C++ 性能优化实战：内存、CPU 与 I/O 效率提升

预分配减少碎片

频繁的小内存分配会导致堆碎片化。对于已知大小的容器，提前 reserve 能显著减少重分配开销。

#include <iostream>
#include <vector>
#include <chrono>

void preallocateMemory() {
    const int size = 10000;
    std::vector<int> vec;
    vec.reserve(size); // 预分配内存
    for (int i = 0; i < size; ++i) {
        vec.push_back(i);
    }
}

void notPreallocateMemory() {
    const int size = 10000;
    std::vector<int> vec;
    for (int i = 0; i < size; ++i) {
        vec.push_back(i);
    }
}

int main() {
    auto start = std::chrono::high_resolution_clock::now();
    preallocateMemory();
    auto end = std::chrono::high_resolution_clock::now();
    std::cout << "预分配耗时：" 
              << std::chrono::duration_cast<std::chrono::microseconds>(end - start).count() 
              << "微秒" << std::endl;
    return 0;
}

CPU 优化技巧

循环合并与缓存友好

多次遍历数组会增加缓存未命中概率。尽量将操作合并到一个循环中，提高数据局部性。

#include <iostream>
#include <vector>
#include <chrono>

void optimizedLoop() {
    const int size = 10000;
    std::vector<int> vec1(size, 1);
    std::vector<int> vec2(size, 2);
    std::vector<int> result(size, 0);
    for (int i = 0; i < size; ++i) {
        result[i] = vec1[i] + vec2[i]; // 单次遍历
    }
}

void unoptimizedLoop() {
    const int size = 10000;
    std::vector<int> vec1(size, 1);
    std::vector<int> vec2(size, 2);
    std::vector<int> result(size, 0);
    for (int i = 0; i < size; ++i) result[i] = vec1[i];
    for (int i = 0; i < size; ++i) result[i] += vec2[i]; // 两次遍历
}

int main() {
    auto start = std::chrono::high_resolution_clock::now();
    optimizedLoop();
    auto end = std::chrono::high_resolution_clock::now();
    std::cout << "优化循环耗时：" 
              << std::chrono::duration_cast<std::chrono::microseconds>(end - start).count() 
              << "微秒" << std::endl;
    return 0;
}

内联函数减少调用开销

对于高频调用的短函数，使用 inline 关键字或编译器优化可减少栈帧建立和返回的开销。

#include <iostream>
#include <chrono>

inline int add(int a, int b) {
    return a + b;
}

int addNotInline(int a, int b) {
    return a + b;
}

void testFunctionCallOverhead() {
    const int size = 1000000;
    int result = 0;
    auto start = std::chrono::high_resolution_clock::now();
    for (int i = 0; i < size; ++i) {
        result += add(i, i);
    }
    auto end = std::chrono::high_resolution_clock::now();
    std::cout << "内联函数调用耗时：" 
              << std::chrono::duration_cast<std::chrono::microseconds>(end - start).count() 
              << "微秒" << std::endl;
}

int main() {
    testFunctionCallOverhead();
    return 0;
}

I/O 操作优化

文件 I/O 缓冲策略

默认情况下流会缓冲，但在某些场景下显式控制缓冲区大小或禁用缓冲能提升吞吐量。注意，频繁关闭文件也是开销来源。

#include <iostream>
#include <fstream>
#include <chrono>

void optimizedFileIO() {
    const std::string filename = "test.txt";
    const int size = 10000;
    std::ofstream file(filename);
    // 实际生产中建议保留默认缓冲，此处仅为演示
    for (int i = 0; i < size; ++i) {
        file << i << std::endl;
    }
    file.close();
}

int main() {
    auto start = std::chrono::high_resolution_clock::now();
    optimizedFileIO();
    auto end = std::chrono::high_resolution_clock::now();
    std::cout << "文件 I/O 耗时：" 
              << std::chrono::duration_cast<std::chrono::milliseconds>(end - start).count() 
              << "毫秒" << std::endl;
    return 0;
}

网络 I/O 异步处理

同步阻塞会浪费线程资源。使用 Boost.Asio 等库实现异步 I/O 可以显著提升并发能力。

#include <iostream>
#include <boost/asio.hpp>
#include <boost/asio/ip/tcp.hpp>
#include <boost/asio/streambuf.hpp>
#include <sstream>

using boost::asio::ip::tcp;

void optimizedNetworkIO() {
    try {
        boost::asio::io_service io_service;
        tcp::resolver resolver(io_service);
        tcp::resolver::query query("example.com", "http");
        tcp::resolver::iterator endpoint_iterator = resolver.resolve(query);
        tcp::socket socket(io_service);
        boost::asio::connect(socket, endpoint_iterator);
        
        std::string request = "GET / HTTP/1.1\r\n";
        request += "Host: example.com\r\n";
        request += "Connection: close\r\n\r\n";
        
        boost::asio::write(socket, boost::asio::buffer(request));
        boost::asio::streambuf response;
        boost::asio::read_until(socket, response, "\r\n");
        
        std::istream is(&response);
        std::string status_line;
        is >> status_line;
    } catch (const std::exception& e) {
        std::cerr << "错误：" << e.what() << std::endl;
    }
}

int main() {
    optimizedNetworkIO();
    return 0;
}

综合案例：矩阵乘法算法优化

项目结构

MatrixMultiplicationOptimization/
├── include/
│   └── Matrix.h
├── src/
│   ├── Matrix.cpp
│   └── main.cpp
└── build/

