry_array[GetCurrentThreadIndex()];
-}
-
-template< typename GuardType >
-void HazardPointer< GuardType >::HelpScan() {
-  // This is a little bit different than in the paper. In the paper,
-  // the retired nodes from other threads are added to our retired list.
-  // To be able to give a bound on memory consumption, we execute scan
-  // for those threads, without moving elements. The effect shall be
-  // the same.
-
-  for (size_t i = 0; i != hazard_pointers; ++i) {
-    // Try to find non active lists...
-    if (!hazard_pointer_thread_entry_array[i].IsActive() &&
-      hazard_pointer_thread_entry_array[i].TryReserve()) {
-      // Here: grab retired things, first check if there are any...
-      if (hazard_pointer_thread_entry_array[i].GetRetiredCounter() > 0) {
-        Scan(&hazard_pointer_thread_entry_array[i]);
-      }
-
-      // We are done, mark it as deactivated again
-      hazard_pointer_thread_entry_array[i].Deactivate();
-    }
-  }
-}
-
-template< typename GuardType >
-void HazardPointer< GuardType >::
-Scan(HazardPointerThreadEntry_t* currentHazardPointerEntry) {
-#ifdef EMBB_DEBUG
-  // scan should only be executed by one thread at a time, otherwise we have
-  // a bug... this assertions checks that
+  template< typename GuardType >
+  unsigned int HazardPointer< GuardType >::GetObjectLocalThreadIndex() {
+    // first, get the EMBB native thread id.
+    unsigned int embb_thread_index;
+
+    int return_val = embb_internal_thread_index(&embb_thread_index);
+
+    if (return_val != EMBB_SUCCESS) {
+      EMBB_THROW(embb::base::ErrorException, "Could not get thread id");
+    }
+
+    // iterate over the mappings array
+    for (unsigned int i = 0; i != max_accessors_count_; ++i) {
+      // end of mappings? then we need to write our id
+      if (thread_id_mapping_[i] == -1) {
+        // try to CAS the initial value with out thread id
         int expected = -1;
-  if (!currentHazardPointerEntry->GetScanningThread().CompareAndSwap(
-    expected, static_cast<int>(GetCurrentThreadIndex()))) {
-    assert(false);
+        if (thread_id_mapping_[i].CompareAndSwap(expected,
+          static_cast<int>(embb_thread_index))) {
+          //successful, return our mapping
+          return i;
+        }
+      }
+
+      if (thread_id_mapping_[i] == static_cast<int>(embb_thread_index)) {
+        // found our mapping!
+        return i;
+      }
+    }
+
+    // when we reach this point, we have too many accessors
+    // (no mapping possible)
+    EMBB_THROW(embb::base::ErrorException, "Too many accessors");
+
+    return 0;
   }
+#ifdef EMBB_PLATFORM_COMPILER_MSVC
+#pragma warning(pop)
 #endif
-  // In this function, we compute the intersection between local retired
-  // pointers and all hazard pointers. This intersection cannot be deleted and
-  // forms the new local retired pointers list.
-  // It is assumed that the union of all retired pointers contains no two
-  // pointers with the same value. However, the union of all hazard guards
-  // might.
-
-  // Here, we store the temporary hazard pointers. We have to store them,
-  // as iterating multiple time over them might be expensive, as this
-  // atomic array is shared between threads.
-  currentHazardPointerEntry->GetHazardTemp().clear();
-
-  // Get all active hazard pointers!
-  for (unsigned int i = 0; i != hazard_pointers; ++i) {
-    // Only consider guards of active threads
-    if (hazard_pointer_thread_entry_array[i].IsActive()) {
-      // For each guard in an hazard pointer entry
-      for (int pos = 0; pos != guards_per_thread; ++pos) {
-        GuardType guard = hazard_pointer_thread_entry_array[i].GetGuard(pos);
-
-        // UndefinedGuard means not guarded
-        if (guard == undefined_guard)
-          continue;
-
-        currentHazardPointerEntry->GetHazardTemp().PushBack(guard);
-      }
-    }
-  }
-
-  currentHazardPointerEntry->GetRetiredTemp().clear();
-
-  // Sort them, we will do a binary search on each entry from the retired list
-  std::sort(
-    currentHazardPointerEntry->GetHazardTemp().begin(),
-    currentHazardPointerEntry->GetHazardTemp().end());
-
-  for (
-    EMBB_CONTAINERS_CPP_DEPENDANT_TYPENAME FixedSizeList< GuardType >::iterator
-      it = currentHazardPointerEntry->GetRetired().begin();
-  it != currentHazardPointerEntry->GetRetired().end(); ++it) {
-    if (false == ::std::binary_search(
-      currentHazardPointerEntry->GetHazardTemp().begin(),
-      currentHazardPointerEntry->GetHazardTemp().end(), *it)) {
-      this->free_guard_callback(*it);
+
+  template< typename GuardType >
+  void HazardPointer< GuardType >::RemoveGuard(int guard_position) {
+    const unsigned int my_thread_id = GetObjectLocalThreadIndex();
+
+    // check invariants...
+    assert(guard_position < max_guards_per_thread_);
+    assert(my_thread_id < max_accessors_count_);
+
+    // set guard
+    guards_[guard_position*max_accessors_count_ + my_thread_id] =
+      undefined_guard_;
+  }
+
+  template< typename GuardType >
+  HazardPointer< GuardType >::HazardPointer(
+    embb::base::Function<void, GuardType> freeGuardCallback,
+    GuardType undefined_guard, int guardsPerThread, int accessors) :
+    max_accessors_count_(accessors < 0 ?
+    embb::base::Thread::GetThreadsMaxCount() : accessors),
+    undefined_guard_(undefined_guard),
+    max_guards_per_thread_(guardsPerThread),
+    release_object_callback_(freeGuardCallback),
+    thread_id_mapping_(static_cast<embb::base::Atomic<int>*>(
+      embb::base::Allocation::Allocate(sizeof(embb::base::Atomic<int>)
+      *max_accessors_count_))),
+    guards_(static_cast<embb::base::Atomic< GuardType >*>
+      (embb::base::Allocation::Allocate(
+      sizeof(embb::base::Atomic< GuardType >) * max_guards_per_thread_ *
+      max_accessors_count_))),
+    thread_local_retired_lists_temp_(static_cast<GuardType*>
+      (embb::base::Allocation::Allocate(
+      sizeof(GuardType) * max_guards_per_thread_ * max_accessors_count_ *
+      max_accessors_count_
+      ))),
+    thread_local_retired_lists_(static_cast<GuardType*>
+      (embb::base::Allocation::Allocate(
+      sizeof(GuardType) * max_guards_per_thread_ * max_accessors_count_ *
+      max_accessors_count_
+      ))) {
+    const unsigned int count_guards =
+      max_guards_per_thread_ * max_accessors_count_;
+
+    const unsigned int count_ret_elements =
+      count_guards * max_accessors_count_;
+
+    for (unsigned int i = 0; i != max_accessors_count_; ++i) {
+      //in-place new for each cell
+      new (&thread_id_mapping_[i]) embb::base::Atomic < int >(-1);
+    }
+
+    for (unsigned int i = 0; i != count_guards; ++i) {
+      //in-place new for each cell
+      new (&guards_[i]) embb::base::Atomic < GuardType >(undefined_guard);
+    }
+
+    for (unsigned int i = 0; i != count_ret_elements; ++i) {
+      //in-place new for each cell
+      new (&thread_local_retired_lists_temp_[i]) GuardType(undefined_guard);
+    }
+
+    for (unsigned int i = 0; i != count_ret_elements; ++i) {
+      //in-place new for each cell
+      new (&thread_local_retired_lists_[i]) GuardType(undefined_guard);
+    }
+  }
+
+  template< typename GuardType >
+  HazardPointer< GuardType >::~HazardPointer() {
+    const unsigned int count_guards =
+      max_guards_per_thread_ * max_accessors_count_;
+
+    const unsigned int count_ret_elements =
+      count_guards * max_accessors_count_;
+
+    // Release references from all retired lists. Note that for this to work,
+    // the data structure using hazard pointer has still to be active... So
+    // first, the hazard pointer class shall be destructed, then the memory
+    // management class (e.g. some pool). Otherwise, the hazard pointer class
+    // would try to return memory to an already destructed memory manager.
+    for (unsigned int i = 0; i != count_ret_elements; ++i) {
+      GuardType pointerToFree =
+          thread_local_retired_lists_[i];
+      if (pointerToFree == undefined_guard_) {
+        break;
+      }
+      release_object_callback_(pointerToFree);
+    }
+
+    for (unsigned int i = 0; i != max_accessors_count_; ++i) {
+      thread_id_mapping_[i].~Atomic();
+    }
+
+    embb::base::Allocation::Free(thread_id_mapping_);
+
+    for (unsigned int i = 0; i != count_guards; ++i) {
+      guards_[i].~Atomic();
+    }
+
+    embb::base::Allocation::Free(guards_);
+
+    for (unsigned int i = 0; i != count_ret_elements; ++i) {
+      thread_local_retired_lists_temp_[i].~GuardType();
+    }
+
+    embb::base::Allocation::Free(thread_local_retired_lists_temp_);
+
+    for (unsigned int i = 0; i != count_ret_elements; ++i) {
+      thread_local_retired_lists_[i].~GuardType();
+    }
+
+    embb::base::Allocation::Free(thread_local_retired_lists_);
+  }
+
+  template< typename GuardType >
+  void HazardPointer< GuardType >::Guard(int guardPosition,
+    GuardType guardedElement) {
+    const unsigned int my_thread_id = GetObjectLocalThreadIndex();
+
+    // check invariants...
+    assert(guardPosition < max_guards_per_thread_);
+    assert(my_thread_id < max_accessors_count_);
+
+    // set guard
+    guards_[guardPosition*max_accessors_count_ + my_thread_id] = guardedElement;
+  }
+
+  template< typename GuardType >
+  size_t HazardPointer< GuardType >::ComputeMaximumRetiredObjectCount(
+  size_t guardsPerThread, int accessors) {
+    unsigned int accessorCount = (accessors == -1 ?
+      embb::base::Thread::GetThreadsMaxCount() :
+      accessors);
+
+    return static_cast<size_t>(
+        guardsPerThread * accessorCount * accessorCount);
+  }
+
+  /**
+   * Remark: it might be faster to just swap pointers for temp retired list and
+   * retired list. However, with the current implementation (one array for all
+   * retired and retired temp lists, respectively) this is not possible. This is
+   * not changed until this copying accounts for a performance problem. The
+   * copying is not the bottleneck currently.
+   */
+  template< typename GuardType >
+  void HazardPointer< GuardType >::CopyRetiredList(GuardType* sourceList,
+    GuardType* targetList, unsigned int retiredListSize,
+    GuardType undefinedGuard) {
+    bool done = false;
+    for (unsigned int ii = 0; ii != retiredListSize; ++ii) {
+      if (!done) {
+        GuardType guardToCopy = sourceList[ii];
+
+        if (guardToCopy == undefinedGuard) {
+          done = true;
+
+          if (targetList[ii] == undefinedGuard) {
+            // end of target list
+            break;
+          }
+        }
+        targetList[ii] = guardToCopy;
       } else {
-      currentHazardPointerEntry->GetRetiredTemp().PushBack(*it);
+        // we copied the whole source list, remaining values in the target
+        // have to be zeroed.
+        if (targetList[ii] == undefinedGuard) {
+          // end of target list
+          break;
+        } else {
+          targetList[ii] = undefinedGuard;
+        }
+      }
     }
   }
-  currentHazardPointerEntry->SetRetired(
-    currentHazardPointerEntry->GetRetiredTemp());
 
-#ifdef EMBB_DEBUG
-  currentHazardPointerEntry->GetScanningThread().Store(-1);
-#endif
-}
-
-template< typename GuardType >
-size_t HazardPointer< GuardType >::GetRetiredListMaxSize() const {
-  return static_cast<size_t>(RETIRE_THRESHOLD *
-        static_cast<double>(embb::base::Thread::GetThreadsMaxCount()) *
-        static_cast<double>(guards_per_thread)) + 1;
-}
-
-template< typename GuardType >
-HazardPointer< GuardType >::HazardPointer(
-  embb::base::Function<void, GuardType> free_guard_callback,
-  GuardType undefined_guard, int guards_per_thread) :
-  undefined_guard(undefined_guard),
-  guards_per_thread(guards_per_thread),
-  //initially, all potential hazard pointers are active...
-  active_hazard_pointer(embb::base::Thread::GetThreadsMaxCount()),
-  free_guard_callback(free_guard_callback) {
-  hazard_pointers = embb::base::Thread::GetThreadsMaxCount();
-
-  hazard_pointer_thread_entry_array = static_cast<HazardPointerThreadEntry_t*>(
-    embb::base::Allocation::Allocate(sizeof(HazardPointerThreadEntry_t) *
-    hazard_pointers));
-
-  for (size_t i = 0; i != hazard_pointers; ++i) {
-    new (static_cast<void*>(&(hazard_pointer_thread_entry_array[i])))
-      HazardPointerThreadEntry_t(undefined_guard, guards_per_thread,
-      GetRetiredListMaxSize());
-  }
-}
-
-template< typename GuardType >
-HazardPointer< GuardType >::~HazardPointer() {
-  for (size_t i = 0; i != hazard_pointers; ++i) {
-    hazard_pointer_thread_entry_array[i].~HazardPointerThreadEntry_t();
-  }
-
-  embb::base::Allocation::Free(static_cast < void* >
-    (hazard_pointer_thread_entry_array));
-}
-
-template< typename GuardType >
-void HazardPointer< GuardType >::DeactivateCurrentThread() {
-  HazardPointerThreadEntry_t* current_thread_entry =
-    &hazard_pointer_thread_entry_array[GetCurrentThreadIndex()];
-
-  // Deactivating a non-active hazard pointer entry has no effect!
-  if (!current_thread_entry->IsActive()) {
+  template< typename GuardType >
+  void HazardPointer< GuardType >::UpdateRetiredList(GuardType* retired_list,
+    GuardType* updated_retired_list, unsigned int retired_list_size,
+    GuardType guarded_element, GuardType considered_hazard,
+    GuardType undefined_guard) {
+    // no hazard set here
+    if (considered_hazard == undefined_guard)
         return;
-  } else {
-    current_thread_entry->SetActive(false);
-    active_hazard_pointer--;
+
+    // if this hazard is currently in the union of
+    // threadLocalRetiredLists and pointerToRetire, but not yet in
+    // threadLocalRetiredListsTemp, add it to that list
+    bool contained_in_union = false;
+
+    // first iterate over our retired list
+    for (unsigned int i = 0; i != retired_list_size; ++i) {
+      // when reaching 0, we can stop iterating (end of the "list")
+      if (retired_list[i] == 0)
+          break;
+
+      // the hazard is contained in the retired list... it shall go
+      // into the temp list, if not already there
+      if (retired_list[i] == considered_hazard) {
+        contained_in_union = true;
+        break;
+      }
     }
-}
 
