Merge pull request #3456 from ethereum/simplifier

Use simplification rules also for IULIA

6 files changed, 318 insertions, 184 deletions

diff --git a/libevmasm/ExpressionClasses.cpp b/libevmasm/ExpressionClasses.cpp
index fc283b0b..69b33ec5 100644
--- a/libevmasm/ExpressionClasses.cpp
+++ b/libevmasm/ExpressionClasses.cpp
@@ -181,7 +181,7 @@ string ExpressionClasses::fullDAGToString(ExpressionClasses::Id _id) const
 	return str.str();
 }
 
-ExpressionClasses::Id ExpressionClasses::tryToSimplify(Expression const& _expr, bool _secondRun)
+ExpressionClasses::Id ExpressionClasses::tryToSimplify(Expression const& _expr)
 {
 	static Rules rules;
 
@@ -202,14 +202,7 @@ ExpressionClasses::Id ExpressionClasses::tryToSimplify(Expression const& _expr,
 		//cout << "with rule " << match->first.toString() << endl;
 		//ExpressionTemplate t(match->second());
 		//cout << "to " << match->second().toString() << endl;
-		return rebuildExpression(ExpressionTemplate(match->second(), _expr.item->location()));
-	}
-
-	if (!_secondRun && _expr.arguments.size() == 2 && SemanticInformation::isCommutativeOperation(*_expr.item))
-	{
-		Expression expr = _expr;
-		swap(expr.arguments[0], expr.arguments[1]);
-		return tryToSimplify(expr, true);
+		return rebuildExpression(ExpressionTemplate(match->action(), _expr.item->location()));
 	}
 
 	return -1;
diff --git a/libevmasm/ExpressionClasses.h b/libevmasm/ExpressionClasses.h
index 6b426e97..df8082f9 100644
--- a/libevmasm/ExpressionClasses.h
+++ b/libevmasm/ExpressionClasses.h
@@ -108,8 +108,7 @@ public:
 private:
 	/// Tries to simplify the given expression.
 	/// @returns its class if it possible or Id(-1) otherwise.
-	/// @param _secondRun is set to true for the second run where arguments of commutative expressions are reversed
-	Id tryToSimplify(Expression const& _expr, bool _secondRun = false);
+	Id tryToSimplify(Expression const& _expr);
 
