Michael Schellenberger Costa comments

Results 319 comments of


                                            Michael Schellenberger Costa

Add an efficient unstable thread sort, use it in unstable block/device merge/segmented sorts, and improve tests

/ok to test

Add an efficient unstable thread sort, use it in unstable block/device merge/segmented sorts, and improve tests

@Nyrio We have recently applied formatting to the cub subproject. I have merged in main and applied formatting to your changes. I hope that should make this transition as painless...

Add an efficient unstable thread sort, use it in unstable block/device merge/segmented sorts, and improve tests

/ok to test

[BUG]: `reduce_by_key` fails with zip_iterator to const pointers

Are you building against current main or a specific branch?

[BUG]: `reduce_by_key` fails with zip_iterator to const pointers

@upsj Thanks a lot for the bug report. I opened a PR with a fix and added your reproducer to out test suite

There's no GPU-side std::terminate()

Actually we have `cuda::std::terminate()` that we can use internally. https://github.com/NVIDIA/cccl/blob/main/libcudacxx/include/cuda/std/detail/libcxx/include/__exception/terminate.h It calls `__trap()` on device as the next best thing

There's no GPU-side std::terminate()

Funnily enough I have https://github.com/NVIDIA/cccl/pull/1583 open, and that does not use `test_allocator.h`

There's no GPU-side std::terminate()

That should actually work 🤔 `cuda/std/__exception>` pulls in `` which pulls in `` There we have a [definition ](https://github.com/NVIDIA/cccl/blob/main/libcudacxx/include/cuda/std/detail/libcxx/include/__exception/terminate.h#L65)of `cuda::std::terminate()` which is essentially a forward to `__cccl_terminate()`

There's no GPU-side std::terminate()

But you are calling `std::terminate` It works fine with `cuda::std::terminate` https://godbolt.org/z/cP531sj93

There's no GPU-side std::terminate()

Oh now I understand it, that code is inside some test header that was brought in from libcxx in our initial fork. AFAIK it is never actually used anywhere. I...