我正在尝试实施CUDA博客上提供的Unified Vidrtual Memory示例: http://devblogs.nvidia.com/parallelforall/unified-memory-in-cuda-6/
Visual Studio 2013给出了一个错误:类没有默认构造函数" Term"
#include "cuda_runtime.h"
#include "device_launch_parameters.h"
#include <stdio.h>
#include <iostream>
#define NUM_ATTRIBUTES 10
class Managed {
public:
void *operator new(size_t len) {
void *ptr;
cudaMallocManaged(&ptr, len);
std::cout << "custom new for size " << len << '\n';
return ptr;
}
void *operator new[] (size_t len) {
void *ptr;
cudaMallocManaged(&ptr, len);
std::cout << "custom new for size " << len << '\n';
return ptr;
}
void operator delete(void *ptr) {
cudaFree(ptr);
}
};
class Attribute : public Managed {
public:
int num_levels;
int num_active_levels;
int active_levels;
};
class Term : public Managed {
public:
int num_active_attributes;
int active_attributes;
Attribute attribute[NUM_ATTRIBUTES];
// Unified memory copy constructor allows pass-by-value
Term (const Term &orig) {
num_active_attributes = orig.num_active_attributes;
active_attributes = orig.active_attributes;
cudaMallocManaged((void **)&attribute,NUM_ATTRIBUTES * sizeof(Attribute));
memcpy(attribute, orig.attribute, NUM_ATTRIBUTES * sizeof(Attribute));
}
};
int main()
{
Term * A = new Term[10];
getchar();
return 0;
}
这个类是否继承自父类Managed中定义的new运算符?
答案 0 :(得分:1)
这与CUDA无关。编译器没有生成默认构造函数,因为您已经提供了另一个构造函数(副本)。
只需显式请求默认构造函数:
class Term : public Managed {
public:
...
// C++11
Term() = default;
// Pre C++11 alternative (although not strictly equivalent)
Term() {}
...
};
另请注意,operator new
和构造函数是两个不同的东西:new
用于分配存储,构造函数用于初始化该存储到有意义的状态(大致)。
答案 1 :(得分:0)
您需要定义数组元素
所需的默认构造函数class Term : public Managed {
public:
//...
Term () {} // default c-tor
Term() = default; // since C++11, note however that there are some
// differences between Term() = default; and Term() {}
// they are not equal when value initialization
// is considered
};
您没有,因为显式定义的copy-ctor Term( const Term&)
禁止编译器生成默认构造函数。关于默认构造函数方面=default
和{}
之间的差异,您可以看一下: