Merge pull request #12775 from ethereum/refactorDataFlowAnalyzer

Refactor data flow analyzer state access.

Author: chriseth

libyul/optimiser/CommonSubexpressionEliminator.cpp

@@ -95,18 +95,18 @@ void CommonSubexpressionEliminator::visit(Expression& _e)
 	if (Identifier const* identifier = get_if<Identifier>(&_e))
 	{
 		YulString identifierName = identifier->name;
-		if (m_value.count(identifierName))
+		if (AssignedValue const* assignedValue = variableValue(identifierName))
 		{
-			assertThrow(m_value.at(identifierName).value, OptimizerException, "");
-			if (Identifier const* value = get_if<Identifier>(m_value.at(identifierName).value))
+			assertThrow(assignedValue->value, OptimizerException, "");
+			if (Identifier const* value = get_if<Identifier>(assignedValue->value))
 				if (inScope(value->name))
 					_e = Identifier{debugDataOf(_e), value->name};
 		}
 	}
 	else
 	{
 		// TODO this search is rather inefficient.
-		for (auto const& [variable, value]: m_value)
+		for (auto const& [variable, value]: allValues())
 		{
 			assertThrow(value.value, OptimizerException, "");
 			// Prevent using the default value of return variables

libyul/optimiser/DataFlowAnalyzer.cpp

@@ -24,6 +24,7 @@
 
 #include <libyul/optimiser/NameCollector.h>
 #include <libyul/optimiser/Semantics.h>
+#include <libyul/optimiser/KnowledgeBase.h>
 #include <libyul/AST.h>
 #include <libyul/Dialect.h>
 #include <libyul/Exceptions.h>
@@ -47,7 +48,7 @@ DataFlowAnalyzer::DataFlowAnalyzer(
 ):
 	m_dialect(_dialect),
 	m_functionSideEffects(std::move(_functionSideEffects)),
-	m_knowledgeBase(_dialect, m_value)
+	m_knowledgeBase(_dialect, [this](YulString _var) { return variableValue(_var); })
 {
 	if (auto const* builtin = _dialect.memoryStoreFunction(YulString{}))
 		m_storeFunctionName[static_cast<unsigned>(StoreLoadLocation::Memory)] = builtin->name;
@@ -64,20 +65,20 @@ void DataFlowAnalyzer::operator()(ExpressionStatement& _statement)
 	if (auto vars = isSimpleStore(StoreLoadLocation::Storage, _statement))
 	{
 		ASTModifier::operator()(_statement);
-		cxx20::erase_if(m_storage, mapTuple([&](auto&& key, auto&& value) {
+		cxx20::erase_if(m_state.storage, mapTuple([&](auto&& key, auto&& value) {
 			return
 				!m_knowledgeBase.knownToBeDifferent(vars->first, key) &&
 				!m_knowledgeBase.knownToBeEqual(vars->second, value);
 		}));
-		m_storage[vars->first] = vars->second;
+		m_state.storage[vars->first] = vars->second;
 	}
 	else if (auto vars = isSimpleStore(StoreLoadLocation::Memory, _statement))
 	{
 		ASTModifier::operator()(_statement);
-		cxx20::erase_if(m_memory, mapTuple([&](auto&& key, auto&& /* value */) {
+		cxx20::erase_if(m_state.memory, mapTuple([&](auto&& key, auto&& /* value */) {
 			return !m_knowledgeBase.knownToBeDifferentByAtLeast32(vars->first, key);
 		}));
-		m_memory[vars->first] = vars->second;
+		m_state.memory[vars->first] = vars->second;
 	}
 	else
 	{
@@ -116,8 +117,8 @@ void DataFlowAnalyzer::operator()(VariableDeclaration& _varDecl)
 void DataFlowAnalyzer::operator()(If& _if)
 {
 	clearKnowledgeIfInvalidated(*_if.condition);
-	unordered_map<YulString, YulString> storage = m_storage;
-	unordered_map<YulString, YulString> memory = m_memory;
+	unordered_map<YulString, YulString> storage = m_state.storage;
+	unordered_map<YulString, YulString> memory = m_state.memory;
 
 	ASTModifier::operator()(_if);
 
@@ -133,8 +134,8 @@ void DataFlowAnalyzer::operator()(Switch& _switch)
 	set<YulString> assignedVariables;
 	for (auto& _case: _switch.cases)
 	{
-		unordered_map<YulString, YulString> storage = m_storage;
-		unordered_map<YulString, YulString> memory = m_memory;
+		unordered_map<YulString, YulString> storage = m_state.storage;
+		unordered_map<YulString, YulString> memory = m_state.memory;
 		(*this)(_case.body);
 		joinKnowledge(storage, memory);
 
