SSE 原子指令加速矩阵运算

利用原子指令加速矩阵运算

C++里面有个原子指令库，不需要通过内嵌汇编就可以调用cpu内部SIMD的指令，头文件<xmmintrin.h> 包含SSE库，<emmintrin.h> 包含SSE2库，<pmmintrin.h> 包含SSE3库。

原子指令可以利用cpu内部的128位寄存器，同时进行多个数据运算，比如可以同时计算4个float类型数，或2个double类型的数。

运行环境：

win10 下ubuntu18.04
g++ 编译的时候要加关键选项 -msse3 否则识别不了_mm_hadd_pd。非常关键，百度找不到，最终还是借助bing英文搜索。

g++ -msse3 test.cpp -o test

# include <iostream>
# include <cstdlib>
# include <ctime>
# include <pmmintrin.h> 
using namespace std;

//show result
template<typename T>
void show_matrix(int size,T ** M){
    for(int i=0;i<size;i++){
        for(int j = 0;j<size;j++){
            cout<<M[i][j]<<" ";
        }
        cout<<"\n";
    }
}
// roll over matrix b
template<typename T>
void Transpose(int size,T** m)
{
    for(int i=0;i<size;++i){
        for(int j=i+1;j<size;++j){
            std::swap(m[i][j],m[j][i]);
        }
    }
}

//method 1
template<typename T>
void SeqMatrixMult1(int size, T** m1, T** m2, T** result)
{
    Transpose(size, m2);
    for (int i = 0; i < size; i++) {
        for (int j = 0; j < size; j++) {
            // temp parameter can reduce memory access, which is very important
            T c = 0;  
            for (int k = 0; k < size; k++) {
                c += m1[i][k] * m2[j][k];
            }
            result[i][j] = c;
        }
    }
    Transpose(size, m2);
}

//method 2
//template<typename T>
void SeqMatrixMult2(int size, double** m1, double** m2, double** result)
{
    Transpose(size, m2);
    for (int i = 0; i < size; i++) {
        for (int j = 0; j < size; j++) {
            __m128d c = _mm_setzero_pd();
            for (int k = 0; k < size; k += 2) {
                c = _mm_add_pd(c, _mm_mul_pd(_mm_load_pd(&m1[i][k]), _mm_load_pd(&m2[j][k])));
            }
            // horizontal add of the single register
            c = _mm_hadd_pd(c, c);
            _mm_store_sd(&result[i][j], c);
        }
    }
    Transpose(size, m2);
}

int main(){
    int n = 500;
    
    double **a = new double*[n];
    double **b = new double*[n];
    double **result1 = new double*[n];
    double **result2 = new double*[n];

    for(int i=0;i<n;i++){
        a [i]=new double[n];
        b [i]=new double[n];
        result1[i]=new double[n];
        result2[i]=new double[n];
    }
    
    for(int i=0;i<n;++i)
        for(int j=0;j<n;++j){
            a[i][j]=rand()%100/10.0;
            b[i][j]=rand()%100/10.0;
            result1[i][j]=0;
            result2[i][j]=0;
        }

//  show_matrix(n,a);
//  show_matrix(n,b);

    clock_t start,end;

// method 1
    start=clock();
    SeqMatrixMult1(n,a,b,result1);
    end = clock();
    cout<<end-start<<"\n";
//  method 2
    start=clock();
    SeqMatrixMult2(n,a,b,result2);
    end = clock();
    cout<<end-start<<"\n";

//  show_matrix(n,result1);
//  show_matrix(n,result2);

//delete;   
    for(int i=0;i<n;++i){
        delete[]a[i];
        delete[]b[i];
        delete[]result1[i];
        delete[]result2[i];
    }
    delete[]a;
    delete[]b;
    delete[]result1;
    delete[]result2;
    
    return 0;
}

参考:
Optimize Your Code: Matrix Multiplication
'mm_hadd_ps' was not declared in this scope