-template< typename GuardType >
-void HazardPointer< GuardType >::GuardPointer(int guardPosition,
-  GuardType guardedElement) {
-  GetHazardPointerElementForCurrentThread().GuardPointer(
-    guardPosition, guardedElement);
-}
+    // the union also contains pointerToRetire
+    if (!contained_in_union) {
+      contained_in_union = (considered_hazard == guarded_element);
+    }
 
-template< typename GuardType >
-void HazardPointer< GuardType >::EnqueuePointerForDeletion(
-  GuardType guardedElement) {
-  GetHazardPointerElementForCurrentThread().AddRetired(guardedElement);
-  if (IsThresholdExceeded()) {
-    HazardPointerThreadEntry_t* currentHazardPointerEntry =
-      &GetHazardPointerElementForCurrentThread();
+    // add the pointer to temp. retired list, if not already there
+    if (contained_in_union) {
+      for (unsigned int ii = 0; ii != retired_list_size; ++ii) {
+        // is it already there?
+        if (updated_retired_list[ii] == considered_hazard)
+            break;
+
+        // end of the list
+        if (updated_retired_list[ii] == undefined_guard) {
+          // add hazard
+          updated_retired_list[ii] = considered_hazard;
+
+          // we are done here...
+          break;
+        }
+      }
+    }
+  }
+
+  template< typename GuardType >
+  void HazardPointer< GuardType >::EnqueueForDeletion(GuardType toRetire) {
+    unsigned int my_thread_id = GetObjectLocalThreadIndex();
+
+    // check for invariant
+    assert(my_thread_id < max_accessors_count_);
+
+    const unsigned int retired_list_size = max_accessors_count_ *
+      max_guards_per_thread_;
+
+    const unsigned int count_guards = max_accessors_count_ *
+      max_guards_per_thread_;
 
-    Scan(currentHazardPointerEntry);
+    GuardType* retired_list =
+      &thread_local_retired_lists_[my_thread_id * retired_list_size];
 
-    // Help deactivated threads to clean their retired nodes.
-    HelpScan();
+    GuardType* retired_list_temp =
+      &thread_local_retired_lists_temp_[my_thread_id * retired_list_size];
+
+    // wipe my temp. retired list...
+    for (unsigned int i = 0; i < retired_list_size; ++i) {
+      // the list is filled always from left to right, so occurring the first
+      // undefinedGuard, the remaining ones are also undefinedGuard...
+      if (retired_list_temp[i] == undefined_guard_)
+          break;
+
+      retired_list_temp[i] = undefined_guard_;
+    }
+
+    // we test each hazard if it is in the union of retiredList and
+    // guardedElement. If it is, it goes into the new retired list...
+    for (unsigned int i = 0; i != count_guards; ++i) {
+      // consider each current active guard
+      GuardType considered_hazard = guards_[i].Load();
+      UpdateRetiredList(retired_list, retired_list_temp, retired_list_size,
+        toRetire, considered_hazard, undefined_guard_);
     }
-}
 
-template<typename GuardType>
-const double embb::containers::internal::HazardPointer<GuardType>::
-  RETIRE_THRESHOLD = 1.25f;
+    int retired_list_size_signed = static_cast<int>(retired_list_size);
+    assert(retired_list_size_signed >= 0);
+
+    // now we created a a new retired list... the elements that are "removed"
+    // from the old retired list can be safely deleted now...
+    for (int i = -1; i != retired_list_size_signed; ++i) {
+      // we iterate over the current retired list... -1 is used as dummy element
+      // in the iteration, to also iterate over the pointerToRetire, which is
+      // logically also part of the current retired list...
+
+      // end of the list, stop iterating
+      if (i >= 0 && retired_list[i] == undefined_guard_)
+          break;
+
+      GuardType to_check_if_in_new_list = undefined_guard_;
+
+      to_check_if_in_new_list = (i == -1 ? toRetire : retired_list[i]);
+
+      // still in the new retired list?
+      bool still_in_list = false;
+      for (unsigned int ii = 0; ii != retired_list_size; ++ii) {
+        // end of list
+        if (retired_list_temp[ii] == undefined_guard_)
+            break;
+
+        if (to_check_if_in_new_list == retired_list_temp[ii]) {
+          // still in list, cannot delete element!
+          still_in_list = true;
+          break;
+        }
+      }
+
+      if (!still_in_list) {
+        this->release_object_callback_(to_check_if_in_new_list);
+      }
+    }
+
+    // copy the updated retired list (temp) to the retired list...
+    CopyRetiredList(retired_list_temp, retired_list, retired_list_size,
+      undefined_guard_);
+  }
 } // namespace internal
 } // namespace containers
 } // namespace embb

containers_cpp/include/embb/containers/internal/hazard_pointer.h

@@ -40,487 +40,274 @@
 #define EMBB_CONTAINERS_CPP_DEPENDANT_TYPENAME typename
 #endif
 
+// forward declaration for white-box test, used in friend declaration of
+// HazardPointer class.
 namespace embb {
 namespace containers {
-namespace internal {
-/**
- * A list with fixed size, implemented as an array. Replaces std::vector that
- * was used in previous hazard pointer implementation.
- *
- * Provides iterators, so we can apply algorithms from the STL.
- *
- * \tparam ElementT Type of the elements contained in the list.
- */
-template< typename ElementT >
-class FixedSizeList {
- private:
-   /**
-    * Capacity of the list
-    */
-  size_t max_size;
-
-  /**
-   * Size of the list
-   */
-  size_t size;
-
-  /**
-   * Pointer to the array containing the list
-   */
-  ElementT* elementsArray;
-
-  /**
-   * Copy constructor not implemented. Would require dynamic memory allocation.
-   */
-  FixedSizeList(
-    const FixedSizeList &
-    /**< [IN] Other list */);
-
- public:
-  /**
-   * Definition of an iterator
-   */
-  typedef ElementT * iterator;
-
-  /**
-   * Definition of a const iterator
-   */
-  typedef const ElementT * const_iterator;
-
-  /**
-   * Constructor, initializes list with given capacity
-   */
-  FixedSizeList(
-    size_t max_size
-    /**< [IN] Capacity of the list */);
-
-  /**
-   * Gets the current size of the list
-   *
-   * \return Size of the list
-   */
-  inline size_t GetSize() const;
-
-  /**
-   * Gets the capacity of the list
-   *
-   * \return The capacity of the list
-   */
-  inline size_t GetMaxSize() const;
-
-  /**
-   * Removes all elements from the list without changing the capacity
-   */
-  inline void clear();
-
-  /**
-   * Iterator pointing to the first element
-   *
-   * \return Begin iterator
-   */
-  iterator begin() const;
-
-  /**
-   * Iterator pointing beyond the last element
-   *
-   * \return End iterator
-   */
-  iterator end() const;
-
-  /**
-   * Copies the elements of another list to this list. The capacity of
-   * this list has to be greater than or equal to the size of the other list.
-   */
-  FixedSizeList & operator=(
-    const FixedSizeList & other
-    /**< [IN] Other list */);
-
-  /**
-   * Appends an element to the end of the list
-   *
-   * \return \c false if the operation was not successful because the list is
-   *         full, otherwise \c true.
-   */
-  bool PushBack(
-    ElementT const el
-    /**< [IN] Element to append to the list */);
-
-  /**
-   * Destructs the list.
-   */
-  ~FixedSizeList();
-};
+namespace test {
+class HazardPointerTest2;
+}
+}
+}
 
+namespace embb {
+namespace containers {
+namespace internal {
 /**
- * Hazard pointer entry for a single thread. Holds the actual guards that
- * determine if the current thread is about to use the guarded pointer.
- * Guarded pointers are protected and not deleted.
+ * This class contains a hazard pointer implementation following publication:
  *
- * Moreover, the retired list for this thread is contained. It determines
- * the pointers that have been allocated from this thread, but are not used
- * anymore by this thread. However, another thread could have a guard on it,
- * so the pointer cannot be deleted immediately.
- *
- * For the scan operation, the intersection of the guarded pointers from all
- * threads and the retired list has to be computed. For this computation, we
- * need thread local temporary lists which are also contained here.
+ * Maged M. Michael. "Hazard pointers: Safe memory reclamation for lock-free
+ * objects." IEEE Transactions on Parallel and Distributed Systems, 15.6 (2004)
+ * : 491-504.
  *
- * \tparam GuardType The type of guard, usually a pointer.
+ * Hazard pointers are a wait-free memory reclamation scheme for lock-free
+ * algorithms. Loosely speaking, they act as garbage collector. The release of
+ * objects contained within the memory, managed by the hazard pointer class, is
+ * intercepted and possibly delayed to avoid concurrency bugs.
+ *
+ * Before accessing an object, threads announce their intention to do so (i.e.
+ * the intention to dereference the respective pointer) to the hazard pointer
+ * class. This is called guarding. From now on, the hazard pointer class will
+ * prohibit the release or reuse of the guarded object. This is necessary, to
+ * assure that the object is not released or reused while it is accessed and to
+ * assure that it has not unnoticed changed (effectively avoiding the ABA
+ * problem).
+ *
+ * Note that after guarding an object, a consecutive check that the object (i.e.
+ * its pointer) is still valid is necessary; the object release could already
+ * have been started when guarding the object. Guarding is repeated, until this
+ * check eventually succeeds. Note that this "guard-and-check" loop makes the
+ * usage of the hazard pointer class lock-free, even though its implementation
+ * is wait-free.
+ *
+ * Internally, guarding is realized by providing each thread slots, where
+ * pointers can be placed that should not be freed (so called guards). When
+ * trying to release an object, it is checked if the object's pointer is
+ * guarded, and if so this object is not released, but instead put into a
+ * retired list for later release, when all guards for this object have been
+ * removed.
+ *
+ * In contrast to the original implementation, our implementation consumes only
+ * fixed-size memory. Note that the number of threads accessing the hazard
+ * pointer object accounts quadratic for the memory consumption: managed objects
+ * are provided from outside and the number of accessors accounts quadric for
+ * the minimum count of those objects.
+ *
+ * Also in contrast to the original implementation, we do not provide a HelpScan
+ * functionality, which gives threads the possibility, to not participate in the
+ * garbage collection anymore: other threads will help to clean-up the objects
+ * protected by the exiting thread. The reason is, that the only use-case would
+ * be a crashing thread, not participating anymore. However, as the thread has
+ * to signal its exit himself, this is not possible to realize anyways. In the
+ * end, it is still guaranteed that all memory is properly returned (in the
+ * destructor).
+ *
+ * Additionally, the original implementation holds a threshold, which determines
+ * when objects shall be freed. In this implementation, we free whenever it is
+ * possibly to do so, as we want to keep the memory footprint as low as
+ * possible. We also don't see a performance drop in the current algorithms that
+ * are using hazard pointers, when not using a threshold.
+ *
+ * \tparam GuardType the type of the guards. Usually the pointer type of some
+ *         object to protect.
  */
 template< typename GuardType >
-class HazardPointerThreadEntry {
-#ifdef EMBB_DEBUG
-
- public:
-    embb::base::Atomic<int>& GetScanningThread() {
-      return who_is_scanning;
-    }
-
- private:
-    embb::base::Atomic<int> who_is_scanning;
-#endif
-
- private:
-   /**
-    * Value of the undefined guard (means that no guard is set).
-    */
-  GuardType undefined_guard;
-
-  /**
-   * The number of guards per thread. Determines the size of the guard array.
-   */
-  int guards_per_thread;
-
-  /**
-   * The capacity of the retired list. It is determined by number of guards,
-   * retired threshold, and maximum number of threads.
-   */
-  size_t max_size_retired_list;
-
-  /**
-   * Set to true if the current thread is active. Is used for a thread to
-   * signal that it is leaving. If a thread has left, the other threads are
-   * responsible for cleaning up its retired list.
-   */
-  embb::base::Atomic< bool > is_active;
-
-  /**
-   * The guarded pointer of this thread, has size \c guard_per_thread.
-   */
-  embb::base::Atomic< GuardType >* guarded_pointers;
-
-  /**
-   * The retired list of this thread, contains pointer that shall be released
-   * when no thread holds a guard on it anymore.
-   */
-  FixedSizeList< GuardType > retired_list;
-
-  /**
-   * Temporary retired list, has same capacity as \c retired_list, It is used to
-   * compute the intersection of all guards and the \c retired list.
-   */
-  FixedSizeList< GuardType > retired_list_temp;
-
-  /**
-   * Temporary guards list. Used to compute the intersection of all guards and
-   * the \c retired_list.
-   */
-  FixedSizeList< GuardType > hazard_pointer_list_temp;
-
-  /**
-   * HazardPointerThreadEntry shall not be copied
-   */
-  HazardPointerThreadEntry(const HazardPointerThreadEntry&);
-
-  /**
-   * HazardPointerThreadEntry shall not be assigned
-   */
-  HazardPointerThreadEntry & operator= (const HazardPointerThreadEntry&);
-
+class HazardPointer  {
  public:
   /**
-   * Checks if current thread is active (with respect to participating in hazard
-   * pointer management)
-   *
-   * \return \c true if the current thread is active, otherwise \c false.
-   */
-  bool IsActive();
-
-  /**
-   * Tries to set the active flag to true (atomically). Used if the current
-   * thread is not active anymore as lock for another thread to help cleaning
-   * up hazard pointer.
+   * The user of the hazard pointer class has to provide the memory that is
+   * managed here. The user has to take into account, that releasing of memory
+   * might be delayed. He has therefore to provide more memory than he wants to
+   * guarantee at each point in time. More specific, on top of the guaranteed
+   * count of objects, he has to provide the additional count of objects that
+   * can be (worst-case) contained in the retired lists and therefore are not
+   * released yet. The size sum of all retired lists is guardsPerThread *
+   * accessorCount * accessorCount, which is computed using this function. So
+   * the result of function denotes to the user, how many objects he has to
+   * allocate additionally to the guaranteed count.
    *
-   * \return \c true if this thread was successful setting the active flag,
-   *         otherwise \c false.
-   */
-  bool TryReserve();
-
-  /**
-   * Deactivates current thread by atomically setting active flag to false.
+   * \waitfree
    */
-  void Deactivate();
+  static size_t ComputeMaximumRetiredObjectCount(
+    size_t guardsPerThread,
+      /**<[IN] the count of guards per thread*/
+      int accessors = -1
+      /**<[IN] Number of accessors. Determines, how many threads will access
+               the hazard pointer object. Default value -1 will allow the
+               maximum amount of threads as defined with
+               \c embb::base::Thread::GetThreadsMaxCount()*/
+      );
 
