Is there `cuda::aligned_size_t` on pre-Volta devices ?

[fkallen]

cooperative_groups::memcpy_async(
  const TyGroup &group,
  TyElem *__restrict__ dst,
  const DstLayout &dstLayout,
  const TyElem *__restrict__ src,
  const SrcLayout &srcLayout
);

requires Compute Capability 3.5 minimum, Compute Capability 8.0 for asynchronicity, C++11

cuda::aligned_size_t is only defined in <cuda/barrier> and <cuda/pipeline> which are only available for CC >= 7.0.

#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 700
#  error "CUDA synchronization primitives are only supported for sm_70 and up."
#endif

To use aligned_size_t on older devices, one needs to manually define aligned_size_t .

Problem example: https://cuda.godbolt.org/z/zv5v15z5W