From d049a1339e7bb80ac8dac381488d25a2f81058a2 Mon Sep 17 00:00:00 2001 From: Victor Zarubkin Date: Thu, 28 Dec 2017 22:21:54 +0300 Subject: [PATCH] #75 [Core] No more waiting behavior when dumping blocks. Current solution resolves possible dead-lock but restricts any Events out of Frame bounds (You can see that the last event for LoadingResources thread in the profiler_sample is always absent). --- easy_profiler_core/chunk_allocator.h | 178 +++++++++++++++---------- easy_profiler_core/profile_manager.cpp | 76 ++++++----- easy_profiler_core/profile_manager.h | 2 +- easy_profiler_core/thread_storage.cpp | 21 ++- easy_profiler_core/thread_storage.h | 10 +- 5 files changed, 169 insertions(+), 118 deletions(-) diff --git a/easy_profiler_core/chunk_allocator.h b/easy_profiler_core/chunk_allocator.h index c189cf3..82e1c39 100644 --- a/easy_profiler_core/chunk_allocator.h +++ b/easy_profiler_core/chunk_allocator.h @@ -91,7 +91,7 @@ The Apache License, Version 2.0 (the "License"); template EASY_FORCE_INLINE bool is_aligned(void* ptr) { - static_assert(ALIGNMENT % 2 == 0, "Alignment must be a power of two."); + static_assert((ALIGNMENT & 1) == 0, "Alignment must be a power of two."); return ((uintptr_t)ptr & (ALIGNMENT-1)) == 0; } @@ -119,9 +119,7 @@ EASY_FORCE_INLINE void unaligned_zero32(void* ptr) EASY_FORCE_INLINE void unaligned_zero64(void* ptr) { -#ifndef EASY_ENABLE_STRICT_ALIGNMENT - *(uint64_t*)ptr = 0; -#else +#ifdef EASY_ENABLE_STRICT_ALIGNMENT // Assume unaligned is more common. if (!is_aligned(ptr)) { ((char*)ptr)[0] = 0; @@ -133,20 +131,21 @@ EASY_FORCE_INLINE void unaligned_zero64(void* ptr) ((char*)ptr)[6] = 0; ((char*)ptr)[7] = 0; } - else { - *(uint64_t*)ptr = 0; - } + else #endif + + *(uint64_t*)ptr = 0; } template EASY_FORCE_INLINE void unaligned_store16(void* ptr, T val) { static_assert(sizeof(T) == 2, "16 bit type required."); + #ifndef EASY_ENABLE_STRICT_ALIGNMENT *(T*)ptr = val; #else - const char* const temp = (char*)&val; + const char* const temp = (const char*)&val; ((char*)ptr)[0] = temp[0]; ((char*)ptr)[1] = temp[1]; #endif @@ -156,10 +155,11 @@ template EASY_FORCE_INLINE void unaligned_store32(void* ptr, T val) { static_assert(sizeof(T) == 4, "32 bit type required."); + #ifndef EASY_ENABLE_STRICT_ALIGNMENT *(T*)ptr = val; #else - const char* const temp = (char*)&val; + const char* const temp = (const char*)&val; ((char*)ptr)[0] = temp[0]; ((char*)ptr)[1] = temp[1]; ((char*)ptr)[2] = temp[2]; @@ -171,12 +171,11 @@ template EASY_FORCE_INLINE void unaligned_store64(void* ptr, T val) { static_assert(sizeof(T) == 8, "64 bit type required."); -#ifndef EASY_ENABLE_STRICT_ALIGNMENT - *(T*)ptr = val; -#else - const char* const temp = (char*)&val; + +#ifdef EASY_ENABLE_STRICT_ALIGNMENT // Assume unaligned is more common. if (!is_aligned(ptr)) { + const char* const temp = (const char*)&val; ((char*)ptr)[0] = temp[0]; ((char*)ptr)[1] = temp[1]; ((char*)ptr)[2] = temp[2]; @@ -186,22 +185,23 @@ EASY_FORCE_INLINE void unaligned_store64(void* ptr, T val) ((char*)ptr)[6] = temp[6]; ((char*)ptr)[7] = temp[7]; } - else { - *(T*)ptr = val; - } + else #endif + + *(T*)ptr = val; } template EASY_FORCE_INLINE T