opencv 使用cuda加速

请帮我把中间这个for循环用cuda在.cu中写出来 ，谢谢啦
descriptors1的数据类型是Mat

for (int  a = 0; a < descriptors1.rows; a++)
{
        float *mass = new float[vecdim];

        for (size_t b = 0; b < vecdim; b++)
        {
            mass[b] = descriptors1.at<float>(a, b);
        }

        hnsw.addPoint((void*)mass, count);
        ++count;
        delete[] mass;
}