@@ -153,11 +154,8 @@ void DataFlowAnalyzer::operator()(FunctionDefinition& _fun)
 {
 	// Save all information. We might rather reinstantiate this class,
 	// but this could be difficult if it is subclassed.
-	ScopedSaveAndRestore valueResetter(m_value, {});
+	ScopedSaveAndRestore stateResetter(m_state, {});
 	ScopedSaveAndRestore loopDepthResetter(m_loopDepth, 0u);
-	ScopedSaveAndRestore referencesResetter(m_references, {});
-	ScopedSaveAndRestore storageResetter(m_storage, {});
-	ScopedSaveAndRestore memoryResetter(m_memory, {});
 	pushScope(true);
 
 	for (auto const& parameter: _fun.parameters)
@@ -218,6 +216,22 @@ void DataFlowAnalyzer::operator()(Block& _block)
 	assertThrow(numScopes == m_variableScopes.size(), OptimizerException, "");
 }
 
+optional<YulString> DataFlowAnalyzer::storageValue(YulString _key) const
+{
+	if (YulString const* value = util::valueOrNullptr(m_state.storage, _key))
+		return *value;
+	else
+		return nullopt;
+}
+
+optional<YulString> DataFlowAnalyzer::memoryValue(YulString _key) const
+{
+	if (YulString const* value = util::valueOrNullptr(m_state.memory, _key))
+		return *value;
+	else
+		return nullopt;
+}
+
 void DataFlowAnalyzer::handleAssignment(set<YulString> const& _variables, Expression* _value, bool _isDeclaration)
 {
 	if (!_isDeclaration)
@@ -242,17 +256,17 @@ void DataFlowAnalyzer::handleAssignment(set<YulString> const& _variables, Expres
 	auto const& referencedVariables = movableChecker.referencedVariables();
 	for (auto const& name: _variables)
 	{
-		m_references[name] = referencedVariables;
+		m_state.references[name] = referencedVariables;
 		if (!_isDeclaration)
 		{
 			// assignment to slot denoted by "name"
-			m_storage.erase(name);
+			m_state.storage.erase(name);
 			// assignment to slot contents denoted by "name"
-			cxx20::erase_if(m_storage, mapTuple([&name](auto&& /* key */, auto&& value) { return value == name; }));
+			cxx20::erase_if(m_state.storage, mapTuple([&name](auto&& /* key */, auto&& value) { return value == name; }));
 			// assignment to slot denoted by "name"
-			m_memory.erase(name);
+			m_state.memory.erase(name);
 			// assignment to slot contents denoted by "name"
-			cxx20::erase_if(m_memory, mapTuple([&name](auto&& /* key */, auto&& value) { return value == name; }));
+			cxx20::erase_if(m_state.memory, mapTuple([&name](auto&& /* key */, auto&& value) { return value == name; }));
 		}
 	}
 
@@ -265,9 +279,9 @@ void DataFlowAnalyzer::handleAssignment(set<YulString> const& _variables, Expres
 			// On the other hand, if we knew the value in the slot
 			// already, then the sload() / mload() would have been replaced by a variable anyway.
 			if (auto key = isSimpleLoad(StoreLoadLocation::Memory, *_value))
-				m_memory[*key] = variable;
+				m_state.memory[*key] = variable;
 			else if (auto key = isSimpleLoad(StoreLoadLocation::Storage, *_value))
-				m_storage[*key] = variable;
+				m_state.storage[*key] = variable;
 		}
 	}
 }
@@ -281,8 +295,8 @@ void DataFlowAnalyzer::popScope()
 {
 	for (auto const& name: m_variableScopes.back().variables)
 	{
-		m_value.erase(name);
-		m_references.erase(name);
+		m_state.value.erase(name);
+		m_state.references.erase(name);
 	}
 	m_variableScopes.pop_back();
 }
@@ -308,53 +322,53 @@ void DataFlowAnalyzer::clearValues(set<YulString> _variables)
 	auto eraseCondition = mapTuple([&_variables](auto&& key, auto&& value) {
 		return _variables.count(key) || _variables.count(value);
 	});
-	cxx20::erase_if(m_storage, eraseCondition);
-	cxx20::erase_if(m_memory, eraseCondition);
+	cxx20::erase_if(m_state.storage, eraseCondition);
+	cxx20::erase_if(m_state.memory, eraseCondition);
 
