Merge pull request #7251 from ethereum/sideEffectsForUserDefinedFunctions

Side effects for user defined functions

Changelog.md

@@ -6,6 +6,7 @@ Language Features:
 
 Compiler Features:
  * Optimizer: Add rule that replaces the BYTE opcode by 0 if the first argument is larger than 31.
+ * Yul Optimizer: Take side-effect-freeness of user-defined functions into account.
 
 
 Bugfixes:

libyul/optimiser/CommonSubexpressionEliminator.cpp

@@ -23,6 +23,9 @@
 
 #include <libyul/optimiser/Metrics.h>
 #include <libyul/optimiser/SyntacticalEquality.h>
+#include <libyul/optimiser/CallGraphGenerator.h>
+#include <libyul/optimiser/Semantics.h>
+#include <libyul/SideEffects.h>
 #include <libyul/Exceptions.h>
 #include <libyul/AsmData.h>
 #include <libyul/Dialect.h>
@@ -31,6 +34,23 @@ using namespace std;
 using namespace dev;
 using namespace yul;
 
+void CommonSubexpressionEliminator::run(Dialect const& _dialect, Block& _ast)
+{
+	CommonSubexpressionEliminator cse{
+		_dialect,
+		SideEffectsPropagator::sideEffects(_dialect, CallGraphGenerator::callGraph(_ast))
+	};
+	cse(_ast);
+}
+
+CommonSubexpressionEliminator::CommonSubexpressionEliminator(
+	Dialect const& _dialect,
+	map<YulString, SideEffects> _functionSideEffects
+):
+	DataFlowAnalyzer(_dialect, std::move(_functionSideEffects))
+{
+}
+
 void CommonSubexpressionEliminator::visit(Expression& _e)
 {
 	bool descend = true;

libyul/optimiser/CommonSubexpressionEliminator.h

@@ -27,6 +27,7 @@ namespace yul
 {
 
 struct Dialect;
+struct SideEffects;
 
 /**
  * Optimisation stage that replaces expressions known to be the current value of a variable
@@ -37,7 +38,14 @@ struct Dialect;
 class CommonSubexpressionEliminator: public DataFlowAnalyzer
 {
 public:
-	CommonSubexpressionEliminator(Dialect const& _dialect): DataFlowAnalyzer(_dialect) {}
+	/// Runs the CSE pass. @a _ast needs to be the complete AST of the program!
+	static void run(Dialect const& _dialect, Block& _ast);
+
+private:
+	CommonSubexpressionEliminator(
+		Dialect const& _dialect,
+		std::map<YulString, SideEffects> _functionSideEffects
+	);
 
 protected:
 	using ASTModifier::visit;

libyul/optimiser/DataFlowAnalyzer.cpp

@@ -211,7 +211,7 @@ void DataFlowAnalyzer::handleAssignment(set<YulString> const& _variables, Expres
 {
 	clearValues(_variables);
 
-	MovableChecker movableChecker{m_dialect};
+	MovableChecker movableChecker{m_dialect, &m_functionSideEffects};
 	if (_value)
 		movableChecker.visit(*_value);
 	else

libyul/optimiser/DataFlowAnalyzer.h

@@ -26,6 +26,7 @@
 #include <libyul/optimiser/KnowledgeBase.h>
 #include <libyul/YulString.h>
 #include <libyul/AsmData.h>
+#include <libyul/SideEffects.h>
 
 // TODO avoid
 #include <libevmasm/Instruction.h>
@@ -38,6 +39,7 @@
 namespace yul
 {
 struct Dialect;
+struct SideEffects;
 
 /**
  * Base class to perform data flow analysis during AST walks.
@@ -67,8 +69,16 @@ struct Dialect;
 class DataFlowAnalyzer: public ASTModifier
 {
 public:
-	explicit DataFlowAnalyzer(Dialect const& _dialect):
+	/// @param _functionSideEffects
+	///            Side-effects of user-defined functions. Worst-case side-effects are assumed
+	///            if this is not provided or the function is not found.
+	///            The parameter is mostly used to determine movability of expressions.
+	explicit DataFlowAnalyzer(
+		Dialect const& _dialect,
+		std::map<YulString, SideEffects> _functionSideEffects = {}
+	):
 		m_dialect(_dialect),
+		m_functionSideEffects(std::move(_functionSideEffects)),
 		m_knowledgeBase(_dialect, m_value)
 	{}
 
@@ -124,6 +134,9 @@ protected:
 	) const;
 
 	Dialect const& m_dialect;
+	/// Side-effects of user-defined functions. Worst-case side-effects are assumed
+	/// if this is not provided or the function is not found.
+	std::map<YulString, SideEffects> m_functionSideEffects;
 
 	/// Current values of variables, always movable.
 	std::map<YulString, Expression const*> m_value;

libyul/optimiser/Semantics.cpp

@@ -35,8 +35,12 @@ using namespace dev;
 using namespace yul;
 
