diff --git a/Makefile b/Makefile index 0b113c1b5..68bc489c3 100644 --- a/Makefile +++ b/Makefile @@ -49,6 +49,7 @@ VALGRIND_VER := $(join $(VALGRIND_VER),valgrind) VALGRIND_OPTS = --error-exitcode=$(VALGRIND_ERROR) --leak-check=full TESTS = \ + autovector_test \ db_test \ table_properties_collector_test \ arena_test \ @@ -226,6 +227,9 @@ signal_test: util/signal_test.o $(LIBOBJECTS) arena_test: util/arena_test.o $(LIBOBJECTS) $(TESTHARNESS) $(CXX) util/arena_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS) +autovector_test: util/autovector_test.o $(LIBOBJECTS) $(TESTHARNESS) + $(CXX) util/autovector_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS) + table_properties_collector_test: db/table_properties_collector_test.o $(LIBOBJECTS) $(TESTHARNESS) $(CXX) db/table_properties_collector_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS) diff --git a/util/autovector.h b/util/autovector.h new file mode 100644 index 000000000..2b9cb40e9 --- /dev/null +++ b/util/autovector.h @@ -0,0 +1,329 @@ +// Copyright (c) 2013, Facebook, Inc. All rights reserved. +// This source code is licensed under the BSD-style license found in the +// LICENSE file in the root directory of this source tree. An additional grant +// of patent rights can be found in the PATENTS file in the same directory. +#pragma once + +#include +#include +#include +#include +#include + +namespace rocksdb { + +// A vector that leverages pre-allocated stack-based array to achieve better +// performance for array with small amount of items. +// +// The interface resembles that of vector, but with less features since we aim +// to solve the problem that we have in hand, rather than implementing a +// full-fledged generic container. +// +// Currently we don't support: +// * reserve()/shrink_to_fit()/resize() +// If used correctly, in most cases, people should not touch the +// underlying vector at all. +// * random insert()/erase(), please only use push_back()/pop_back(). +// * No move/swap operations. Each autovector instance has a +// stack-allocated array and if we want support move/swap operations, we +// need to copy the arrays other than just swapping the pointers. In this +// case we'll just explicitly forbid these operations since they may +// lead users to make false assumption by thinking they are inexpensive +// operations. +// +// Naming style of public methods almost follows that of the STL's. +template +class autovector { + public: + // General STL-style container member types. + typedef T value_type; + typedef typename std::vector::difference_type difference_type; + typedef typename std::vector::size_type size_type; + typedef value_type& reference; + typedef const value_type& const_reference; + typedef value_type* pointer; + typedef const value_type* const_pointer; + + // This class is the base for regular/const iterator + template + class iterator_impl { + public: + // -- iterator traits + typedef iterator_impl self_type; + typedef TValueType value_type; + typedef TValueType& reference; + typedef TValueType* pointer; + typedef typename TAutoVector::difference_type difference_type; + typedef std::random_access_iterator_tag iterator_category; + + iterator_impl(TAutoVector* vect, size_t index) + : vect_(vect) + , index_(index) { + }; + iterator_impl(const iterator_impl&) = default; + ~iterator_impl() { } + iterator_impl& operator=(const iterator_impl&) = default; + + // -- Advancement + // iterator++ + self_type& operator++() { + ++index_; + return *this; + } + + // ++iterator + self_type operator++(int) { + auto old = *this; + ++index_; + return old; + } + + // iterator-- + self_type& operator--() { + --index_; + return *this; + } + + // --iterator + self_type operator--(int) { + auto old = *this; + --index_; + return old; + } + + self_type operator-(difference_type len) { + return self_type(vect_, index_ - len); + } + + difference_type operator-(const self_type& other) { + assert(vect_ == other.vect_); + return index_ - other.index_; + } + + self_type operator+(difference_type len) { + return self_type(vect_, index_ + len); + } + + self_type& operator+=(difference_type len) { + index_ += len; + return *this; + } + + self_type& operator-=(difference_type len) { + index_ -= len; + return *this; + } + + // -- Reference + reference operator*() { + assert(vect_->size() >= index_); + return (*vect_)[index_]; + } + pointer operator->() { + assert(vect_->size() >= index_); + return &(*vect_)[index_]; + } + + // -- Logical Operators + bool operator==(const self_type& other) const { + assert(vect_ == other.vect_); + return index_ == other.index_; + } + + bool operator!