[cherubrock-seb]

Fixes #126.

Warning-flag support is now checked independently for the C++ compiler and the NVIDIA CUDA host compiler.

For CUDA compilation, each forwarded -Xcompiler flag is added only when a CUDA-language compiler check confirms that nvcc with its configured host compiler accepts it. The existing top-level developer-build policy requiring matching CXX and CUDA host compilers is unchanged.

Validation:

• Dependency-mode mixed-toolchain probe: clang++ 18.1.3 as CXX, nvcc 13.2.86 with g++ 12.4 as CUDA host.
• -Wgnu: accepted for CXX, rejected for CUDA host, and therefore omitted from the nvcc command.
• -Wall: accepted and retained for both CXX and CUDA host.
• A mixed CXX/CUDA static target built successfully with the resulting flags.
• No GPU kernel was executed.
• Full current-tree NVBench configure was not run on this host because its CMake is 3.28.3 while current NVBench requires CMake 3.30.4.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[oleksandr-pavlyk]

Please provide configuration cmake command used that was broken before, and work now.

On my attempt to check the fix I tried:

cmake -B build-498 --preset nvbench-dev -G Ninja \
    -DCMAKE_CXX_COMPILER=$(which g++) \
    -DCMAKE_CUDA_COMPILER=/usr/local/cuda/bin/nvcc \
    -DCMAKE_CUDA_ARCHITECTURES=native 
    -DCMAKE_CUDA_HOST_COMPILER=$(which clang++)

This configuration failed at cmake/NVBenchConfigTarget.cmake:66 with "NVBench developer builds require that CMAKE_CUDA_HOST_COMPILER matches CMAKE_CXX_COMPILER when using nvcc".

Same for --preset nvbench-ci.