 
-SideEffectsCollector::SideEffectsCollector(Dialect const& _dialect, Expression const& _expression):
+SideEffectsCollector::SideEffectsCollector(
-	SideEffectsCollector(_dialect)
+		Dialect const& _dialect,
+		Expression const& _expression,
+		map<YulString, SideEffects> const* _functionSideEffects
+):
+	SideEffectsCollector(_dialect, _functionSideEffects)
 {
 	visit(_expression);
 }
@@ -64,8 +68,11 @@ void SideEffectsCollector::operator()(FunctionCall const& _functionCall)
 {
 	ASTWalker::operator()(_functionCall);
 
-	if (BuiltinFunction const* f = m_dialect.builtin(_functionCall.functionName.name))
+	YulString functionName = _functionCall.functionName.name;
+	if (BuiltinFunction const* f = m_dialect.builtin(functionName))
 		m_sideEffects += f->sideEffects;
+	else if (m_functionSideEffects && m_functionSideEffects->count(functionName))
+		m_sideEffects += m_functionSideEffects->at(functionName);
 	else
 		m_sideEffects += SideEffects::worst();
 }

libyul/optimiser/Semantics.h

@@ -22,6 +22,7 @@
 
 #include <libyul/optimiser/ASTWalker.h>
 #include <libyul/SideEffects.h>
+#include <libyul/AsmData.h>
 
 #include <set>
 
@@ -36,8 +37,15 @@ struct Dialect;
 class SideEffectsCollector: public ASTWalker
 {
 public:
-	explicit SideEffectsCollector(Dialect const& _dialect): m_dialect(_dialect) {}
+	explicit SideEffectsCollector(
-	SideEffectsCollector(Dialect const& _dialect, Expression const& _expression);
+		Dialect const& _dialect,
+		std::map<YulString, SideEffects> const* _functionSideEffects = nullptr
+	): m_dialect(_dialect), m_functionSideEffects(_functionSideEffects) {}
+	SideEffectsCollector(
+		Dialect const& _dialect,
+		Expression const& _expression,
+		std::map<YulString, SideEffects> const* _functionSideEffects = nullptr
+	);
 	SideEffectsCollector(Dialect const& _dialect, Statement const& _statement);
 	SideEffectsCollector(Dialect const& _dialect, Block const& _ast);
 
@@ -59,6 +67,7 @@ public:
 
 private:
 	Dialect const& m_dialect;
+	std::map<YulString, SideEffects> const* m_functionSideEffects = nullptr;
 	SideEffects m_sideEffects;
 };
 
@@ -108,7 +117,10 @@ private:
 class MovableChecker: public SideEffectsCollector
 {
 public:
-	explicit MovableChecker(Dialect const& _dialect): SideEffectsCollector(_dialect) {}
+	explicit MovableChecker(
+		Dialect const& _dialect,
+		std::map<YulString, SideEffects> const* _functionSideEffects = nullptr
+	): SideEffectsCollector(_dialect, _functionSideEffects) {}
 	MovableChecker(Dialect const& _dialect, Expression const& _expression);
 
 	void operator()(Identifier const& _identifier) override;

libyul/optimiser/Suite.cpp

@@ -23,6 +23,7 @@
 #include <libyul/optimiser/Disambiguator.h>
 #include <libyul/optimiser/VarDeclInitializer.h>
 #include <libyul/optimiser/BlockFlattener.h>
+#include <libyul/optimiser/CallGraphGenerator.h>
 #include <libyul/optimiser/ControlFlowSimplifier.h>
 #include <libyul/optimiser/DeadCodeEliminator.h>
 #include <libyul/optimiser/FunctionGrouper.h>
@@ -37,6 +38,7 @@
 #include <libyul/optimiser/UnusedPruner.h>
 #include <libyul/optimiser/ExpressionSimplifier.h>
 #include <libyul/optimiser/CommonSubexpressionEliminator.h>
+#include <libyul/optimiser/Semantics.h>
 #include <libyul/optimiser/SSAReverser.h>
 #include <libyul/optimiser/SSATransform.h>
 #include <libyul/optimiser/StackCompressor.h>
@@ -85,7 +87,7 @@ void OptimiserSuite::run(
 	DeadCodeEliminator{_dialect}(ast);
 	FunctionGrouper{}(ast);
 	EquivalentFunctionCombiner::run(ast);
-	UnusedPruner::runUntilStabilised(_dialect, ast, reservedIdentifiers);
+	UnusedPruner::runUntilStabilisedOnFullAST(_dialect, ast, reservedIdentifiers);
 	BlockFlattener{}(ast);
 	ControlFlowSimplifier{_dialect}(ast);
 	StructuralSimplifier{_dialect}(ast);
@@ -114,7 +116,8 @@ void OptimiserSuite::run(
 			RedundantAssignEliminator::run(_dialect, ast);
 
 			ExpressionSimplifier::run(_dialect, ast);
-			CommonSubexpressionEliminator{_dialect}(ast);
+
+			CommonSubexpressionEliminator::run(_dialect, ast);
 		}
 