 	// Also clear variables that reference variables to be cleared.
 	for (auto const& variableToClear: _variables)
-		for (auto const& [ref, names]: m_references)
+		for (auto const& [ref, names]: m_state.references)
 			if (names.count(variableToClear))
 				_variables.emplace(ref);
 
 	// Clear the value and update the reference relation.
 	for (auto const& name: _variables)
 	{
-		m_value.erase(name);
-		m_references.erase(name);
+		m_state.value.erase(name);
+		m_state.references.erase(name);
 	}
 }
 
 void DataFlowAnalyzer::assignValue(YulString _variable, Expression const* _value)
 {
-	m_value[_variable] = {_value, m_loopDepth};
+	m_state.value[_variable] = {_value, m_loopDepth};
 }
 
 void DataFlowAnalyzer::clearKnowledgeIfInvalidated(Block const& _block)
 {
 	SideEffectsCollector sideEffects(m_dialect, _block, &m_functionSideEffects);
 	if (sideEffects.invalidatesStorage())
-		m_storage.clear();
+		m_state.storage.clear();
 	if (sideEffects.invalidatesMemory())
-		m_memory.clear();
+		m_state.memory.clear();
 }
 
 void DataFlowAnalyzer::clearKnowledgeIfInvalidated(Expression const& _expr)
 {
 	SideEffectsCollector sideEffects(m_dialect, _expr, &m_functionSideEffects);
 	if (sideEffects.invalidatesStorage())
-		m_storage.clear();
+		m_state.storage.clear();
 	if (sideEffects.invalidatesMemory())
-		m_memory.clear();
+		m_state.memory.clear();
 }
 
 void DataFlowAnalyzer::joinKnowledge(
 	unordered_map<YulString, YulString> const& _olderStorage,
 	unordered_map<YulString, YulString> const& _olderMemory
 )
 {
-	joinKnowledgeHelper(m_storage, _olderStorage);
-	joinKnowledgeHelper(m_memory, _olderMemory);
+	joinKnowledgeHelper(m_state.storage, _olderStorage);
+	joinKnowledgeHelper(m_state.memory, _olderMemory);
 }
 
 void DataFlowAnalyzer::joinKnowledgeHelper(
@@ -364,8 +378,8 @@ void DataFlowAnalyzer::joinKnowledgeHelper(
 {
 	// We clear if the key does not exist in the older map or if the value is different.
 	// This also works for memory because _older is an "older version"
-	// of m_memory and thus any overlapping write would have cleared the keys
-	// that are not known to be different inside m_memory already.
+	// of m_state.memory and thus any overlapping write would have cleared the keys
+	// that are not known to be different inside m_state.memory already.
 	cxx20::erase_if(_this, mapTuple([&_older](auto&& key, auto&& currentValue){
 		YulString const* oldValue = util::valueOrNullptr(_older, key);
 		return !oldValue || *oldValue != currentValue;
@@ -386,8 +400,8 @@ bool DataFlowAnalyzer::inScope(YulString _variableName) const
 
 optional<u256> DataFlowAnalyzer::valueOfIdentifier(YulString const& _name)
 {
-	if (m_value.count(_name))
-		if (Literal const* literal = get_if<Literal>(m_value.at(_name).value))
+	if (AssignedValue const* value = variableValue(_name))
+		if (Literal const* literal = get_if<Literal>(value->value))
 			return valueOfLiteral(*literal);
 	return nullopt;
 }

libyul/optimiser/DataFlowAnalyzer.h

@@ -29,6 +29,7 @@
 #include <libyul/AST.h> // Needed for m_zero below.
 #include <libyul/SideEffects.h>
 
+#include <libsolutil/Numeric.h>
 #include <libsolutil/Common.h>
 
 #include <map>
@@ -38,6 +39,7 @@ namespace solidity::yul
 {
 struct Dialect;
 struct SideEffects;
+class KnowledgeBase;
 
 /// Value assigned to a variable.
 struct AssignedValue
@@ -98,6 +100,13 @@ class DataFlowAnalyzer: public ASTModifier
 	void operator()(ForLoop&) override;
 	void operator()(Block& _block) override;
 
