用向量乘矩阵的最快方法?

时间:2019-03-13 10:53:21

标签: c++ matrix-multiplication

有人可以在此函数内建议任何更快速的方法将矩阵乘以矢量吗?

  inline void multiply(
        std::vector< std::vector<double> > &matrix,
        std::vector<double> &vector,
        std::vector<double> &result
    ){
        int size = (int) vector.size();

        result.resize(size);

        #pragma omp parallel for
        for(int i = 0; i < size; ++i){
              int j = 0;

            for(; j <= size - 16; j += 16){
                result[i] += matrix[i][j] * vector[j]
                    + matrix[i][j + 1] * vector[j + 1]
                    + matrix[i][j + 2] * vector[j + 2]
                    + matrix[i][j + 3] * vector[j + 3]
                    + matrix[i][j + 4] * vector[j + 4]
                    + matrix[i][j + 5] * vector[j + 5]
                    + matrix[i][j + 6] * vector[j + 6]
                    + matrix[i][j + 7] * vector[j + 7]
                    + matrix[i][j + 8] * vector[j + 8]
                    + matrix[i][j + 9] * vector[j + 9]
                    + matrix[i][j + 10] * vector[j + 10]
                    + matrix[i][j + 11] * vector[j + 11]
                    + matrix[i][j + 12] * vector[j + 12]
                    + matrix[i][j + 13] * vector[j + 13]
                    + matrix[i][j + 14] * vector[j + 14]
                    + matrix[i][j + 15] * vector[j + 15];
            }

            for(; j < size; ++j){
                result[i] += matrix[i][j] * vector[j];
            }
        }
    }

该函数在运行时被调用了很多次,因此对整个计算时间有非常关键的影响。

1 个答案:

答案 0 :(得分:0)

根据您的硬件,使用GPU并行化(例如:CUDA)可能会大有帮助。