Feeding raw CUDA pointer to thrust::lower_bound fails

[lodhb]

Despite specifying the thrust::cuda::par parallel execution policy, thrust::lower_bound returns the following error:

error: no suitable constructor exists to convert from "float *" to "thrust::detail::normal_iteratorthrust::device_ptr"

Code is the following:

float * raw_ptr;
int length;
...
... (allocate memory on the device at raw_ptr)
...
thrust::device_vector::iterator iter = thrust::lower_bound(thrust::cuda::par, raw_ptr, raw_ptr+length, some_value, some_comp);

Using pgcpp 14.3-0 64-bit target on x86-64 Linux and Cuda compilation tools, release 6.5, V6.5.12.

[jaredhoberock]

Hi lodhb,

When I compile the following program against the latest development version of Thrust, it succeeds without error on my system:

$ cat test.cu 
#include <thrust/device_vector.h>
#include <thrust/sequence.h>
#include <thrust/binary_search.h>
#include <thrust/system/cuda/execution_policy.h>

int main()
{
  thrust::device_vector<int> v(10);

  thrust::sequence(v.begin(), v.end());

  int *ptr = raw_pointer_cast(v.data());

  thrust::lower_bound(thrust::cuda::par, ptr, ptr + v.size(), 0);

  return 0;
}

$ nvcc -arch=sm_35 -I. test.cu -run

If you still observe the error with the latest development version of Thrust, please submit a small, self-contained program which reproduces the error.