   /**
-   * Gets the count of current retired pointer for the current thread.
+   * Initializes the hazard pointer object
    *
-   * \return Count of current retired pointer
-   */
-  size_t GetRetiredCounter();
-
-  /**
-   * Gets the retired list.
+   * \notthreadsafe
    *
-   * \return Reference to \c retired_list
-   */
-  FixedSizeList< GuardType >& GetRetired();
-
-  /**
-   * Gets the temporary retired list.
+   * \memory We dynamically allocate the following:
    *
-   * \return Reference to \c retired_list_temp
-   */
-  FixedSizeList< GuardType >& GetRetiredTemp();
-
-  /**
-   * Gets the temporary hazard pointer list.
+   * (sizeof(Atomic<int>) * accessors) + (sizeof(Atomic<GuardType>) *
+   * guards_per_thread * accessors) + (2*sizeof(GuardType) *
+   * guards_per_thread * accessors^2)
    *
-   * \return Reference to \c hazard_pointer_list_temp
-   */
-  FixedSizeList< GuardType >& GetHazardTemp();
-
-  /**
-   * Sets the retired list.
+   * The last addend is the dominant one, as accessorCount accounts
+   * quadratically for it.
    */
-  void SetRetired(
-    embb::containers::internal::FixedSizeList< GuardType > const & retired_list
-    /**< [IN] Retired list */);
-
-  /**
-   * Constructor
-   */
-  HazardPointerThreadEntry(
+  HazardPointer(
+    embb::base::Function<void, GuardType> free_guard_callback,
+    /**<[IN] Callback to the function that shall be called when a retired
+             guard can be deleted */
     GuardType undefined_guard,
-    /**< [IN] Value of the undefined guard (e.g. NULL) */
+    /**<[IN] The guard value denoting "not guarded"*/
     int guards_per_thread,
-    /**< [IN] Number of guards per thread */
-    size_t max_size_retired_list
-    /**< [IN] The capacity of the retired list(s) */);
+    /**<[IN] Number of guards per thread*/
+    int accessors = -1
+    /**<[IN] Number of accessors. Determines, how many threads will access
+              this hazard pointer object. Default value -1 will allow the
+              maximum amount of threads as defined with
+              \c embb::base::Thread::GetThreadsMaxCount()*/
+    );
 
   /**
-   * Destructor
+   * Deallocates internal data structures. Additionally releases all objects
+   * currently held in the retired lists, using the release functor passed in
+   * the constructor.
    *
-   * Deallocate lists
-   */
-  ~HazardPointerThreadEntry();
-
-  /**
-   * Gets the guard at the specified position.
-   * Positions are numbered, beginning with 0.
-   */
-  GuardType GetGuard(
-    int pos
-    /**< [IN] Position of the guard */) const;
-
-  /**
-   * Adds pointer to the retired list
-   */
-  void AddRetired(
-    GuardType pointerToGuard
-    /**< [IN] Guard to retire */);
-
-  /**
-   * Guards pointer
+   * \notthreadsafe
    */
-  void GuardPointer(
-    int guardNumber,
-    /**< [IN] Position of guard */
-    GuardType pointerToGuard
-    /**<[IN] Pointer to guard */);
+  ~HazardPointer();
 
   /**
-   * Sets the current thread active, i.e., announce that the thread
-   * participates in managing hazard pointer.
-   */
-  void SetActive(
-    bool active
-    /**<[IN] \c true for active, \c false for inactive */);
-};
-
-/**
- * HazardPointer implementation as presented in:
- *
- * Maged M. Michael. "Hazard pointers: Safe memory reclamation for lock-free
- * objects." IEEE Transactions on Parallel and Distributed Systems, 15.6 (2004)
- * : 491-504.
+   * Guards \c to_guard. If the guarded_element is passed to \c EnqueueForDeletion
+   * it is prevented from release from now on. The user must have a check, that
+   * EnqueueForDeletion has not been called on to_guard, before the guarding took
+   * effect.
    *
- * In contrast to the original implementation, our implementation only uses
- * fixed-size memory. There is a safe upper limit, hazard pointer are guaranteed
- * to not consume more memory. Memory is allocated solely at initialization.
- *
- * Hazard pointers solve the ABA problem for lock-free algorithms. Before
- * accessing a pointer, threads announce that they want to access this pointer
- * and then check if the pointer is still valid. This announcement is done by
- * placing a guard. It is guaranteed that the pointer is not reused until all
- * threads remove their guards to this pointer. Objects, these pointers are
- * pointing to, can therefore not be deleted directly. Instead, these pointers
- * are put into a list for later deletion (retired list). Regularly, this list
- * is processed to check which pointers can be deleted. If a pointer can be
- * deleted, a callback function provided by the user is called. The user can
- * then, e.g., free the respective object, so that the pointer can be safely
- * reused.
- */
-template< typename GuardType >
-class HazardPointer  {
- private:
-   /**
-    * Concrete hazard pointer entry type
+   * \waitfree
    */
-  typedef HazardPointerThreadEntry < GuardType >
-    HazardPointerThreadEntry_t;
+  void Guard(
+    int guard_position,
+    /**<[IN] position to place guard*/
+    GuardType to_guard
+    /**<[IN] element to guard*/
+    );
 
   /**
-   * The guard value denoting "not guarding"
+   * Enqueue guarded element for deletion. If not guarded, it is deleted
+   * immediately. If it is guarded, it is added to a thread local retired list,
+   * and deleted in a subsequent call to \c EnqueueForDeletion, when no guard is
+   * placed on it anymore.
    */
-  GuardType undefined_guard;
+  void EnqueueForDeletion(
+    GuardType guarded_element
+    /**<[IN] element to logically delete*/
+    );
 
   /**
-   * The capacity of the retired list (safe upper bound for retired list size)
+   * Explicitly remove guard from thread local slot.
+   *
+   * \waitfree
    */
-  int retired_list_max_size;
+  void RemoveGuard(int guard_position);
 
+ private:
   /**
-   * Guards that can be set per thread
+   * HazardPointerTest2 is a white-box test, needing access to private members
+   * of this class. So declaring it as friend.
    */
-  int guards_per_thread;
+  friend class embb::containers::test::HazardPointerTest2;
 
   /**
-   * Array of HazardPointerElements. Each thread is assigned to one.
+   * This number determines the amount of maximal accessors (threads) that
+   * will access this hazard pointer instance. Note that a thread once
+   * accessing this object will be permanently count as accessor, even if not
+   * participating anymore. If too many threads access this object, an
+   * exception is thrown.
    */
-  HazardPointerThreadEntry_t* hazard_pointer_thread_entry_array;
+  unsigned int max_accessors_count_;
 
   /**
-   * The threshold, determines at which size of the retired list pointers
-   * are tried to be deleted.
+   * The guard value denoting "not guarded"
    */
-  static const double RETIRE_THRESHOLD;
+  GuardType undefined_guard_;
 
   /**
-   * Each thread is assigned a thread index (starting with 0).
-   * Get the index of the current thread.
+   * The maximal count of guards that can be set per thread.
    */
-  static unsigned int GetCurrentThreadIndex();
+  int max_guards_per_thread_;
 
   /**
-   * The number of hazard pointers currently active.
+   * The functor that is called to release an object. This is called by this
+   * class, when it is safe to do so, i.e., no thread accesses this object
+   * anymore.
    */
-  size_t active_hazard_pointer;
+  embb::base::Function<void, GuardType> release_object_callback_;
 
   /**
-   * Count of all hazard pointers.
+   * Mapping from EMBB thread id to hazard pointer thread ids. Hazard pointer
+   * thread ids are in range [0;accesor_count-1]. The position of a EMBB thread
+   * id in that array determines the respective hazard pointer thread id.
    */
-  size_t hazard_pointers;
+  embb::base::Atomic<int>* thread_id_mapping_;
 
   /**
-   * The callback that is triggered when a retired guard can be
-   * freed. Usually, the user will call a free here.
+   * The hazard pointer guards, represented as array. Each thread has a fixed
+   * set of slots (guardsPerThread) within this array.
    */
-  embb::base::Function<void, GuardType> free_guard_callback;
+  embb::base::Atomic<GuardType>* guards_;
 
   /**
-   * Checks if the current size of the retired list exceeds the threshold, so
-   * that each retired guard is checked for being not hazardous anymore.
-   *
-   * \return \c true is threshold is exceeded, otherwise \c false.
+   * \see threadLocalRetiredLists documentation
    */
-  bool IsThresholdExceeded();
+  GuardType* thread_local_retired_lists_temp_;
 
   /**
-   * Gets the number of hazard pointe, currently active
-   *
-   * \return Number of active hazard pointers
+   * A list of lists, represented as single array. Each thread maintains a list
+   * of retired pointers, that are objects that are logically released but not
+   * released because some thread placed a guard on it.
    */
-  size_t GetActiveHazardPointers();
+  GuardType* thread_local_retired_lists_;
 
   /**
-   * Gets the hazard pointer entry for the current thread
+   * Each thread is assigned a thread index (starting with 0). Get the index of
+   * the current thread. Note that this is not the global index, but an hazard
+   * pointer class internal one. The user is free to define less accessors than
+   * the amount of default threads. This is useful, as the number of accessors
+   * accounts quadratic for the memory consumption, so the user should have the
+   * possibility to avoid memory wastage when only having a small, fixed size,
+   * number of accessors.
    *
-   * \return Hazard pointer entry for current thread
-   */
-  HazardPointerThreadEntry_t&
-    GetHazardPointerElementForCurrentThread();
-
-  /**
-   * Threads might leave from participating in hazard pointer management.
-   * This method helps all those threads processing their retired list.
-   */
-  void HelpScan();
-
-  /**
-   * Checks the retired list of a hazard pointer entry for elements of the
-   * retired list that can be freed, and executes the delete callback for those
-   * elements.
+   * @return current (hazard pointer object local) thread index
    */
-  void Scan(
-    HazardPointerThreadEntry_t* currentHazardPointerEntry
-    /**<[IN] Hazard pointer entry that should be checked for elements that
-             can be deleted*/);
+  unsigned int GetObjectLocalThreadIndex();
 
- public:
   /**
-   * Gets the capacity of one retired list
-   *
-   * \waitfree
+   * Copy retired list \c sourceList to retired list \c targetList
    */
-  size_t GetRetiredListMaxSize() const;
+  static void CopyRetiredList(
+    GuardType* source_list,
+    /**<[IN] the source retired list*/
+    GuardType* target_list,
+    /**<[IN] the target retired list*/
+    unsigned int single_retired_list_size,
+    /**<[IN] the size of a thread local retired list*/
+    GuardType undefined_guard
+    /**<[IN] the undefined guard (usually the NULL pointer)*/
+    );
 
