Error compiling CUDA kernel

Question

I'm trying to compile a CUDA kernel but nvcc returns me these errors:

error: identifier "__shfl_xor" is undefined
error: identifier "__any" is undefined

Clearly some header some argument for the compiler is missing... can you help me? I'm running the CUDA toolkit 5.0 on Ubuntu 13.04

That looks like a gcc incompatibility, are you using the default gcc 4.7? Have you read the Linux release notes? — talonmies
Yes I'm using the default gcc and the CUDA toolkit that comes from the repositories — Filippo Bistaffa
gcc 4.7 isn't supported by CUDA 5 and neither is Ubuntu 13.04, so why would you expect it to work? — talonmies
I thought it was supported, since it's in the official repository. By the way it is the first problem of this kind, all other programs works OK — Filippo Bistaffa
Are you sure that you are compiling for the correct GPU architecture? Warp vote functions are only supported by devices of compute capability 1.2 and higher. Warp shuffle functions are only supported by devices of compute capability 3.x. — RoBiK

Filippo Bistaffa Filippo Bistaffa · Accepted Answer · 2013-06-07T10:13:51

3

votes

I solved this issue adding -arch=sm_30 to the command line