Robert Maynard

@gpucibot merge

> If we want to compile to PTX I think the cleanest CMake solution is to create an object library with just this file and then set [CUDA_PTX_COMPILATION](https://cmake.org/cmake/help/latest/prop_tgt/CUDA_PTX_COMPILATION.html). > >...

I don't like the PARENT_SCOPE solution for a couple of reasons: - We will have to determine if CPM is already in the root project scope. If so we can't...

I will draft up a PR to fix this, but I think I will keep the policies being setup for CPM.cmake itself as well. That will ensure that CPM options...

> > libcudacxx headers will need to be marked as non-system for users of rmm. This is necessary so that the version that rmm installs is used instead of the...

With cudf updated to `rapids_cpm_libcudacxx` you can now use that logic to simplify the process of using a custom `libcudacxx` in this PR. For example to use your custom fork...

Can you try 3.16.2 and see if that works? If it does it means that the following fix is at fault: https://gitlab.kitware.com/cmake/cmake/merge_requests/4191/

@henryiii Do you have time to make a small test case that can reproduce this issue? That would really help me track down the root issue, and propose a solution...

Basic thresholding and comparing multiple files was added in NVIDIA/thrust#48

As far as I am aware Windows recommends placing the dlls in the same directory as the executable. In theory you can use a CMake POST_BUILD custom command that invokes...