libcudacxx
libcudacxx copied to clipboard
NVIDIA
[ARCHIVED] The C++ Standard Library for your entire system. See https://github.com/NVIDIA/cccl
This was the original design intent, and that's what `cuda::atomic` does.
In internal CI, we're seeing failures of some tuple CTAD tests with NVCC + GCC and NVCC + Clang. Clang 8 (the only C++17 Clang in CI) fails: ``` std/utilities/tuple/tuple.tuple/tuple.cnstr/deduct.pass.cpp...
# Build failures by VC version: Executed with: `lit -sv .\test\std\utilities\tuple\ -Dcompute_archs=70` ## VS 2017 Compilers: 19.11, 19.12, 19.13, 19.14 Failing tests seem related to type conversions ```c++ struct A...
When we turn on gpuCI testing, we'll be testing with an older version of CUDA - probably CUDA 11.0. That means pipeline won't be available. We should mark our pipeline...
I believe it's currently broken.
Today, we [`#if` away `atomic_flag`'s wait/notify API for < SM70](https://github.com/NVIDIA/libcudacxx/blob/feature/nvcxx-compatibility/libcxx/include/atomic#L2600), but we don't do the same for `atomic`. We should be consistent here. We should also consider whether `#if` away...
libcu++ should provide a standalone C++ Standard Library for NVC++ where most things in `std::` work heterogeneously.
We currently use `__CUDA_ARCH__`/`__NVCOMPILER_CUDA_ARCH__` in a few places that are difficult to replace with `if target`: * For some headers like `` and ``, we need to produce a compile...