-  /**
-   * Initializes hazard pointer
-   *
-   * \notthreadsafe
-   *
-   * \memory
-   *  - Let \c t be the number of maximal threads determined by EMBB
-   *  - Let \c g be the number of guards per thread
-   *  - Let \c x be 1.25*t*g + 1
-   *
-   * We dynamically allocate \c x*(3*t+1) elements of size \c sizeof(void*).
-   */
-  HazardPointer(
-    embb::base::Function<void, GuardType> free_guard_callback,
-    /**<[IN] Callback to the function that shall be called when a retired
-             guard can be deleted */
-    GuardType undefined_guard,
-    /**<[IN] The guard value denoting "not guarded"*/
-    int guards_per_thread
-    /**<[IN] Number of guards per thread*/);
-
-  /**
-   * Deallocates lists for hazard pointer management. Note that no objects
-   * currently in the retired lists are deleted. This is the responsibility
-   * of the user. Usually, HazardPointer manages pointers of an object pool.
-   * After destructing HazardPointer, the object pool is deleted, so that
-   * everything is properly cleaned up.
-   */
-  ~HazardPointer();
-
-  /**
-   * Announces that the current thread stops participating in hazard pointer
-   * management. The other threads now take care of his retired list.
-   *
-   * \waitfree
-   */
-  void DeactivateCurrentThread();
-
-  /**
-   * Guards \c guardedElement with the guard at position \c guardPosition
-   */
-  void GuardPointer(int guardPosition, GuardType guardedElement);
-  /**
-   * Enqueue a pointer for deletion. It is added to the retired list and
-   * deleted when no thread accesses it anymore.
-   */
-  void EnqueuePointerForDeletion(GuardType guardedElement);
+  static void UpdateRetiredList(
+    GuardType* retired_list,
+    /**<[IN] the old retired list*/
+    GuardType* updated_retired_list,
+    /**<[IN] the updated retired list*/
+    unsigned int retired_list_size,
+    /**<[IN] the size of a thread local retired list*/
+    GuardType to_retire,
+    /**<[IN] the element to retire*/
+    GuardType considered_hazard,
+    /**<[IN] the currently considered hazard*/
+    GuardType undefined_guard
+    /**<[IN] the undefined guard (usually the NULL pointer)*/
+    );
 };
 } // namespace internal
 } // namespace containers

containers_cpp/include/embb/containers/internal/lock_free_mpmc_queue-inl.h

@@ -77,7 +77,12 @@ LockFreeMPMCQueue<Type, ValuePool>::~LockFreeMPMCQueue() {
 
 template< typename Type, typename ValuePool >
 LockFreeMPMCQueue<Type, ValuePool>::LockFreeMPMCQueue(size_t capacity) :
-capacity(capacity),
+  capacity(capacity),
+  // Object pool, size with respect to the maximum number of retired nodes not
+  // eligible for reuse. +1 for dummy node.
+  objectPool(
+  MPMCQueueNodeHazardPointer_t::ComputeMaximumRetiredObjectCount(2) +
+  capacity + 1),
 // Disable "this is used in base member initializer" warning.
 // We explicitly want this.
 #ifdef EMBB_PLATFORM_COMPILER_MSVC
@@ -89,13 +94,7 @@ delete_pointer_callback(*this,
 #ifdef EMBB_PLATFORM_COMPILER_MSVC
 #pragma warning(pop)
 #endif
-  hazardPointer(delete_pointer_callback, NULL, 2),
-  // Object pool, size with respect to the maximum number of retired nodes not
-  // eligible for reuse. +1 for dummy node.
-  objectPool(
-  hazardPointer.GetRetiredListMaxSize()*
-  embb::base::Thread::GetThreadsMaxCount() +
-  capacity + 1) {
+  hazardPointer(delete_pointer_callback, NULL, 2) {
   // Allocate dummy node to reduce the number of special cases to consider.
   internal::LockFreeMPMCQueueNode<Type>* dummyNode = objectPool.Allocate();
   // Initially, head and tail point to the dummy node.
@@ -120,7 +119,7 @@ bool LockFreeMPMCQueue<Type, ValuePool>::TryEnqueue(Type const& element) {
   for (;;) {
     my_tail = tail;
 
-    hazardPointer.GuardPointer(0, my_tail);
+    hazardPointer.Guard(0, my_tail);
 
     // Check if pointer is still valid after guarding.
     if (my_tail != tail) {
@@ -163,12 +162,12 @@ bool LockFreeMPMCQueue<Type, ValuePool>::TryDequeue(Type & element) {
   Type data;
   for (;;) {
     my_head = head;
-    hazardPointer.GuardPointer(0, my_head);
+    hazardPointer.Guard(0, my_head);
     if (my_head != head) continue;
 
     my_tail = tail;
     my_next = my_head->GetNext();
-    hazardPointer.GuardPointer(1, my_next);
+    hazardPointer.Guard(1, my_next);
     if (head != my_head) continue;
 
     if (my_next == NULL)
@@ -187,7 +186,7 @@ bool LockFreeMPMCQueue<Type, ValuePool>::TryDequeue(Type & element) {
       break;
   }
 
-  hazardPointer.EnqueuePointerForDeletion(my_head);
+  hazardPointer.EnqueueForDeletion(my_head);
   element = data;
   return true;
 }

containers_cpp/include/embb/containers/internal/lock_free_stack-inl.h

@@ -81,13 +81,12 @@ capacity(capacity),
 #ifdef EMBB_PLATFORM_COMPILER_MSVC
 #pragma warning(pop)
 #endif
-  hazardPointer(delete_pointer_callback, NULL, 1),
   // Object pool, size with respect to the maximum number of retired nodes not
   // eligible for reuse:
   objectPool(
-  hazardPointer.GetRetiredListMaxSize()*
-  embb::base::Thread::GetThreadsMaxCount() +
-  capacity) {
+  StackNodeHazardPointer_t::ComputeMaximumRetiredObjectCount(1) +
+  capacity),
+  hazardPointer(delete_pointer_callback, NULL, 1) {
 }
 
 template< typename Type, typename ValuePool >
@@ -128,7 +127,7 @@ bool LockFreeStack< Type, ValuePool >::TryPop(Type & element) {
       return false;
 
     // Guard top_cached
-    hazardPointer.GuardPointer(0, top_cached);
+    hazardPointer.Guard(0, top_cached);
 
     // Check if top is still top. If this is the case, it has not been
     // retired yet (because before retiring that thing, the retiring thread
@@ -144,16 +143,16 @@ bool LockFreeStack< Type, ValuePool >::TryPop(Type & element) {
       break;
     } else {
       // We continue with the next and can unguard top_cached
-      hazardPointer.GuardPointer(0, NULL);
+      hazardPointer.Guard(0, NULL);
     }
   }
 
   Type data = top_cached->GetElement();
 
   // We don't need to read from this reference anymore, unguard it
-  hazardPointer.GuardPointer(0, NULL);
+  hazardPointer.Guard(0, NULL);
 
-  hazardPointer.EnqueuePointerForDeletion(top_cached);
+  hazardPointer.EnqueueForDeletion(top_cached);
 
   element = data;
   return true;

containers_cpp/include/embb/containers/internal/lock_free_tree_value_pool-inl.h

@@ -42,7 +42,7 @@ template<typename Type, Type Undefined, class PoolAllocator,
   class TreeAllocator >
 bool LockFreeTreeValuePool<Type, Undefined, PoolAllocator, TreeAllocator>::
 IsLeaf(int node) {
-  if (node >= size - 1 && node <= 2 * size - 1) {
+  if (node >= size_ - 1 && node <= 2 * size_ - 1) {
     return true;
   }
   return false;
@@ -52,7 +52,7 @@ template<typename Type, Type Undefined, class PoolAllocator,
   class TreeAllocator >
 bool LockFreeTreeValuePool<Type, Undefined, PoolAllocator, TreeAllocator>::
 IsValid(int node) {
-  return (node >= 0 && node <= 2 * size - 1);
+  return (node >= 0 && node <= 2 * size_ - 1);
 }
 
 template<typename Type, Type Undefined, class PoolAllocator,
@@ -77,14 +77,14 @@ template<typename T, T Undefined, class PoolAllocator, class TreeAllocator >
 int LockFreeTreeValuePool<T, Undefined, PoolAllocator, TreeAllocator>::
 NodeIndexToPoolIndex(int node) {
   assert(IsLeaf(node));
-  return(node - (size - 1));
+  return(node - (size_ - 1));
 }
 
 template<typename Type, Type Undefined, class PoolAllocator,
   class TreeAllocator >
 int LockFreeTreeValuePool<Type, Undefined, PoolAllocator, TreeAllocator>::
 PoolIndexToNodeIndex(int index) {
-  int node = index + (size - 1);
+  int node = index + (size_ - 1);
   assert(IsLeaf(node));
   return node;
 }
@@ -100,7 +100,7 @@ template<typename T, T Undefined, class PoolAllocator, class TreeAllocator >
 int LockFreeTreeValuePool<T, Undefined, PoolAllocator, TreeAllocator>::
 GetParentNode(int node) {
   int parent = (node - 1) / 2;
-  assert(parent >= 0 && parent < size - 1);
+  assert(parent >= 0 && parent < size_ - 1);
   return parent;
 }
 
@@ -112,11 +112,11 @@ allocate_rec(int node, Type& element) {
   if (IsLeaf(node)) {
     int pool_index = NodeIndexToPoolIndex(node);
 
-    Type expected = pool[pool_index];
+    Type expected = pool_[pool_index];
     if (expected == Undefined)
       return -1;
 
-    if (pool[pool_index].CompareAndSwap(expected, Undefined)) {
+    if (pool_[pool_index].CompareAndSwap(expected, Undefined)) {
       element = expected;
       return pool_index;
     }
@@ -131,11 +131,11 @@ allocate_rec(int node, Type& element) {
   // atomically decrement the value in the node if the result is greater than
   // or equal to zero. This cannot be done atomically.
   do {
-    current = tree[node];
+    current = tree_[node];
     desired = current - 1;
     if (desired < 0)
       return -1;
-  } while (!tree[node].CompareAndSwap(current, desired));
+  } while (!tree_[node].CompareAndSwap(current, desired));
 
   int leftResult = allocate_rec(GetLeftChildIndex(node), element);
   if (leftResult != -1) {
@@ -156,7 +156,7 @@ Fill(int node, int elementsToStore, int power2Value) {
   if (IsLeaf(node))
     return;
 
-  tree[node] = elementsToStore;
+  tree_[node] = elementsToStore;
 
   int postPower2Value = power2Value >> 1;
 
@@ -188,14 +188,14 @@ Free(Type element, int index) {
   assert(element != Undefined);
 
   // Put the element back
-  pool[index].Store(element);
+  pool_[index].Store(element);
 
-  assert(index >= 0 && index < size);
+  assert(index >= 0 && index < size_);
   int node = PoolIndexToNodeIndex(index);
 
   while (!IsRoot(node)) {
     node = GetParentNode(node);
-    tree[node].FetchAndAdd(1);
+    tree_[node].FetchAndAdd(1);
   }
 }
 
@@ -205,37 +205,76 @@ template< typename ForwardIterator >
 LockFreeTreeValuePool<Type, Undefined, PoolAllocator, TreeAllocator>::
 LockFreeTreeValuePool(ForwardIterator first, ForwardIterator last) {
   // Number of elements to store
-  real_size = static_cast<int>(::std::distance(first, last));
+  real_size_ = static_cast<int>(::std::distance(first, last));
 
   // Let k be smallest number so that real_size <= 2^k, size = 2^k
-  size = GetSmallestPowerByTwoValue(real_size);
+  size_ = GetSmallestPowerByTwoValue(real_size_);
 
   // Size of binary tree without the leaves
-  tree_size = size - 1;
+  tree_size_ = size_ - 1;
+
+  // make sure, signed values are not negative
+  assert(tree_size_ >= 0);
+  assert(real_size_ >= 0);
+
+  size_t tree_size_unsigned = static_cast<size_t>(tree_size_);
+  size_t real_size_unsigned = static_cast<size_t>(real_size_);
 
   // Pool stores elements of type T
-  pool = poolAllocator.allocate(static_cast<size_t>(real_size));
+  pool_ = pool_allocator_.allocate(real_size_unsigned);
+
+  // invoke inplace new for each pool element
+  for (size_t i = 0; i != real_size_unsigned; ++i) {
+    new (&pool_[i]) embb::base::Atomic<Type>();
+  }
 
   // Tree holds the counter of not allocated elements
-  tree = treeAllocator.allocate(static_cast<size_t>(tree_size));
+  tree_ = tree_allocator_.allocate(tree_size_unsigned);
+
+  // invoke inplace new for each tree element
+  for (size_t i = 0; i != tree_size_unsigned; ++i) {
+    new (&tree_[i]) embb::base::Atomic<int>();
+  }
 
   int i = 0;
 
   // Store the elements from the range
   for (ForwardIterator curIter(first); curIter != last; ++curIter) {
-    pool[i++] = *curIter;
+    pool_[i++] = *curIter;
   }
 
   // Initialize the binary tree without leaves (counters)
-  Fill(0, static_cast<int>(::std::distance(first, last)), size);
+  Fill(0, static_cast<int>(::std::distance(first, last)), size_);
 }
 
 template<typename Type, Type Undefined, class PoolAllocator,
   class TreeAllocator >
 LockFreeTreeValuePool<Type, Undefined, PoolAllocator, TreeAllocator>::
 ~LockFreeTreeValuePool() {
-  poolAllocator.deallocate(pool, static_cast<size_t>(real_size));
-  treeAllocator.deallocate(tree, static_cast<size_t>(tree_size));
+  size_t tree_size_unsigned = static_cast<size_t>(tree_size_);
+  size_t real_size_unsigned = static_cast<size_t>(real_size_);
+
+  // invoke destructor for each pool element
+  for (size_t i = 0; i != real_size_unsigned; ++i) {
+    pool_[i].~Atomic();
+  }
+
+  pool_allocator_.deallocate(pool_, real_size_unsigned);
+
+  // invoke destructor for each tree element
+  for (size_t i = 0; i != tree_size_unsigned; ++i) {
+    tree_[i].~Atomic();
+  }
+
+  tree_allocator_.deallocate(tree_, tree_size_unsigned);
+}
+
+template<typename Type, Type Undefined, class PoolAllocator,
+class TreeAllocator >
+size_t LockFreeTreeValuePool<Type, Undefined, PoolAllocator, TreeAllocator>::
+GetMinimumElementCountForGuaranteedCapacity(size_t capacity) {
+  // for this value pool, this is just capacity...
+  return capacity;
 }
 
