Make local_fill_to_alloc in fast_run, not fast_compile. Register it after…

theano/tensor/opt.py

@@ -1672,13 +1672,15 @@ def local_fill_to_alloc(node):
             node,)  # theano.printing.debugprint(node.outputs[0], file='str'))
         return rval
 
-compile.optdb['canonicalize'].register('local_fill_to_alloc',
-                                       in2out(local_fill_to_alloc),
-                                       1.1, 'fast_run')
+# Register this after stabilize at 1.5 to make sure stabilize don't
+# get affected by less canonicalized graph due to alloc.
+compile.optdb.register('local_fill_to_alloc',
+                       in2out(local_fill_to_alloc),
+                       1.51, 'fast_run')#, 'canonicalize')
 # Needed to clean some extra alloc added by local_fill_to_alloc
-compile.optdb['canonicalize'].register('local_elemwise_alloc',
-                                       in2out(local_elemwise_alloc),
-                                       1.11, 'fast_run')
+compile.optdb.register('local_elemwise_alloc',
+                       in2out(local_elemwise_alloc),
+                       1.52, 'fast_run')#, 'canonicalize')
 
 
 @register_canonicalize("fast_compile")

theano/tensor/tests/test_opt.py

@@ -481,7 +481,7 @@ class test_canonize(unittest.TestCase):
         mode = compile.mode.get_default_mode()
 
         opt = gof.Query(["canonicalize"])
-        opt = opt.including('ShapeOpt')
+        opt = opt.including('ShapeOpt', 'local_fill_to_alloc')
         opt = opt.excluding(
             'local_elemwise_fusion')
         mode = mode.__class__(linker=mode.linker, optimizer=opt)