separate operators and ir, create new abstraction for element wise ops

hpvm/projects/onnx/frontend/ir.py

@@ -120,64 +120,21 @@ class DFGNode:
         self.outputs = []
         self.name = layer.name
         self.op_type = layer.op_type
-        
-class AddNode(DFGNode):
-    pass
-class MatMul(DFGNode):
+
+'''
+Element wise operatos that is for activation function
+e.g. HardSigmoid, LeakyRelu, PRelu, Pow, Reciprocal,
+Relu, Selu, Sigmoid, Softplus, Sqrt, ThresholdedRelu,
+Abs, Ceil, Elu, Floor, Neg
+'''
+class ActivationNode(DFGNode):
     pass
-class SoftMax(DFGNode):
+'''
+ELement wise operators that is not for activation function.
+In other words, they are logical comparison operators 
+e.g. And, Equal, Greater, GreaterOrEqual, Less, LessOrEqual, 
+Or, Xor
+'''
+class LogicalOpNode(DFGNode):
     pass
-class Conv2DNode(DFGNode):
-    def __init__(self, layer):
-        DFGNode.__init__(self, layer)
-        self.weights = layer.get_weights()[0]
-        print("\t", self.weights.shape)
-        self.use_bias = layer.use_bias
-        self.padding = layer.padding
-        self.strides = layer.strides
-        print("\t", self.strides)
-        print("\tPadding = ", self.padding)
-class DepthwiseConv2DNode(DFGNode):
-    def __init__(self, layer):
-        DFGNode.__init__(self, layer)
-        self.weights = layer.get_weights()[0]
-        print("\t", self.weights.shape)
-        self.use_bias = layer.use_bias
-        self.padding = layer.padding
-        self.strides = layer.strides
-        print("\t", self.strides)
-        print("\tPadding = ", self.padding)
-class DenseNode(DFGNode):
-    def __init__(self, layer):
-        DFGNode.__init__(self, layer)
-        self.weights = layer.get_weights()[0]
-        print("\t", self.weights.shape)
-        self.use_bias = layer.use_bias
-class MaxPool2DNode(DFGNode):
-    def __init__(self, layer):
-        DFGNode.__init__(self, layer)
-        self.pool_size = layer.pool_size
-        self.strides = layer.strides
-        print("\t pool_size = ", self.pool_size)
-        print("\t strides = ", self.strides)
-class AveragePooling2DNode(DFGNode):
-    def __init__(self, layer):
-        DFGNode.__init__(self, layer)
-        self.pool_size = layer.pool_size
-        self.strides = layer.strides
-        print("\t pool_size = ", self.pool_size)
-        print("\t strides = ", self.strides)
-class ZeroPadding2DNode(DFGNode):
-    def __init__(self, layer):
-        DFGNode.__init__(self, layer)
-        print ("***ZeroPaddding \n");
-        self.padding = layer.padding
-        print ("padding = ", self.padding)
-class BatchNormalizationNode(DFGNode):
-    def __init__(self, layer):
-        DFGNode.__init__(self, layer)
-        self.epsilon = layer.epsilon
-        self.beta = layer.beta
-        self.gamma = layer.gamma
-        self.moving_mean = layer.moving_mean
-        self.moving_variance = layer.moving_variance
\ No newline at end of file
+        
\ No newline at end of file

hpvm/projects/onnx/frontend/operators.py

+from .ir import DFGNode, ActivationNode, LogicalOpNode
+
+class AddNode(DFGNode):
+    pass
+class MatMul(DFGNode):
+    pass
+class SoftMax(DFGNode):
+    pass
+class Conv2DNode(DFGNode):
+    def __init__(self, layer):
+        DFGNode.__init__(self, layer)
+        self.weights = layer.get_weights()[0]
+        print("\t", self.weights.shape)
+        self.use_bias = layer.use_bias
+        self.padding = layer.padding
+        self.strides = layer.strides
+        print("\t", self.strides)
+        print("\tPadding = ", self.padding)
+class DepthwiseConv2DNode(DFGNode):
+    def __init__(self, layer):
+        DFGNode.__init__(self, layer)
+        self.weights = layer.get_weights()[0]
+        print("\t", self.weights.shape)
+        self.use_bias = layer.use_bias
+        self.padding = layer.padding
+        self.strides = layer.strides
+        print("\t", self.strides)
+        print("\tPadding = ", self.padding)
+class DenseNode(DFGNode):
+    def __init__(self, layer):
+        DFGNode.__init__(self, layer)
+        self.weights = layer.get_weights()[0]
+        print("\t", self.weights.shape)
+        self.use_bias = layer.use_bias
+class MaxPool2DNode(DFGNode):
+    def __init__(self, layer):
+        DFGNode.__init__(self, layer)
+        self.pool_size = layer.pool_size
+        self.strides = layer.strides
+        print("\t pool_size = ", self.pool_size)
+        print("\t strides = ", self.strides)
+class AveragePooling2DNode(DFGNode):
+    def __init__(self, layer):
+        DFGNode.__init__(self, layer)
+        self.pool_size = layer.pool_size
+        self.strides = layer.strides
+        print("\t pool_size = ", self.pool_size)
+        print("\t strides = ", self.strides)
+class ZeroPadding2DNode(DFGNode):
+    def __init__(self, layer):
+        DFGNode.__init__(self, layer)
+        print ("***ZeroPaddding \n");
+        self.padding = layer.padding
+        print ("padding = ", self.padding)
+class BatchNormalizationNode(DFGNode):
+    def __init__(self, layer):
+        DFGNode.__init__(self, layer)
+        self.epsilon = layer.epsilon
+        self.beta = layer.beta
+        self.gamma = layer.gamma
+        self.moving_mean = layer.moving_mean
+        self.moving_variance = layer.moving_variance
\ No newline at end of file