Unnecessary compiler warning with nvcc?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi!

I have that piece of code at the bottom. When compiling with nvcc as a .cu file, compiler warnings are issued. When compiled as .cpp file, no warnings occur. Two things bother me:

  1. I have a static_assert in the class which shows that the type is actually POD (see program output). So why would it warn me that is is non-POD?
  2. g++ doesn't issue that warning. Also, I could disable it with '-Wno-invalid-offsetof'. However, nvcc does not accept that flag. What can I use with nvcc to get rid of the (unnecessary?) warnings?

nvcc is V6.5.12. Output:

$ nvcc -std=c++11 offsetof_error.cu -o app
offsetof_error.cu(30): warning: offsetof applied to non-POD types is nonstandard

offsetof_error.cu(30): warning: offsetof applied to non-POD types is nonstandard

$ ./app 
4data is pod? 1
Offset of head is: 0

$ nvcc -std=c++11 offsetof_error.cpp -o app
$ ./app 
4data is pod? 1
Offset of head is: 0

#include <iostream>
#include <typeinfo>
#include <type_traits>
#include <cstddef>

using namespace std;

template <typename T> struct queue_t
{
    T *head;
    T *tail;
};

struct data
{
    int foo;
};

template<typename T> class myclass
{
    static_assert(std::is_pod<T>::value,            "Queue Payload must be POD!");
    static_assert(std::is_pod< queue_t<T> >::value, "Queue Payload must be POD!");

public:
    myclass()
    {
        cout << typeid(T).name() << " is pod? " << is_pod< queue_t<T> >::value << endl;
        cout << "Offset of head is: " << offsetof(queue_t<T>, head) << endl;
    }
};

int main()
{
    myclass<data> test;
    return 0;
}

Thanks for your help,
nargin

2

You can pass

-Xcudafe "--diag_suppress=1427"

to nvcc. To get various error/warning ID:s, pass

--display_error_number

to cudafe.

(Tested with NVCC 7.0)

3

That worked. Thanks :)