/** * @file lltraceaccumulators.h * @brief Storage for accumulating statistics * * $LicenseInfo:firstyear=2001&license=viewerlgpl$ * Second Life Viewer Source Code * Copyright (C) 2012, Linden Research, Inc. * * This library is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public * License as published by the Free Software Foundation; * version 2.1 of the License only. * * This library is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * Lesser General Public License for more details. * * You should have received a copy of the GNU Lesser General Public * License along with this library; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA * * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA * $/LicenseInfo$ */ #ifndef LL_LLTRACEACCUMULATORS_H #define LL_LLTRACEACCUMULATORS_H #include "stdtypes.h" #include "llpreprocessor.h" #include "llunits.h" #include "lltimer.h" #include "llrefcount.h" #include "llthreadlocalstorage.h" #include "llmemory.h" #include namespace LLTrace { const F64 NaN = std::numeric_limits::quiet_NaN(); enum EBufferAppendType { SEQUENTIAL, NON_SEQUENTIAL }; template class AccumulatorBuffer : public LLRefCount { typedef AccumulatorBuffer self_t; static const S32 ACCUMULATOR_BUFFER_SIZE_INCREMENT = 16; private: struct StaticAllocationMarker { }; AccumulatorBuffer(StaticAllocationMarker m) : mStorageSize(0), mStorage(NULL) {} public: AccumulatorBuffer(const AccumulatorBuffer& other = *getDefaultBuffer()) : mStorageSize(0), mStorage(NULL) { resize(other.mStorageSize); for (S32 i = 0; i < sNextStorageSlot; i++) { mStorage[i] = other.mStorage[i]; } } ~AccumulatorBuffer() { if (isCurrent()) { LLThreadLocalSingletonPointer::setInstance(NULL); } delete[] mStorage; } LL_FORCE_INLINE ACCUMULATOR& operator[](size_t index) { return mStorage[index]; } LL_FORCE_INLINE const ACCUMULATOR& operator[](size_t index) const { return mStorage[index]; } void addSamples(const AccumulatorBuffer& other, EBufferAppendType append_type) { llassert(mStorageSize >= sNextStorageSlot && other.mStorageSize >= sNextStorageSlot); for (size_t i = 0; i < sNextStorageSlot; i++) { mStorage[i].addSamples(other.mStorage[i], append_type); } } void copyFrom(const AccumulatorBuffer& other) { llassert(mStorageSize >= sNextStorageSlot && other.mStorageSize >= sNextStorageSlot); for (size_t i = 0; i < sNextStorageSlot; i++) { mStorage[i] = other.mStorage[i]; } } void reset(const AccumulatorBuffer* other = NULL) { llassert(mStorageSize >= sNextStorageSlot); for (size_t i = 0; i < sNextStorageSlot; i++) { mStorage[i].reset(other ? &other->mStorage[i] : NULL); } } void sync(F64SecondsImplicit time_stamp) { llassert(mStorageSize >= sNextStorageSlot); for (size_t i = 0; i < sNextStorageSlot; i++) { mStorage[i].sync(time_stamp); } } void makeCurrent() { LLThreadLocalSingletonPointer::setInstance(mStorage); } bool isCurrent() const { return LLThreadLocalSingletonPointer::getInstance() == mStorage; } static void clearCurrent() { LLThreadLocalSingletonPointer::setInstance(NULL); } // NOTE: this is not thread-safe. We assume that slots are reserved in the main thread before any child threads are spawned size_t reserveSlot() { size_t next_slot = sNextStorageSlot++; if (next_slot >= mStorageSize) { // don't perform doubling, as this should only happen during startup // want to keep a tight bounds as we will have a lot of these buffers resize(mStorageSize + ACCUMULATOR_BUFFER_SIZE_INCREMENT); } llassert(mStorage && next_slot < mStorageSize); return next_slot; } void resize(size_t new_size) { if (new_size <= mStorageSize) return; ACCUMULATOR* old_storage = mStorage; mStorage = new ACCUMULATOR[new_size]; if (old_storage) { for (S32 i = 0; i < mStorageSize; i++) { mStorage[i] = old_storage[i]; } } mStorageSize = new_size; delete[] old_storage; self_t* default_buffer = getDefaultBuffer(); if (this != default_buffer && new_size > default_buffer->size()) { //NB: this is not thread safe, but we assume that all resizing occurs during static initialization default_buffer->resize(new_size); } } size_t size() const { return getNumIndices(); } static size_t getNumIndices() { return sNextStorageSlot; } static self_t* getDefaultBuffer() { static bool sInitialized = false; if (!sInitialized) { // this buffer is allowed to leak so that trace calls from global destructors have somewhere to put their data // so as not to trigger an access violation sDefaultBuffer = new AccumulatorBuffer(StaticAllocationMarker()); sInitialized = true; sDefaultBuffer->resize(ACCUMULATOR_BUFFER_SIZE_INCREMENT); } return sDefaultBuffer; } private: ACCUMULATOR* mStorage; size_t mStorageSize; static size_t sNextStorageSlot; static self_t* sDefaultBuffer; }; template size_t AccumulatorBuffer::sNextStorageSlot = 0; template AccumulatorBuffer* AccumulatorBuffer::sDefaultBuffer = NULL; class EventAccumulator { public: typedef F64 value_t; EventAccumulator() : mSum(0), mMin(NaN), mMax(NaN), mMean(NaN), mSumOfSquares(0), mNumSamples(0), mLastValue(NaN) {} void record(F64 value) { if (mNumSamples == 0) { mSum = value; mMean = value; mMin = value; mMax = value; } else { mSum += value; F64 old_mean = mMean; mMean += (value - old_mean) / (F64)mNumSamples; mSumOfSquares += (value - old_mean) * (value - mMean); if (value < mMin) { mMin = value; } else if (value > mMax) { mMax = value; } } mNumSamples++; mLastValue = value; } void addSamples(const EventAccumulator& other, EBufferAppendType append_type); void reset(const EventAccumulator* other); void sync(F64SecondsImplicit) {} F64 getSum() const { return mSum; } F64 getMin() const { return mMin; } F64 getMax() const { return mMax; } F64 getLastValue() const { return mLastValue; } F64 getMean() const { return mMean; } F64 getStandardDeviation() const { return sqrtf(mSumOfSquares / mNumSamples); } F64 getSumOfSquares() const { return mSumOfSquares; } S32 getSampleCount() const { return mNumSamples; } bool hasValue() const { return mNumSamples > 0; } private: F64 mSum, mMin, mMax, mLastValue; F64 mMean, mSumOfSquares; S32 mNumSamples; }; class SampleAccumulator { public: typedef F64 value_t; SampleAccumulator() : mSum(0), mMin(NaN), mMax(NaN), mMean(NaN), mSumOfSquares(0), mLastSampleTimeStamp(0), mTotalSamplingTime(0), mNumSamples(0), mLastValue(NaN), mHasValue(false) {} void sample(F64 value) { F64SecondsImplicit time_stamp = LLTimer::getTotalSeconds(); // store effect of last value sync(time_stamp); if (!mHasValue) { mHasValue = true; mMin = value; mMax = value; mMean = value; mLastSampleTimeStamp = time_stamp; } else { if (value < mMin) { mMin = value; } else if (value > mMax) { mMax = value; } } mLastValue = value; mNumSamples++; } void addSamples(const SampleAccumulator& other, EBufferAppendType append_type); void reset(const SampleAccumulator* other); void sync(F64SecondsImplicit time_stamp) { if (mHasValue && time_stamp != mLastSampleTimeStamp) { F64SecondsImplicit delta_time = time_stamp - mLastSampleTimeStamp; mSum += mLastValue * delta_time; mTotalSamplingTime += delta_time; F64 old_mean = mMean; mMean += (delta_time / mTotalSamplingTime) * (mLastValue - old_mean); mSumOfSquares += delta_time * (mLastValue - old_mean) * (mLastValue - mMean); } mLastSampleTimeStamp = time_stamp; } F64 getSum() const { return mSum; } F64 getMin() const { return mMin; } F64 getMax() const { return mMax; } F64 getLastValue() const { return mLastValue; } F64 getMean() const { return mMean; } F64 getStandardDeviation() const { return sqrtf(mSumOfSquares / mTotalSamplingTime); } F64 getSumOfSquares() const { return mSumOfSquares; } F64SecondsImplicit getSamplingTime() { return mTotalSamplingTime; } S32 getSampleCount() const { return mNumSamples; } bool hasValue() const { return mHasValue; } private: F64 