thrust::sort Compare warnings

I’m getting some very odd warnings using thrust::sort… This code compiles fine:

thrust::device_vector<int> indices(count);
thrust::sort(indices.begin(), indices.end(),  thrust::greater<int>());

Now, if I go to header file where thrust::greater is defined, copy it into my code and rename the template class to “MySorter”, eg.

template<typename T>
struct Sorter
{
 ...

and then try

thrust::sort(indices.begin(), indices.end(),  Sorter<int>());

I get pages and page of compiler warnings about converting signed to unsigned 64-bit ints:

warning C4244: 'argument' : conversion from '__int64' to 'const size_t', possible loss of data

Am I missing something? Is there some compiler magic which allows the SDK version to compile warning-free? Compiling under Visual Studio 2012, Cuda 6.5.

can you provide a simple, complete example? Which file did you pull the thrust::greater definition out of?

The file is thrust/functional.h, line 557. Here’s a standalone sample:

#include <stdio.h>
#include <thrust/host_vector.h>
#include <thrust/device_vector.h>
#include <thrust/copy.h>
#include <thrust/fill.h>
#include <thrust/sequence.h>
#include <thrust/sort.h>

template<typename T>
struct Sorter
{
  /*! \typedef first_argument_type
   *  \brief The type of the function object's first argument.
   */
  typedef T first_argument_type;

  /*! \typedef second_argument_type
   *  \brief The type of the function object's second argument.
   */
  typedef T second_argument_type;

  /*! \typedef result_type
   *  \brief The type of the function object's result;
   */
  typedef bool result_type;

  /*! Function call operator. The return value is <tt>lhs > rhs</tt>.
   */
  __host__ __device__ bool operator()(const T &lhs, const T &rhs) const {return lhs > rhs;}
}; // end greater

int main()
{
#if 1
    // i get warnings
        const int count = 100;
        thrust::device_vector<int> indices(count);
        thrust::sort(indices.begin(), indices.end(),  Sorter<int>());
#else
    // i am warning free
        const int count = 100;
        thrust::device_vector<int> indices(count);
        thrust::sort(indices.begin(), indices.end(), thrust::greater<int>());
#endif

    return 0;
}

I was able to see the issue with cuda 6.5 and VS2013 (also). I was not able to reproduce it with CUDA 5.0/VS2010. I was not able to reproduce the issue on linux. I can’t explain it.

You might get more direct assistance by asking about it on the thrust users group mailing list:

[url]Redirecting to Google Groups