统计C++程序运行时间

前言

为了分析C++程序的效率，一个非常重要的指标就是其算法的运行时间，我们C++11提供了chrono库专门进行时间的测量，具体使用方法如下

chrono

首先我们要说明chrono库中有自己的chrono命名空间，且这个命名空间包含在std中，所以我们要使用他先指定在std命名空间中，再指定在chrono命名空间中，比如我们计算现在的时间是多少

#include 
std::chrono::high_resoulution_clock::now();

我们测算的基本思想是先统计当前时间，再在算法执行完毕后再统计一下当前时间，最后用chrono的内置转换函数，将其转换成我们想要的单位返回，最后输出，例子如下

#include 
#include 

using namespace std;

int main()
{
    auto start = chrono::high_resolution_clock::now();
   // do somthine 
    auto end = chrono::high_resolution_clock::now();
    auto ms = chrono::duration_cast< std::chrono::milliseconds>(start - end);
    
    cout << ms.count() << endl;

    return 0;

具体例子如下，我们写一个矩阵相乘算法， 矩阵相乘算法具体放到device(GPU)上执行

#include 
#include 
#include 

void
__global__
matrix_miltiple(int *a, int *b, int *c, int n){
    int row = blockIdx.y * blockDim.y + threadIdx.y; //每个block中的thread_id都是从0开始
    int col = blockIdx.x * blockDim.x + threadIdx.x;

    int temp = 0;

    if((row < n) && (col < n)){ //one thread iterate over matrix a's row and matrix b'x column
        for(int i = 0; i < n; i++) temp += a[row * n + i] * b[i * n + col];
    }

    c[row * n + col] = temp;
}

void
init_matrix(int *a, int n){
    for(int i = 0; i < n; i++){
        a[n] = rand() % 100;
    }
}

int
main(){
    //matrix size of row or col (1024)
    int n = 1 << 10;

    // size of matrix n * n (1024*1024)
    size_t bytes = n * n * sizeof(int);

    //host pointer
    int *h_a, *h_b, *h_c;

    //malloc host
    h_a = (int*)malloc(bytes);
    h_b = (int*)malloc(bytes);
    h_c = (int*)malloc(bytes);

    //device pointer 
    int *d_a, *d_b, *d_c;

    //alloc device mem
    cudaMalloc(&d_a, bytes);
    cudaMalloc(&d_b, bytes);
    cudaMalloc(&d_c, bytes);

    //init martrix
    init_matrix(h_a, n);
    init_matrix(h_b, n);

    //memcopy from host to device
    cudaMemcpy(d_a, h_a, bytes, cudaMemcpyHostToDevice);
    cudaMemcpy(d_b, h_b, bytes, cudaMemcpyHostToDevice);

    //thread pre block
    int BLOCK_SIZE = 16;

    //how many block in each demension
    int GRID_SIZE = (int)ceil(n / BLOCK_SIZE);

    //开启二维
    dim3 grid(GRID_SIZE, GRID_SIZE);
    dim3 threads(BLOCK_SIZE, BLOCK_SIZE);

    //计时
    auto start = std::chrono::high_resolution_clock::now();

    matrix_miltiple<<<grid, threads>>>(d_a, d_b, d_c, n);

    //wait 
    cudaDeviceSynchronize();

    //copy back to host
    cudaMemcpy(h_c, d_c, bytes, cudaMemcpyDeviceToHost);

    //计时
    auto end = std::chrono::high_resolution_clock::now();

    //转换
    auto ms = std::chrono::duration_cast< std::chrono::milliseconds>(end - start);

    std::cout << ms.count() << "ms" << std::endl; 
    return 0;
}