=(const self_type& other) const { + return !(*this == other); + } + + bool operator>(const self_type& other) const { + assert(vect_ == other.vect_); + return index_ > other.index_; + } + + bool operator<(const self_type& other) const { + assert(vect_ == other.vect_); + return index_ < other.index_; + } + + bool operator>=(const self_type& other) const { + assert(vect_ == other.vect_); + return index_ >= other.index_; + } + + bool operator<=(const self_type& other) const { + assert(vect_ == other.vect_); + return index_ <= other.index_; + } + + private: + TAutoVector* vect_ = nullptr; + size_t index_ = 0; + }; + + typedef iterator_impl iterator; + typedef iterator_impl const_iterator; + typedef std::reverse_iterator reverse_iterator; + typedef std::reverse_iterator const_reverse_iterator; + + autovector() = default; + ~autovector() = default; + + // -- Immutable operations + // Indicate if all data resides in in-stack data structure. + bool only_in_stack() const { + // If no element was inserted at all, the vector's capacity will be `0`. + return vect_.capacity() == 0; + } + + size_type size() const { + return num_stack_items_ + vect_.size(); + } + + bool empty() const { + return size() == 0; + } + + // will not check boundry + const_reference operator[](size_type n) const { + return n < kSize ? values_[n] : vect_[n - kSize]; + } + + reference operator[](size_type n) { + return n < kSize ? values_[n] : vect_[n - kSize]; + } + + // will check boundry + const_reference at(size_type n) const { + if (n >= size()) { + throw std::out_of_range("autovector: index out of range"); + } + return (*this)[n]; + } + + reference at(size_type n) { + if (n >= size()) { + throw std::out_of_range("autovector: index out of range"); + } + return (*this)[n]; + } + + reference front() { + assert(!empty()); + return *begin(); + } + + const_reference front() const { + assert(!empty()); + return *begin(); + } + + reference back() { + assert(!empty()); + return *(end() - 1); + } + + const_reference back() const { + assert(!empty()); + return *(end() - 1); + } + + // -- Mutable Operations + void push_back(T&& item) { + if (num_stack_items_ < kSize) { + values_[num_stack_items_++] = std::move(item); + } else { + vect_.push_back(item); + } + } + + void push_back(const T& item) { + push_back(value_type(item)); + } + + template + void emplace_back(Args&&... args) { + push_back(value_type(args...)); + } + + void pop_back() { + assert(!empty()); + if (!vect_.empty()) { + vect_.pop_back(); + } else { + --num_stack_items_; + } + } + + void clear() { + num_stack_items_ = 0; + vect_.clear(); + } + + // -- Copy and Assignment + autovector& assign(const autovector& other); + + autovector(const autovector& other) { + assign(other); + } + + autovector& operator=(const autovector& other) { + return assign(other); + } + + // move operation are disallowed since it is very hard to make sure both + // autovectors are allocated from the same function stack. + autovector& operator=(autovector&& other) = delete; + autovector(autovector&& other) = delete; + + // -- Iterator Operations + iterator begin() { + return iterator(this, 0); + } + + const_iterator begin() const { + return const_iterator(this, 0); + } + + iterator end() { + return iterator(this, this->size()); + } + + const_iterator end() const { + return const_iterator(this, this->size()); + } + + reverse_iterator rbegin() { + return reverse_iterator(end()); + } + + const_reverse_iterator rbegin() const { + return const_reverse_iterator(end()); + } + + reverse_iterator rend() { + return reverse_iterator(begin()); + } + + const_reverse_iterator rend() const { + return const_reverse_iterator(begin()); + } + + private: + size_type num_stack_items_ = 0; // current number of