[Fix] Failed ASSET GPU backend selection test when running in an environment with CUDA

elephant/test/test_asset.py

@@ -22,6 +22,7 @@
 
 from elephant import statistics, kernels
 from elephant.spike_train_generation import homogeneous_poisson_process
+from elephant.utils import get_cuda_capability_major, get_opencl_capability
 
 try:
     import sklearn
@@ -33,17 +34,8 @@
     HAVE_SKLEARN = True
     stretchedmetric2d = asset._stretched_metric_2d
 
-try:
-    import pyopencl
-    HAVE_PYOPENCL = asset.get_opencl_capability()
-except ImportError:
-    HAVE_PYOPENCL = False
-
-try:
-    import pycuda
-    HAVE_CUDA = asset.get_cuda_capability_major() > 0
-except ImportError:
-    HAVE_CUDA = False
+HAVE_PYOPENCL = get_opencl_capability()
+HAVE_CUDA = get_cuda_capability_major() != 0
 
 
 class AssetBinningTestCase(unittest.TestCase):
@@ -513,6 +505,7 @@ def test_intersection_matrix(self):
 
     #  regression test Issue #481
     #  see: https://github.com/NeuralEnsemble/elephant/issues/481
+    @unittest.skipIf(HAVE_CUDA, "CUDA available, will be used instead of CPU")
     def test_asset_choose_backend_opencl(self):
         class TestClassBackend(asset._GPUBackend):
 

elephant/utils.py

@@ -337,40 +337,30 @@ def round_binning_errors(values, tolerance=1e-8):
 
 def get_cuda_capability_major():
     """
-    Extracts CUDA capability major version of the first available Nvidia GPU
-    card, if detected. Otherwise, return 0.
+    If PyCUDA is available, extracts CUDA capability major version of the
+    first available Nvidia GPU card, if detected. Otherwise, returns 0.
 
     Returns
     -------
     int
         CUDA capability major version.
     """
-    cuda_success = 0
-    for libname in ("libcuda.so", "libcuda.dylib", "cuda.dll"):
-        try:
-            cuda = ctypes.CDLL(libname)
-        except OSError:
-            continue
-        else:
-            break
-    else:
-        # not found
-        return 0
-    result = cuda.cuInit(0)
-    if result != cuda_success:
+    try:
+        import pycuda.driver as cuda
+    except ImportError:
         return 0
-    device = ctypes.c_int()
-    # parse the first GPU card only
-    result = cuda.cuDeviceGet(ctypes.byref(device), 0)
-    if result != cuda_success:
+
+    try:
+        import pycuda.autoinit
+    except (cuda.RuntimeError, AttributeError):
         return 0
 
-    cc_major = ctypes.c_int()
-    cc_minor = ctypes.c_int()
-    cuda.cuDeviceComputeCapability(
-        ctypes.byref(cc_major), ctypes.byref(cc_minor), device
-    )
-    return cc_major.value
+    try:
+        device = cuda.Device(0)
+        major, _ = device.compute_capability()
+        return major
+    except cuda.Error:
+        return 0
 
 
 def get_opencl_capability():
@@ -380,7 +370,7 @@ def get_opencl_capability():
     Returns
     -------
     bool
-        True: if openCL platform detected and at least one device is found,
+        True: if OpenCL platform detected and at least one device is found,
         False: if OpenCL is not found or if no OpenCL devices are found
     """
     try: