Intel C++ 11.1 与 g++ 4.3.3 编译性能小测试

 首先帖出最简单的一个矩阵乘法的cpp代码:

#include <stdio.h>
#include <sys/time.h>

int main(int argc,char** argv)
{
    int Width = 835;
    int Height = 835;
   
    struct timeval s;
    long  _start;
    long  _end;
    int i;
    int j;
   
    //声明三个浮点数组
    float inputA[Height][Width];
    float inputB[Width][Height];
    float output[Height][Height];
  
    //gettimeofday(&s, 0);
    //_start = (long )s.tv_sec * 1000 + (long)s.tv_usec / 1000;
   //初始化

    for(i = 0;i < Width; i = i + 1)
        for(j = 0;j<Height; j = j+1)
        {
        inputA[i][j] = (float)(i*0.01f+j*0.03f);
        inputB[i][j] = (float)(i*0.2f+j*0.1f);
        output[i][j] = 0.0f;
       }
     //开始计时   
    gettimeofday(&s, 0);
    _start = (long )s.tv_sec * 1000 + (long)s.tv_usec / 1000;

    //矩阵乘法
    for(i = 0;i < Width; i = i + 1)
        for(j = 0;j<Height; j = j+1)
        for(int k=0;k<Height;++k)
            output[i][j] += inputA[i][k]*inputB[k][j];
       //停止计时
      gettimeofday(&s, 0);
    _end = (long)s.tv_sec * 1000 + (long)s.tv_usec / 1000;
   //输出结果
        printf("The Program cost time:/n");
        printf("%f/n",(double)(_end - _start)/1000);
          return 0;
}

 

就这么个没有任何技术含量的程序，分别用icc和g++编译，运行的结果是

(时间单位为秒)

1.不开任何选项，

icc:

The Program cost time:
7.859000

 

g++:

The Program cost time:
13.468000

2.开-O2选项

icc:

The Program cost time:
7.913000

 

g++:

The Program cost time:
4.579000

 

3.开-O3选项

icc:

The Program cost time:
7.661000

 

g++:

The Program cost time:
4.849000

 

结果实在是令人困惑啊！难道Intel的优化还没有g++做的好吗？期盼高人给出答案。