 		{
@@ -124,19 +127,20 @@ void OptimiserSuite::run(
 			ControlFlowSimplifier{_dialect}(ast);
 			BlockFlattener{}(ast);
 			DeadCodeEliminator{_dialect}(ast);
-			UnusedPruner::runUntilStabilised(_dialect, ast, reservedIdentifiers);
+			UnusedPruner::runUntilStabilisedOnFullAST(_dialect, ast, reservedIdentifiers);
 		}
+
 		{
 			// simplify again
-			CommonSubexpressionEliminator{_dialect}(ast);
+			CommonSubexpressionEliminator::run(_dialect, ast);
-			UnusedPruner::runUntilStabilised(_dialect, ast, reservedIdentifiers);
+			UnusedPruner::runUntilStabilisedOnFullAST(_dialect, ast, reservedIdentifiers);
 		}
 
 		{
 			// reverse SSA
 			SSAReverser::run(ast);
-			CommonSubexpressionEliminator{_dialect}(ast);
+			CommonSubexpressionEliminator::run(_dialect, ast);
-			UnusedPruner::runUntilStabilised(_dialect, ast, reservedIdentifiers);
+			UnusedPruner::runUntilStabilisedOnFullAST(_dialect, ast, reservedIdentifiers);
 
 			ExpressionJoiner::run(ast);
 			ExpressionJoiner::run(ast);
@@ -147,7 +151,7 @@ void OptimiserSuite::run(
 		{
 			// run functional expression inliner
 			ExpressionInliner(_dialect, ast).run();
-			UnusedPruner::runUntilStabilised(_dialect, ast, reservedIdentifiers);
+			UnusedPruner::runUntilStabilisedOnFullAST(_dialect, ast, reservedIdentifiers);
 		}
 
 		{
@@ -156,7 +160,7 @@ void OptimiserSuite::run(
 			SSATransform::run(ast, dispenser);
 			RedundantAssignEliminator::run(_dialect, ast);
 			RedundantAssignEliminator::run(_dialect, ast);
-			CommonSubexpressionEliminator{_dialect}(ast);
+			CommonSubexpressionEliminator::run(_dialect, ast);
 		}
 
 		{
@@ -177,12 +181,12 @@ void OptimiserSuite::run(
 			BlockFlattener{}(ast);
 			DeadCodeEliminator{_dialect}(ast);
 			ControlFlowSimplifier{_dialect}(ast);
-			CommonSubexpressionEliminator{_dialect}(ast);
+			CommonSubexpressionEliminator::run(_dialect, ast);
 			SSATransform::run(ast, dispenser);
 			RedundantAssignEliminator::run(_dialect, ast);
 			RedundantAssignEliminator::run(_dialect, ast);
-			UnusedPruner::runUntilStabilised(_dialect, ast, reservedIdentifiers);
+			UnusedPruner::runUntilStabilisedOnFullAST(_dialect, ast, reservedIdentifiers);
-			CommonSubexpressionEliminator{_dialect}(ast);
+			CommonSubexpressionEliminator::run(_dialect, ast);
 		}
 	}
 
@@ -190,19 +194,19 @@ void OptimiserSuite::run(
 
 	ExpressionJoiner::run(ast);
 	Rematerialiser::run(_dialect, ast);
-	UnusedPruner::runUntilStabilised(_dialect, ast, reservedIdentifiers);
+	UnusedPruner::runUntilStabilisedOnFullAST(_dialect, ast, reservedIdentifiers);
 	ExpressionJoiner::run(ast);
-	UnusedPruner::runUntilStabilised(_dialect, ast, reservedIdentifiers);
+	UnusedPruner::runUntilStabilisedOnFullAST(_dialect, ast, reservedIdentifiers);
 	ExpressionJoiner::run(ast);
-	UnusedPruner::runUntilStabilised(_dialect, ast, reservedIdentifiers);
+	UnusedPruner::runUntilStabilisedOnFullAST(_dialect, ast, reservedIdentifiers);
 
 	SSAReverser::run(ast);
-	CommonSubexpressionEliminator{_dialect}(ast);
+	CommonSubexpressionEliminator::run(_dialect, ast);
-	UnusedPruner::runUntilStabilised(_dialect, ast, reservedIdentifiers);
+	UnusedPruner::runUntilStabilisedOnFullAST(_dialect, ast, reservedIdentifiers);
 
 	ExpressionJoiner::run(ast);
 	Rematerialiser::run(_dialect, ast);
-	UnusedPruner::runUntilStabilised(_dialect, ast, reservedIdentifiers);
+	UnusedPruner::runUntilStabilisedOnFullAST(_dialect, ast, reservedIdentifiers);
 
 	// This is a tuning parameter, but actually just prevents infinite loops.
 	size_t stackCompressorMaxIterations = 16;

libyul/optimiser/UnusedPruner.cpp

@@ -20,12 +20,14 @@
 
 #include <libyul/optimiser/UnusedPruner.h>
 
+#include <libyul/optimiser/CallGraphGenerator.h>
 #include <libyul/optimiser/NameCollector.h>
 #include <libyul/optimiser/Semantics.h>
 #include <libyul/optimiser/OptimizerUtilities.h>
 #include <libyul/Exceptions.h>
 #include <libyul/AsmData.h>
 #include <libyul/Dialect.h>
+#include <libyul/SideEffects.h>
 
