Here's a smoketest which compiles correctly with NVCC given all of the same compiler options as RAJA

iterDeductionSmokeTest.cpp

/*
  $ /opt/cuda/bin/nvcc --keep  iterDeductionSmokeTest.cpp -m64 -DRAJA_ENABLE_NESTED -O2 -restrict -arch compute_35 -std c++11 --expt-extended-lambda -x cu -ccbin /opt/cuda/bin/g++ -Xcompiler -fopenmp -DNVCC -I/opt/cuda/include -I/opt/cuda/include -I/home/wkillian/RAJA_build/tpl/src/googletest/include -I/home/wkillian/RAJA_build/include/RAJA -I/home/wkillian/RAJA_build/include -I/home/wkillian/RAJA/include -I/home/wkillian/RAJA/test/include
*/

#include <iostream>
#include <iterator>
#include <type_traits>
#include <algorithm>

#include <RAJA/RAJA.hxx>

template <typename IterType>
using OutputIter = std::ostream_iterator <typename std::iterator_traits<IterType>::value_type>;

int main() {
  const int N = 1024;
  const int LIMIT = 10;

  float* a;
  cudaMallocManaged (&a, sizeof (float) * N);
  std::fill (a, a + N, 1);

  std::copy (a, a + LIMIT, OutputIter<decltype(a)> (std::cout, " "));
  std::cout << std::endl;
  for (int loopCount = 0; loopCount < 3; ++loopCount) {
    RAJA::inclusive_scan_inplace <RAJA::seq_exec> (a, a + N);
    std::copy (a, a + LIMIT, OutputIter<decltype(a)> (std::cout, " "));
    std::cout << std::endl;
    RAJA::inclusive_scan_inplace <RAJA::omp_parallel_for_exec> (a, a + N);
    std::copy (a, a + LIMIT, OutputIter<decltype(a)> (std::cout, " "));
    std::cout << std::endl;
    RAJA::inclusive_scan_inplace <RAJA::cuda_exec<128>> (a, a + N);
    std::copy (a, a + LIMIT, OutputIter<decltype(a)> (std::cout, " "));
    std::cout << std::endl;
  }
  cudaFree (a);

  return 0;
}