CUDAEnsemble / RunPlan / RunPlanVector tests

[ptheywood]

Testing of CUDAEnsembles and related features is sparse / missing, with minor coverage provided in the test_logging files for each interface.

Some functions such as setRandomSimulationSeed do not appear in either test suite.

• CUDA C++ Tests
• CUDA C++ Fixes
• Python tests
• Python Fixes

[ptheywood]

Potential issues to ensure there are test for, as highligthed by JP (snippets written by me so may not be exact match):

Set seeds on RunPlans and maybe elsewhere appear to only support little numbers that fit int uint32_t, but most RNG implementations have 48 bits of entropy or more. We should probably modify the c++ to allow this, and not need a per int template specialisation. for it.

vec = pyflamegpu.RunPlanVector()
# This is a > 32 bit value, so will need a type specialised version of setSeed, or make it take 64 bit values all the time which is likely the better choice. 
x = 4628484339127465486
vec.setRandomSimulationSeed(x)
# OverflowError: in method 'RunPlan_setRandomSimulationSeed', argument 2 of type 'unsigned int'

SetRandomSimulationSeed methods may also have unsigned vs signed issues from python, but this may already have been resolved

vec = pyflamegpu.RunPlanVector()
# Use a small number. May encounter signed / unsigned issues.
x = 46
vec.setRandomSimulationSeed(x)

Calling SetPropertyInt(100) followed by GetPropertyInt returns 1 (the default).

[ptheywood]

re: Random seeds of more than 32 bits:

CuRAND accepts unsigned long long for seeds, i.e. uint64_t

The current host rand implementation uses std::mt19937 which only accepts uint32_t seeds, hence seeds throuhgout flame are 32 bit values. Instead, we can use std::mt19937_64 for a host rng with more than 32 bits of entropy, that accepts uint64_t as seeds.

This should address the seed type issue in python, but will require changes throughout many files where seeds are currently considered uint32_t

[ptheywood]

Have now confirmed that there are issues with:

• RunPlan::getProperty when setProperty has not been called (segfault in the c++ end)
  • Segfault occurs at RunPlan.h::314:

    FLAMEGPU2/include/flamegpu/sim/RunPlan.h

    Line 314 in c7fd9e8

    return *static_cast<T *>(it2->second.ptr);

• RunPlan::getProperty after calling setProperty returns the old value.