I am having some problems when using a POSIX threads program and CUDA at the same time.
I am obeying the one thread per context rule.
Are there any known problems with interactions between my (creation of) POSIX threads and cudaMalloc/general CUDA usage (i see in the crash dumps i get that libcuda.so.1 is on top of libpthread)
Sorry to be vague but i cannot track this puppy down…