I got the following while running nvcc to compile some CUDA code:
0. Running pass 'Ldu/Ldg Transformation' on function '@_ZN14dense_odometry11CalcJKernelINS_13PinholeCameraEEEvmmT_iyyyyyPK6float3PKhPPf'
Segmentation fault (core dumped)
Googling seems to yield no relevant results… Does anyone have an idea about where I could look to find solutions to this issue?
I’m using nvcc 6.0.1 and the relevant kernel has the following signature:
__global__ void CalcJKernel(const size_t, const size_t,
const PinholeCamera, const int, cudaTextureObject_t, cudaTextureObject_t,
cudaTextureObject_t, cudaTextureObject_t, cudaTextureObject_t,
const float3*, const unsigned char*, float**);
You should retest your compilation case against the latest version of nvcc (7.5, or 8.0RC if you wish).
If it still fails, the best thing to do is to create the smallest code set you can that still causes the failure, and submit that as a bug to nvidia, at developer.nvidia.com
If it doesn’t fail with a newer nvcc compiler, then the solution would be to upgrade. I actually think 6.0.1 is an odd nvcc version, it doesn’t sound like the production release version of CUDA 6 nvcc, it sounds like maybe a release candidate version. (plus, it’s pretty old now ~ 4 years old I think)
But in any event, if you don’t wish to upgrade, filing a bug won’t be that useful for you. NVIDIA is not likely to investigate an issue that is fixed in a newer version.
Hi, thanks a lot for the tips.
I’m generally restricted by compiler version but have some flexibility for the short term. Fortunately, the code compiles with nvcc 7.5.
I will return to this issue when I have to, and file an issue as necessary.
If it compiles with nvcc 7.5, there is basically no point in filing an issue.
The only response you’ll get is “switch to nvcc 7.5”