+	/// @returns the current value of the given variable, if known - always movable.
+	AssignedValue const* variableValue(YulString _variable) const { return util::valueOrNullptr(m_state.value, _variable); }
+	std::set<YulString> const* references(YulString _variable) const { return util::valueOrNullptr(m_state.references, _variable); }
+	std::map<YulString, AssignedValue> const& allValues() const { return m_state.value; }
+	std::optional<YulString> storageValue(YulString _key) const;
+	std::optional<YulString> memoryValue(YulString _key) const;
+
 protected:
 	/// Registers the assignment.
 	void handleAssignment(std::set<YulString> const& _names, Expression* _value, bool _isDeclaration);
@@ -164,14 +173,20 @@ class DataFlowAnalyzer: public ASTModifier
 	/// 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, AssignedValue> m_value;
-	/// m_references[a].contains(b) <=> the current expression assigned to a references b
-	std::unordered_map<YulString, std::set<YulString>> m_references;
+private:
+	struct State
+	{
+		/// Current values of variables, always movable.
+		std::map<YulString, AssignedValue> value;
+		/// m_references[a].contains(b) <=> the current expression assigned to a references b
+		std::unordered_map<YulString, std::set<YulString>> references;
 
-	std::unordered_map<YulString, YulString> m_storage;
-	std::unordered_map<YulString, YulString> m_memory;
+		std::unordered_map<YulString, YulString> storage;
+		std::unordered_map<YulString, YulString> memory;
+	};
+	State m_state;
 
+protected:
 	KnowledgeBase m_knowledgeBase;
 
 	YulString m_storeFunctionName[static_cast<unsigned>(StoreLoadLocation::Last) + 1];

libyul/optimiser/EqualStoreEliminator.cpp

@@ -54,13 +54,13 @@ void EqualStoreEliminator::visit(Statement& _statement)
 	{
 		if (auto vars = isSimpleStore(StoreLoadLocation::Storage, *expression))
 		{
-			if (auto const* currentValue = util::valueOrNullptr(m_storage, vars->first))
+			if (optional<YulString> currentValue = storageValue(vars->first))
 				if (*currentValue == vars->second)
 					m_pendingRemovals.insert(&_statement);
 		}
 		else if (auto vars = isSimpleStore(StoreLoadLocation::Memory, *expression))
 		{
-			if (auto const* currentValue = util::valueOrNullptr(m_memory, vars->first))
+			if (optional<YulString> currentValue = memoryValue(vars->first))
 				if (*currentValue == vars->second)
 					m_pendingRemovals.insert(&_statement);
 		}

libyul/optimiser/ExpressionSimplifier.cpp

@@ -38,6 +38,10 @@ void ExpressionSimplifier::visit(Expression& _expression)
 {
 	ASTModifier::visit(_expression);
 
-	while (auto const* match = SimplificationRules::findFirstMatch(_expression, m_dialect, m_value))
+	while (auto const* match = SimplificationRules::findFirstMatch(
+		_expression,
+		m_dialect,
+		[this](YulString _var) { return variableValue(_var); }
+	))
 		_expression = match->action().toExpression(debugDataOf(_expression));
 }

libyul/optimiser/KnowledgeBase.cpp

@@ -80,8 +80,8 @@ bool KnowledgeBase::knownToBeZero(YulString _a)
 
 optional<u256> KnowledgeBase::valueIfKnownConstant(YulString _a)
 {
-	if (m_variableValues.count(_a))
-		if (Literal const* literal = get_if<Literal>(m_variableValues.at(_a).value))
+	if (AssignedValue const* value = m_variableValues(_a))
+		if (Literal const* literal = get_if<Literal>(value->value))
 			return valueOfLiteral(*literal);
 	return {};
 }

libyul/optimiser/KnowledgeBase.h

@@ -28,6 +28,7 @@
 #include <libsolutil/Numeric.h>
 
 #include <map>
+#include <functional>
 
 namespace solidity::yul
 {
@@ -37,15 +38,16 @@ struct AssignedValue;
 
 /**
  * Class that can answer questions about values of variables and their relations.
- *
- * The reference to the map of values provided at construction is assumed to be updating.
  */
 class KnowledgeBase
 {
 public:
-	KnowledgeBase(Dialect const& _dialect, std::map<YulString, AssignedValue> const& _variableValues):
+	KnowledgeBase(
+		Dialect const& _dialect,
+		std::function<AssignedValue const*(YulString)> _variableValues
+	):
 		m_dialect(_dialect),
-		m_variableValues(_variableValues)
+		m_variableValues(std::move(_variableValues))
 	{}
 
 	bool knownToBeDifferent(YulString _a, YulString _b);
@@ -60,7 +62,7 @@ class KnowledgeBase
 	Expression simplifyRecursively(Expression _expression);
 
 	Dialect const& m_dialect;
-	std::map<YulString, AssignedValue> const& m_variableValues;
+	std::function<AssignedValue const*(YulString)> m_variableValues;
 	size_t m_counter = 0;
 };