Open
Description
#1160 added checks that comm radii are a factor of the environment when using spacial comms with seatbelts enabled.
This seems to be incorrectly triggering in some cases, including the python_native example.
python3 ../examples/python_native/boids_spatial3D_wrapped/boids_spatial3D.py -t -v
FLAME GPU 2.0.0-rc.2+25a37160
Simulation configuration:
Random Seed: 1706546446
Steps: 1
RTC Initialisation Processing time: 26.107000 s
Init Function Processing time: 0.000000 s
Processing Simulation Step 0
Traceback (most recent call last):
File "/users/pheywood/flamegpu/FLAMEGPU2/build-gh-123/../examples/python_native/boids_spatial3D_wrapped/boids_spatial3D.py", line 430, in <module>
cudaSimulation.simulate()
File "/users/pheywood/flamegpu/FLAMEGPU2/build-gh-123/lib/Release/python/venv/lib64/python3.9/site-packages/pyflamegpu/pyflamegpu.py", line 9255, in simulate
return _pyflamegpu.CUDASimulation_simulate(self, *args)
pyflamegpu.pyflamegpu.FLAMEGPURuntimeException: (DeviceError) Device function 'inputdata' reported 40000 errors.
First error:
flamegpu/runtime/messaging/MessageSpatial3D/MessageSpatial3DDevice.cuh(545)[52,0,0][0,0,0]:
Spatial messaging radius (0.05) is not a factor of environment dimensions (1, 1, 1), this is unsupported for the wrapped iterator, MessageSpatial3D::In::wrap().
Probably due to inexact floating point stuff.
This needs fixing, and the test suite(s) expanding to cover a broader range of cases / edge cases.
Additionally, this check is currenrlty doing expensive ops per thread on device (when seatbelts enabled) which could instead be done once on the host, then a single bool / bit being set on device and read when the iterator is requested.