Add tests for host only TUs and fix several found issues. #340
Conversation
| @@ -9,6 +9,10 @@ | |||
| #ifndef _CUDA_ARRAY | |||
| #define _CUDA_ARRAY | |||
|
|
|||
| #ifndef __CUDACC_RTC__ | |||
| # include <cstdlib> | |||
There was a problem hiding this comment.
Out of curiosity, what happens when using NVRTC with things like this? Will the functionality that depends on stuff from <cstdlib> just not work?
There was a problem hiding this comment.
Usually NVRTC defines its own equivalent. It's all just __device__ only.
| ROOT_IMAGE: "nvcr.io/nvidia/cuda:11.8.0-devel-ubuntu18.04" | ||
| COMPILERS: "g++-5 g++-6 g++-7 clang-7" | ||
| COMPILERS: "g++-6 g++-7 clang-7" |
There was a problem hiding this comment.
AFAIK we settled on gcc-7 as the minimal supported version.
Do we want to drop gcc-6 while we are at it?
There was a problem hiding this comment.
Will consider it dropping later in another patch. In tandem with #341 would be good timing.
include/cuda/std/cmath
Outdated
| @@ -9,19 +9,19 @@ | |||
| #ifndef _CUDA_CMATH | |||
| #define _CUDA_CMATH | |||
|
|
|||
| #ifndef __CUDACC_RTC__ | |||
| #include <math.h> | |||
| #include <cmath> | |||
There was a problem hiding this comment.
I am considerably scared that we need both of them
Technically one should suffice
There was a problem hiding this comment.
Sorry, didn't mean to leave it in, we do only need host math.h.
No description provided.