 #include <boost/algorithm/cxx11/none_of.hpp>
 
@@ -37,10 +39,12 @@ UnusedPruner::UnusedPruner(
 	Dialect const& _dialect,
 	Block& _ast,
 	bool _allowMSizeOptimization,
+	map<YulString, SideEffects> const* _functionSideEffects,
 	set<YulString> const& _externallyUsedFunctions
 ):
 	m_dialect(_dialect),
-	m_allowMSizeOptimization(_allowMSizeOptimization)
+	m_allowMSizeOptimization(_allowMSizeOptimization),
+	m_functionSideEffects(_functionSideEffects)
 {
 	m_references = ReferencesCounter::countReferences(_ast);
 	for (auto const& f: _externallyUsedFunctions)
@@ -88,7 +92,10 @@ void UnusedPruner::operator()(Block& _block)
 			{
 				if (!varDecl.value)
 					statement = Block{std::move(varDecl.location), {}};
-				else if (SideEffectsCollector(m_dialect, *varDecl.value).sideEffectFree(m_allowMSizeOptimization))
+				else if (
+					SideEffectsCollector(m_dialect, *varDecl.value, m_functionSideEffects).
+					sideEffectFree(m_allowMSizeOptimization)
+				)
 				{
 					subtractReferences(ReferencesCounter::countReferences(*varDecl.value));
 					statement = Block{std::move(varDecl.location), {}};
@@ -104,7 +111,10 @@ void UnusedPruner::operator()(Block& _block)
 		else if (statement.type() == typeid(ExpressionStatement))
 		{
 			ExpressionStatement& exprStmt = boost::get<ExpressionStatement>(statement);
-			if (SideEffectsCollector(m_dialect, exprStmt.expression).sideEffectFree(m_allowMSizeOptimization))
+			if (
+				SideEffectsCollector(m_dialect, exprStmt.expression, m_functionSideEffects).
+				sideEffectFree(m_allowMSizeOptimization)
+			)
 			{
 				subtractReferences(ReferencesCounter::countReferences(exprStmt.expression));
 				statement = Block{std::move(exprStmt.location), {}};
@@ -120,26 +130,31 @@ void UnusedPruner::runUntilStabilised(
 	Dialect const& _dialect,
 	Block& _ast,
 	bool _allowMSizeOptimization,
+	map<YulString, SideEffects> const* _functionSideEffects,
 	set<YulString> const& _externallyUsedFunctions
 )
 {
 	while (true)
 	{
-		UnusedPruner pruner(_dialect, _ast, _allowMSizeOptimization, _externallyUsedFunctions);
+		UnusedPruner pruner(
+			_dialect, _ast, _allowMSizeOptimization, _functionSideEffects,
+							_externallyUsedFunctions);
 		pruner(_ast);
 		if (!pruner.shouldRunAgain())
 			return;
 	}
 }
 
-void UnusedPruner::runUntilStabilised(
+void UnusedPruner::runUntilStabilisedOnFullAST(
 	Dialect const& _dialect,
 	Block& _ast,
 	set<YulString> const& _externallyUsedFunctions
 )
 {
+	map<YulString, SideEffects> functionSideEffects =
+		SideEffectsPropagator::sideEffects(_dialect, CallGraphGenerator::callGraph(_ast));
 	bool allowMSizeOptimization = !MSizeFinder::containsMSize(_dialect, _ast);
-	runUntilStabilised(_dialect, _ast, allowMSizeOptimization, _externallyUsedFunctions);
+	runUntilStabilised(_dialect, _ast, allowMSizeOptimization, &functionSideEffects, _externallyUsedFunctions);
 }
 
 void UnusedPruner::runUntilStabilised(

libyul/optimiser/UnusedPruner.h

@@ -29,6 +29,7 @@
 namespace yul
 {
 struct Dialect;
+struct SideEffects;
 
 /**
  * Optimisation stage that removes unused variables and functions and also
@@ -50,6 +51,7 @@ public:
 		Dialect const& _dialect,
 		Block& _ast,
 		bool _allowMSizeOptimization,
+		std::map<YulString, SideEffects> const* _functionSideEffects = nullptr,
 		std::set<YulString> const& _externallyUsedFunctions = {}
 	);
 	UnusedPruner(
@@ -70,10 +72,15 @@ public:
 		Dialect const& _dialect,
 		Block& _ast,
 		bool _allowMSizeOptimization,
+		std::map<YulString, SideEffects> const* _functionSideEffects = nullptr,
 		std::set<YulString> const& _externallyUsedFunctions = {}
 	);
 
-	static void runUntilStabilised(
+	/// Run the pruner until the code does not change anymore.
+	/// The provided block has to be a full AST.
+	/// The pruner itself determines if msize is used and which user-defined functions
+	/// are side-effect free.
+	static void runUntilStabilisedOnFullAST(
 		Dialect const& _dialect,
 		Block& _ast,
 		std::set<YulString> const& _externallyUsedFunctions = {}
@@ -97,6 +104,7 @@ private:
 