 } // namespace containers

containers_cpp/include/embb/containers/internal/object_pool-inl.h

@@ -83,7 +83,8 @@ ReturningTrueIterator::operator!=(const self_type& rhs) {
 template<class Type, typename ValuePool, class ObjectAllocator>
 bool ObjectPool<Type, ValuePool, ObjectAllocator>::
 IsContained(const Type &obj) const {
-  if ((&obj < &objects[0]) || (&obj > &objects[capacity - 1])) {
+  if ((&obj < &objects_array_[0]) ||
+    (&obj > &objects_array_[value_pool_size_ - 1])) {
     return false;
   } else {
     return true;
@@ -94,17 +95,17 @@ template<class Type, typename ValuePool, class ObjectAllocator>
 int ObjectPool<Type, ValuePool, ObjectAllocator>::
 GetIndexOfObject(const Type &obj) const {
   assert(IsContained(obj));
-  return(static_cast<int>(&obj - &objects[0]));
+  return(static_cast<int>(&obj - &objects_array_[0]));
 }
 
 template<class Type, typename ValuePool, class ObjectAllocator>
 Type* ObjectPool<Type, ValuePool, ObjectAllocator>::AllocateRaw() {
   bool val;
-  int allocated_index = p.Allocate(val);
+  int allocated_index = value_pool_.Allocate(val);
   if (allocated_index == -1) {
     return NULL;
   } else {
-    Type* ret_pointer = &(objects[allocated_index]);
+    Type* ret_pointer = &(objects_array_[allocated_index]);
 
     return ret_pointer;
   }
@@ -112,15 +113,17 @@ Type* ObjectPool<Type, ValuePool, ObjectAllocator>::AllocateRaw() {
 
 template<class Type, typename ValuePool, class ObjectAllocator>
 size_t ObjectPool<Type, ValuePool, ObjectAllocator>::GetCapacity() {
-  return capacity;
+  return capacity_;
 }
 
 template<class Type, typename ValuePool, class ObjectAllocator>
 ObjectPool<Type, ValuePool, ObjectAllocator>::ObjectPool(size_t capacity) :
-capacity(capacity),
-  p(ReturningTrueIterator(0), ReturningTrueIterator(capacity)) {
-  // Allocate the objects (without construction, just get the memory)
-  objects = objectAllocator.allocate(capacity);
+  capacity_(capacity),
+  value_pool_size_(
+  ValuePool::GetMinimumElementCountForGuaranteedCapacity(capacity)),
+  value_pool_(ReturningTrueIterator(0), ReturningTrueIterator(
+  value_pool_size_)),
+  objects_array_(object_allocator_.allocate(value_pool_size_)) {
 }
 
 template<class Type, typename ValuePool, class ObjectAllocator>
@@ -128,7 +131,7 @@ void ObjectPool<Type, ValuePool, ObjectAllocator>::Free(Type* obj) {
   int index = GetIndexOfObject(*obj);
   obj->~Type();
 
-  p.Free(true, index);
+  value_pool_.Free(true, index);
 }
 
 template<class Type, typename ValuePool, class ObjectAllocator>
@@ -189,7 +192,7 @@ Type* ObjectPool<Type, ValuePool, ObjectAllocator>::Allocate(
 template<class Type, typename ValuePool, class ObjectAllocator>
 ObjectPool<Type, ValuePool, ObjectAllocator>::~ObjectPool() {
   // Deallocate the objects
-  objectAllocator.deallocate(objects, capacity);
+  object_allocator_.deallocate(objects_array_, value_pool_size_);
 }
 } // namespace containers
 } // namespace embb

containers_cpp/include/embb/containers/internal/wait_free_array_value_pool-inl.h

@@ -35,21 +35,21 @@ Free(Type element, int index) {
   assert(element != Undefined);
 
   // Just put back the element
-  pool[index].Store(element);
+  pool_array_[index].Store(element);
 }
 
 template<typename Type, Type Undefined, class Allocator >
 int WaitFreeArrayValuePool<Type, Undefined, Allocator>::
 Allocate(Type & element) {
-  for (int i = 0; i != size; ++i) {
+  for (int i = 0; i != size_; ++i) {
     Type expected;
 
     // If the memory cell is not available, go ahead
-    if (Undefined == (expected = pool[i].Load()))
+    if (Undefined == (expected = pool_array_[i].Load()))
       continue;
 
     // Try to get the memory cell
-    if (pool[i].CompareAndSwap(expected, Undefined)) {
+    if (pool_array_[i].CompareAndSwap(expected, Undefined)) {
       // When the CAS was successful, this element is ours
       element = expected;
       return i;
@@ -64,23 +64,45 @@ WaitFreeArrayValuePool<Type, Undefined, Allocator>::
 WaitFreeArrayValuePool(ForwardIterator first, ForwardIterator last) {
   size_t dist = static_cast<size_t>(std::distance(first, last));
 
-  size = static_cast<int>(dist);
+  size_ = static_cast<int>(dist);
+
+  // conversion may result in negative number. check!
+  assert(size_ >= 0);
 
   // Use the allocator to allocate an array of size dist
-  pool = allocator.allocate(dist);
+  pool_array_ = allocator_.allocate(dist);
+
+  // invoke inplace new for each pool element
+  for ( size_t i = 0; i != dist; ++i ) {
+    new (&pool_array_[i]) embb::base::Atomic<Type>();
+  }
 
   int i = 0;
 
   // Store the elements of the range
   for (ForwardIterator curIter(first); curIter != last; ++curIter) {
-    pool[i++] = *curIter;
+    pool_array_[i++] = *curIter;
   }
 }
 
 template<typename Type, Type Undefined, class Allocator >
 WaitFreeArrayValuePool<Type, Undefined, Allocator>::~WaitFreeArrayValuePool() {
-  allocator.deallocate(pool, (size_t)size);
+  // invoke destructor for each pool element
+  for (int i = 0; i != size_; ++i) {
+    pool_array_[i].~Atomic();
+  }
+
+  // free memory
+  allocator_.deallocate(pool_array_, static_cast<size_t>(size_));
 }
+
+template<typename Type, Type Undefined, class Allocator >
+size_t WaitFreeArrayValuePool<Type, Undefined, Allocator>::
+GetMinimumElementCountForGuaranteedCapacity(size_t capacity) {
+  // for this value pool, this is just capacity...
+  return capacity;
+}
+
 } // namespace containers
 } // namespace embb
 

containers_cpp/include/embb/containers/lock_free_mpmc_queue.h

@@ -113,8 +113,17 @@ class LockFreeMPMCQueue {
    * least as many elements, maybe more.
    */
   size_t capacity;
-  // Do not change the ordering of class local variables.
-  // Important for initialization.
+
+  /**
+   * The object pool, used for lock-free memory allocation.
+   *
+   * Warning: the objectPool has to be initialized before the hazardPointer
+   * object, to be sure that the hazardPointer object is destructed before the
+   * Pool as the hazardPointer object might return elements to the pool in its
+   * destructor. So the ordering of the members objectPool and hazardPointer is
+   * important here!
+   */
+  ObjectPool< internal::LockFreeMPMCQueueNode<Type>, ValuePool > objectPool;
 
   /**
    * Callback to the method that is called by hazard pointers if a pointer is
@@ -124,15 +133,17 @@ class LockFreeMPMCQueue {
     delete_pointer_callback;
 
   /**
-   * The hazard pointer object, used for memory management.
+   * Definition of the used hazard pointer type
    */
-  embb::containers::internal::HazardPointer
-    < internal::LockFreeMPMCQueueNode<Type>* > hazardPointer;
+  typedef embb::containers::internal::HazardPointer
+    < internal::LockFreeMPMCQueueNode<Type>* >
+    MPMCQueueNodeHazardPointer_t;
 
   /**
-   * The object pool, used for lock-free memory allocation.
+   * The hazard pointer object, used for memory management.
    */
-  ObjectPool< internal::LockFreeMPMCQueueNode<Type>, ValuePool > objectPool;
+  MPMCQueueNodeHazardPointer_t hazardPointer;
+
 
   /**
    * Atomic pointer to the head node of the queue

containers_cpp/include/embb/containers/lock_free_stack.h

@@ -187,11 +187,6 @@ class LockFreeStack {
     delete_pointer_callback;
 
   /**
-   * The hazard pointer object, used for memory management.
-   */
-  internal::HazardPointer<internal::LockFreeStackNode<Type>*> hazardPointer;
-
-  /**
    * The callback function, used to cleanup non-hazardous pointers.
    * \see delete_pointer_callback
    */
@@ -199,10 +194,27 @@ class LockFreeStack {
 
   /**
    * The object pool, used for lock-free memory allocation.
+   *
+   * Warning: the objectPool has to be initialized before the hazardPointer
+   * object, to be sure that the hazardPointer object is destructed before the
+   * Pool as the hazardPointer object might return elements to the pool in its
+   * destructor. So the ordering of the members objectPool and hazardPointer is
+   * important here!
    */
   ObjectPool< internal::LockFreeStackNode<Type>, ValuePool > objectPool;
 
   /**
+   * Definition of the used hazard pointer type
+   */
+  typedef internal::HazardPointer < internal::LockFreeStackNode<Type>* >
+    StackNodeHazardPointer_t;
+
+  /**
+   * The hazard pointer object, used for memory management.
+   */
+  StackNodeHazardPointer_t hazardPointer;
+
+  /**
    * Atomic pointer to the top node of the stack (element that is popped next)
    */
   embb::base::Atomic<internal::LockFreeStackNode<Type>*> top;

containers_cpp/include/embb/containers/lock_free_tree_value_pool.h

@@ -123,22 +123,25 @@ class LockFreeTreeValuePool {
   LockFreeTreeValuePool& operator=(const LockFreeTreeValuePool&);
 
   // See algorithm description above
-  int size;
+  int size_;
 
   // See algorithm description above
-  int tree_size;
+  int tree_size_;
 
   // See algorithm description above
-  int real_size;
+  int real_size_;
 
   // The tree above the pool
-  embb::base::Atomic<int>* tree;
+  embb::base::Atomic<int>* tree_;
 
   // The actual pool
-  embb::base::Atomic<Type>* pool;
+  embb::base::Atomic<Type>* pool_;
 
-  PoolAllocator poolAllocator;
-  TreeAllocator treeAllocator;
+  // respective allocator
+  PoolAllocator pool_allocator_;
+
+  // respective allocator
+  TreeAllocator tree_allocator_;
 
   /**
    * Computes smallest power of two fitting the specified value
@@ -278,6 +281,18 @@ class LockFreeTreeValuePool {
   );
 
   /**
+   * Due to concurrency effects, a pool might provide less elements than managed
+   * by it. However, usually one wants to guarantee a minimal capacity. The
+   * count of elements, that must be given to the pool when to guarantee \c
+   * capacity elements is computed using this function.
+   *
+   * \return count of indices the pool has to be initialized with
+   */
+  static size_t GetMinimumElementCountForGuaranteedCapacity(
+    size_t capacity
+    /**< [IN] count of indices that shall be guaranteed */);
+
+  /**
    * Destructs the pool.
    *
    * \notthreadsafe

containers_cpp/include/embb/containers/object_pool.h

@@ -35,7 +35,6 @@
 
 namespace embb {
 namespace containers {
-
 /**
  * \defgroup CPP_CONTAINERS_POOLS Pools
  * Concurrent pools
@@ -62,22 +61,29 @@ class ObjectPool {
   /**
    * Allocator used to allocate elements of the object pool
    */
-  ObjectAllocator objectAllocator;
+  ObjectAllocator object_allocator_;
 
   /**
-   * Array holding the allocated object
+   * Capacity of the object pool
    */
-  Type* objects;
+  size_t capacity_;
 
   /**
-   * Capacity of the object pool
+   * The size of the underlying value pool. This is also the size of the object
+   * array in this class. It is assumed, that the valuepool manages indices in
+   * range [0;value_pool_size_-1].
    */
-  size_t capacity;
+  size_t value_pool_size_;
 
   /**
    * Underlying value pool
    */
-  ValuePool p;
+  ValuePool value_pool_;
+
+  /**
+   * Array holding the allocated object
+   */
+  Type* objects_array_;
 
