Conversation
This comment has been minimized.
This comment has been minimized.
| #include <cuda/std/__type_traits/is_same.h> | ||
| #include <cuda/std/array> | ||
| #include <cuda/std/climits> | ||
| #ifndef _CUB_DISABLE_CMATH |
There was a problem hiding this comment.
I thought we wanted to drop that?
There was a problem hiding this comment.
We decided to drop _CCCL one. This is _CUB version that only disables reproducible reduction, which is needed to guard against includes of cmath parts that do not work in freestanding at the moment.
There was a problem hiding this comment.
No the ask was to split the PR into small chunks that can be individually reviewed I am 90% sure that we can actually just use the cmath code fine with any reasonably modern compiler, so please lets move this PR to only the ostream changes as discussed and open a followup for the other parts
| template <typename Tag> | ||
| _CCCL_HOST inline PerDeviceAttributeCache& GetPerDeviceAttributeCache() | ||
| { | ||
| static PerDeviceAttributeCache cache; | ||
| return cache; | ||
| } | ||
|
|
||
| struct PtxVersionCacheTag | ||
| {}; | ||
| struct SmVersionCacheTag | ||
| {}; | ||
|
|
There was a problem hiding this comment.
Question: Is there a need for the code to be moved?
|
|
||
| #include <cub/agent/agent_reduce.cuh> | ||
| #include <cub/detail/rfa.cuh> | ||
| #ifndef _CUB_DISABLE_CMATH |
🥳 CI Workflow Results🟩 Finished in 1h 52m: Pass: 100%/349 | Total: 4d 19h | Max: 1h 11m | Hits: 99%/477209See results here. |
|
This turns into too much work for people to agree on for MVP. Will keep the changes in my fork and ask others to help productize it once MVP is ready. |
3 participants
Description
contributes towards https://github.com/NVIDIA/cccl_private/issues/622
We decided to split #7735 into multiple PRs. This one is the first PR that just introduces new macro and provide trivial changes.
Checklist