C++ 性能优化实战：内存、CPU 与 I/O 效率提升指南

CPU 优化技巧

循环合并与局部性

减少遍历次数能显著降低缓存未命中。尽量在一个循环内完成相关操作。

#include <iostream>
#include <vector>
#include <algorithm>
#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;
    return 0;
}

内联函数减少调用开销

短小且频繁调用的函数适合设为 inline，避免栈帧切换成本。

#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;
}

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

I/O 操作优化

文件 I/O 缓冲控制

默认缓冲区可能带来额外开销，根据场景调整缓冲策略。注意示例中禁用缓冲是为了演示对比，生产环境通常保留缓冲。

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

void optimizedFileIO() {
    const std::string filename = "test.txt";
    const int size = 10000;
    std::ofstream file(filename);
    // 实际场景中通常不建议完全禁用缓冲，此处仅为演示
    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;
    return 0;
}

网络 I/O 异步模型

高并发场景下，异步 I/O 能大幅提升吞吐量，避免线程阻塞等待。

#include <iostream>
#include <boost/asio.hpp>
#include <chrono>

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

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\nHost: example.com\r\nConnection: 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");
    } catch (const std::exception& e) {
        cerr << "错误：" << e.what() << endl;
    }
}

int main() {
    auto start = std::chrono::high_resolution_clock::now();
    optimizedNetworkIO();
    auto end = std::chrono::high_resolution_clock::now();
    std::cout << "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

项目结构

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

核心实现

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

#endif

// 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("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("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() {
    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();
    cout << "朴素算法耗时：" << duration_cast<milliseconds>(end - start).count() << "毫秒" << endl;

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

总结与实践

性能优化不是魔法，而是基于数据的科学决策。记住'先测量后优化'，善用工具定位瓶颈。内存上优先使用智能指针和预分配，CPU 上关注循环局部性和内联，I/O 上选择合适缓冲与异步模型。通过矩阵乘法案例可以看到，算法层面的改进往往比单纯加速更有价值。后续可以尝试 SIMD 指令集或并发编程进一步挖掘潜力。