   /**
    * Helper providing a virtual iterator that just returns true in each

containers_cpp/include/embb/containers/wait_free_array_value_pool.h

@@ -39,12 +39,30 @@ namespace containers {
  * \ingroup CPP_CONCEPT
  * \{
  * \par Description
- * A value pool is a fixed-size multiset of elements, where each element has a
- * unique index. The elements cannot be modified and are given at construction
- * time (by providing first/last iterators). A value pool provides two
- * operations: \c Allocate and \c Free. \c Allocate removes an element from the
- * pool, and \c Free returns an element to the pool. It is only allowed to
- * free elements that have previously been allocated.
+ * A value pool is a multi-set of elements, where each element has a unique,
+ * continuous (starting with 0) index. The elements cannot be modified and are
+ * given at construction time by providing first/last iterators.
+ *
+ * \par
+ * A value pool provides two primary operations: \c Allocate and \c Free. \c
+ * Allocate allocates an element/index "pair" (index via return, element via
+ * reference parameter) from the pool, and \c Free returns an element/index pair
+ * to the pool. To guarantee linearizability, \c element is not allowed to be
+ * modified between \c Allocate and \c Free. It is only allowed to free elements
+ * that have previously been allocated. The \c Allocate function does not
+ * guarantee an order on which indices are allocated. The count of elements that
+ * can be allocated with \c Allocate might be smaller than the count of
+ * elements, the pool is initialized with. This might be because of
+ * implementation details and respective concurrency effects: for example, if
+ * indices are managed within a queue, one has to protect queue elements from
+ * concurrency effects (reuse and access). As long as a thread potentially
+ * accesses a node (and with that an index), the respective index cannot not be
+ * given out to the user, even if being logically not part of the pool anymore.
+ * However, the user might want to guarantee a certain amount of indices to the
+ * user. Therefore, the static \c GetMinimumElementCountForGuaranteedCapacity
+ * method is used. The user passes the count of indices to this method, that
+ * shall be guaranteed by the pool. The method returns the count on indices, the
+ * pool has to be initialized with in order to guarantee this count on indices.
  *
  * \par Requirements
  * - Let \c Pool be the pool class
@@ -54,6 +72,7 @@ namespace containers {
  * - Let \c i, j be forward iterators supporting \c std::distance.
  * - Let \c c be an object of type \c Type&
  * - Let \c e be a value of type \c int
+ * - Let \c f be a value of type \c int
  *
  * \par Valid Expressions
  *
@@ -72,7 +91,7 @@ namespace containers {
  *     the bottom element. The bottom element cannot be stored in the pool, it
  *     is exclusively used to mark empty cells. The pool initially contains
  *     \c std::distance(i, j) elements which are copied during construction from
- *     the range \c [i, j). A concrete class satisfying the value pool concept
+ *     the range \c [i, j]. A concrete class satisfying the value pool concept
  *     might provide additional template parameters for specifying allocators.
  *     </td>
  *   </tr>
@@ -80,9 +99,10 @@ namespace containers {
  *     <td>\code{.cpp} Allocate(c) \endcode</td>
  *     <td>\c int</td>
  *     <td>
- *     Gets an element from the pool. Returns -1, if no element is available,
- *     i.e., the pool is empty. Otherwise, returns the index of the element in
- *     the pool. The value of the pool element is written into reference \c c.
+ *     Allocates an element/index "pair" from the pool. Returns -1, if no
+ *     element is available, i.e., the pool is empty. Otherwise, returns the
+ *     index of the element in the pool. The value of the pool element is
+ *     written into parameter reference \c c.
  *     </td>
  *   </tr>
  *   <tr>
@@ -93,6 +113,15 @@ namespace containers {
  *     \c Allocate. For each allocated element, \c Free must be called exactly
  *     once.</td>
  *   </tr>
+ *   <tr>
+ *     <td>\code{.cpp} GetMinimumElementCountForGuaranteedCapacity(f)
+ *     \endcode</td>
+ *     <td>\c void</td>
+ *     <td>Static method, returns the count of indices, the user has to
+ *     initialize the pool with in order to guarantee a count of \c f elements
+ *     (irrespective of concurrency effects).
+ *      </td>
+ *   </tr>
  * </table>
  *
  * \}
@@ -116,10 +145,10 @@ template<typename Type,
   class Allocator = embb::base::Allocator< embb::base::Atomic<Type> > >
 class WaitFreeArrayValuePool {
  private:
-  int size;
-  embb::base::Atomic<Type>* pool;
+  int size_;
+  embb::base::Atomic<Type>* pool_array_;
   WaitFreeArrayValuePool();
-  Allocator allocator;
+  Allocator allocator_;
 
   // Prevent copy-construction
   WaitFreeArrayValuePool(const WaitFreeArrayValuePool&);
@@ -150,6 +179,18 @@ class WaitFreeArrayValuePool {
   );
 
   /**
+   * Due to concurrency effects, a pool might provide less elements than managed
+   * by it. However, usually one wants to guarantee a minimal capacity. The
+   * count of elements, that must be given to the pool when to guarantee \c
+   * capacity elements is computed using this function.
+   *
+   * \return count of indices the pool has to be initialized with
+   */
+  static size_t GetMinimumElementCountForGuaranteedCapacity(
+    size_t capacity
+    /**< [IN] count of indices that shall be guaranteed */);
+
+  /**
    * Destructs the pool.
    *
    * \notthreadsafe

containers_cpp/test/hazard_pointer_test.cc

@@ -31,24 +31,71 @@
 namespace embb {
 namespace containers {
 namespace test {
+IntObjectTestPool::IntObjectTestPool(unsigned int pool_size) :
+poolSize(pool_size) {
+  simplePoolObjects = static_cast<int*>(
+    embb::base::Allocation::Allocate(sizeof(int)*pool_size));
+
+  simplePool = static_cast<embb::base::Atomic<int>*> (
+    embb::base::Allocation::Allocate(sizeof(embb::base::Atomic<int>)*
+    pool_size));
+
+  for (unsigned int i = 0; i != pool_size; ++i) {
+    // in-place new for each array cell
+    new (&simplePool[i]) embb::base::Atomic<int>;
+  }
+
+  for (unsigned int i = 0; i != pool_size; ++i) {
+    simplePool[i] = FREE_MARKER;
+    simplePoolObjects[i] = 0;
+  }
+}
+
+IntObjectTestPool::~IntObjectTestPool() {
+  embb::base::Allocation::Free(simplePoolObjects);
+
+  for (unsigned int i = 0; i != poolSize; ++i) {
+    // in-place new for each array cell
+    simplePool[i].~Atomic();
+  }
+
+  embb::base::Allocation::Free(simplePool);
+}
+
+int* IntObjectTestPool::Allocate() {
+  for (unsigned int i = 0; i != poolSize; ++i) {
+    int expected = FREE_MARKER;
+    if (simplePool[i].CompareAndSwap
+      (expected, ALLOCATED_MARKER)) {
+      return &simplePoolObjects[i];
+    }
+  }
+  return 0;
+}
+
+void IntObjectTestPool::Release(int* object_pointer) {
+  int cell = object_pointer - simplePoolObjects;
+  simplePool[cell].Store(FREE_MARKER);
+}
+
 HazardPointerTest::HazardPointerTest() :
 #ifdef EMBB_PLATFORM_COMPILER_MSVC
 #pragma warning(push)
 #pragma warning(disable:4355)
 #endif
-  delete_pointer_callback(*this, &HazardPointerTest::DeletePointerCallback),
+delete_pointer_callback_(*this, &HazardPointerTest::DeletePointerCallback),
 #ifdef EMBB_PLATFORM_COMPILER_MSVC
 #pragma warning(pop)
 #endif
-  object_pool(NULL),
-  stack(NULL),
-  hp(NULL),
-n_threads(static_cast<int>
+  object_pool_(NULL),
+  stack_(NULL),
+  hazard_pointer_(NULL),
+  n_threads_(static_cast<int>
   (partest::TestSuite::GetDefaultNumThreads())) {
-  n_elements_per_thread = 100;
-  n_elements = n_threads*n_elements_per_thread;
+  n_elements_per_thread_ = 100;
+  n_elements_ = n_threads_*n_elements_per_thread_;
   embb::base::Function < void, embb::base::Atomic<int>* >
-    delete_pointer_callback(
+    deletePointerCallback(
     *this,
     &HazardPointerTest::DeletePointerCallback);
 
@@ -59,45 +106,52 @@ n_threads(static_cast<int>
   // placed, the pointer is not allowed to be deleted until the second thread
   // removes this guard.
   CreateUnit("HazardPointerTestThatGuardWorks").
-    Pre(&HazardPointerTest::HazardPointerTest1_Pre, this).
+    Pre(&HazardPointerTest::HazardPointerTest1Pre, this).
     Add(
-    &HazardPointerTest::HazardPointerTest1_ThreadMethod,
-    this, static_cast<size_t>(n_threads)).
-    Post(&HazardPointerTest::HazardPointerTest1_Post, this);
+    &HazardPointerTest::HazardPointerTest1ThreadMethod,
+    this, static_cast<size_t>(n_threads_)).
+    Post(&HazardPointerTest::HazardPointerTest1Post, this);
 }
 
-void HazardPointerTest::HazardPointerTest1_Pre() {
+void HazardPointerTest::HazardPointerTest1Pre() {
   embb_internal_thread_index_reset();
-  object_pool = new embb::containers::ObjectPool< embb::base::Atomic<int> >
-    (static_cast<size_t>(n_elements));
-  stack = new embb::containers::LockFreeStack< embb::base::Atomic<int>* >
-    (static_cast<size_t>(n_elements));
-  hp = new embb::containers::internal::HazardPointer< embb::base::Atomic<int>*>
-    (delete_pointer_callback,
-    NULL,
+
+  object_pool_ =
+     embb::base::Allocation::
+     New<embb::containers::ObjectPool< embb::base::Atomic<int> > >
+     (static_cast<size_t>(n_elements_));
+
+  stack_ = embb::base::Allocation::
+    New<embb::containers::LockFreeStack< embb::base::Atomic<int>* > >
+    (static_cast<size_t>(n_elements_));
+
+  hazard_pointer_ = embb::base::Allocation::
+    New<embb::containers::internal::HazardPointer < embb::base::Atomic<int>* > >
+    (delete_pointer_callback_,
+    static_cast<embb::base::Atomic<int>*>(NULL),
     1);
 }
 
-void HazardPointerTest::HazardPointerTest1_Post() {
-  delete object_pool;
-  delete stack;
-  delete hp;
+void HazardPointerTest::HazardPointerTest1Post() {
+  embb::base::Allocation::Delete(hazard_pointer_);
+  embb::base::Allocation::Delete(object_pool_);
+  embb::base::Allocation::Delete(stack_);
 }
 
-void HazardPointerTest::HazardPointerTest1_ThreadMethod() {
+void HazardPointerTest::HazardPointerTest1ThreadMethod() {
   unsigned int thread_index;
   embb_internal_thread_index(&thread_index);
 
-  for (int i = 0; i != n_elements_per_thread; ++i) {
-    embb::base::Atomic<int>* allocated_object = object_pool->Allocate(0);
+  for (int i = 0; i != n_elements_per_thread_; ++i) {
+    embb::base::Atomic<int>* allocated_object = object_pool_->Allocate(0);
 
-    hp->GuardPointer(0, allocated_object);
+    hazard_pointer_->Guard(0, allocated_object);
 
-    bool success = stack->TryPush(allocated_object);
+    bool success = stack_->TryPush(allocated_object);
 
     PT_ASSERT(success == true);
 
-    embb::base::Atomic<int>* allocated_object_from_different_thread;
+    embb::base::Atomic<int>* allocated_object_from_different_thread(0);
 
     int diff_count = 0;
 
@@ -105,50 +159,365 @@ void HazardPointerTest::HazardPointerTest1_ThreadMethod() {
     bool success_pop;
 
     while (
-      (success_pop = stack->TryPop(allocated_object_from_different_thread))
+      (success_pop = stack_->TryPop(allocated_object_from_different_thread))
       == true
       && allocated_object_from_different_thread == allocated_object
       ) {
-      //try to make it probable to get an element from a different thread
-      //however, can be the same. Try 10000 times to get a different element.
+      // try to make it probable to get an element from a different thread
+      // however, can be the same. Try 10000 times to get a different element.
       if (diff_count++ > 10000) {
         same = true;
         break;
       }
-      bool success = stack->TryPush(allocated_object_from_different_thread);
+      bool success = stack_->TryPush(allocated_object_from_different_thread);
       PT_ASSERT(success == true);
     }
     PT_ASSERT(success_pop == true);
     allocated_object->Store(1);
-    hp->EnqueuePointerForDeletion(allocated_object);
+    hazard_pointer_->EnqueueForDeletion(allocated_object);
 
     if (!same) {
-      hp->GuardPointer(0, allocated_object_from_different_thread);
+      hazard_pointer_->Guard(0, allocated_object_from_different_thread);
 
       // if this holds, we were successful in guarding... otherwise we
       // were to late, because the pointer has already been added
       // to the retired list.
       if (*allocated_object_from_different_thread == 0) {
         // the pointer must not be deleted here!
-        vector_mutex.Lock();
+        vector_mutex_.Lock();
         for (std::vector< embb::base::Atomic<int>* >::iterator
-          it = deleted_vector.begin();
-          it != deleted_vector.end();
+          it = deleted_vector_.begin();
+          it != deleted_vector_.end();
         ++it) {
           PT_ASSERT(*it != allocated_object_from_different_thread);
         }
-        vector_mutex.Unlock();
+        vector_mutex_.Unlock();
       }
-      hp->GuardPointer(0, NULL);
+      hazard_pointer_->Guard(0, NULL);
     }
   }
 }
 
