nvcc Compiler Bug?

If I attempt to compile this code:

#include <iostream>
#include <utility>

#include <thrust/tuple.h>
#include <thrust/device_vector.h>
#include <thrust/execution_policy.h>
#include <thrust/iterator/zip_iterator.h>
#include <thrust/iterator/counting_iterator.h>

namespace T = thrust;

template <typname ...Ts>
auto reduce_vectors(Ts&&... vectors) -> T::tuple<Ts...>
  auto const zip_begin = T::make_zip_iterator(

  return T::reduce();

int main(void)
  auto const num_vals = size_t{3000};

  using T::device_vector;

  auto const count_begin = T::make_counting_iterator(0);
  auto const count_end   = count_begin + num_vals;

  auto const vec_a = device_vector<int>{count_begin, count_end};
  auto const vec_b = device_vector<int>{count_begin, count_end};
  auto const vec_c = device_vector<int>{count_begin, count_end};
  auto const vec_d = device_vector<int>{count_begin, count_end};
  auto const vec_e = device_vector<int>{count_begin, count_end};

  auto const zip_begin = T::make_zip_iterator(

  using tuple_t = decltype(*zip_begin);

  auto const zip_end = zip_begin + num_vals;

  auto const tuple_sum = T::reduce(
    zip_begin, zip_end, 
    T::make_tuple(0, 0), 
    [] __host__ __device__ (tuple_t const& a, tuple_t const& b) -> tuple_t
      return T::make_tuple(
        T::get<0>(a) + T::get<0>(b),
        T::get<1>(a) + T::get<1>(b));

  std::cout << T::get<0>(tuple_sum) << ", " << T::get<1>(tuple_sum) << "\n";

  return 0;

I receive this in the console:

C:\Some\Path>nvcc --expt-extended-lambda -O3 -gencode arch=compute_61,code=sm_61 -o zip-reduce zip-reduce.cu
zip-reduce.cu(12): error: identifier "typname" is undefined

zip-reduce.cu(13): error: cannot deduce "auto" type (initializer required)

At end of source: internal error: assertion failed: check_use_of_auto_type: bad symbol (C:/dvs/p4/build/sw/rel/gpu_drv/r361/r361_00/drivers/compiler/edg/EDG_4.10/src/decls.c, line 18644)

2 errors and 1 catastrophic error detected in the compilation of "C:/Some/PATH~1/AppData/Local/Temp/tmpxft_000038dc_00000000-13_zip-reduce.cpp1.ii".
Compilation aborted.

This application has requested the Runtime to terminate it in an unusual way.
Please contact the application's support team for more information.

I’ve never seen nvcc hit a “catastrophic” error before so I figured I’d post it here and see if other people can replicated it. I’m using CUDA 8 on Windows 10 with a gtx 1060, using MSVC 2015 specifically.

“internal error: assertion failed” is a pretty good indication that this is a bug inside the compiler itself, so that seems to be worth reporting to NVIDIA. From the path spit out along with the error message it appears to originate in the EDG frontend used by the CUDA toolchain, that is, a third-party component. But these are internal details you don’t need to worry about even if my reading is correct.

In practical terms, I wonder whether fixing the first two errors reported will avoid hitting this internal error. Have you tried that?

Yeah, I got things working. Turns out, I spelled typname instead of typename and that’s what caused the bug.