items + value_type values_[kSize]; // the first `kSize` items + // used only if there are more than `kSize` items. + std::vector vect_; +}; + +template +autovector& autovector::assign(const autovector& other) { + // copy the internal vector + vect_.assign(other.vect_.begin(), other.vect_.end()); + + // copy array + num_stack_items_ = other.num_stack_items_; + std::copy(other.values_, other.values_ + num_stack_items_, values_); + + return *this; +} + +} // rocksdb diff --git a/util/autovector_test.cc b/util/autovector_test.cc new file mode 100644 index 000000000..31ce4ed19 --- /dev/null +++ b/util/autovector_test.cc @@ -0,0 +1,286 @@ +// Copyright (c) 2013, Facebook, Inc. All rights reserved. +// This source code is licensed under the BSD-style license found in the +// LICENSE file in the root directory of this source tree. An additional grant +// of patent rights can be found in the PATENTS file in the same directory. + +#include +#include + +#include "rocksdb/env.h" +#include "util/autovector.h" +#include "util/testharness.h" +#include "util/testutil.h" + +namespace rocksdb { + +using namespace std; + +class AutoVectorTest { }; + +const size_t kSize = 8; +TEST(AutoVectorTest, PushBackAndPopBack) { + autovector vec; + ASSERT_TRUE(vec.empty()); + ASSERT_EQ(0ul, vec.size()); + + for (size_t i = 0; i < 1000 * kSize; ++i) { + vec.push_back(i); + ASSERT_TRUE(!vec.empty()); + if (i < kSize) { + ASSERT_TRUE(vec.only_in_stack()); + } else { + ASSERT_TRUE(!vec.only_in_stack()); + } + ASSERT_EQ(i + 1, vec.size()); + ASSERT_EQ(i, vec[i]); + ASSERT_EQ(i, vec.at(i)); + } + + size_t size = vec.size(); + while (size != 0) { + vec.pop_back(); + // will always be in heap + ASSERT_TRUE(!vec.only_in_stack()); + ASSERT_EQ(--size, vec.size()); + } + + ASSERT_TRUE(vec.empty()); +} + +TEST(AutoVectorTest, EmplaceBack) { + typedef std::pair ValueType; + autovector vec; + + for (size_t i = 0; i < 1000 * kSize; ++i) { + vec.emplace_back(i, std::to_string(i + 123)); + ASSERT_TRUE(!vec.empty()); + if (i < kSize) { + ASSERT_TRUE(vec.only_in_stack()); + } else { + ASSERT_TRUE(!vec.only_in_stack()); + } + + ASSERT_EQ(i + 1, vec.size()); + ASSERT_EQ(i, vec[i].first); + ASSERT_EQ(std::to_string(i + 123), vec[i].second); + } + + vec.clear(); + ASSERT_TRUE(vec.empty()); + ASSERT_TRUE(!vec.only_in_stack()); +} + +void AssertEqual( + const autovector& a, const autovector& b) { + ASSERT_EQ(a.size(), b.size()); + ASSERT_EQ(a.empty(), b.empty()); + ASSERT_EQ(a.only_in_stack(), b.only_in_stack()); + for (size_t i = 0; i < a.size(); ++i) { + ASSERT_EQ(a[i], b[i]); + } +} + +TEST(AutoVectorTest, CopyAndAssignment) { + // Test both heap-allocated and stack-allocated cases. + for (auto size : { kSize / 2, kSize * 1000 }) { + autovector vec; + for (size_t i = 0; i < size; ++i) { + vec.push_back(i); + } + + { + autovector other; + other = vec; + AssertEqual(other, vec); + } + + { + autovector other(vec); + AssertEqual(other, vec); + } + } +} + +TEST(AutoVectorTest, Iterators) { + autovector vec; + for (size_t i = 0; i < kSize * 1000; ++i) { + vec.push_back(std::to_string(i)); + } + + // basic operator test + ASSERT_EQ(vec.front(), *vec.begin()); + ASSERT_EQ(vec.back(), *(vec.end() - 1)); + ASSERT_TRUE(vec.begin() < vec.end()); + + // non-const iterator + size_t index = 0; + for (const auto& item : vec) { + ASSERT_EQ(vec[index++], item); + } + + index = vec.size() - 1; + for (auto pos = vec.rbegin(); pos != vec.rend(); ++pos) { + ASSERT_EQ(vec[index--], *pos); + } + + // const iterator + const auto& cvec = vec; + index = 0; + for (const auto& item : cvec) { + ASSERT_EQ(cvec[index++], item); + } + + index = vec.size() - 1; + for (auto pos = cvec.rbegin(); pos != cvec.rend(); ++pos) { + ASSERT_EQ(cvec[index--], *pos); + } + + // forward and backward + auto pos = vec.begin(); + while (pos != vec.end()) { + auto old_val = *pos; + auto old = pos++; + // HACK: make sure -> works + ASSERT_TRUE(!old->empty()); + ASSERT_EQ(old_val, *old); + ASSERT_TRUE(old_val != *pos); + } + + pos = vec.begin(); + for (size_t i = 0; i < vec.size(); i += 2) { + // Cannot use ASSERT_EQ since that macro depends on iostream serialization + ASSERT_TRUE(pos + 2 - 2 == pos); + pos += 