change final_opt to be a keyword argument

theano/gof/optdb.py

@@ -230,16 +230,18 @@ class EquilibriumDB(DB):
     def register(self, name, obj, *tags, **kwtags):
         # if name == 'cut_gpua_constant_transfers':
         #     import ipdb;ipdb.set_trace()
-        if 'final_opt' in tags:
-            final_opt = True
+        if 'final_opt' in kwtags:
+            final_opt = kwtags['final_opt']
+            kwtags.pop('final_opt', None)
         else:
             final_opt = False
         super(EquilibriumDB, self).register(name, obj, *tags, **kwtags)
         self.__final__[name] = final_opt
 
     def query(self, *tags, **kwtags):
-        opts = super(EquilibriumDB, self).query(*tags, **kwtags)
-        final_opts = [o for o in opts if self.__final__.get(o.name, False)]
+        _opts = super(EquilibriumDB, self).query(*tags, **kwtags)
+        final_opts = [o for o in _opts if self.__final__.get(o.name, False)]
+        opts = [o for o in _opts if o not in final_opts]
         if len(final_opts) == 0:
             final_opts = None
         return opt.EquilibriumOptimizer(

theano/sandbox/cuda/__init__.py

@@ -27,10 +27,11 @@ def register_opt(*tags, **kwargs):
     if any([not isinstance(t, str) for t in tags]):
         raise RuntimeError("Bad call to register_opt."
                            " All tags must be strings.", tags)
+
     def f(local_opt):
         name = (kwargs and kwargs.pop('name')) or local_opt.__name__
         gpu_optimizer.register(name, local_opt, 'fast_run', 'fast_compile',
-                               'gpu', *tags)
+                               'gpu', *tags, **kwargs)
         return local_opt
     return f
 

theano/sandbox/cuda/opt.py

@@ -94,13 +94,13 @@ optdb.register('gpu_after_fusion',
 
 # Register merge_optimizer as a global opt
 gpu_optimizer.register('gpu_merge', theano.gof.opt.merge_optimizer,
-                       'fast_run', 'fast_compile', 'final_opt')
+                       'fast_run', 'fast_compile', final_opt=True)
 
 
 # register local_track_shape_i at this level too
 # to make multi-level lift of shape work.
 register_opt()(theano.tensor.opt.local_track_shape_i)
-register_opt('final_opt', name='gpu_constant_folding')(
+register_opt(final_opt=True, name='gpu_constant_folding')(
     tensor.opt.constant_folding)
 register_opt()(theano.tensor.opt.local_subtensor_make_vector)