[Feature Request] support -std=c++17
#755
Labels
Comments
|
I just found the reason: CUDA 10.1 doesn't support C++ 17. |
njzjz
added a commit
to njzjz/deepmd-kit
that referenced
this issue
Jun 16, 2021
cuda<=10 doesn't support `-std=c++17`, but conda compilers will add it into the flag. We can friendly remove such flag to prevent users go here and ask deepmodeling#428. fix deepmodeling#755.
|
It seems that CUDA 11.1 does not support c++17 either. I was building with gcc 8.4 and cudatoolkit 11.1 and got the same error. Changing it to std=c++11 solved the problem. |
I can reproduce |
|
NVIDIA announced CUDA 11 supports c++17. I don't know why it still doesn't work. |
njzjz
added a commit
to njzjz/deepmd-kit
that referenced
this issue
Jul 17, 2021
CUDA 11 has included cub, so we don't need to include it any more. cub can be upgraded along with the CUDA Toolkit to obtain bugfixes. See https://github.com/NVIDIA/cub#releases. See also https://github.com/NVIDIA/cub/blob/main/CHANGELOG.md. P.S. It's still unclear to me that which version of C++ it supports. From the changelog, maybe C++ 17 is supported in CUDA 11.2? (deepmodeling#755)
njzjz
added a commit
to njzjz/deepmd-kit
that referenced
this issue
Jul 17, 2021
CUDA 11 has included cub, so we don't need to include it any more. cub can be upgraded along with the CUDA Toolkit to obtain bugfixes. See https://github.com/NVIDIA/cub#releases. See also https://github.com/NVIDIA/cub/blob/main/CHANGELOG.md. P.S. It's still unclear to me that which version of C++ it supports. From the changelog, maybe C++ 17 is supported in CUDA 11.2? (deepmodeling#755)
amcadmus
pushed a commit
that referenced
this issue
Jul 19, 2021
CUDA 11 has included cub, so we don't need to include it any more. cub can be upgraded along with the CUDA Toolkit to obtain bugfixes. See https://github.com/NVIDIA/cub#releases. See also https://github.com/NVIDIA/cub/blob/main/CHANGELOG.md. P.S. It's still unclear to me that which version of C++ it supports. From the changelog, maybe C++ 17 is supported in CUDA 11.2? (#755)
gzq942560379
pushed a commit
to HPC-AI-Team/deepmd-kit
that referenced
this issue
Sep 1, 2021
cuda<=10 doesn't support `-std=c++17`, but conda compilers will add it into the flag. We can friendly remove such flag to prevent users go here and ask deepmodeling#428. fix deepmodeling#755.
gzq942560379
pushed a commit
to HPC-AI-Team/deepmd-kit
that referenced
this issue
Sep 2, 2021
CUDA 11 has included cub, so we don't need to include it any more. cub can be upgraded along with the CUDA Toolkit to obtain bugfixes. See https://github.com/NVIDIA/cub#releases. See also https://github.com/NVIDIA/cub/blob/main/CHANGELOG.md. P.S. It's still unclear to me that which version of C++ it supports. From the changelog, maybe C++ 17 is supported in CUDA 11.2? (deepmodeling#755)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Summary
When add
-std=c++17toCXXFLAGS(this is the default environment variables of conda compilers), CUDA files will throw the following errors (CPU version works well). It works after changing-std=c++17to-std=c++11.It's not a bug, but I hope
-std=c++17can be supported on some day.UPDATE: I am not sure whether it's a bug. Maybe CMake should set the proper C++ standard option?
Detailed Description
Further Information, Files, and Links
The text was updated successfully, but these errors were encountered: