Compiling nvopencc from source code

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi all,

I’m currently trying to compile nvopencc from the sources available at ftp://download.nvidia.com/CUDAOpen64/nvopencc_4.0_src_10003849.tar.gz on Ubuntu precise(32).
I’m using the target “targia32_nvisa” and I’m having a really bad time trying to make it work. First I had to edit ROOT_DIR in linux/make/gcommondefs to point the right directory. Then a header called “syscall_names.h” was missing as well as an entire source directory (namely “src/be/cmc”).
I must admit I have no idea what most of these modules do and I just want the whole project to compile. However for some reason the building looks totally broken. Can anyone tell me how I’m supposed to compile nvopencc ?

Thanks in advance,

Hadrien

ps: I already searched and it looks like very few people succeeded in this task…

2

Is there any reason in particular why you want to use the Open64 CUDA compiler? If you are just looking for an open source compiler that targets NVIDIA, CLANG/LLVM has a PTX backend in the trunk.

3

Thanks for your answer, I’ll look into that.