Intermittant RTCLayerConcurrency Test Failure

[ptheywood]

The test TestCUDASimulationConcurrency.RTCLayerConcurrency is interrmittantly failing on my 3060ti machine, for -DCMAKE_BUILD_TYPE=Release -DCUDA_ARCH=86. This appears to be the case with SEATBELTS ON or OFF.

This is a performance test, so it's not a strict failure and may have been introduced by a driver change or just be an unreliable test.

./bin/Release/tests --gtest_filter="TestCUDASimulationConcurrency.RTCLayerConcurrency"

This failed ~ 40% of the time after running a numeber of times in quick succession.

Expected: (speedup) >= (SPEEDUP_THRESHOLD), actual: 0.718858 vs 1.5

This may be a side effect of the same GPU being used to drive my displays, however it is the only test in TestCUDASimulationConcurrency which failed. It does only take 25ms in total, which may be a factor (a longer running kernel might be needed).

I've not looked into this yet, just making a note for future investigation if needed.

/home/ptheywood/code/flamegpu/FLAMEGPU2/tests/test_cases/gpu/test_cuda_simulation_concurrency.cu:1288: Failure
Expected: (speedup) >= (SPEEDUP_THRESHOLD), actual: 0.698954 vs 1.5

[ptheywood]

@Robadob encountered this on Windows, using a 1080 driving display(s), so this is not linux exclusive.