 	/// Rebuilds an expression from a (matched) pattern.
 	Id rebuildExpression(ExpressionTemplate const& _template);
diff --git a/libevmasm/RuleList.h b/libevmasm/RuleList.h
new file mode 100644
index 00000000..2312d673
--- /dev/null
+++ b/libevmasm/RuleList.h
@@ -0,0 +1,255 @@
+/*
+	This file is part of solidity.
+
+	solidity is free software: you can redistribute it and/or modify
+	it under the terms of the GNU General Public License as published by
+	the Free Software Foundation, either version 3 of the License, or
+	(at your option) any later version.
+
+	solidity is distributed in the hope that it will be useful,
+	but WITHOUT ANY WARRANTY; without even the implied warranty of
+	MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+	GNU General Public License for more details.
+
+	You should have received a copy of the GNU General Public License
+	along with solidity.  If not, see <http://www.gnu.org/licenses/>.
+*/
+/**
+ * @date 2018
+ * Templatized list of simplification rules.
+ */
+
+#pragma once
+
+#include <vector>
+#include <functional>
+
+#include <libevmasm/Instruction.h>
+#include <libevmasm/SimplificationRule.h>
+
+#include <libdevcore/CommonData.h>
+
+namespace dev
+{
+namespace solidity
+{
+
+template <class S> S divWorkaround(S const& _a, S const& _b)
+{
+	return (S)(bigint(_a) / bigint(_b));
+}
+
+template <class S> S modWorkaround(S const& _a, S const& _b)
+{
+	return (S)(bigint(_a) % bigint(_b));
+}
+
+/// @returns a list of simplification rules given certain match placeholders.
+/// A, B and C should represent constants, X and Y arbitrary expressions.
+/// The simplifications should neven change the order of evaluation of
+/// arbitrary operations.
+template <class Pattern>
+std::vector<SimplificationRule<Pattern>> simplificationRuleList(
+	Pattern A,
+	Pattern B,
+	Pattern C,
+	Pattern X,
+	Pattern Y
+)
+{
+	std::vector<SimplificationRule<Pattern>> rules;
+	rules += std::vector<SimplificationRule<Pattern>>{
+		// arithmetics on constants
+		{{Instruction::ADD, {A, B}}, [=]{ return A.d() + B.d(); }, false},
+		{{Instruction::MUL, {A, B}}, [=]{ return A.d() * B.d(); }, false},
+		{{Instruction::SUB, {A, B}}, [=]{ return A.d() - B.d(); }, false},
+		{{Instruction::DIV, {A, B}}, [=]{ return B.d() == 0 ? 0 : divWorkaround(A.d(), B.d()); }, false},
+		{{Instruction::SDIV, {A, B}}, [=]{ return B.d() == 0 ? 0 : s2u(divWorkaround(u2s(A.d()), u2s(B.d()))); }, false},
+		{{Instruction::MOD, {A, B}}, [=]{ return B.d() == 0 ? 0 : modWorkaround(A.d(), B.d()); }, false},
+		{{Instruction::SMOD, {A, B}}, [=]{ return B.d() == 0 ? 0 : s2u(modWorkaround(u2s(A.d()), u2s(B.d()))); }, false},
+		{{Instruction::EXP, {A, B}}, [=]{ return u256(boost::multiprecision::powm(bigint(A.d()), bigint(B.d()), bigint(1) << 256)); }, false},
+		{{Instruction::NOT, {A}}, [=]{ return ~A.d(); }, false},
+		{{Instruction::LT, {A, B}}, [=]() -> u256 { return A.d() < B.d() ? 1 : 0; }, false},
+		{{Instruction::GT, {A, B}}, [=]() -> u256 { return A.d() > B.d() ? 1 : 0; }, false},
+		{{Instruction::SLT, {A, B}}, [=]() -> u256 { return u2s(A.d()) < u2s(B.d()) ? 1 : 0; }, false},
+		{{Instruction::SGT, {A, B}}, [=]() -> u256 { return u2s(A.d()) > u2s(B.d()) ? 1 : 0; }, false},
+		{{Instruction::EQ, {A, B}}, [=]() -> u256 { return A.d() == B.d() ? 1 : 0; }, false},
+		{{Instruction::ISZERO, {A}}, [=]() -> u256 { return A.d() == 0 ? 1 : 0; }, false},
+		{{Instruction::AND, {A, B}}, [=]{ return A.d() & B.d(); }, false},
+		{{Instruction::OR, {A, B}}, [=]{ return A.d() | B.d(); }, false},
+		{{Instruction::XOR, {A, B}}, [=]{ return A.d() ^ B.d(); }, false},
+		{{Instruction::BYTE, {A, B}}, [=]{ return A.d() >= 32 ? 0 : (B.d() >> unsigned(8 * (31 - A.d()))) & 0xff; }, false},
+		{{Instruction::ADDMOD, {A, B, C}}, [=]{ return C.d() == 0 ? 0 : u256((bigint(A.d()) + bigint(B.d())) % C.d()); }, false},
+		{{Instruction::MULMOD, {A, B, C}}, [=]{ return C.d() == 0 ? 0 : u256((bigint(A.d()) * bigint(B.d())) % C.d()); }, false},
+		{{Instruction::MULMOD, {A, B, C}}, [=]{ return A.d() * B.d(); }, false},
+		{{Instruction::SIGNEXTEND, {A, B}}, [=]() -> u256 {
+			if (A.d() >= 31)
+				return B.d();
+			unsigned testBit = unsigned(A.d()) * 8 + 7;
+			u256 mask = (u256(1) << testBit) - 1;