 	Dialect const& m_dialect;
 	bool m_allowMSizeOptimization = false;
+	std::map<YulString, SideEffects> const* m_functionSideEffects = nullptr;
 	bool m_shouldRunAgain = false;
 	std::map<YulString, size_t> m_references;
 };

test/cmdlineTests/standard_eWasm_requested/output.json

@@ -7,24 +7,16 @@
     (local $_1 i64)
     (local $pos i64)
     (local $_2 i64)
-    (local $_3 i64)
     (local $hi i64)
-    (local $_4 i64)
+    (local $y i64)
     (local $hi_1 i64)
-    (local $hi_2 i64)
-    (local $hi_3 i64)
-    (local $hi_4 i64)
     (set_local $_1 (i64.const 0))
     (set_local $pos (call $u256_to_i32 (get_local $_1) (get_local $_1) (get_local $_1) (i64.const 64)))
-    (set_local $_2 (i64.const 32))
+    (set_local $_2 (i64.const 65280))
-    (set_local $_3 (i64.shr_u (get_local $_1) (i64.const 16)))
+    (set_local $hi (i64.shl (i64.or (i64.shl (i64.or (i64.and (i64.shl (get_local $_1) (i64.const 8)) (get_local $_2)) (i64.and (i64.shr_u (get_local $_1) (i64.const 8)) (i64.const 255))) (i64.const 16)) (call $endian_swap_16 (i64.shr_u (get_local $_1) (i64.const 16)))) (i64.const 32)))
-    (set_local $hi (i64.shl (i64.or (i64.shl (i64.or (i64.and (i64.shl (get_local $_1) (i64.const 8)) (i64.const 65280)) (i64.and (i64.shr_u (get_local $_1) (i64.const 8)) (i64.const 255))) (i64.const 16)) (call $endian_swap_16 (get_local $_3))) (get_local $_2)))
+    (set_local $y (i64.or (get_local $hi) (call $endian_swap_32 (i64.shr_u (get_local $_1) (i64.const 32)))))
-    (set_local $_4 (i64.shr_u (get_local $_1) (get_local $_2)))
+    (i64.store (get_local $pos) (get_local $y)) (i64.store (i64.add (get_local $pos) (i64.const 8)) (get_local $y)) (i64.store (i64.add (get_local $pos) (i64.const 16)) (get_local $y)) (set_local $hi_1 (i64.shl (i64.or (i64.shl (i64.or (i64.and (i64.shl (i64.const 64) (i64.const 8)) (get_local $_2)) (i64.and (i64.shr_u (i64.const 64) (i64.const 8)) (i64.const 255))) (i64.const 16)) (call $endian_swap_16 (i64.shr_u (i64.const 64) (i64.const 16)))) (i64.const 32)))
-    (i64.store (get_local $pos) (i64.or (get_local $hi) (call $endian_swap_32 (get_local $_4)))) (set_local $hi_1 (i64.shl (call $endian_swap_16 (get_local $_1)) (i64.const 16)))
+    (i64.store (i64.add (get_local $pos) (i64.const 24)) (i64.or (get_local $hi_1) (call $endian_swap_32 (i64.shr_u (i64.const 64) (i64.const 32))))) (call $eth.revert (call $u256_to_i32 (get_local $_1) (get_local $_1) (get_local $_1) (get_local $_1)) (call $u256_to_i32 (get_local $_1) (get_local $_1) (get_local $_1) (get_local $_1)))
-    (set_local $hi_2 (i64.shl (i64.or (get_local $hi_1) (call $endian_swap_16 (get_local $_3))) (get_local $_2)))
-    (i64.store (i64.add (get_local $pos) (i64.const 8)) (i64.or (get_local $hi_2) (call $endian_swap_32 (get_local $_4)))) (set_local $hi_3 (i64.shl (call $endian_swap_32 (get_local $_1)) (get_local $_2)))
-    (i64.store (i64.add (get_local $pos) (i64.const 16)) (i64.or (get_local $hi_3) (call $endian_swap_32 (get_local $_4)))) (set_local $hi_4 (i64.shl (call $endian_swap_32 (i64.const 64)) (get_local $_2)))
-    (i64.store (i64.add (get_local $pos) (i64.const 24)) (i64.or (get_local $hi_4) (call $endian_swap_32 (i64.shr_u (i64.const 64) (get_local $_2))))) (call $eth.revert (call $u256_to_i32 (get_local $_1) (get_local $_1) (get_local $_1) (get_local $_1)) (call $u256_to_i32 (get_local $_1) (get_local $_1) (get_local $_1) (get_local $_1)))
 )
 
