fix for GpuCAReduceCuda

theano/gpuarray/opt.py

@@ -2408,8 +2408,8 @@ def local_gpu_elemwise_careduce(node):
         inp = node.inputs[0].owner.inputs[0]
         props = node.op._props_dict()
         props["pre_scalar_op"] = scalar.basic.sqr
-        out = GpuCAReduceCuda(**props)(inp)
         with inherit_stack_trace(node.outputs):
+            out = GpuCAReduceCuda(**props)(inp)
             return [out]
 
 

theano/gpuarray/tests/test_opt.py

@@ -40,7 +40,6 @@ def _check_stack_trace(thing):
                                    theano.tensor.elemwise.Elemwise,
                                    theano.ifelse.IfElse,
                                    GpuFromHost, HostFromGpu,
-                                   GpuCAReduceCuda,
                                    basic_ops.GpuContiguous,
                                    GpuElemwise,
                                    theano.printing.Print,