2; + ASSERT_TRUE(i + 2 == pos - vec.begin()); + ASSERT_TRUE(pos >= vec.begin()); + ASSERT_TRUE(pos <= vec.end()); + } +} + +vector GetTestKeys(size_t size) { + vector keys; + keys.resize(size); + + int index = 0; + for (auto& key : keys) { + key = "item-" + to_string(index++); + } + return keys; +} + +template +void BenchmarkVectorCreationAndInsertion( + string name, size_t ops, size_t item_size, + const std::vector& items) { + auto env = Env::Default(); + + int index = 0; + auto start_time = env->NowNanos(); + auto ops_remaining = ops; + while(ops_remaining--) { + TVector v; + for (size_t i = 0; i < item_size; ++i) { + v.push_back(items[index++]); + } + } + auto elapsed = env->NowNanos() - start_time; + cout << "created " << ops << " " << name << " instances:\n\t" + << "each was inserted with " << item_size << " elements\n\t" + << "total time elapsed: " << elapsed << " (ns)" << endl; +} + +template +void BenchmarkSequenceAccess(string name, size_t ops, size_t elem_size) { + TVector v; + for (const auto& item : GetTestKeys(elem_size)) { + v.push_back(item); + } + auto env = Env::Default(); + + auto ops_remaining = ops; + auto start_time = env->NowNanos(); + size_t total = 0; + while (ops_remaining--) { + auto end = v.end(); + for (auto pos = v.begin(); pos != end; ++pos) { + total += pos->size(); + } + } + auto elapsed = env->NowNanos() - start_time; + cout << "performed " << ops << " sequence access against " << name << "\n\t" + << "size: " << elem_size << "\n\t" + << "total time elapsed: " << elapsed << " (ns)" << endl; +} + +// This test case only reports the performance between std::vector +// and autovector. We chose string for comparison because in most +// o our use cases we used std::vector. +TEST(AutoVectorTest, PerfBench) { + // We run same operations for kOps times in order to get a more fair result. + size_t kOps = 100000; + + // Creation and insertion test + // Test the case when there is: + // * no element inserted: internal array of std::vector may not really get + // initialize. + // * one element inserted: internal array of std::vector must have + // initialized. + // * kSize elements inserted. This shows the most time we'll spend if we + // keep everything in stack. + // * 2 * kSize elements inserted. The internal vector of + // autovector must have been initialized. + cout << "=====================================================" << endl; + cout << "Creation and Insertion Test (value type: std::string)" << endl; + cout << "=====================================================" << endl; + + // pre-generated unique keys + auto string_keys = GetTestKeys(kOps * 2 * kSize); + for (auto insertions : { 0ul, 1ul, kSize / 2, kSize, 2 * kSize }) { + BenchmarkVectorCreationAndInsertion>( + "vector", kOps, insertions, string_keys + ); + BenchmarkVectorCreationAndInsertion>( + "autovector", kOps, insertions, string_keys + ); + cout << "-----------------------------------" << endl; + } + + cout << "=====================================================" << endl; + cout << "Creation and Insertion Test (value type: uint64_t)" << endl; + cout << "=====================================================" << endl; + + // pre-generated unique keys + vector int_keys(kOps * 2 * kSize); + for (size_t i = 0; i < kOps * 2 * kSize; ++i) { + int_keys[i] = i; + } + for (auto insertions : { 0ul, 1ul, kSize / 2, kSize, 2 * kSize }) { + BenchmarkVectorCreationAndInsertion>( + "vector", kOps, insertions, int_keys + ); + BenchmarkVectorCreationAndInsertion>( + "autovector", kOps, insertions, int_keys + ); + cout << "-----------------------------------" << endl; + } + + // Sequence Access Test + cout << "=====================================================" << endl; + cout << "Sequence Access Test" << endl; + cout << "=====================================================" << endl; + for (auto elem_size : { kSize / 2, kSize, 2 * kSize }) { + BenchmarkSequenceAccess>( + "vector", kOps, elem_size + ); + BenchmarkSequenceAccess>( + "autovector", kOps, elem_size + ); + cout << "-----------------------------------" << endl; + } +} + +} // namespace rocksdb + +int main(int argc, char** argv) { + return rocksdb::test::RunAllTests(); +}