核心代码

头文件中需声明转置方法以支持优化算法。

#ifndef MATRIX_H
#define MATRIX_H
#include <vector>
#include <chrono>

using namespace std;
using namespace chrono;

class Matrix {
public:
    Matrix(int rows, int cols);
    Matrix(const vector<vector<int>>& data);
    int getRows() const;
    int getCols() const;
    int& operator()(int row, int col);
    const int& operator()(int row, int col) const;
    Matrix multiplyNaive(const Matrix& other) const;
    Matrix multiplyOptimized(const Matrix& other) const;
    void print() const;
    vector<vector<int>> getTransposed() const; // 新增声明
private:
    int rows_;
    int cols_;
    vector<vector<int>> data_;
};
#endif

实现文件中包含转置逻辑及乘法对比。

#include "Matrix.h"
#include <iostream>
#include <stdexcept>

Matrix::Matrix(int rows, int cols)
    : rows_(rows), cols_(cols), data_(rows, vector<int>(cols, 0)) {}

Matrix::Matrix(const vector<vector<int>>& data)
    : rows_(data.size()), cols_(data[0].size()), data_(data) {}

int Matrix::getRows() const { return rows_; }
int Matrix::getCols() const { return cols_; }

int& Matrix::operator()(int row, int col) { return data_[row][col]; }
const int& Matrix::operator()(int row, int col) const { return data_[row][col]; }

Matrix Matrix::multiplyNaive(const Matrix& other) const {
    if (cols_ != other.rows_) throw invalid_argument("矩阵尺寸不兼容");
    Matrix result(rows_, other.cols_);
    for (int i = 0; i < rows_; ++i) {
        for (int j = 0; j < other.cols_; ++j) {
            for (int k = 0; k < cols_; ++k) {
                result(i, j) += data_[i][k] * other.data_[k][j];
            }
        }
    }
    return result;
}

Matrix Matrix::multiplyOptimized(const Matrix& other) const {
    if (cols_ != other.rows_) throw invalid_argument("矩阵尺寸不兼容");
    Matrix result(rows_, other.cols_);
    vector<vector<int>> otherTransposed = other.getTransposed();
    for (int i = 0; i < rows_; ++i) {
        for (int j = 0; j < other.cols_; ++j) {
            int sum = 0;
            for (int k = 0; k < cols_; ++k) {
                sum += data_[i][k] * otherTransposed[j][k];
            }
            result(i, j) = sum;
        }
    }
    return result;
}

vector<vector<int>> Matrix::getTransposed() const {
    vector<vector<int>> transposed(cols_, vector<int>(rows_));
    for (int i = 0; i < rows_; ++i) {
        for (int j = 0; j < cols_; ++j) {
            transposed[j][i] = data_[i][j];
        }
    }
    return transposed;
}

void Matrix::print() const {
    for (const auto& row : data_) {
        for (int value : row) cout << value << " ";
        cout << endl;
    }
}

主程序进行性能对比。

#include <iostream>
#include <vector>
#include <chrono>
#include "Matrix.h"

using namespace std;
using namespace chrono;

int main() {
    std::cout << "=== 矩阵乘法优化示例 ===" << std::endl;
    const int size = 100;
    Matrix matrix1(size, size);
    Matrix matrix2(size, size);
    for (int i = 0; i < size; ++i) {
        for (int j = 0; j < size; ++j) {
            matrix1(i, j) = i + j;
            matrix2(i, j) = i * j;
        }
    }

    auto start = high_resolution_clock::now();
    Matrix resultNaive = matrix1.multiplyNaive(matrix2);
    auto end = high_resolution_clock::now();
    std::cout << "朴素算法耗时：" 
              << duration_cast<milliseconds>(end - start).count() 
              << "毫秒" << std::endl;

    start = high_resolution_clock::now();
    Matrix resultOptimized = matrix1.multiplyOptimized(matrix2);
    end = high_resolution_clock::now();
    std::cout << "优化算法耗时：" 
              << duration_cast<milliseconds>(end - start).count() 
              << "毫秒" << std::endl;
    return 0;
}

构建与运行：

mkdir -p build && cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
cmake --build . --config Release
./MatrixMultiplicationOptimization

总结与实践

性能优化是一个持续迭代的过程。通过智能指针管理内存、利用预分配减少碎片、合并循环提升缓存命中率、以及采用异步 I/O，我们可以显著提升程序效率。矩阵乘法的案例展示了算法复杂度对实际运行的巨大影响。

建议后续尝试以下练习巩固知识：

1. 使用 GProf 或 Perf 分析实际项目的性能热点。
2. 编写函数封装智能指针，彻底杜绝内存泄漏。
3. 研究 SIMD 指令集进一步优化循环计算。
4. 探索内存池技术在高频分配场景下的应用。
5. 学习并发编程模型处理 CPU 密集型任务。