Hello,
I have a very simple code (in src.cu)
int main()
{
return 0;
}
which I am trying to compile with nvcc, with intel icpc compiler set as the cuda host compiler.
nvcc -ccbin=icpc src.cu -o program.x
It fails with an error
/apps/all/GCCcore/11.3.0/include/c++/11.3.0/bits/stl_pair.h(460): error: argument list for class template "std::pair" is missing
/apps/all/GCCcore/11.3.0/include/c++/11.3.0/bits/stl_pair.h(460): error: expected a ")"
/apps/all/GCCcore/11.3.0/include/c++/11.3.0/bits/stl_pair.h(460): error: template parameter "_T1" may not be redeclared in this scope
/apps/all/GCCcore/11.3.0/include/c++/11.3.0/bits/stl_pair.h(460): error: expected a ";"
4 errors detected in the compilation of "src.cu".
These are the lines in the stl_pair.h header file causing the error:
#if __cpp_deduction_guides >= 201606
template<typename _T1, typename _T2> pair(_T1, _T2) -> pair<_T1, _T2>;
#endif
I am using intel/2022a module we have on our Karolina cluster (icpc --version 2021.6.0 20220226), CUDA 11.4.1 with
$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Wed_Jul_14_19:41:19_PDT_2021
Cuda compilation tools, release 11.4, V11.4.100
Build cuda_11.4.r11.4/compiler.30188945_0
The system is running CentOS 7.
If I use intel/2021b (icpc 2021.4.0 20210910) or newer, it fails as described. But if I use intel/2021a (icpc 2021.2.0 20210228) or older, it then works with no problem.
How can I solve this issue?
Why are gcc headers included, if I am using icpc as a host compiler?
Why does it even produce an error?
Adding the -std=c++11 flag (or newer C++ standard) solves the issue, but I do not use any c++11 features anywhere in my code, so it should not be necessary. And according to nvcc docs and icpc docs, the C++ standard should be set to C++14 or higher by default. Running only icpc and testing the __cplusplus macro, it is C++17.
I would appreciate any help with this.