I currently need to write a project paper for my studies. My challenge will be to find some calculations in a given software, that can be put on a graphics card to speed up the whole software. The problem I currently have is to decide whether to use CUDA or OpenCL. I wanted to start checking if CUDA would be a good choice for my aims, but i cannot find any posts about the following problem: The software I should analyze uses GCC 4.3 and c++03 standard (I know that's very old, but these are requirements by the client).
Is it possible to take a newer CUDA version and make it work with software compiled by gcc4.3?
I hope somebody might help me with this.
-
The supported compiler matrix for CUDA 9 is here: docs.nvidia.com/cuda/cuda-installation-guide-linux/index.htmltalonmies– talonmies2017年11月06日 08:34:47 +00:00Commented Nov 6, 2017 at 8:34
-
I know I need gcc 4.8 for installing CUDA but is it possible to LINK compiled CUDA Code against gcc4.3 compiled c++ code?M. Bremer– M. Bremer2017年11月06日 14:20:41 +00:00Commented Nov 6, 2017 at 14:20
-
You are asking the wrong question. It is libC and libC++ which define binary compatibility, not the compiler. Do you have a platform which has a supported version of CUDA and both GCC 4.3 and a supported host compiler built on the sale ABI/libC versions?talonmies– talonmies2017年11月06日 16:06:18 +00:00Commented Nov 6, 2017 at 16:06
-
Well, I see your point but I have no idea how to look up, which libc both compilers are using.M. Bremer– M. Bremer2017年11月07日 07:33:14 +00:00Commented Nov 7, 2017 at 7:33
1 Answer 1
Okay, I tested linking of nvcc 9 compiled cuda kernel against gcc4.3 compiled C++ code and it worked fine.