dot(zeros,x) -> zeros

theano/tensor/opt.py

@@ -249,6 +249,57 @@ def register_specialize_device(lopt, *tags, **kwargs):
     return lopt
 
 
+#####################
+# Dot optimizations #
+#####################
+
+@register_canonicalize
+@register_stabilize
+@gof.local_optimizer([None])
+def local_0_dot_x(node):
+
+    if isinstance(node.op, T.Dot):
+        x = node.inputs[0]
+        y = node.inputs[1]
+        replace = False
+        if x.owner and isinstance(x.owner.op, T.Alloc):
+            try:
+                val = get_constant_value(x.owner.inputs[0])
+                if numpy.all(val == 0):
+                    replace = True
+            except TypeError:
+                pass
+
+        if y.owner and isinstance(y.owner.op, T.Alloc):
+            try:
+                val = get_constant_value(y.owner.inputs[0])
+                if numpy.all(val == 0):
+                    replace = True
+            except TypeError:
+                pass
+        if isinstance(x, T.TensorConstant) and (x.tag.unique_value == 0):
+            replace = True
+        if isinstance(y, T.TensorConstant) and (y.tag.unique_value == 0):
+            replace = True
+
+        if replace:
+            if x.ndim ==2 and y.ndim == 2:
+                return [T.alloc( T.constant(0,dtype = node.outputs[0].type.dtype),
+                               x.shape[0], y.shape[1])]
+            elif x.ndim==1 and y.ndim == 2:
+                return [T.alloc( T.constant(0,dtype = node.outputs[0].type.dtype),
+                               y.shape[1])]
+            elif x.ndim==2 and y.ndim ==1:
+                return [T.alloc( T.constant(0,dtype = node.outputs[0].type.dtype),
+                               x.shape[0])]
+            elif x.ndim==1 and y.ndim==1:
+                return [T.constant(0,dtype = node.outputs[0].type.dtype)]
+        else:
+            return False
+
+    else:
+        return False
+
 ######################
 # DimShuffle lifters #
 ######################