 void HazardPointerTest::DeletePointerCallback
 (embb::base::Atomic<int>* to_delete) {
-  vector_mutex.Lock();
-  deleted_vector.push_back(to_delete);
-  vector_mutex.Unlock();
+  vector_mutex_.Lock();
+  deleted_vector_.push_back(to_delete);
+  vector_mutex_.Unlock();
+}
+
+void HazardPointerTest2::DeletePointerCallback(int* to_delete) {
+  test_pool_->Release(to_delete);
+}
+
+bool HazardPointerTest2::SetRelativeGuards() {
+  unsigned int thread_index;
+  embb_internal_thread_index(&thread_index);
+
+  unsigned int my_begin = guards_per_phread_count_*thread_index;
+  int guard_number = 0;
+  unsigned int alreadyGuarded = 0;
+
+  for (unsigned int i = my_begin; i != my_begin + guards_per_phread_count_;
+    ++i) {
+    if (shared_guarded_[i] != 0) {
+      alreadyGuarded++;
+      guard_number++;
+      continue;
+    }
+
+    int * to_guard = shared_allocated_[i];
+    if (to_guard) {
+      hazard_pointer_->Guard(guard_number, to_guard);
+
+      // changed in the meantime?
+      if (to_guard == shared_allocated_[i].Load()) {
+        // guard was successful. Communicate to other threads.
+        shared_guarded_[i] = to_guard;
+      } else {
+        // reset the guard, couldn't guard...
+        hazard_pointer_->RemoveGuard(guard_number);
+      }
+    }
+    guard_number++;
+  }
+  return(alreadyGuarded == guards_per_phread_count_);
+}
+
+void HazardPointerTest2::HazardPointerTest2Master() {
+  // while the hazard pointer guard array is not full
+  int** allocatedLocal = static_cast<int**>(
+  embb::base::Allocation::Allocate(sizeof(int*)*guaranteed_capacity_pool_));
+
+  bool full = false;
+  while (!full) {
+    full = true;
+    for (unsigned int i = 0; i != guaranteed_capacity_pool_; ++i) {
+      if (shared_guarded_[i] == 0) {
+        full = false;
+        break;
+      }
+    }
+
+    // not all guards set
+    for (unsigned int i = 0; i != guaranteed_capacity_pool_; ++i) {
+      allocatedLocal[i] = test_pool_->Allocate();
+      shared_allocated_[i].Store(allocatedLocal[i]);
+    }
+
+    // set my hazards. We do not have to check, this must be successful
+    // here.
+    SetRelativeGuards();
+
+    // free
+    for (unsigned int i = 0; i != guaranteed_capacity_pool_; ++i) {
+      shared_allocated_[i].Store(0);
+      hazard_pointer_->EnqueueForDeletion(allocatedLocal[i]);
+    }
+  }
+
+  embb::base::Allocation::Free(allocatedLocal);
+}
+
+void HazardPointerTest2::HazardPointerTest2Slave() {
+  unsigned int thread_index;
+  embb_internal_thread_index(&thread_index);
+
+  while (!SetRelativeGuards()) {}
+}
+
+void HazardPointerTest2::HazardPointerTest2Pre() {
+  embb_internal_thread_index_reset();
+  current_master_ = 0;
+  sync1_ = 0;
+  sync2_ = 0;
+
+  // first the test pool has to be created
+  test_pool_ = embb::base::Allocation::New<IntObjectTestPool>
+    (pool_size_using_hazard_pointer_);
+
+  // after the pool has been created, we create the hp class
+  hazard_pointer_ = embb::base::Allocation::New <
+    embb::containers::internal::HazardPointer<int*> >
+    (delete_pointer_callback_, static_cast<int*>(NULL),
+    static_cast<int>(guards_per_phread_count_), n_threads);
+
+  shared_guarded_ = static_cast<embb::base::Atomic<int*>*>(
+    embb::base::Allocation::Allocate(sizeof(embb::base::Atomic<int*>)*
+    guaranteed_capacity_pool_));
+
+  for (unsigned int i = 0; i != guaranteed_capacity_pool_; ++i) {
+    // in-place new for each array cell
+    new (&shared_guarded_[i]) embb::base::Atomic < int* >;
+  }
+
+  shared_allocated_ = static_cast<embb::base::Atomic<int*>*>(
+    embb::base::Allocation::Allocate(sizeof(embb::base::Atomic<int*>)*
+    guaranteed_capacity_pool_));
+
+  for (unsigned int i = 0; i !=
+  guaranteed_capacity_pool_; ++i) {
+    // in-place new for each array cell
+    new (&shared_allocated_[i]) embb::base::Atomic < int* >;
+  }
+
+  for (unsigned int i = 0; i != guaranteed_capacity_pool_; ++i) {
+    shared_guarded_[i] = 0;
+    shared_allocated_[i] = 0;
+  }
+}
+
+void HazardPointerTest2::HazardPointerTest2Post() {
+  for (unsigned int i = 0; i != static_cast<unsigned int>(n_threads); ++i) {
+    for (unsigned int i2 = 0; i2 != static_cast<unsigned int>(n_threads)*
+      guards_per_phread_count_; ++i2) {
+      if (hazard_pointer_->thread_local_retired_lists_
+        [i2 + i*n_threads*guards_per_phread_count_] == NULL) {
+        // all retired lists must be completely filled
+        PT_ASSERT(false);
+      }
+    }
+  }
+
+  unsigned int checks = 0;
+  for (unsigned int i = 0; i != static_cast<unsigned int>(n_threads); ++i) {
+    for (unsigned int i2 = 0; i2 != static_cast<unsigned int>(n_threads)*
+      guards_per_phread_count_; ++i2) {
+      for (unsigned int j = 0; j != static_cast<unsigned int>(n_threads); ++j) {
+        for (unsigned int j2 = 0; j2 != static_cast<unsigned int>(n_threads)*
+          guards_per_phread_count_; ++j2) {
+          if (i2 == j2 && i == j)
+            continue;
+
+          // all retired elements have to be disjoint
+          PT_ASSERT(
+            hazard_pointer_->thread_local_retired_lists_
+            [i2 + i*n_threads*guards_per_phread_count_] !=
+            hazard_pointer_->thread_local_retired_lists_
+            [j2 + j*n_threads*guards_per_phread_count_]);
+
+          checks++;
+        }
+      }
+    }
+  }
+
+  // sanity check on the count of expected comparisons.
+  PT_ASSERT(
+    checks ==
+    n_threads*n_threads*guards_per_phread_count_ *
+    (n_threads*n_threads*guards_per_phread_count_ - 1));
+
+  std::vector< int* > additionallyAllocated;
+
+  // we should be able to still allocate the guaranteed capacity of
+  // elements from the pool.
+  for (unsigned int i = 0; i != guaranteed_capacity_pool_; ++i) {
+    int* allocated = test_pool_->Allocate();
+
+    // allocated is not allowed to be zero
+    PT_ASSERT(allocated != NULL);
+
+    // push to vector, to check if elements are disjunctive and to release
+    // afterwards.
+    additionallyAllocated.push_back(allocated);
+  }
+
+  // the pool should now be empty
+  PT_ASSERT(test_pool_->Allocate() == NULL);
+
+  // release allocated elements...
+  for (unsigned int i = 0; i != additionallyAllocated.size(); ++i) {
+    test_pool_->Release(additionallyAllocated[i]);
+  }
+
+  // the additionallyAllocated elements shall be disjoint
+  for (unsigned int i = 0; i != additionallyAllocated.size(); ++i) {
+    for (unsigned int i2 = 0; i2 != additionallyAllocated.size(); ++i2) {
+      if (i == i2)
+        continue;
+      PT_ASSERT(additionallyAllocated[i] !=
+        additionallyAllocated[i2]);
+    }
+  }
+
+  // no allocated element should be in any retired list...
+  for (unsigned int a = 0; a != additionallyAllocated.size(); ++a) {
+    for (unsigned int i = 0; i != static_cast<unsigned int>(n_threads); ++i) {
+      for (unsigned int i2 = 0; i2 != static_cast<unsigned int>(n_threads)*
+        guards_per_phread_count_; ++i2) {
+        PT_ASSERT(
+          hazard_pointer_->thread_local_retired_lists_
+          [i2 + i*n_threads*guards_per_phread_count_] !=
+          additionallyAllocated[a]);
+      }
+    }
+  }
+
+  for (unsigned int i = 0; i != guaranteed_capacity_pool_; ++i) {
+    // in-place new for each array cell
+    shared_guarded_[i].~Atomic();
+  }
+
+  embb::base::Allocation::Free(shared_guarded_);
+
+  for (unsigned int i = 0; i != guaranteed_capacity_pool_; ++i) {
+    // in-place new for each array cell
+    shared_allocated_[i].~Atomic();
+  }
+
+  embb::base::Allocation::Free(shared_allocated_);
+  embb::base::Allocation::Delete(hazard_pointer_);
+
+  // after deleting the hazard pointer object, all retired pointers have
+  // to be returned to the pool!
+  std::vector<int*> elementsInPool;
+
+  int* nextElement;
+  while ((nextElement = test_pool_->Allocate()) != NULL) {
+    for (unsigned int i = 0; i != elementsInPool.size(); ++i) {
+      // all elements need to be disjoint
+      PT_ASSERT(elementsInPool[i] != nextElement);
+    }
+    elementsInPool.push_back(nextElement);
+  }
+
+  // all elements should have been returned by the hp object, so we should be
+  // able to acquire all elements.
+  PT_ASSERT(elementsInPool.size() == pool_size_using_hazard_pointer_);
+
+  embb::base::Allocation::Delete(test_pool_);
+}
+
+void HazardPointerTest2::HazardPointerTest2ThreadMethod() {
+  for (;;) {
+    unsigned int thread_index;
+    embb_internal_thread_index(&thread_index);
+
+    if (thread_index == current_master_) {
+      HazardPointerTest2Master();
+    } else {
+      HazardPointerTest2Slave();
+    }
+
+    sync1_.FetchAndAdd(1);
+
+    // wait until cleanup thread signals to be finished
+    while (sync1_ != 0) {
+      int expected = n_threads;
+      int desired = FINISH_MARKER;
+      // select thread, responsible for cleanup
+      if (sync1_.CompareAndSwap(expected, desired)) {
+        // wipe arrays!
+        for (unsigned int i = 0; i != guaranteed_capacity_pool_; ++i) {
+          shared_guarded_[i] = 0;
+          shared_allocated_[i] = 0;
+        }
+
+        // increase master
+        current_master_.FetchAndAdd(1);
+        sync2_ = 0;
+        sync1_.Store(0);
+      }
+    }
+
+    // wait for all threads to reach this position
+    sync2_.FetchAndAdd(1);
+    while (sync2_ != static_cast<unsigned int>(n_threads)) {}
+
+    // if each thread was master once, terminate.
+    if (current_master_ == static_cast<unsigned int>(n_threads)) {
+      return;
+    }
+  }
+}
+
+HazardPointerTest2::HazardPointerTest2() :
+n_threads(static_cast<int>
+(partest::TestSuite::GetDefaultNumThreads())),
+
+#ifdef EMBB_PLATFORM_COMPILER_MSVC
+#pragma warning(push)
+#pragma warning(disable:4355)
+#endif
+  delete_pointer_callback_(
+  *this,
+  &HazardPointerTest2::DeletePointerCallback)
+#ifdef EMBB_PLATFORM_COMPILER_MSVC
+#pragma warning(pop)
+#endif
+{
+  guards_per_phread_count_ = 5;
+  guaranteed_capacity_pool_ = guards_per_phread_count_*n_threads;
+  pool_size_using_hazard_pointer_ = guaranteed_capacity_pool_ +
+    guards_per_phread_count_*n_threads*n_threads;
+
+  embb::base::Thread::GetThreadsMaxCount();
+  CreateUnit("HazardPointerTestSimulateMemoryWorstCase").
+    Pre(&HazardPointerTest2::HazardPointerTest2Pre, this).
+    Add(
+    &HazardPointerTest2::HazardPointerTest2ThreadMethod,
+    this, static_cast<size_t>(n_threads)).
+    Post(&HazardPointerTest2::HazardPointerTest2Post, this);
 }
 }  // namespace test
 }  // namespace containers

containers_cpp/test/hazard_pointer_test.h

@@ -36,32 +36,112 @@
 namespace embb {
 namespace containers {
 namespace test {
-class HazardPointerTest : public partest::TestCase {
+/**
+ * @brief a very simple wait-free object pool implementation to have tests
+ * being independent of the EMBB object pool implementation.
+ */
+class IntObjectTestPool {
  private:
-  embb::base::Function<void, embb::base::Atomic<int>*> delete_pointer_callback;
+  int* simplePoolObjects;
+  embb::base::Atomic<int>* simplePool;
 
-  //used to allocate random stuff, we will just use the pointers, not the
-  //contents
-  embb::containers::ObjectPool< embb::base::Atomic<int> >* object_pool;
+ public:
+  static const int ALLOCATED_MARKER = 1;
+  static const int FREE_MARKER = 0;
+  unsigned int poolSize;
 
-  //used to move pointer between threads
-  embb::containers::LockFreeStack< embb::base::Atomic<int>* >* stack;
-  embb::base::Mutex vector_mutex;
-  embb::containers::internal::HazardPointer<embb::base::Atomic<int>*>* hp;
-  std::vector< embb::base::Atomic<int>* > deleted_vector;
-  int n_threads;
-  int n_elements_per_thread;
-  int n_elements;
+  explicit IntObjectTestPool(unsigned int pool_size);
+
+  ~IntObjectTestPool();
 
+  /**
+   * Allocate object from the pool
+   *
+   * @return the allocated object
+   */
+  int* Allocate();
+
+  /**
+   * Return an element to the pool
+   *
+   * @param objectPointer the object to be freed
+   */
+  void Release(int* object_pointer);
+};
+
+class HazardPointerTest : public partest::TestCase {
  public:
   /**
   * Adds test methods.
   */
   HazardPointerTest();
-  void HazardPointerTest1_Pre();
-  void HazardPointerTest1_Post();
-  void HazardPointerTest1_ThreadMethod();
+  void HazardPointerTest1Pre();
+  void HazardPointerTest1Post();
+  void HazardPointerTest1ThreadMethod();
   void DeletePointerCallback(embb::base::Atomic<int>* to_delete);
+
+ private:
+  embb::base::Function<void, embb::base::Atomic<int>*> delete_pointer_callback_;
+
+  //used to allocate random stuff, we will just use the pointers, not the
+  //contents
+  embb::containers::ObjectPool< embb::base::Atomic<int> >* object_pool_;
+
+  //used to move pointer between threads
+  embb::containers::LockFreeStack< embb::base::Atomic<int>* >* stack_;
+  embb::base::Mutex vector_mutex_;
+  embb::containers::internal::HazardPointer<embb::base::Atomic<int>*>*
+    hazard_pointer_;
+  std::vector< embb::base::Atomic<int>* > deleted_vector_;
+  int n_threads_;
+  int n_elements_per_thread_;
+  int n_elements_;
+};
+
+class HazardPointerTest2 : public partest::TestCase {
+ public:
+  void DeletePointerCallback(int* to_delete);
+  bool SetRelativeGuards();
+  void HazardPointerTest2Master();
+  void HazardPointerTest2Slave();
+
+  void HazardPointerTest2Pre();
+  void HazardPointerTest2Post();
+
+  void HazardPointerTest2ThreadMethod();
+
+  HazardPointerTest2();
+
+ private:
+  // number of threads, participating in that test
+  int n_threads;
+
+  embb::base::Function<void, int*> delete_pointer_callback_;
+  // the thread id of the master
+  embb::base::Atomic<unsigned int> current_master_;
+
+  // variables, to synchronize threads. At each point in time, one master,
+  // the master changes each round until each thread was assigned master once.
+  embb::base::Atomic<int> sync1_;
+  embb::base::Atomic<unsigned int> sync2_;
+
+  unsigned int guards_per_phread_count_;
+  unsigned int guaranteed_capacity_pool_;
+  unsigned int pool_size_using_hazard_pointer_;
+
+  // The threads write here, if they guarded an object successfully. Used to
+  // determine when all allocated objects were guarded successfully.
+  embb::base::Atomic<int*>* shared_guarded_;
+
+  // This array is used by the master, to communicate and share what he has
+  // allocated with the slaves.
+  embb::base::Atomic<int*>* shared_allocated_;
+
+  // Reference to the object pool
+  IntObjectTestPool* test_pool_;
+
+  embb::containers::internal::HazardPointer<int*>* hazard_pointer_;
+  static const int FINISH_MARKER = -1;
 };
 } // namespace test
 } // namespace containers

containers_cpp/test/main.cc

@@ -55,6 +55,7 @@ using embb::containers::test::HazardPointerTest;
 using embb::containers::test::QueueTest;
 using embb::containers::test::StackTest;
 using embb::containers::test::ObjectPoolTest;
+using embb::containers::test::HazardPointerTest2;
 