 (func $u256_to_i32
@@ -74,44 +66,38 @@
     (local $_1 i64)
     (local $pos i64)
     (local $hi i64)
-    (local $_2 i64)
+    (local $y i64)
     (local $hi_1 i64)
-    (local $_3 i64)
     (local $hi_2 i64)
-    (local $hi_3 i64)
+    (local $_2 i64)
-    (local $hi_4 i64)
+    (local $_3 i64)
     (local $_4 i64)
     (local $_5 i64)
     (local $_6 i64)
     (local $_7 i64)
     (local $_8 i64)
     (local $_9 i64)
-    (local $_10 i64)
-    (local $_11 i64)
     (set_local $_1 (i64.const 0))
     (set_local $pos (call $u256_to_i32 (get_local $_1) (get_local $_1) (get_local $_1) (i64.const 64)))
-    (set_local $hi (i64.shl (call $endian_swap_16 (get_local $_1)) (i64.const 16)))
+    (set_local $hi (i64.shl (i64.or (i64.shl (i64.or (i64.and (i64.shl (get_local $_1) (i64.const 8)) (i64.const 65280)) (i64.and (i64.shr_u (get_local $_1) (i64.const 8)) (i64.const 255))) (i64.const 16)) (call $endian_swap_16 (i64.shr_u (get_local $_1) (i64.const 16)))) (i64.const 32)))
-    (set_local $_2 (i64.shr_u (get_local $_1) (i64.const 16)))
+    (set_local $y (i64.or (get_local $hi) (call $endian_swap_32 (i64.shr_u (get_local $_1) (i64.const 32)))))
-    (set_local $hi_1 (i64.shl (i64.or (get_local $hi) (call $endian_swap_16 (get_local $_2))) (i64.const 32)))
+    (i64.store (get_local $pos) (get_local $y)) (i64.store (i64.add (get_local $pos) (i64.const 8)) (get_local $y)) (i64.store (i64.add (get_local $pos) (i64.const 16)) (get_local $y)) (set_local $hi_1 (i64.shl (call $endian_swap_16 (i64.const 64)) (i64.const 16)))
-    (set_local $_3 (i64.shr_u (get_local $_1) (i64.const 32)))
+    (set_local $hi_2 (i64.shl (i64.or (get_local $hi_1) (call $endian_swap_16 (i64.shr_u (i64.const 64) (i64.const 16)))) (i64.const 32)))
-    (i64.store (get_local $pos) (i64.or (get_local $hi_1) (call $endian_swap_32 (get_local $_3)))) (set_local $hi_2 (i64.shl (call $endian_swap_16 (get_local $_1)) (i64.const 16)))
+    (i64.store (i64.add (get_local $pos) (i64.const 24)) (i64.or (get_local $hi_2) (call $endian_swap_32 (i64.shr_u (i64.const 64) (i64.const 32))))) (block
-    (set_local $hi_3 (i64.shl (i64.or (get_local $hi_2) (call $endian_swap_16 (get_local $_2))) (i64.const 32)))
+        (set_local $_2 (datasize \"C_2_deployed\"))
-    (i64.store (i64.add (get_local $pos) (i64.const 8)) (i64.or (get_local $hi_3) (call $endian_swap_32 (get_local $_3)))) (set_local $hi_4 (i64.shl (call $endian_swap_32 (get_local $_1)) (i64.const 32)))
+        (set_local $_3 (get_global $global_))
-    (i64.store (i64.add (get_local $pos) (i64.const 16)) (i64.or (get_local $hi_4) (call $endian_swap_32 (get_local $_3)))) (i64.store (i64.add (get_local $pos) (i64.const 24)) (call $endian_swap (i64.const 64))) (block
+        (set_local $_4 (get_global $global__1))
-        (set_local $_4 (datasize \"C_2_deployed\"))
+        (set_local $_5 (get_global $global__2))
-        (set_local $_5 (get_global $global_))
-        (set_local $_6 (get_global $global__1))
-        (set_local $_7 (get_global $global__2))
 
     )
     (block
-        (set_local $_8 (dataoffset \"C_2_deployed\"))
+        (set_local $_6 (dataoffset \"C_2_deployed\"))
-        (set_local $_9 (get_global $global_))
+        (set_local $_7 (get_global $global_))
-        (set_local $_10 (get_global $global__1))
+        (set_local $_8 (get_global $global__1))
-        (set_local $_11 (get_global $global__2))
+        (set_local $_9 (get_global $global__2))
 
     )
-    (call $eth.codeCopy (call $u256_to_i32 (get_local $_1) (get_local $_1) (get_local $_1) (get_local $_1)) (call $u256_to_i32 (get_local $_8) (get_local $_9) (get_local $_10) (get_local $_11)) (call $u256_to_i32 (get_local $_4) (get_local $_5) (get_local $_6) (get_local $_7))) (call $eth.finish (call $u256_to_i32 (get_local $_1) (get_local $_1) (get_local $_1) (get_local $_1)) (call $u256_to_i32 (get_local $_4) (get_local $_5) (get_local $_6) (get_local $_7)))
+    (call $eth.codeCopy (call $u256_to_i32 (get_local $_1) (get_local $_1) (get_local $_1) (get_local $_1)) (call $u256_to_i32 (get_local $_6) (get_local $_7) (get_local $_8) (get_local $_9)) (call $u256_to_i32 (get_local $_2) (get_local $_3) (get_local $_4) (get_local $_5))) (call $eth.finish (call $u256_to_i32 (get_local $_1) (get_local $_1) (get_local $_1) (get_local $_1)) (call $u256_to_i32 (get_local $_2) (get_local $_3) (get_local $_4) (get_local $_5)))
 )
 
