Fix Advanced Activation Layers Bugs and Create New Test

keras2c/layer2c.py

@@ -355,7 +355,7 @@ def _write_layer_AdvancedActivation(self, layer, inputs, outputs, i):
         if is_model_input:
             inp = inputs + '->'
         else:
-            inp = inputs + '.'
+            inp = inputs[1:] + '.' #remove & prefix that is added to output_layers to pass tensor by reference
 
         if layer_type(layer) == 'LeakyReLU':
             self.layers += 'k2c_LeakyReLU(' + inp + 'array,' + \

keras2c/weights2c.py

@@ -118,9 +118,17 @@ def write_weights(self, verbose=True):
                 - **static_vars** (*str*): code fora C struct containing static variables
                     (eg, states of a stateful RNN)
         """
+        if verbose:
+            print(__class__.__name__+"."+__name__)
+
         for layer in self.model.layers:
             method = getattr(self, '_write_weights_' + layer_type(layer))
             method(layer)
+            if(verbose):
+                print("")
+                print(__class__.__name__+"."+__name__+" : layer : "+layer.name+" : "+method.__name__)
+                print("Stack: " + str(self.stack_vars))
+                print("Malloc: " + str(self.malloc_vars))
         return self.stack_vars, self.malloc_vars, self._write_static_vars()
 
     def _write_static_vars(self):
@@ -647,7 +655,7 @@ def _write_weights_LeakyReLU(self, layer):
 
     def _write_weights_ThresholdedReLU(self, layer):
         theta = layer.get_config()['theta']
-        self.stack_vars = 'float ' + layer.name + \
+        self.stack_vars += 'float ' + layer.name + \
             '_theta = ' + str(theta) + '; \n'
         self.stack_vars += '\n\n'
 

tests/test_advanced_activation_layers.py

@@ -82,3 +82,25 @@ def test_ReLU(self):
         keras2c_main.k2c(model, name)
         rcode = build_and_run(name)
         self.assertEqual(rcode, 0)
+
+    def test_AdvancedActivationLayers_NonInputLayer(self):
+        inshp = (9, 7, 6, 3)
+        alpha = 0.5
+        theta = 0.3
+        max_value = 1.0
+        negative_slope = 1.0
+        threshold = 0.3
+        input_layer = keras.layers.Input(inshp)
+        first = keras.layers.LeakyReLU(alpha=0.3)(input_layer)
+        middle1 = keras.layers.LeakyReLU(alpha=alpha)(first)
+        middle2 = keras.layers.PReLU(alpha_initializer='glorot_uniform')(middle1)
+        middle3 = keras.layers.ELU(alpha=alpha)(middle2)
+        middle4 = keras.layers.ThresholdedReLU(theta=theta)(middle3)
+        output_layer  = keras.layers.ReLU(max_value=max_value,
+                              negative_slope=negative_slope,
+                              threshold=threshold)(middle4)
+        model = keras.models.Model(inputs=input_layer, outputs=output_layer)
+        name = 'test___AdvancedActivationLayers_NonInputLayers' + str(int(time.time()))
+        keras2c_main.k2c(model, name)
+        rcode = build_and_run(name)
+        self.assertEqual(rcode, 0)