Describe the bug
Running Louvain example code produces CUDA illegal memory access error. We got the same error in both CUDA 11.0 and 11.4.
Steps/Code to reproduce bug
>>> import cugraph
>>> import cudf
>>> M = cudf.read_csv('karate.csv',
... delimiter = ' ',
... dtype=['int32', 'int32', 'float32'],
... header=None)
>>> G = cugraph.Graph()
>>> G.from_cudf_edgelist(M, source='0', destination='1')
>>> parts, modularity_score = cugraph.louvain(G)
terminate called after throwing an instance of 'raft::cuda_error'
what(): CUDA error encountered at: file=/opt/conda/envs/rapids/include/raft/handle.hpp line=268: call='cudaEventDestroy(event_)', Reason=cudaErrorIllegalAddress:an illegal memory access was encountered
Obtained 23 stack frames
#0 in /opt/conda/envs/rapids/lib/python3.8/site-packages/cugraph/community/louvain_wrapper.cpython-38-x86_64-linux-gnu.so(_ZN4raft9exception18collect_call_stackEv+0x3b) [0x7f9a1c6dbd0b]
#1 in /opt/conda/envs/rapids/lib/python3.8/site-packages/cugraph/community/louvain_wrapper.cpython-38-x86_64-linux-gnu.so(_ZN4raft10cuda_errorC1ERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE+0xbd) [0x7f9a1c6dc4bd]
#2 in /opt/conda/envs/rapids/lib/python3.8/site-packages/cugraph/community/louvain_wrapper.cpython-38-x86_64-linux-gnu.so(_ZN4raft8handle_t17destroy_resourcesEv+0x4a5) [0x7f9a1c6e4475]
#3 in /opt/conda/envs/rapids/lib/python3.8/site-packages/cugraph/community/louvain_wrapper.cpython-38-x86_64-linux-gnu.so(+0x4391a) [0x7f9a1c6cf91a]
#4 in /opt/conda/envs/rapids/lib/python3.8/site-packages/cugraph/community/louvain_wrapper.cpython-38-x86_64-linux-gnu.so(+0x48992) [0x7f9a1c6d4992]
#5 in python(PyCFunction_Call+0x54) [0x5564d3df6f44]
#6 in python(_PyObject_MakeTpCall+0x31e) [0x5564d3e0630e]
#7 in python(_PyEval_EvalFrameDefault+0x53cf) [0x5564d3e9c6ff]
#8 in python(_PyEval_EvalCodeWithName+0x2c3) [0x5564d3e79db3]
#9 in python(_PyFunction_Vectorcall+0x378) [0x5564d3e7b198]
#10 in python(_PyEval_EvalFrameDefault+0x4e03) [0x5564d3e9c133]
#11 in python(_PyEval_EvalCodeWithName+0x2c3) [0x5564d3e79db3]
#12 in python(PyEval_EvalCodeEx+0x39) [0x5564d3e7ae19]
#13 in python(PyEval_EvalCode+0x1b) [0x5564d3f1d24b]
#14 in python(+0x2522e3) [0x5564d3f1d2e3]
#15 in python(+0x26e543) [0x5564d3f39543]
#16 in python(+0xf4a4b) [0x5564d3dbfa4b]
#17 in python(PyRun_InteractiveLoopFlags+0xeb) [0x5564d3dbfbf6]
#18 in python(+0xf4c93) [0x5564d3dbfc93]
#19 in python(+0xf4e4a) [0x5564d3dbfe4a]
#20 in python(Py_BytesMain+0x39) [0x5564d3f3ee79]
#21 in /lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xe7) [0x7f9a1c9aebf7]
#22 in python(+0x1e6d69) [0x5564d3eb1d69]
Aborted (core dumped)
Environment overview (please complete the following information)
Describe the bug Running Louvain example code produces CUDA illegal memory access error. We got the same error in both CUDA 11.0 and 11.4.
Steps/Code to reproduce bug
Environment overview (please complete the following information)
gpuci/rapidsai:21.12-cuda11.0-devel-ubuntu18.04-py3.8
mamba install -c nvidia -c rapidsai-nightly -c numba -c conda-forge cugraph=21.12
Environment details
Click here to see environment details
Additional context Discovered this in CLX nightly build where we use cuGraph Louvain in our Network Mapping demo notebook.