C++ 性能优化实战：从内存管理到 CPU 指令的效率提升

对于频繁增长的容器，预分配空间可以显著减少动态内存分配的开销和碎片化。

#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;

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

CPU 优化技巧

循环优化

合并循环可以减少遍历次数，提高缓存命中率。此外，编译器优化标志（如 -O2 或 -O3）通常能自动处理部分循环展开，但理解原理有助于编写更友好的代码。

#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];
    }
}

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() {
    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;

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

内联函数

对于短小且频繁调用的函数，使用 inline 关键字可以避免函数调用开销。不过，过度使用内联可能会增加代码体积，需权衡利弊。

#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 << "内联函数耗时：" 
              << std::chrono::duration_cast<std::chrono::microseconds>(end - start).count()
              << "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();
    std::cout << "普通函数耗时：" 
              << 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

默认的文件流缓冲区可能导致频繁的磁盘写入。在某些场景下，调整缓冲区大小或使用二进制模式可以提升性能。注意示例中禁用了缓冲是为了演示极端情况，实际生产中请谨慎关闭缓冲。

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

void optimizedFileIO() {
    const std::string filename = "test.txt";
    const int size = 10000;
    std::ofstream file(filename);
    // 禁用缓冲区仅用于测试，生产环境建议保留默认或自定义大缓冲
    file.rdbuf()->pubsetbuf(nullptr, 0);
    for (int i = 0; i < size; ++i) {
        file << i << std::endl;
    }
    file.close();
}

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

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

网络 I/O

异步 I/O 模型（如 Boost.Asio）在处理高并发连接时优于同步阻塞模型。它允许线程在等待网络响应时执行其他任务。

注：运行此代码需要安装并链接 Boost 库。

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

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::string status_line;
        std::istringstream response_stream(&response);
        response_stream >> status_line;
    } catch (const std::exception& e) {
        std::cerr << "错误：" << e.what() << std::endl;
    }
}

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));
        std::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) {
        std::cerr << "错误：" << e.what() << std::endl;
    }
}

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

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

综合案例：矩阵乘法优化

通过一个完整的矩阵乘法项目，我们可以直观地看到算法优化带来的性能差异。这里展示了一个简单的结构对比朴素算法与转置优化的实现。

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 <stdexcept>
#include <iostream>

using namespace std;

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; // 声明转置方法

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

#endif // MATRIX_H

src/Matrix.cpp

#include "Matrix.h"

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() {
    cout << "=== 矩阵乘法优化示例 ===" << 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();
    cout << "朴素算法耗时：" << duration << "毫秒" << endl;

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

    return 0;
}

构建与运行

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

结语

性能优化是一个持续的过程。除了上述提到的技术点，还可以关注 SIMD 指令集、内存池以及缓存局部性等高级主题。保持对代码行为的敏感度，善用工具测量，才能在效率与可维护性之间找到最佳平衡点。