specification of minimal device capability

[moritzaugustin]

from the used cuda features alone the minimal compute capability is 2.0 when running some tests on asterope (device capability 2.1):

1. CUBA does not show any spikes with cuda_standalone (but not cpp_standalone)
2. the feature test test_cuda_standalone.py (copied from test_cpp_standalone.py replacing all cpp_standalone occurences with cuda_standalone) does not work as expected: AssertionError in line 424)

running cuda-memcheck ./main -- e.g. for the 1. example above -- a cudaErrorLaunchOutofResources error is reported due to too many resources requested for launch on CUDA API call to cudaLaunch

it is likely that compute capability 3.5 is required due to the available registers per thread (c.f. https://en.wikipedia.org/wiki/CUDA) -- we have to check which kernels are problematic if we wanted to relax the required cc version no

this is related with iss #4

[moritzaugustin]

do not support very old architectures such as fermi

[moritzaugustin]

first test for practical relevance: add warning to code and info to user documentation that archs up to some compute capability do not support hardware atomicadd (then our custom software atomicadd is used) which makes them potentially slower

[denisalevi]

Also check what compute capability the spatialneuron implementation needs. See this comment