unaligned_load16(const void* ptr) { static_assert(sizeof(T) == 2, "16 bit type required."); + #ifndef EASY_ENABLE_STRICT_ALIGNMENT - return *(T*)ptr; + return *(const T*)ptr; #else T value; - ((char*)&value)[0] = ((char*)ptr)[0]; - ((char*)&value)[1] = ((char*)ptr)[1]; + ((char*)&value)[0] = ((const char*)ptr)[0]; + ((char*)&value)[1] = ((const char*)ptr)[1]; return value; #endif } @@ -210,28 +210,30 @@ template EASY_FORCE_INLINE T unaligned_load16(const void* ptr, T* val) { static_assert(sizeof(T) == 2, "16 bit type required."); + #ifndef EASY_ENABLE_STRICT_ALIGNMENT - *val = *(T*)ptr; - return *val; + *val = *(const T*)ptr; #else - ((char*)val)[0] = ((char*)ptr)[0]; - ((char*)val)[1] = ((char*)ptr)[1]; - return *val; + ((char*)val)[0] = ((const char*)ptr)[0]; + ((char*)val)[1] = ((const char*)ptr)[1]; #endif + + return *val; } template EASY_FORCE_INLINE T unaligned_load32(const void* ptr) { static_assert(sizeof(T) == 4, "32 bit type required."); + #ifndef EASY_ENABLE_STRICT_ALIGNMENT - return *(T*)ptr; + return *(const T*)ptr; #else T value; - ((char*)&value)[0] = ((char*)ptr)[0]; - ((char*)&value)[1] = ((char*)ptr)[1]; - ((char*)&value)[2] = ((char*)ptr)[2]; - ((char*)&value)[3] = ((char*)ptr)[3]; + ((char*)&value)[0] = ((const char*)ptr)[0]; + ((char*)&value)[1] = ((const char*)ptr)[1]; + ((char*)&value)[2] = ((const char*)ptr)[2]; + ((char*)&value)[3] = ((const char*)ptr)[3]; return value; #endif } @@ -240,65 +242,63 @@ template EASY_FORCE_INLINE T unaligned_load32(const void* ptr, T* val) { static_assert(sizeof(T) == 4, "32 bit type required."); + #ifndef EASY_ENABLE_STRICT_ALIGNMENT - *val = *(T*)ptr; + *val = *(const T*)ptr; #else - ((char*)&val)[0] = ((char*)ptr)[0]; - ((char*)&val)[1] = ((char*)ptr)[1]; - ((char*)&val)[2] = ((char*)ptr)[2]; - ((char*)&val)[3] = ((char*)ptr)[3]; - return *val; + ((char*)&val)[0] = ((const char*)ptr)[0]; + ((char*)&val)[1] = ((const char*)ptr)[1]; + ((char*)&val)[2] = ((const char*)ptr)[2]; + ((char*)&val)[3] = ((const char*)ptr)[3]; #endif + + return *val; } template EASY_FORCE_INLINE T unaligned_load64(const void* ptr) { static_assert(sizeof(T) == 8, "64 bit type required."); -#ifndef EASY_ENABLE_STRICT_ALIGNMENT - return *(T*)ptr; -#else + +#ifdef EASY_ENABLE_STRICT_ALIGNMENT if (!is_aligned(ptr)) { T value; - ((char*)&value)[0] = ((char*)ptr)[0]; - ((char*)&value)[1] = ((char*)ptr)[1]; - ((char*)&value)[2] = ((char*)ptr)[2]; - ((char*)&value)[3] = ((char*)ptr)[3]; - ((char*)&value)[4] = ((char*)ptr)[4]; - ((char*)&value)[5] = ((char*)ptr)[5]; - ((char*)&value)[6] = ((char*)ptr)[6]; - ((char*)&value)[7] = ((char*)ptr)[7]; + ((char*)&value)[0] = ((const char*)ptr)[0]; + ((char*)&value)[1] = ((const char*)ptr)[1]; + ((char*)&value)[2] = ((const char*)ptr)[2]; + ((char*)&value)[3] = ((const char*)ptr)[3]; + ((char*)&value)[4] = ((const char*)ptr)[4]; + ((char*)&value)[5] = ((const char*)ptr)[5]; + ((char*)&value)[6] = ((const char*)ptr)[6]; + ((char*)&value)[7] = ((const char*)ptr)[7]; return value; } - else { - return *(T*)ptr; - } #endif + + return *(const T*)ptr; } template EASY_FORCE_INLINE T unaligned_load64(const void* ptr, T* val) { static_assert(sizeof(T) == 8, "64 bit type