我正在修补CUDA提供的Thrust库。我试图在用户定义的结构的设备向量上执行包含和独占扫描。这是代码。
#include <iostream>
#include <thrust/copy.h>
#include <thrust/count.h>
#include <thrust/device_vector.h>
#include <thrust/fill.h>
#include <thrust/functional.h>
#include <thrust/host_vector.h>
#include <thrust/replace.h>
#include <thrust/scan.h>
#include <thrust/sequence.h>
#include <thrust/transform.h>
#include <thrust/version.h>
#include <vector>
struct mystruct
{
int first;
int second;
};
//Overload the + operator for the used defined struct
__host__ __device__
mystruct operator + (mystruct a, mystruct b)
{
mystruct c;
c.first =a.first +b.first;
c.second=a.second+b.second;
return c;
}
int main(void)
{
thrust::host_vector<mystruct> host_vec(5);
thrust::device_vector<mystruct> dev_vec(5);
host_vec[0].first=2 ; host_vec[0].second=2 ;
host_vec[1].first=2 ; host_vec[1].second=2 ;
host_vec[2].first=2 ; host_vec[2].second=2 ;
host_vec[3].first=2 ; host_vec[3].second=2 ;
host_vec[4].first=2 ; host_vec[4].second=2 ;
thrust::copy(host_vec.begin(), host_vec.end(), dev_vec.begin());//copy to device
thrust::inclusive_scan(dev_vec.begin(), dev_vec.end(), dev_vec.begin()); //In-place inclusive scan
//thrust::exclusive_scan(dev_vec.begin(), dev_vec.end(), dev_vec.begin()); //In-place exclusive scan
std::cout<<"The inclusive scanned mystruct vector is "<<std::endl;//Print the scan
thrust::copy(dev_vec.begin(),dev_vec.end(),host_vec.begin());//copy back to host
for (int i = 0; i < host_vec.size(); ++i)//print the scan
{
std::cout<< host_vec[i].first<<" "<< host_vec[i].second << std::endl;
}
return 0;
}
以上代码执行包含运行完美,给我所需的结果。
现在在上面的代码中我已经注释掉了独占扫描。 如果我尝试运行包含扫描的到位,那么我会收到以下编译错误。
Desktop: nvcc temp.cu
/usr/local/cuda/bin/../include/thrust/detail/scan.inl(68): error: no suitable constructor exists to convert from "int" to "mystruct"
detected during instantiation of "OutputIterator thrust::exclusive_scan(InputIterator, InputIterator, OutputIterator) [with InputIterator=thrust::detail::normal_iterator<thrust::device_ptr<mystruct>>, OutputIterator=thrust::detail::normal_iterator<thrust::device_ptr<mystruct>>]"
temp.cu(54): here
1 error detected in the compilation of "/tmp/tmpxft_00003330_00000000-4_temp.cpp1.ii".
我该怎么办?作为参考,独占扫描的结果必须是
0 0
2 2
4 4
6 6
8 8
答案 0 :(得分:1)
更改此行
thrust::inclusive_scan(dev_vec.begin(), dev_vec.end(), dev_vec.begin());
与
thrust::inclusive_scan(dev_vec.begin(), dev_vec.end(),
dev_vec.begin(), mystruct(), thrust::plus<mystruct>());
否则,将构造函数放在接受整数值的struct中,因此scan操作中的默认值可能会隐式地转换为您的struct。
struct mystruct {
mystruct(int a) : first(a), second(a) {}
...
};