CUDA Segfault

[danem]

I've encountered this weird bug that I haven't been able to quite track down. I'm trying to use GLM with a program compiled by nvcc 8.0.54. A call glm::lookAt in another source file causes a segfault here. If step through the program with lldb I see the value passed to normalize (ie: center - eye) is null, causing the segfault. Oddly enough I have found that this only segfaults if I reference certain GLM functions in the main cuda source file. Here is a minimal program that demonstrates the behavior.

This is on OSX 10.11.6 using GLM 0.9.8.3.

edit: I've also discovered that if I change the extension of the Camera.cc to Camera.cu everything works nicely. Suggesting the issue has something to do with how nvcc interprets the files.

[blahgeek]

Hi @danem could you test if #613 fixes your issue? (or you can just go to your .../include/glm/simd/platform.h line 175, change GLM_COMPILER_CUDA to GLM_COMPILER_CUDA80 to see if it works)

[blahgeek]

#define GLM_FORCE_CXX03 before including glm.hpp also works for me.

[danem]

@blahgeek That did the trick! Thanks! Just as a note, only the first suggestion worked for me, the second suggestion still resulted in a segfault.

[Groovounet]

Thanks @blahgeek for the fix!
Christophe