required."); -#ifndef EASY_ENABLE_STRICT_ALIGNMENT - *val = *(T*)ptr; -#else + +#ifdef EASY_ENABLE_STRICT_ALIGNMENT if (!is_aligned(ptr)) { - ((char*)&val)[0] = ((char*)ptr)[0]; - ((char*)&val)[1] = ((char*)ptr)[1]; - ((char*)&val)[2] = ((char*)ptr)[2]; - ((char*)&val)[3] = ((char*)ptr)[3]; - ((char*)&val)[4] = ((char*)ptr)[4]; - ((char*)&val)[5] = ((char*)ptr)[5]; - ((char*)&val)[6] = ((char*)ptr)[6]; - ((char*)&val)[7] = ((char*)ptr)[7]; - return *val; - } - else { - *val = *(T*)ptr; - return *val; + ((char*)&val)[0] = ((const char*)ptr)[0]; + ((char*)&val)[1] = ((const char*)ptr)[1]; + ((char*)&val)[2] = ((const char*)ptr)[2]; + ((char*)&val)[3] = ((const char*)ptr)[3]; + ((char*)&val)[4] = ((const char*)ptr)[4]; + ((char*)&val)[5] = ((const char*)ptr)[5]; + ((char*)&val)[6] = ((const char*)ptr)[6]; + ((char*)&val)[7] = ((const char*)ptr)[7]; } + else #endif + + *val = *(const T*)ptr; + return *val; } ////////////////////////////////////////////////////////////////////////// @@ -380,16 +380,19 @@ class chunk_allocator }; // Used in serialize(): workaround for no constexpr support in MSVC 2013. - static const int_fast32_t MAX_CHUNK_OFFSET = N - sizeof(uint16_t); - static const uint16_t N_MINUS_ONE = N - 1; + EASY_STATIC_CONSTEXPR int_fast32_t MAX_CHUNK_OFFSET = N - sizeof(uint16_t); + EASY_STATIC_CONSTEXPR uint16_t N_MINUS_ONE = N - 1; - chunk_list m_chunks; ///< List of chunks. - uint32_t m_size; ///< Number of elements stored(# of times allocate() has been called.) - uint16_t m_chunkOffset; ///< Number of bytes used in the current chunk. + chunk_list m_chunks; ///< List of chunks. + const chunk* m_markedChunk; ///< Chunk marked by last closed frame + uint32_t m_size; ///< Number of elements stored(# of times allocate() has been called.) + uint32_t m_markedSize; ///< Number of elements to the moment when put_mark() has been called. + uint16_t m_chunkOffset; ///< Number of bytes used in the current chunk. + uint16_t m_markedChunkOffset; ///< Last byte in marked chunk for serializing. public: - chunk_allocator() : m_size(0), m_chunkOffset(0) + chunk_allocator() : m_markedChunk(nullptr), m_size(0), m_markedSize(0), m_chunkOffset(0), m_markedChunkOffset(0) { } @@ -451,10 +454,22 @@ public: return m_size == 0; } + uint32_t markedSize() const + { + return m_markedSize; + } + + bool markedEmpty() const + { + return m_markedSize == 0; + } + void clear() { m_size = 0; + m_markedSize = 0; m_chunkOffset = 0; + m_markedChunk = nullptr; m_chunks.clear_all_except_last(); // There is always at least one chunk } @@ -479,11 +494,18 @@ public: // too small to cary more than a zero-sized element. chunk* current = m_chunks.last; + bool isMarked; do { + + isMarked = (current == m_markedChunk); const char* data = current->data; + + const int_fast32_t maxOffset = isMarked ? m_markedChunkOffset : MAX_CHUNK_OFFSET; int_fast32_t chunkOffset = 0; // signed int so overflow is not checked. - uint16_t payloadSize = unaligned_load16(data); - while (chunkOffset < MAX_CHUNK_OFFSET && payloadSize != 0) { + auto payloadSize = unaligned_load16(data); + + while (chunkOffset < maxOffset && payloadSize != 0) + { const uint16_t chunkSize = sizeof(uint16_t) + payloadSize; _outputStream.write(data, chunkSize); data += chunkSize; @@ -492,11 +514,19 @@ public: } current = current->prev; - } while (current != nullptr); + + } while (current != nullptr && !isMarked); clear(); } + void put_mark() + { + m_markedChunk = m_chunks.last; + m_markedSize = m_size; + m_markedChunkOffset = m_chunkOffset; + } + private: chunk_allocator(const chunk_allocator&) = delete; diff --git a/easy_profiler_core/profile_manager.cpp b/easy_profiler_core/profile_manager.cpp index d61246d..b09ecb6 100644 --- a/easy_profiler_core/profile_manager.cpp +++ b/easy_profiler_core/profile_manager.cpp @@ -160,9 +160,9 @@ extern const uint32_t EASY_CURRENT_VERSION = EASY_VERSION_INT(EASY_PROFILER_VERS ////////////////////////////////////////////////////////////////////////// -# define EASY_PROF_DISABLED 0 -# define EASY_PROF_ENABLED 1 -# define EASY_PROF_DUMP 2 +# define EASY_PROF_DISABLED false//0 +# define EASY_PROF_ENABLED true//1 +//# define EASY_PROF_DUMP 2 ////////////////////////////////////////////////////////////////////////// @@ -644,7 +644,7 @@ ThreadGuard::~ThreadGuard() { bool isMarked = false; EASY_EVENT_RES(isMarked, "ThreadFinished", EASY_COLOR_THREAD_END, ::profiler::FORCE_ON); - THIS_THREAD->profiledFrameOpened.store(false, std::memory_order_release); + THIS_THREAD->markProfilingFrameEnded(); THIS_THREAD->expired.store(isMarked ? 2 : 1, std::memory_order_release); THIS_THREAD = nullptr; } @@ -810,12 +810,13 @@ bool ProfileManager::storeBlock(const profiler::BaseBlockDescriptor* _desc, cons if (state == EASY_PROF_DISABLED || (_desc->m_status & profiler::ON) == 0) return false; - if (state == EASY_PROF_DUMP) + /*if (state == EASY_PROF_DUMP) { if (THIS_THREAD == nullptr || THIS_THREAD->halt) return false; } - else if (THIS_THREAD == nullptr) + else*/ + if (THIS_THREAD == nullptr) { registerThread(); } @@ -837,12 +838,13 @@ bool ProfileManager::storeBlock(const profiler::BaseBlockDescriptor* _desc, cons if (state == EASY_PROF_DISABLED || (_desc->m_status & profiler::ON) == 0) return false; - if (state == EASY_PROF_DUMP) + /*if (state == EASY_PROF_DUMP) { if (THIS_THREAD == nullptr || THIS_THREAD->halt) return false; } - else if (THIS_THREAD == nullptr) + else*/ + if (THIS_THREAD == nullptr) { registerThread(); } @@ -915,18 +917,19 @@ void ProfileManager::beginBlock(Block& _block) bool empty = true; const auto state = m_profilerStatus.load(std::memory_order_acquire); - switch (state) - { - case EASY_PROF_DISABLED: + if (state == + //switch (state) + //{ + EASY_PROF_DISABLED) { _block.m_status = profiler::OFF; - THIS_THREAD->halt = false; + //THIS_THREAD->halt = false; THIS_THREAD->blocks.openedList.emplace_back(_block); beginFrame(); return; } - case EASY_PROF_DUMP: + /*case EASY_PROF_DUMP: { const bool halt = THIS_THREAD->halt; if (halt || THIS_THREAD->blocks.openedList.empty()) @@ -945,17 +948,18 @@ void ProfileManager::beginBlock(Block& _block) empty = false; break; - } + }*/ - default: + //default: + else { empty = THIS_THREAD->blocks.openedList.empty(); - break; + //break; } - } + //} THIS_THREAD->stackSize = 0; - THIS_THREAD->halt = false; + //THIS_THREAD->halt = false; auto blockStatus = _block.m_status; #if EASY_ENABLE_BLOCK_STATUS != 0 @@ -981,7 +985,7 @@ void ProfileManager::beginBlock(Block& _block) if (empty) { beginFrame(); - THIS_THREAD->profiledFrameOpened.store(true, std::memory_order_release); + THIS_THREAD->markProfilingFrameStarted(); } THIS_THREAD->blocks.openedList.emplace_back(_block); @@ -1017,7 +1021,7 @@ void ProfileManager::endBlock() } THIS_THREAD->stackSize = 0; - if (THIS_THREAD->halt || m_profilerStatus.load(std::memory_order_acquire) == EASY_PROF_DISABLED) + if (/*THIS_THREAD->halt ||*/ m_profilerStatus.load(std::memory_order_acquire) == EASY_PROF_DISABLED) { THIS_THREAD->popSilent(); endFrame(); @@ -1046,7 +1050,7 @@ void ProfileManager::endBlock() const bool empty = THIS_THREAD->blocks.openedList.empty(); if (empty) { - THIS_THREAD->profiledFrameOpened.store(false, std::memory_order_release); + THIS_THREAD->markProfilingFrameEnded(); endFrame(); #if EASY_ENABLE_BLOCK_STATUS != 0 THIS_THREAD->allowChildren = true; @@ -1192,7 +1196,7 @@ void ProfileManager::setEnabled(bool isEnable) auto time = getCurrentTime(); const auto status = isEnable ? EASY_PROF_ENABLED : EASY_PROF_DISABLED; - const auto prev = m_profilerStatus.exchange(status, std::memory_order_release); + const auto prev = m_profilerStatus.exchange(status, std::memory_order_acq_rel); if (prev == status) return; @@ -1286,24 +1290,19 @@ uint32_t ProfileManager::dumpBlocksToStream(profiler::OStream& _outputStream, bo #endif if (state == EASY_PROF_ENABLED) { - m_profilerStatus.store(EASY_PROF_DUMP, std::memory_order_release); + //m_profilerStatus.store(EASY_PROF_DUMP, std::memory_order_release); + m_profilerStatus.store(EASY_PROF_DISABLED, std::memory_order_release); disableEventTracer(); m_endTime = getCurrentTime(); } - - // This is to make sure that no new descriptors or new threads will be - // added until we finish sending data. - //m_spin.lock(); - // This is the only place using both spins, so no dead-lock will occur - if (_async && m_stopDumping.load(std::memory_order_acquire)) { if (_lockSpin) m_dumpSpin.unlock(); return 0; } - +/* // Wait for some time to be sure that all operations which began before setEnabled(false) will be finished. // This is much better than inserting spin-lock or atomic variable check into each store operation. std::this_thread::sleep_for(std::chrono::milliseconds(20)); @@ -1345,9 +1344,13 @@ uint32_t ProfileManager::dumpBlocksToStream(profiler::OStream& _outputStream, bo EASY_LOGMSG("All threads have closed frames\n"); EASY_LOGMSG("Disabled profiling\n"); +*/ + // This is to make sure that no new descriptors or new threads will be + // added until we finish sending data. m_spin.lock(); m_storedSpin.lock(); + // This is the only place using both spins, so no dead-lock will occur // TODO: think about better solution because this one is not 100% safe... const profiler::timestamp_t now = getCurrentTime(); @@ -1421,7 +1424,7 @@ uint32_t ProfileManager::dumpBlocksToStream(profiler::OStream& _outputStream, bo } auto& thread = thread_it->second; - uint32_t num = static_cast(thread.blocks.closedList.size()) + static_cast(thread.sync.closedList.size()); + uint32_t num = thread.blocks.closedList.markedSize() + thread.sync.closedList.size(); const char expired = ProfileManager::checkThreadExpired(thread); #ifdef _WIN32 @@ -1523,8 +1526,8 @@ uint32_t ProfileManager::dumpBlocksToStream(profiler::OStream& _outputStream, bo if (!thread.sync.closedList.empty()) thread.sync.closedList.serialize(_outputStream); - _outputStream.write(thread.blocks.closedList.size()); - if (!thread.blocks.closedList.empty()) + _outputStream.write(thread.blocks.closedList.markedSize()); + if (!thread.blocks.closedList.markedEmpty()) thread.blocks.closedList.serialize(_outputStream); thread.clearClosed(); @@ -1680,8 +1683,6 @@ void ProfileManager::stopListen() if (m_listenThread.joinable()) m_listenThread.join(); m_isAlreadyListening.store(false, std::memory_order_release); - - EASY_LOGMSG("Listening stopped\n"); } bool ProfileManager::isListening() const @@ -1873,7 +1874,8 @@ void ProfileManager::listen(uint16_t _port) m_dumpSpin.lock(); auto time = getCurrentTime(); - const auto prev = m_profilerStatus.exchange(EASY_PROF_DUMP, std::memory_order_release); + //const auto prev = m_profilerStatus.exchange(EASY_PROF_DUMP, std::memory_order_release); + const auto prev = m_profilerStatus.exchange(EASY_PROF_DISABLED, std::memory_order_acq_rel); if (prev == EASY_PROF_ENABLED) { disableEventTracer(); m_endTime = time; @@ -2009,6 +2011,8 @@ void ProfileManager::listen(uint16_t _port) m_stopDumping.store(true, std::memory_order_release); join(dumpingResult); } + + EASY_LOGMSG("Listening stopped\n"); } ////////////////////////////////////////////////////////////////////////// diff --git a/easy_profiler_core/profile_manager.h b/easy_profiler_core/profile_manager.h index 963f9da..ebed050 100644 --- a/easy_profiler_core/profile_manager.h +++ b/easy_profiler_core/profile_manager.h @@ -104,7 +104,7 @@ class ProfileManager profiler::spin_lock m_storedSpin; profiler::spin_lock m_dumpSpin; std::atomic m_mainThreadId; - std::atomic m_profilerStatus; + std::atomic_bool m_profilerStatus; std::atomic_bool m_isEventTracingEnabled; std::atomic_bool m_isAlreadyListening; std::atomic_bool m_frameMaxReset; diff --git a/easy_profiler_core/thread_storage.cpp b/easy_profiler_core/thread_storage.cpp index beb6839..4acb832 100644 --- a/easy_profiler_core/thread_storage.cpp +++ b/easy_profiler_core/thread_storage.cpp @@ -53,7 +53,7 @@ ThreadStorage::ThreadStorage() , named(false) , guarded(false) , frameOpened(false) - , halt(false) + //, halt(false) { expired = ATOMIC_VAR_INIT(0); profiledFrameOpened = ATOMIC_VAR_INIT(false); @@ -69,7 +69,7 @@ void ThreadStorage::storeValue(profiler::timestamp_t _timestamp, profiler::block char* cdata = reinterpret_cast(data); memcpy(cdata + sizeof(profiler::ArbitraryValue), _data, _size); - blocks.usedMemorySize += serializedDataSize; + blocks.frameMemorySize += serializedDataSize; } void ThreadStorage::storeBlock(const profiler::Block& block) @@ -98,7 +98,7 @@ void ThreadStorage::storeBlock(const profiler::Block& block) #endif ::new (data) profiler::SerializedBlock(block, nameLength); - blocks.usedMemorySize += serializedDataSize; + blocks.frameMemorySize += serializedDataSize; #if EASY_OPTION_MEASURE_STORAGE_EXPAND != 0 if (expanded) @@ -109,7 +109,7 @@ void ThreadStorage::storeBlock(const profiler::Block& block) serializedDataSize = static_cast(sizeof(profiler::BaseBlockData) + 1); data = blocks.closedList.allocate(serializedDataSize); ::new (data) profiler::SerializedBlock(b, 0); - blocks.usedMemorySize += serializedDataSize; + blocks.frameMemorySize += serializedDataSize; } #endif } @@ -155,3 +155,16 @@ profiler::timestamp_t ThreadStorage::endFrame() frameOpened = false; return getCurrentTime() - frameStartTime; } + +void ThreadStorage::markProfilingFrameStarted() +{ + profiledFrameOpened.store(true, std::memory_order_release); +} + +void ThreadStorage::markProfilingFrameEnded() +{ + profiledFrameOpened.store(false, std::memory_order_release); + blocks.closedList.put_mark(); + blocks.usedMemorySize += blocks.frameMemorySize; + blocks.frameMemorySize = 0; +} diff --git a/easy_profiler_core/thread_storage.h b/easy_profiler_core/thread_storage.h index bdfc08f..1c63832 100644 --- a/easy_profiler_core/thread_storage.h +++ b/easy_profiler_core/thread_storage.h @@ -63,10 +63,12 @@ struct BlocksList std::vector openedList; chunk_allocator closedList; uint64_t usedMemorySize = 0; + uint64_t frameMemorySize = 0; void clearClosed() { //closedList.clear(); usedMemorySize = 0; + frameMemorySize = 0; } private: @@ -90,8 +92,8 @@ public: ////////////////////////////////////////////////////////////////////////// -const uint16_t SIZEOF_BLOCK = sizeof(profiler::BaseBlockData) + 1 + sizeof(uint16_t); // SerializedBlock stores BaseBlockData + at least 1 character for name ('\0') + 2 bytes for size of serialized data -const uint16_t SIZEOF_CSWITCH = sizeof(profiler::CSwitchEvent) + 1 + sizeof(uint16_t); // SerializedCSwitch also stores additional 4 bytes to be able to save 64-bit thread_id +EASY_CONSTEXPR uint16_t SIZEOF_BLOCK = sizeof(profiler::BaseBlockData) + 1 + sizeof(uint16_t); // SerializedBlock stores BaseBlockData + at least 1 character for name ('\0') + 2 bytes for size of serialized data +EASY_CONSTEXPR uint16_t SIZEOF_CSWITCH = sizeof(profiler::CSwitchEvent) + 1 + sizeof(uint16_t); // SerializedCSwitch also stores additional 4 bytes to be able to save 64-bit thread_id struct ThreadStorage EASY_FINAL { @@ -109,7 +111,7 @@ struct ThreadStorage EASY_FINAL bool named; ///< True if thread name was set bool guarded; ///< True if thread has been registered using ThreadGuard bool frameOpened; ///< Is new frame opened (this does not depend on profiling status) \sa profiledFrameOpened - bool halt; ///< This is set to true when new frame started while dumping blocks. Used to restrict collecting blocks during dumping process. + //bool halt; ///< This is set to true when new frame started while dumping blocks. Used to restrict collecting blocks during dumping process. void storeValue(profiler::timestamp_t _timestamp, profiler::block_id_t _id, profiler::DataType _type, const void* _data, size_t _size, bool _isArray, profiler::ValueId _vin); void storeBlock(const profiler::Block& _block); @@ -119,6 +121,8 @@ struct ThreadStorage EASY_FINAL void beginFrame(); profiler::timestamp_t endFrame(); + void markProfilingFrameStarted(); + void markProfilingFrameEnded(); ThreadStorage();