basic implementation of forward pass

theano/gof/opt.py

@@ -1131,13 +1131,20 @@ class LocalMetaOptimizer(LocalOptimizer):
 
     """
 
-    def __init__(self, tracks=None, optimizers=()):
-        self._tracks = tracks
+    def __init__(self, optimizers=()):
+        self._tracks = [x for o in optimizers for x in o.tracks()]
         self.optimizers = list(optimizers)
         self.verbose = config.metaopt.verbose
+        self.track_dict = defaultdict(lambda: [])
+
+        for o in optimizers:
+            for c in o.tracks():
+                self.track_dict[c].append(o)
 
     def register(self, optimizer):
         self.optimizers.append(optimizer)
+        for c in optimizer.tracks():
+            self.track_dict[c].append(optimizer)
 
     def tracks(self):
         return self._tracks
@@ -1178,7 +1185,7 @@ class LocalMetaOptimizer(LocalOptimizer):
             print(("%s meta-optimizing %s (%d choices):" %
                    (self.__class__.__name__, node, len(self.optimizers))))
         timings = []
-        for opt in self.optimizers:
+        for opt in (self.track_dict[type(node.op)] + self.track_dict[node.op]):
             outputs = opt.transform(node)
             if outputs:
                 try:
@@ -2313,7 +2320,6 @@ class EquilibriumOptimizer(NavigatorOptimizer):
         self.final_optimizers = []
         self.cleanup_optimizers = []
         self.tracks_on_change_inputs = tracks_on_change_inputs
-
         for opt in optimizers:
             if isinstance(opt, LocalOptimizer):
                 if opt.tracks() is None:

theano/gpuarray/opt.py

@@ -15,6 +15,7 @@ from theano.compile.ops import shape_i
 from theano.gof import (local_optimizer, EquilibriumDB, TopoOptimizer,
                         LocalGroupDB,
                         SequenceDB, Optimizer, DB, toolbox, graph)
+from theano.gof.opt import LocalMetaOptimizer
 from theano.ifelse import IfElse
 from theano.misc.ordered_set import OrderedSet
 
@@ -1776,6 +1777,35 @@ def local_abstractconv3d_gradinputs_gemm(node):
     return [rval]
 
 
+class LocalCudaMetaOptimizer(LocalMetaOptimizer):
+
+    def time_call(self, fn):
+        start = time.time()
+        fn()[0].sync()
+        return time.time() - start
+
+
+class ConvMetaOptimizer(LocalCudaMetaOptimizer):
+
+    def __init__(self, optimizers):
+        super(ConvMetaOptimizer, self).__init__(optimizers)
+
+    def provide_inputs(self, node, inputs):
+        result = {}
+        img, kern = node.inputs
+        vars = (img, kern)
+
+        shapes = (node.op.imshp, node.op.kshp)
+        if(node.op.imshp is None or node.op.kshp is None or
+                any([s is None for shape in shapes for s in shape])):
+            return result
+
+        for(var, shape) in zip(vars, shapes):
+            result[var] = theano.shared(np.random.random(shape).astype(theano.config.floatX),
+                                        var.name, borrow=True)
+        return result
+
+
 # This deals with any abstract convs that have a transfer somewhere
 @register_opt('fast_compile', 'conv_dnn', 'cudnn')
 @op_lifter([AbstractConv2d,
@@ -2356,6 +2386,12 @@ register_opt('fast_compile')(abstractconv_groupopt)
 # to avoid a circular dependency problem with dnn
 from .dnn import (local_abstractconv_cudnn, local_abstractconv_gw_cudnn,
                   local_abstractconv_gi_cudnn)     # noqa: 402
+
+conv_metaopt = ConvMetaOptimizer((local_abstractconv_gemm,
+                                  local_abstractconv_cudnn))
+abstractconv_groupopt.register('conv_metaopt', conv_metaopt, 0, 'conv_meta')
+
+
 abstractconv_groupopt.register('local_abstractconv_dnn',
                                local_abstractconv_cudnn, 20,
                                'conv_dnn',