 PT_MAIN("Data Structures C++") {
   unsigned int max_threads = static_cast<unsigned int>(
@@ -64,6 +65,7 @@ PT_MAIN("Data Structures C++") {
   PT_RUN(PoolTest< WaitFreeArrayValuePool<int COMMA -1> >);
   PT_RUN(PoolTest< LockFreeTreeValuePool<int COMMA -1> >);
   PT_RUN(HazardPointerTest);
+  PT_RUN(HazardPointerTest2);
   PT_RUN(QueueTest< WaitFreeSPSCQueue< ::std::pair<size_t COMMA int> > >);
   PT_RUN(QueueTest< LockFreeMPMCQueue< ::std::pair<size_t COMMA int> >
     COMMA true COMMA true >);

dataflow_cpp/test/dataflow_cpp_test_simple.cc

@@ -39,7 +39,7 @@
 #define NUM_SLICES 8
 #define TEST_COUNT 12
 
-typedef embb::dataflow::Network<8> MyNetwork;
+typedef embb::dataflow::Network<NUM_SLICES> MyNetwork;
 typedef MyNetwork::ConstantSource< int > MyConstantSource;
 typedef MyNetwork::Source< int > MySource;
 typedef MyNetwork::SerialProcess< MyNetwork::Inputs<int>::Type,
@@ -156,9 +156,7 @@ void SimpleTest::TestBasic() {
     core_set,
     1024, // max tasks (default: 1024)
     128,  // max groups (default: 128)
-    // Currently needs to be initialized
-    // with (max_queues + 1), see defect embb449
-    num_cores + 1, // max queues (default: 16)
+    num_cores, // max queues (default: 16)
     1024, // queue capacity (default: 1024)
     4);   // num priorities (default: 4)
 

mtapi_c/src/embb_mtapi_id_pool_t.c

@@ -71,7 +71,7 @@ mtapi_uint_t embb_mtapi_id_pool_allocate(embb_mtapi_id_pool_t * that) {
       /* acquire position to fetch id from */
       mtapi_uint_t id_position = that->get_id_position;
       that->get_id_position++;
-      if (that->capacity <= that->get_id_position) {
+      if (that->capacity < that->get_id_position) {
         that->get_id_position = 0;
       }
 
@@ -97,7 +97,7 @@ void embb_mtapi_id_pool_deallocate(
       /* acquire position to put id to */
       mtapi_uint_t id_position = that->put_id_position;
       that->put_id_position++;
-      if (that->capacity <= that->put_id_position) {
+      if (that->capacity < that->put_id_position) {
         that->put_id_position = 0;
       }
 

mtapi_c/test/embb_mtapi_test_id_pool.cc

+/*
+ * Copyright (c) 2014-2015, Siemens AG. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ * this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ * this list of conditions and the following disclaimer in the documentation
+ * and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <embb_mtapi_test_id_pool.h>
+#include <vector>
+
+IdPoolTest::IdPoolTest() {
+  CreateUnit("mtapi id pool test single threaded").
+    Add(&IdPoolTest::TestBasic, this, 1, 1000).
+    Pre(&IdPoolTest::TestBasicPre, this).
+    Post(&IdPoolTest::TestBasicPost, this);
+
+  CreateUnit("mtapi id pool test concurrent").
+    Add(&IdPoolTest::TestParallel, this, concurrent_accessors_id_pool_2
+    , 20).
+    Post(&IdPoolTest::TestParallelPost, this).
+    Pre(&IdPoolTest::TestParallelPre, this);
+}
+
+void IdPoolTest::TestParallel() {
+  // allocate ID_ELEMENTS_PER_ACCESSOR elements. Each test thread is
+  // guaranteed to be able to allocate this amount of elements.
+  TestAllocateDeallocateNElementsFromPool(id_pool_parallel,
+    id_elements_per_accessor);
+}
+
+void IdPoolTest::TestParallelPre() {
+  // create second id pool with CONCURRENT_ACCESSORS_ID_POOL_2*
+  // ID_ELEMENTS_PER_ACCESSOR elements
+  embb_mtapi_id_pool_initialize(&id_pool_parallel,
+    concurrent_accessors_id_pool_2*id_elements_per_accessor);
+}
+
+void IdPoolTest::TestParallelPost() {
+  // after the parallel tests, try to again allocate and deallocate all
+  // elements sequentially.
+  TestAllocateDeallocateNElementsFromPool(id_pool_parallel,
+    concurrent_accessors_id_pool_2*id_elements_per_accessor, true);
+
+  // finalize pool
+  embb_mtapi_id_pool_finalize(&id_pool_parallel);
+}
+
+void IdPoolTest::TestBasic() {
+  TestAllocateDeallocateNElementsFromPool(id_pool, id_pool_size_1, true);
+}
+
+void IdPoolTest::TestBasicPre() {
+  // create id pool with ID_POOL_SIZE_1 elements
+  embb_mtapi_id_pool_initialize(&id_pool, id_pool_size_1);
+}
+
+void IdPoolTest::TestBasicPost() {
+  // finalize pool
+  embb_mtapi_id_pool_finalize(&id_pool);
+}
+
+void IdPoolTest::TestAllocateDeallocateNElementsFromPool(
+  embb_mtapi_id_pool_t &pool,
+  int count_elements,
+  bool empty_check) {
+  std::vector<unsigned int> allocated;
+
+  for (int i = 0; i != count_elements; ++i) {
+    allocated.push_back(embb_mtapi_id_pool_allocate(&pool));
+  }
+
+  // the allocated elements should be disjunctive, and never invalid element
+  for (unsigned int x = 0; x != allocated.size(); ++x) {
+    PT_ASSERT(allocated[x] != EMBB_MTAPI_IDPOOL_INVALID_ID);
+    for (unsigned int y = 0; y != allocated.size(); ++y) {
+      if (x == y) {
+        continue;
+      }
+      PT_ASSERT(allocated[x] != allocated[y]);
+    }
+  }
+
+  // now the id pool should be empty... try ten times to get an id,
+  // we should always get the invalid element
+  if (empty_check) {
+    for (int i = 0; i != 10; ++i) {
+      PT_ASSERT_EQ(embb_mtapi_id_pool_allocate(&pool),
+        static_cast<unsigned int>(EMBB_MTAPI_IDPOOL_INVALID_ID)
+        )
+    }
+  }
+
+  // now return allocated elements in a shuffled manner.
+  ::std::random_shuffle(allocated.begin(), allocated.end());
+
+  for (int i = 0; i != count_elements; ++i) {
+    embb_mtapi_id_pool_deallocate(&pool,
+      allocated[static_cast<unsigned int>(i)]);
+  }
+}
+

mtapi_c/test/embb_mtapi_test_id_pool.h

+/*
+ * Copyright (c) 2014-2015, Siemens AG. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ * this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ * this list of conditions and the following disclaimer in the documentation
+ * and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef MTAPI_C_TEST_EMBB_MTAPI_TEST_ID_POOL_H_
+#define MTAPI_C_TEST_EMBB_MTAPI_TEST_ID_POOL_H_
+
+#include <partest/partest.h>
+#include <embb_mtapi_id_pool_t.h>
+
+// for shuffling a vector
+#include <algorithm>
+
+class IdPoolTest : public partest::TestCase {
+ public:
+  embb_mtapi_id_pool_t id_pool;
+  embb_mtapi_id_pool_t id_pool_parallel;
+
+  IdPoolTest();
+
+ private:
+  static const unsigned int id_pool_size_1 = 100;
+  static const unsigned int concurrent_accessors_id_pool_2 = 10;
+  static const unsigned int id_elements_per_accessor = 10;
+
+  /**  
+   * We create a pool of size number_accessors*elements_per_accessor, so
+   * at each time we can guarantee each thread to be able to allocate 
+   * elements_per_accessor elements.
+   * We create number_accessor threads, where each thread iteratively 
+   * allocates and frees elements_per_accessor elements, which in each case
+   * has to be successful. Additionally, the sanity checks from the basic tests
+   * are repeated. The TestParallelPost function also repeats all
+   * sequential tests.
+   */
+  void TestParallel();
+  void TestParallelPre();
+  void TestParallelPost();
+
+  /**
+   * Create a pool of size N. We repeatedly allocate and free N elements, check
+   * if the pool always returns disjunctive ids and check that the pool never
+   * returns the invalid element, if the pool is not empty. Check that the 
+   * invalid element is returned if the pool is empty.
+   */
+  void TestBasic();
+  void TestBasicPre();
+  void TestBasicPost();
+
+  static void TestAllocateDeallocateNElementsFromPool(
+    embb_mtapi_id_pool_t &pool,
+    int count_elements,
+    bool empty_check = false);
+};
+
+#endif // MTAPI_C_TEST_EMBB_MTAPI_TEST_ID_POOL_H_

mtapi_c/test/main.cc

@@ -37,6 +37,9 @@
 #include <embb_mtapi_test_group.h>
 #include <embb_mtapi_test_queue.h>
 #include <embb_mtapi_test_error.h>
+#include <embb_mtapi_test_id_pool.h>
+
+#include <embb/base/c/memory_allocation.h>
 
 PT_MAIN("MTAPI C") {
   embb_log_set_log_level(EMBB_LOG_LEVEL_NONE);
@@ -48,4 +51,7 @@ PT_MAIN("MTAPI C") {
   PT_RUN(InitFinalizeTest);
   PT_RUN(GroupTest);
   PT_RUN(QueueTest);
+  PT_RUN(IdPoolTest);
+
+  PT_EXPECT(embb_get_bytes_allocated() == 0);
 }

mtapi_cpp/CMakeLists.txt

@@ -5,14 +5,10 @@ file(GLOB_RECURSE EMBB_MTAPI_CPP_HEADERS "include/*.h")
 file(GLOB_RECURSE EMBB_MTAPI_CPP_TEST_SOURCES "test/*.cc" "test/*.h")
 
 if (USE_AUTOMATIC_INITIALIZATION STREQUAL ON)
-  message("-- Automatic initialization enabled (default)")
   set(MTAPI_CPP_AUTOMATIC_INITIALIZE 1)
 else()
   set(MTAPI_CPP_AUTOMATIC_INITIALIZE 0)
-  message("-- Automatic initialization disabled")
 endif()
-message("   (set with command line option -DUSE_AUTOMATIC_INITIALIZATION=ON/OFF)")
-
 
 # Execute the GroupSources macro
 include(${CMAKE_SOURCE_DIR}/CMakeCommon/GroupSourcesMSVC.cmake)

tasks_cpp/CMakeLists.txt

@@ -5,13 +5,10 @@ file(GLOB_RECURSE EMBB_TASKS_CPP_HEADERS "include/*.h")
 file(GLOB_RECURSE EMBB_TASKS_CPP_TEST_SOURCES "test/*.cc" "test/*.h")
 
 if (USE_AUTOMATIC_INITIALIZATION STREQUAL ON)
-  message("-- Automatic initialization enabled (default)")
   set(TASKS_CPP_AUTOMATIC_INITIALIZE 1)
 else()
   set(TASKS_CPP_AUTOMATIC_INITIALIZE 0)
-  message("-- Automatic initialization disabled")
 endif()
-message("   (set with command line option -DUSE_AUTOMATIC_INITIALIZATION=ON/OFF)")
 
 configure_file("include/embb/tasks/internal/cmake_config.h.in"
                "include/embb/tasks/internal/cmake_config.h")

tasks_cpp/test/tasks_cpp_test_task.cc

@@ -78,13 +78,19 @@ void TaskTest::TestBasic() {
   PT_EXPECT_EQ(policy.GetPriority(), 0u);
   policy.AddWorker(0u);
   PT_EXPECT_EQ(policy.GetAffinity(), 1u);
+
+  if (policy.GetCoreCount() > 1) {
     policy.AddWorker(1u);
     PT_EXPECT_EQ(policy.GetAffinity(), 3u);
+  }
+
   policy.RemoveWorker(0u);
-  PT_EXPECT_EQ(policy.GetAffinity(), 2u);
   PT_EXPECT_EQ(policy.IsSetWorker(0), false);
-  PT_EXPECT_EQ(policy.IsSetWorker(1), true);
 
+  if (policy.GetCoreCount() > 1) {
+    PT_EXPECT_EQ(policy.GetAffinity(), 2u);
+    PT_EXPECT_EQ(policy.IsSetWorker(1), true);
+  }
   std::string test;
   embb::tasks::Task task = node.Spawn(
     embb::base::Bind(