Initialize TransientBuffers on allocation instead of on top of network execution code

Author: lukamac

Deeploy/CommonExtensions/CodeTransformationPasses/MemoryAllocation.py

@@ -130,6 +130,8 @@ def apply(self,
                 ctxt._dynamicSize[memoryLevel] += int(buffer.sizeInBytes())
 
             executionBlock.addLeft(buffer.allocTemplate, buffer._bufferRepresentation())
+            if isinstance(buffer, TransientBuffer):
+                executionBlock.addLeft(buffer.initTemplate, buffer._bufferRepresentation())
 
         for levels in ctxt._dynamicSize.keys():
             if levels not in ctxt._maxDynamicSize:

Deeploy/DeeployTypes.py

@@ -2893,7 +2893,7 @@ def generateInferenceInitializationCode(self) -> str:
         callStack = ''
         for node in self.ctxt.localObjects.values():
             # WIESEP: We don't want to initialize the struct buffers as this should be handled by the ArgumentStructGeneration
-            if isinstance(node, StructBuffer):
+            if isinstance(node, (StructBuffer, TransientBuffer)):
                 continue
 
             name = node.name