[Execution] Differentiation between custom and standard nodes implemented and...

[Execution] Differentiation between custom and standard nodes implemented and standard node path completed

src/finn/transformation/infer_shapes.py

 import onnx.shape_inference as si
+import onnx.helper as helper
+
 
 
 def infer_shapes(model):
     """Ensure every tensor in the model has a specified shape (ValueInfo)."""
-    # currently just calls ONNX shape inference, but in the future we will
-    # have to handle shape inference for custom ops ourselves
-    model.model = si.infer_shapes(model.model)
+    for node in model.graph.node:
+        if node.domain == 'finn':
+            print("finn shape inference was called for node with op_type: " + str(node.op_type))
+        else:
+            # onnx shape inference unfortunately does not implement shape inference on a single node,
+            # it can only check the shapes of the node in an entire models 
+            # -- create a model which solely consists of current node (see also onnx_exec.py).
+            node_inputs = list(filter(lambda x: x.name in node.input, model.graph.input))
+            node_inputs += list(filter(lambda x: x.name in node.input, model.graph.value_info))
+            node_outputs = list(filter(lambda x: x.name in node.output, model.graph.output))
+            node_outputs += list(filter(lambda x: x.name in node.output, model.graph.value_info))
+            node_graph = helper.make_graph(
+                nodes=[node],
+                name="single-node-exec",
+                inputs=node_inputs,
+                outputs=node_outputs,
+            )
+            node_model = helper.make_model(node_graph)
+
+            node_model = si.infer_shapes(node_model)
+            print("onnx shape inference was called for node with op_type: " + str(node.op_type))
     # single-step operation, no need to call multiple times so return
     # model_was_changed = false
     return (model, False)