Fix cudaDeviceGetAttribute(81) and cudart version for cudaPopConfiguration

[gangmul12]

1. cudaPushCallConfiguration is used in CUDA 9.2 so i changed preprocessor.
2. I found that cudaDeviceGetAttribute(cudaDevAttrMaxSharedMemoryPerMultiprocessor) should return the maximum amount of shared memory per "Multiprecessor"(which is same as a 'cluster' in gpgpu-sim ) but it currently return the maximum amount of shared memory per 'core'. so i add multiplication at shared_mem_size()
3. i found that gpgpu_shmem_size option is duplicated and only the first option parser works, so i removed second one and assign gpgpu_shmem_sizeDefault at its init() function.