[BUG]: Including cuda/std/atomic disables deprecation warnings in MSVC 2019

[bsadri]

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this bug and that I agree to the Code of Conduct

Type of Bug

Something else

Component

libcu++

Describe the bug

If I compile the following simple program using MSVC 2019 (latest)

#include <iostream>

struct Foo
{
    [[deprecated]] void bar()
    {
        std::cout << "Bar!" << std::endl;
    }
};

int main()
{
    Foo foo;
    foo.bar();
}

I get the following (expected) error: error C4996: 'Foo::bar': was declared deprecated. That this is an error and not a warning may have to do with my other compiler flags and is not the issue here. However, if I add the following include to the top of the code:

#include <cuda/std/atomic>

Then I get no warnings or errors at all and everything goes through.

For comparison, this is not the case if I included <cuda/std/tuple> instead. It also doesn't happen when compiling with either gcc or clang in WSL.

How to Reproduce

Put the code mentioned above in a cpp source file and compile it with typical flags using MSVC. In case I'm wrong about typical flags, I'm using the following relevant flags: -std:c++17 -MD -W4 -WX -wd4505 -bigobj -permissive- -wd4003 -wd4127 -wd4324.

Expected behavior

I think the problem is self-evident. I would never expect the inclusion of a header from CCCL to have a side effect as big as suppression of deprecation warnings. It took me a fair amount of time to trace this back to CCCL, going through a fair bit of trouble to check the compiler flags and even update the compiler itself.

Reproduction link

No response

Operating System

Windows 11

nvidia-smi output

No response

NVCC version

No response

[jrhemstad]

@miscco can you take a look at this?

[miscco]

I was not able to reproduce this locally with VS2022, passing to @wmaxey to check with 2019 because it is a pain to install

[wmaxey]

@bsadri What exact version of MSVC and NVCC are you working with? I need the version dumps for both to have a shot at reproducing.

With CUDA 12.6 + 14.16 (2017 latest) and 14.29 (2019 latest), both compilers fail in the expected way.

[bsadri]

I'm using the latest Visual Studio 2019 (16.11.42) and running the compiler from the command line:

C:\Program Files (x86)\Microsoft Visual Studio\2019\Professional>cl
Microsoft (R) C/C++ Optimizing Compiler Version 19.29.30157 for x64
Copyright (C) Microsoft Corporation.  All rights reserved.

I'm not using nvcc to build the example, but I have 12.6 installed on the system.

[miscco]

I'm not using nvcc to build the example, but I have 12.6 installed on the system.

Can you still provide the exact version of cccl you are using? I cannot reproduce this with VS2022 or VS2017:

C:\src\cccl>cl -Ilibcudacxx/include main.cpp -MD -W4 -WX -wd4505 -bigobj -permissive- -wd4003 -wd4127 -wd4324
Microsoft (R) C/C++ Optimizing Compiler Version 19.43.34604 for x64
Copyright (C) Microsoft Corporation.  All rights reserved.

main.cpp
main.cpp(16): error C2220: the following warning is treated as an error
main.cpp(16): warning C4996: 'Foo::bar': was declared deprecated
C:\Program Files\Microsoft Visual Studio\2022\Preview\VC\Tools\MSVC\14.43.34604\include\__msvc_ostream.hpp(781): warning C4530: C++ exception handler used, but unwind semantics are not enabled. Specify /EHsc
C:\Program Files\Microsoft Visual Studio\2022\Preview\VC\Tools\MSVC\14.43.34604\include\__msvc_ostream.hpp(781): note: the template instantiation context (the oldest one first) is
main.cpp(9): note: see reference to function template instantiation 'std::basic_ostream<char,std::char_traits<char>> &std::operator <<<std::char_traits<char>>(std::basic_ostream<char,std::char_traits<char>> &,const char *)' being compiled

C:\src\cccl>

[bsadri]

Sorry, for the delayed response. I'm on cccl 2.5.0.