mSum, mMin, mMax, mLastValue; bool mHasValue; // distinct from mNumSamples, since we might have inherited an old sample F64 mMean, mSumOfSquares; F64SecondsImplicit mLastSampleTimeStamp, mTotalSamplingTime; S32 mNumSamples; }; class CountAccumulator { public: typedef F64 value_t; CountAccumulator() : mSum(0), mNumSamples(0) {} void add(F64 value) { mNumSamples++; mSum += value; } void addSamples(const CountAccumulator& other, EBufferAppendType /*type*/) { mSum += other.mSum; mNumSamples += other.mNumSamples; } void reset(const CountAccumulator* other) { mNumSamples = 0; mSum = 0; } void sync(F64SecondsImplicit) {} F64 getSum() const { return mSum; } S32 getSampleCount() const { return mNumSamples; } private: F64 mSum; S32 mNumSamples; }; class TimeBlockAccumulator { public: typedef F64Seconds value_t; typedef TimeBlockAccumulator self_t; // fake classes that allows us to view different facets of underlying statistic struct CallCountFacet { typedef S32 value_t; }; struct SelfTimeFacet { typedef F64Seconds value_t; }; // arrays are allocated with 32 byte alignment void *operator new [](size_t size) { return ll_aligned_malloc(32, size); } void operator delete[](void* ptr, size_t size) { ll_aligned_free(32, ptr); } TimeBlockAccumulator(); void addSamples(const self_t& other, EBufferAppendType append_type); void reset(const self_t* other); void sync(F64SecondsImplicit) {} // // members // U64 mTotalTimeCounter, mSelfTimeCounter; S32 mCalls; class TimeBlock* mParent; // last acknowledged parent of this time block class TimeBlock* mLastCaller; // used to bootstrap tree construction U16 mActiveCount; // number of timers with this ID active on stack bool mMoveUpTree; // needs to be moved up the tree of timers at the end of frame }; class TimeBlock; class TimeBlockTreeNode { public: TimeBlockTreeNode(); void setParent(TimeBlock* parent); TimeBlock* getParent() { return mParent; } TimeBlock* mBlock; TimeBlock* mParent; std::vector mChildren; bool mCollapsed; bool mNeedsSorting; }; struct BlockTimerStackRecord { class BlockTimer* mActiveTimer; class TimeBlock* mTimeBlock; U64 mChildTime; }; struct MemStatAccumulator { typedef MemStatAccumulator self_t; // fake classes that allows us to view different facets of underlying statistic struct AllocationFacet { typedef F64Bytes value_t; }; struct DeallocationFacet { typedef F64Bytes value_t; }; void addSamples(const MemStatAccumulator& other, EBufferAppendType append_type) { mFootprintAllocations.addSamples(other.mFootprintAllocations, append_type); mFootprintDeallocations.addSamples(other.mFootprintDeallocations, append_type); if (append_type == SEQUENTIAL) { mSize.addSamples(other.mSize, SEQUENTIAL); } else { F64 allocation_delta(other.mFootprintAllocations.getSum() - other.mFootprintDeallocations.getSum()); mSize.sample(mSize.hasValue() ? mSize.getLastValue() + allocation_delta : allocation_delta); } } void reset(const MemStatAccumulator* other) { mSize.reset(other ? &other->mSize : NULL); mFootprintAllocations.reset(other ? &other->mFootprintAllocations : NULL); mFootprintDeallocations.reset(other ? &other->mFootprintDeallocations : NULL); } void sync(F64SecondsImplicit time_stamp) { mSize.sync(time_stamp); } SampleAccumulator mSize; EventAccumulator mFootprintAllocations; CountAccumulator mFootprintDeallocations; }; struct AccumulatorBufferGroup : public LLRefCount { AccumulatorBufferGroup(); void handOffTo(AccumulatorBufferGroup& other); void makeCurrent(); bool isCurrent() const; static void clearCurrent(); void append(const AccumulatorBufferGroup& other); void merge(const AccumulatorBufferGroup& other); void reset(AccumulatorBufferGroup* other = NULL); void sync(); AccumulatorBuffer mCounts; AccumulatorBuffer mSamples; AccumulatorBuffer mEvents; AccumulatorBuffer mStackTimers; AccumulatorBuffer mMemStats; }; } #endif // LL_LLTRACEACCUMULATORS_H