 (func $u256_to_i32
@@ -147,16 +133,6 @@
     (get_local $y)
 )
 
-(func $endian_swap
-    (param $x i64)
-    (result i64)
-    (local $y i64)
-    (local $hi i64)
-    (set_local $hi (i64.shl (call $endian_swap_32 (get_local $x)) (i64.const 32)))
-    (set_local $y (i64.or (get_local $hi) (call $endian_swap_32 (i64.shr_u (get_local $x) (i64.const 32)))))
-    (get_local $y)
-)
-
 )
 "}}}},"errors":[{"component":"general","formattedMessage":"Warning: The Yul optimiser is still experimental. Do not use it in production unless correctness of generated code is verified with extensive tests.
 ","message":"The Yul optimiser is still experimental. Do not use it in production unless correctness of generated code is verified with extensive tests.","severity":"warning","type":"Warning"}],"sources":{"A":{"id":0}}}

test/libyul/YulOptimizerTest.cpp

@@ -26,6 +26,7 @@
 #include <libyul/optimiser/ControlFlowSimplifier.h>
 #include <libyul/optimiser/DeadCodeEliminator.h>
 #include <libyul/optimiser/Disambiguator.h>
+#include <libyul/optimiser/CallGraphGenerator.h>
 #include <libyul/optimiser/CommonSubexpressionEliminator.h>
 #include <libyul/optimiser/NameCollector.h>
 #include <libyul/optimiser/EquivalentFunctionCombiner.h>
@@ -45,6 +46,7 @@
 #include <libyul/optimiser/ExpressionJoiner.h>
 #include <libyul/optimiser/SSAReverser.h>
 #include <libyul/optimiser/SSATransform.h>
+#include <libyul/optimiser/Semantics.h>
 #include <libyul/optimiser/RedundantAssignEliminator.h>
 #include <libyul/optimiser/StructuralSimplifier.h>
 #include <libyul/optimiser/StackCompressor.h>
@@ -149,7 +151,7 @@ TestCase::TestResult YulOptimizerTest::run(ostream& _stream, string const& _line
 	else if (m_optimizerStep == "commonSubexpressionEliminator")
 	{
 		disambiguate();
-		(CommonSubexpressionEliminator{*m_dialect})(*m_ast);
+		CommonSubexpressionEliminator::run(*m_dialect, *m_ast);
 	}
 	else if (m_optimizerStep == "expressionSplitter")
 	{
@@ -218,9 +220,9 @@ TestCase::TestResult YulOptimizerTest::run(ostream& _stream, string const& _line
 		NameDispenser nameDispenser{*m_dialect, *m_ast};
 		ExpressionSplitter{*m_dialect, nameDispenser}(*m_ast);
 		ForLoopInitRewriter{}(*m_ast);
-		CommonSubexpressionEliminator{*m_dialect}(*m_ast);
+		CommonSubexpressionEliminator::run(*m_dialect, *m_ast);
 		ExpressionSimplifier::run(*m_dialect, *m_ast);
-		UnusedPruner::runUntilStabilised(*m_dialect, *m_ast);
+		UnusedPruner::runUntilStabilisedOnFullAST(*m_dialect, *m_ast);
 		DeadCodeEliminator{*m_dialect}(*m_ast);
 		ExpressionJoiner::run(*m_ast);
 		ExpressionJoiner::run(*m_ast);
@@ -228,7 +230,7 @@ TestCase::TestResult YulOptimizerTest::run(ostream& _stream, string const& _line
 	else if (m_optimizerStep == "unusedPruner")
 	{
 		disambiguate();
-		UnusedPruner::runUntilStabilised(*m_dialect, *m_ast);
+		UnusedPruner::runUntilStabilisedOnFullAST(*m_dialect, *m_ast);
 	}
 	else if (m_optimizerStep == "deadCodeEliminator")
 	{
@@ -260,12 +262,12 @@ TestCase::TestResult YulOptimizerTest::run(ostream& _stream, string const& _line
 		ForLoopInitRewriter{}(*m_ast);
 		NameDispenser nameDispenser{*m_dialect, *m_ast};
 		ExpressionSplitter{*m_dialect, nameDispenser}(*m_ast);
-		CommonSubexpressionEliminator{*m_dialect}(*m_ast);
+		CommonSubexpressionEliminator::run(*m_dialect, *m_ast);
 		ExpressionSimplifier::run(*m_dialect, *m_ast);
 
 		LoadResolver::run(*m_dialect, *m_ast);
 
