Made constant canonicalization [more?] correctly handle non-scalar constants.

theano/tensor/opt.py

@@ -597,13 +597,18 @@ def local_upcast_elemwise_constant_inputs(node):
 
     Rationale: it helps merge things like (1-x) and (1.0 - x).
     """
+    if len(node.outputs)>1:
+        return
+    try:
+        shape_i = node.env.shape_feature.shape_i
+    except AttributeError:
+        shape_i = None
     if isinstance(node.op, T.Elemwise):
         scalar_op = node.op.scalar_op
         #print "aa", scalar_op.output_types_preference
         if getattr(scalar_op,'output_types_preference',None) in (T.scal.upgrade_to_float, T.scal.upcast_out):
             # this is the kind of op that we can screw with the input dtypes by upcasting
             # explicitly
-            #print "HELLO??"
             output_dtype = node.outputs[0].type.dtype
             new_inputs = []
             for i in node.inputs:
@@ -615,8 +620,11 @@ def local_upcast_elemwise_constant_inputs(node):
                         if 0==sum((not b for b in i.broadcastable)): # I mean all() but this might work in python2.4
                             new_inputs.append(T.cast(cval_i, output_dtype))
                         else:
+                            if shape_i is None:
+                                return
                             new_inputs.append(T.alloc(T.cast(cval_i, output_dtype),
-                                *[Shape_i(d)(i) for d in xrange(i.ndim)]))
+                                *[shape_i(d)(i) for d in xrange(i.ndim)]))
+                            #print >> sys.stderr, "AAA", *[Shape_i(d)(i) for d in xrange(i.ndim)]
                     except TypeError:
                         if isinstance(i, T.TensorConstant): #for the case of a non-scalar
                             new_inputs.append(T.cast(i, output_dtype))
@@ -624,8 +632,15 @@ def local_upcast_elemwise_constant_inputs(node):
                             new_inputs.append(i)
 
             if new_inputs != node.inputs:
-                return [node.op(*new_inputs)]
-    
+                rval = [node.op(*new_inputs)]
+                if rval[0].type != node.outputs[0].type:
+                    print >> sys.stderr, "NODE:", node
+                    print >> sys.stderr, "NODE INPUT TYPES:", [i.type for i in node.inputs]
+                    print >> sys.stderr, "RVAL:", rval
+                    print >> sys.stderr, "NEW INPUT TYPES:", [i.type for i in new_inputs]
+                    print >> sys.stderr, "RVAL INPUT TYPES:", [i.type for i in rval[0].owner.inputs]
+                assert rval[0].type == node.outputs[0].type, (node, rval[0])
+                return rval
 
 ##################
 # Subtensor opts #