`cuda::stream_ref` is expensive to include
Open
- Dominant language
- C++
- Stars
- 2.5k
- Forks
- 486
- Avg merge
- 2d 6h
- Merged PRs (30d)
- 295
Description
During some compile-time investigation, I noticed that `cuda::stream_ref` is quite expensive to include, accounting for at least 235ms of compile-time (nvcc 13.0 -fdevice-time-trace).
This is 99% caused by including `cuda/__event/timed_event.h`, which includes `` and `cuda/__device/all_devices.h`.
Contributor guide
Assessment
This issue has not been assessed yet.