-		UnusedPruner::runUntilStabilised(*m_dialect, *m_ast);
+		UnusedPruner::runUntilStabilisedOnFullAST(*m_dialect, *m_ast);
 		ExpressionJoiner::run(*m_ast);
 		ExpressionJoiner::run(*m_ast);
 	}
@@ -298,8 +300,8 @@ TestCase::TestResult YulOptimizerTest::run(ostream& _stream, string const& _line
 		RedundantAssignEliminator::run(*m_dialect, *m_ast);
 		// reverse SSA
 		SSAReverser::run(*m_ast);
-		CommonSubexpressionEliminator{*m_dialect}(*m_ast);
+		CommonSubexpressionEliminator::run(*m_dialect, *m_ast);
-		UnusedPruner::runUntilStabilised(*m_dialect, *m_ast);
+		UnusedPruner::runUntilStabilisedOnFullAST(*m_dialect, *m_ast);
 	}
 	else if (m_optimizerStep == "stackCompressor")
 	{

test/libyul/yulOptimizerTests/commonSubexpressionEliminator/movable_functions.yul

@@ -0,0 +1,26 @@
+{
+    function double(x) -> y { y := add(x, x) }
+    function double_with_se(x) -> y { y := add(x, x) mstore(40, 4) }
+    let i := mload(3)
+    let a := double(i)
+    let b := double(i)
+    let c := double_with_se(i)
+    let d := double_with_se(i)
+}
+// ====
+// step: commonSubexpressionEliminator
+// ----
+// {
+//     function double(x) -> y
+//     { y := add(x, x) }
+//     function double_with_se(x_1) -> y_2
+//     {
+//         y_2 := add(x_1, x_1)
+//         mstore(40, 4)
+//     }
+//     let i := mload(3)
+//     let a := double(i)
+//     let b := a
+//     let c := double_with_se(i)
+//     let d := double_with_se(i)
+// }

test/libyul/yulOptimizerTests/fullSimplify/not_applied_function_call_equality_not_movable.yul

@@ -1,6 +1,6 @@
 // Even if the functions pass the equality check, they are not movable.
 {
-	function f() -> a { }
+	function f() -> a { mstore(1, 2) }
 	let b := sub(f(), f())
 	mstore(0, b)
 }
@@ -9,6 +9,6 @@
 // ----
 // {
 //     function f() -> a
-//     { }
+//     { mstore(1, 2) }
 //     mstore(0, sub(f(), f()))
 // }

test/libyul/yulOptimizerTests/loadResolver/memory_with_different_kinds_of_invalidation.yul

@@ -31,8 +31,5 @@
 //     mstore8(calldataload(_5), 4)
 //     sstore(_5, mload(_2))
 //     mstore(_2, _17)
-//     g()
 //     sstore(_5, mload(_2))
-//     function g()
-//     { }
 // }

test/libyul/yulOptimizerTests/unusedPruner/movable_user_defined_function.yul

@@ -0,0 +1,13 @@
+{
+    function f(x) -> t {
+        let b := 7
+    }
+    function g(x) -> t {
+        t := x
+    }
+    let x := f(g(2))
+}
+// ====
+// step: unusedPruner
+// ----
+// { }

test/libyul/yulOptimizerTests/unusedPruner/multi_declare.yul

@@ -5,8 +5,4 @@
 // ====
 // step: unusedPruner
 // ----
-// {
+// { }
-//     function f() -> x, y
-//     { }
-//     let a, b := f()
-// }

test/tools/yulopti.cpp

@@ -32,6 +32,7 @@
 
 #include <libyul/optimiser/BlockFlattener.h>
 #include <libyul/optimiser/Disambiguator.h>
+#include <libyul/optimiser/CallGraphGenerator.h>
 #include <libyul/optimiser/CommonSubexpressionEliminator.h>
 #include <libyul/optimiser/ControlFlowSimplifier.h>
 #include <libyul/optimiser/NameCollector.h>
@@ -54,6 +55,7 @@
 #include <libyul/optimiser/SSATransform.h>
 #include <libyul/optimiser/StackCompressor.h>
 #include <libyul/optimiser/StructuralSimplifier.h>
+#include <libyul/optimiser/Semantics.h>
 #include <libyul/optimiser/VarDeclInitializer.h>
 #include <libyul/optimiser/VarNameCleaner.h>
 
@@ -151,7 +153,7 @@ public:
 				ForLoopConditionIntoBody{}(*m_ast);
 				break;
 			case 'c':
-				(CommonSubexpressionEliminator{m_dialect})(*m_ast);
+				CommonSubexpressionEliminator::run(m_dialect, *m_ast);
 				break;
 			case 'd':
 				(VarDeclInitializer{})(*m_ast);
@@ -187,7 +189,7 @@ public:
 				(ControlFlowSimplifier{m_dialect})(*m_ast);
 				break;
 			case 'u':
-				UnusedPruner::runUntilStabilised(m_dialect, *m_ast);
+				UnusedPruner::runUntilStabilisedOnFullAST(m_dialect, *m_ast);
 				break;
 			case 'D':
 				DeadCodeEliminator{m_dialect}(*m_ast);