Refactor optimization passes to subfolder

CMakeLists.txt

@@ -29,6 +29,7 @@ add_library(onnx2c_lib STATIC
 	src/node.cc
 	src/tensor.cc
 	src/util.cc
+	src/optimization_passes/unionize_tensors.cpp
 	${CMAKE_CURRENT_BINARY_DIR}/onnx.pb.cc
 	src/nodes/cast.cc
 	src/nodes/constantofshape.cc

src/graph.cc

@@ -584,80 +584,4 @@ void Graph::replaceWithQuantized(std::vector<Tensor*> &inputs)
 }
 
 
-// add t to union. Return allocated union number
-uint32_t Graph::add_to_free_union(Tensor *t)
-{
-	unsigned u=0;
-	// Check if tensor is already allocated to an union
-	for( ; u<tensor_unions.size(); u++ )
-		if( tensor_unions[u] == t)
-			return u;
-
-	// If not, search for free unions
-	u=0;
-	for( ; u<tensor_unions.size(); u++ )
-		if( tensor_unions[u] == NULL ) {
-			tensor_unions[u] = t;
-			t->assign_union(u);
-			return u;
-		}
-
-	// All unions in use, need a new one
-	LOG(TRACE) << "No free unions, creating a new one" << std::endl;
-	tensor_unions.push_back(t);
-	t->assign_union( tensor_unions.size()-1 );
-	return t->union_no;
-}
-void Graph::mark_union_unoccupied(uint32_t u)
-{
-	tensor_unions[u]=NULL;
-}
-// tag intermediate (graph internal) tensors
-// with union numbers so they can share memory
-// in a temporal fashion
-void Graph::unionize_tensors(void)
-{
-	for( auto n : nodes ) {
-		n->isResolved = false;
-	}
-
-	for( auto n : nodes ) {
-
-		// for each output tensor of node
-		for( auto o : n->outputs ) {
-			// assign tensor to next free union
-			// if it is an internal tensor that gets
-			// calculated by a node.
-			if( o->is_used() == false )
-				continue;
-			if( o->isIO == true )
-				continue;
-			// the Constant node is a bit weird - this check must be in
-			if( o->isConst == true )
-				continue;
-			if( o->initialize == true )
-				continue;
-			add_to_free_union(o);
-		}
-		// mark node as resolved
-		n->isResolved = true;
-
-		// Check if union slots can be re-used
-		for( unsigned ui=0; ui < tensor_unions.size(); ui++ ) {
-			Tensor *t = tensor_unions[ui];
-			// skip free slots
-			if ( t==NULL )
-				continue;
-
-			// when all the consumers of this tensors have consumed it
-			// the tensor is nolonger needed, and the union is free to host
-			// a new tensor
-			bool all_resolved = true;
-			for( auto c : t->consumers )
-				all_resolved &= c->isResolved;
-			if (all_resolved)
-				mark_union_unoccupied(ui);
-		}
-	}
-}
 

src/optimization_passes/README.md

@@ -0,0 +1,12 @@
+This folder contains the optimization passes onnx2c runs.
+
+Due to lack of up-front design, and surplus development time later,
+these passes are not stand-alone, but rather a part of the (ever
+increasingly pointless) Graph object.
+
+Setting up a mechanism of stand-alone optimization passes that can
+mutate the Graph via a public API sounds like a cleanup task that
+would make the code look much cleaner.
+
+But add a few optimization passes before starting to design such an API.
+

src/optimization_passes/unionize_tensors.cpp

@@ -0,0 +1,81 @@
+#include "graph.h"
+#include <cstdint>
+
+using namespace toC;
+// add t to union. Return allocated union number
+uint32_t Graph::add_to_free_union(Tensor *t)
+{
+	unsigned u=0;
+	// Check if tensor is already allocated to an union
+	for( ; u<tensor_unions.size(); u++ )
+		if( tensor_unions[u] == t)
+			return u;
+
+	// If not, search for free unions
+	u=0;
+	for( ; u<tensor_unions.size(); u++ )
+		if( tensor_unions[u] == NULL ) {
+			tensor_unions[u] = t;
+			t->assign_union(u);
+			return u;
+		}
+
+	// All unions in use, need a new one
+	LOG(TRACE) << "No free unions, creating a new one" << std::endl;
+	tensor_unions.push_back(t);
+	t->assign_union( tensor_unions.size()-1 );
+	return t->union_no;
+}
+void Graph::mark_union_unoccupied(uint32_t u)
+{
+	tensor_unions[u]=NULL;
+}
+// tag intermediate (graph internal) tensors
+// with union numbers so they can share memory
+// in a temporal fashion
+void Graph::unionize_tensors(void)
+{
+	for( auto n : nodes ) {
+		n->isResolved = false;
+	}
+
+	for( auto n : nodes ) {
+
+		// for each output tensor of node
+		for( auto o : n->outputs ) {
+			// assign tensor to next free union
+			// if it is an internal tensor that gets
+			// calculated by a node.
+			if( o->is_used() == false )
+				continue;
+			if( o->isIO == true )
+				continue;
+			// the Constant node is a bit weird - this check must be in
+			if( o->isConst == true )
+				continue;
+			if( o->initialize == true )
+				continue;
+			add_to_free_union(o);
+		}
+		// mark node as resolved
+		n->isResolved = true;
+
+		// Check if union slots can be re-used
+		for( unsigned ui=0; ui < tensor_unions.size(); ui++ ) {
+			Tensor *t = tensor_unions[ui];
+			// skip free slots
+			if ( t==NULL )
+				continue;
+
+			// when all the consumers of this tensors have consumed it
+			// the tensor is nolonger needed, and the union is free to host
+			// a new tensor
+			bool all_resolved = true;
+			for( auto c : t->consumers )
+				all_resolved &= c->isResolved;
+			if (all_resolved)
+				mark_union_unoccupied(ui);
+		}
+	}
+}
+