fix C++11 CUDA 8 nvcc compile time odeint error

[psychocoderHPC]

fix https://svn.boost.org/trac/boost/ticket/13049

disable usage of noexcept with BOOST_NO_CXX11_NOEXCEPT for nvcc from CUDA 8

[jzmaddock]

Is this only device code that is effected, or does it hit the host meta-compiler as well?

[psychocoderHPC]

This bug effects host code (see example) which is compiled with nvcc out of a *.cu file.

[jzmaddock]

Reproduced and fixed in 9b8de65

[akors]

Hi! Unfortunately, your fix broke compilation with CUDA 9. The reason for that is that you use the symbol __CUDACC_VER__ for version checking, but the symbol has been deprecated in CUDA 9.

Line 64 in crt/common_functions.h of the CUDA toolkit now reads:

#define __CUDACC_VER__ "__CUDACC_VER__ is no longer supported.  Use __CUDACC_VER_MAJOR__, __CUDACC_VER_MINOR__, and __CUDACC_VER_BUILD__ instead."

The fix seems simple enough, instead of checking for the whole version, you could check separately for __CUDACC_VER_MAJOR__ and __CUDACC_VER_MINOR__ .

[psychocoderHPC]

@akors You are right, but the deprecation breaks all checks in include/boost/config/compiler/nvcc.hpp.
Therefore the full file needs an update.

[akors]

Ok, that's a shame. Here's the Trac ticket URL: https://svn.boost.org/trac10/ticket/13152

[psychocoderHPC]

thx for opening the ticket: btw NVIDIA breaks all old boost versions with this change.

[akors]

ouch, it might be worth opening a CUDA bug report as long as CUDA 9 is still in RC!

Good idea! I just did that too, the internal link is https://developer.nvidia.com/nvidia_bug/1971621 , in case anyone has access to that.

[psychocoderHPC]

I tested the beta version of CUDA 9.0067 with a little program:

#include <iostream>
#include <cuda_runtime.h>

int main()
{
        std::cout<<__CUDACC_VER__<<std::endl;
}

compiel and run

nvcc main.cu
./a.out
90067 #this is the output

This means that the define is available in the beta, we need to check the latest rc.

[ax3l]

The problem appears first with the second CUDA 9 release candidate, CUDA 9.0.103.

[jzmaddock]

Can you guys please verify that #175 fixes this?

Many thanks, John Maddock.