Fix non-cudnn reductions that were all subtly broken by 91bc16c3.

theano/gpuarray/dnn.py

@@ -3783,7 +3783,7 @@ def local_dnn_reduction(node):
     with inherit_stack_trace(node.outputs):
         ret = GpuDnnReduction(scal,
                               node.op.axis,
-                              node.op.acc_dtype,
+                              node.op._acc_dtype(node.inputs[0].dtype),
                               node.op.dtype,
                               False)(node.inputs[0])
         return [post(ret)]

theano/gpuarray/opt.py

@@ -1207,7 +1207,7 @@ def local_gpua_careduce(op, context_name, inputs, outputs):
             return False
         x, = inputs
         idtype = x.dtype
-        adtype = getattr(op, 'acc_dtype', idtype)
+        adtype = getattr(op, 'acc_dtype', None)
         odtype = getattr(op, 'dtype', outputs[0].dtype)
 
         # Force accumulator to float32 for float32 inputs since tree