C++ 性能优化实战：从内存到 CPU 的执行效率提升

减少内存碎片

频繁分配释放小对象会导致内存碎片化。对于 std::vector，如果知道大概容量，提前 reserve 能避免多次扩容带来的拷贝开销。

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

// 预分配内存避免内存碎片
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);
    }
}

// 不预分配内存（可能导致多次扩容）
void notPreallocateMemory() {
    const int size = 10000;
    std::vector<int> vec;
    for (int i = 0; i < size; ++i) {
        vec.push_back(i);
    }
}

int main() {
    std::cout << "=== 内存碎片优化示例 ===" << std::endl;
    auto start = std::chrono::high_resolution_clock::now();
    preallocateMemory();
    auto end = std::chrono::high_resolution_clock::now();
    auto duration = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count();
    std::cout << "预分配内存耗时：" << duration << "微秒" << std::endl;

    start = std::chrono::high_resolution_clock::now();
    notPreallocateMemory();
    end = std::chrono::high_resolution_clock::now();
    duration = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count();
    std::cout << "不预分配内存耗时：" << duration << "微秒" << std::endl;
    return 0;
}

CPU 优化技巧

CPU 层面的优化主要集中在指令流水线和缓存命中率上。

循环优化

合并循环可以减少遍历次数，降低分支预测失败的概率。

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

// 优化循环：合并操作
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];
    }
}

// 未优化的循环：多次遍历
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];
    }
}

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

    start = std::chrono::high_resolution_clock::now();
    unoptimizedLoop();
    end = std::chrono::high_resolution_clock::now();
    duration = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count();
    std::cout << "未优化循环耗时：" << duration << "微秒" << std::endl;
    return 0;
}

内联函数

对于短小的函数，使用 inline 关键字可以消除函数调用的栈帧开销。不过编译器通常会自动优化，显式声明时注意不要滥用，否则会增加代码体积。

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

// 优化函数：使用内联函数
inline int add(int a, int b) {
    return a + b;
}

// 未优化的函数：普通函数调用
int addNotInline(int a, int b) {
    return a + b;
}

// 测试函数调用开销
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();
    auto duration = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count();
    std::cout << "内联函数调用耗时：" << duration << "微秒" << std::endl;

    start = std::chrono::high_resolution_clock::now();
    for (int i = 0; i < size; ++i) {
        result += addNotInline(i, i);
    }
    end = std::chrono::high_resolution_clock::now();
    duration = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count();
    std::cout << "普通函数调用耗时：" << duration << "微秒" << std::endl;
}

int main() {
    std::cout << "=== 函数优化示例 ===" << std::endl;
    testFunctionCallOverhead();
    return 0;
}

I/O 操作优化

I/O 往往是系统中最慢的部分，尤其是文件和网络读写。

文件 I/O

默认情况下 ofstream 有缓冲区，但有时禁用缓冲或调整缓冲策略能提升特定场景下的写入速度。注意这里演示的是关闭缓冲的极端情况，通常保留缓冲更好。

#include <iostream>
#include <fstream>
#include <vector>
#include <string>
#include <chrono>

// 优化文件 I/O：调整缓冲区策略
void optimizedFileIO() {
    const std::string filename = "test.txt";
    const int size = 10000;
    std::ofstream file(filename);
    file.rdbuf()->pubsetbuf(nullptr, 0); // 禁用缓冲，视具体需求而定
    for (int i = 0; i < size; ++i) {
        file << i << std::endl;
    }
    file.close();
}

// 未优化的文件 I/O：使用默认缓冲区
void unoptimizedFileIO() {
    const std::string filename = "test.txt";
    const int size = 10000;
    std::ofstream file(filename);
    for (int i = 0; i < size; ++i) {
        file << i << std::endl;
    }
    file.close();
}

int main() {
    std::cout << "=== 文件 I/O 优化示例 ===" << std::endl;
    auto start = std::chrono::high_resolution_clock::now();
    optimizedFileIO();
    auto end = std::chrono::high_resolution_clock::now();
    auto duration = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count();
    std::cout << "优化文件 I/O 耗时：" << duration << "微秒" << std::endl;

    start = std::chrono::high_resolution_clock::now();
    unoptimizedFileIO();
    end = std::chrono::high_resolution_clock::now();
    duration = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count();
    std::cout << "未优化文件 I/O 耗时：" << duration << "微秒" << std::endl;
    return 0;
}

网络 I/O

异步 IO 能让线程在等待网络响应时去处理其他任务，大幅提升吞吐量。这里以 Boost.Asio 为例展示异步与同步的区别。

#include <iostream>
#include <vector>
#include <string>
#include <boost/asio.hpp>
#include <boost/asio/ip/tcp.hpp>
#include <boost/asio/connect.hpp>
#include <boost/asio/write.hpp>
#include <boost/asio/read_until.hpp>
#include <sstream>
#include <exception>
#include <chrono>

using boost::asio::ip::tcp;
using namespace std;

// 优化网络 I/O：使用异步操作
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");
        string status_line;
        istringstream response_stream(&response);
        response_stream >> status_line;
    } catch (const std::exception& e) {
        cerr << "错误：" << e.what() << endl;
    }
}

// 未优化的网络 I/O：使用同步操作
void unoptimizedNetworkIO() {
    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));
        string response;
        char buffer[1024];
        size_t len;
        while ((len = socket.read_some(boost::asio::buffer(buffer))) > 0) {
            response.append(buffer, len);
        }
    } catch (const std::exception& e) {
        cerr << "错误：" << e.what() << endl;
    }
}

int main() {
    std::cout << "=== 网络 I/O 优化示例 ===" << std::endl;
    auto start = std::chrono::high_resolution_clock::now();
    optimizedNetworkIO();
    auto end = std::chrono::high_resolution_clock::now();
    auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(end - start).count();
    std::cout << "优化网络 I/O 耗时：" << duration << "毫秒" << std::endl;

    start = std::chrono::high_resolution_clock::now();
    unoptimizedNetworkIO();
    end = std::chrono::high_resolution_clock::now();
    duration = std::chrono::duration_cast<std::chrono::milliseconds>(end - start).count();
    std::cout << "未优化网络 I/O 耗时：" << duration << "毫秒" << std::endl;
    return 0;
}

综合案例：矩阵乘法优化

最后来看个实际点的例子。矩阵乘法是计算密集型任务的典型代表，优化空间很大。

项目结构

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

核心代码

// include/Matrix.h
#ifndef MATRIX_H
#define MATRIX_H

#include <vector>
#include <chrono>
#include <stdexcept>

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;

private:
    int rows_;
    int cols_;
    vector<vector<int>> data_;
};

#endif // MATRIX_H

// src/Matrix.cpp
#include "Matrix.h"
#include <iostream>

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;
    }
}

// src/main.cpp
#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();
    auto duration = duration_cast<milliseconds>(end - start).count();
    std::cout << "朴素算法耗时：" << duration << "毫秒" << std::endl;

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

    return 0;
}

构建与运行

# 创建构建目录
mkdir -p build && cd build
# 配置 CMake
cmake -DCMAKE_BUILD_TYPE=Release ..
# 编译项目
cmake --build . --config Release
# 运行程序
./MatrixMultiplicationOptimization

总结

性能优化没有银弹，它需要结合具体的业务场景。通过智能指针管理内存，利用向量预分配减少碎片，合并循环减少遍历，以及合理选择同步或异步 IO，都能带来可观的提升。矩阵乘法的例子展示了如何通过转置优化缓存局部性。建议大家在日常开发中多使用性能分析工具，养成'先测量，后优化'的习惯。