+			return u256(boost::multiprecision::bit_test(B.d(), testBit) ? B.d() | ~mask : B.d() & mask);
+		}, false},
+
+		// invariants involving known constants
+		{{Instruction::ADD, {X, 0}}, [=]{ return X; }, false},
+		{{Instruction::ADD, {0, X}}, [=]{ return X; }, false},
+		{{Instruction::SUB, {X, 0}}, [=]{ return X; }, false},
+		{{Instruction::MUL, {X, 0}}, [=]{ return u256(0); }, true},
+		{{Instruction::MUL, {0, X}}, [=]{ return u256(0); }, true},
+		{{Instruction::MUL, {X, 1}}, [=]{ return X; }, false},
+		{{Instruction::MUL, {1, X}}, [=]{ return X; }, false},
+		{{Instruction::MUL, {X, u256(-1)}}, [=]() -> Pattern { return {Instruction::SUB, {0, X}}; }, false},
+		{{Instruction::MUL, {u256(-1), X}}, [=]() -> Pattern { return {Instruction::SUB, {0, X}}; }, false},
+		{{Instruction::DIV, {X, 0}}, [=]{ return u256(0); }, true},
+		{{Instruction::DIV, {0, X}}, [=]{ return u256(0); }, true},
+		{{Instruction::DIV, {X, 1}}, [=]{ return X; }, false},
+		{{Instruction::SDIV, {X, 0}}, [=]{ return u256(0); }, true},
+		{{Instruction::SDIV, {0, X}}, [=]{ return u256(0); }, true},
+		{{Instruction::SDIV, {X, 1}}, [=]{ return X; }, false},
+		{{Instruction::AND, {X, ~u256(0)}}, [=]{ return X; }, false},
+		{{Instruction::AND, {~u256(0), X}}, [=]{ return X; }, false},
+		{{Instruction::AND, {X, 0}}, [=]{ return u256(0); }, true},
+		{{Instruction::AND, {0, X}}, [=]{ return u256(0); }, true},
+		{{Instruction::OR, {X, 0}}, [=]{ return X; }, false},
+		{{Instruction::OR, {0, X}}, [=]{ return X; }, false},
+		{{Instruction::OR, {X, ~u256(0)}}, [=]{ return ~u256(0); }, true},
+		{{Instruction::OR, {~u256(0), X}}, [=]{ return ~u256(0); }, true},
+		{{Instruction::XOR, {X, 0}}, [=]{ return X; }, false},
+		{{Instruction::XOR, {0, X}}, [=]{ return X; }, false},
+		{{Instruction::MOD, {X, 0}}, [=]{ return u256(0); }, true},
+		{{Instruction::MOD, {0, X}}, [=]{ return u256(0); }, true},
+		{{Instruction::EQ, {X, 0}}, [=]() -> Pattern { return {Instruction::ISZERO, {X}}; }, false },
+		{{Instruction::EQ, {0, X}}, [=]() -> Pattern { return {Instruction::ISZERO, {X}}; }, false },
+
+		// operations involving an expression and itself
+		{{Instruction::AND, {X, X}}, [=]{ return X; }, true},
+		{{Instruction::OR, {X, X}}, [=]{ return X; }, true},
+		{{Instruction::XOR, {X, X}}, [=]{ return u256(0); }, true},
+		{{Instruction::SUB, {X, X}}, [=]{ return u256(0); }, true},
+		{{Instruction::EQ, {X, X}}, [=]{ return u256(1); }, true},
+		{{Instruction::LT, {X, X}}, [=]{ return u256(0); }, true},
+		{{Instruction::SLT, {X, X}}, [=]{ return u256(0); }, true},
+		{{Instruction::GT, {X, X}}, [=]{ return u256(0); }, true},
+		{{Instruction::SGT, {X, X}}, [=]{ return u256(0); }, true},
+		{{Instruction::MOD, {X, X}}, [=]{ return u256(0); }, true},
+
+		// logical instruction combinations
+		{{Instruction::NOT, {{Instruction::NOT, {X}}}}, [=]{ return X; }, false},
+		{{Instruction::XOR, {X, {Instruction::XOR, {X, Y}}}}, [=]{ return Y; }, true},
+		{{Instruction::XOR, {X, {Instruction::XOR, {Y, X}}}}, [=]{ return Y; }, true},
+		{{Instruction::XOR, {{Instruction::XOR, {X, Y}}, X}}, [=]{ return Y; }, true},
+		{{Instruction::XOR, {{Instruction::XOR, {Y, X}}, X}}, [=]{ return Y; }, true},
+		{{Instruction::OR, {X, {Instruction::AND, {X, Y}}}}, [=]{ return X; }, true},
+		{{Instruction::OR, {X, {Instruction::AND, {Y, X}}}}, [=]{ return X; }, true},
+		{{Instruction::OR, {{Instruction::AND, {X, Y}}, X}}, [=]{ return X; }, true},
+		{{Instruction::OR, {{Instruction::AND, {Y, X}}, X}}, [=]{ return X; }, true},
+		{{Instruction::AND, {X, {Instruction::OR, {X, Y}}}}, [=]{ return X; }, true},
+		{{Instruction::AND, {X, {Instruction::OR, {Y, X}}}}, [=]{ return X; }, true},
+		{{Instruction::AND, {{Instruction::OR, {X, Y}}, X}}, [=]{ return X; }, true},
+		{{Instruction::AND, {{Instruction::OR, {Y, X}}, X}}, [=]{ return X; }, true},
+		{{Instruction::AND, {X, {Instruction::NOT, {X}}}}, [=]{ return u256(0); }, true},
+		{{Instruction::AND, {{Instruction::NOT, {X}}, X}}, [=]{ return u256(0); }, true},
+		{{Instruction::OR, {X, {Instruction::NOT, {X}}}}, [=]{ return ~u256(0); }, true},
+		{{Instruction::OR, {{Instruction::NOT, {X}}, X}}, [=]{ return ~u256(0); }, true},
+	};
+
+	// Double negation of opcodes with boolean result
+	for (auto const& op: std::vector<Instruction>{
+		Instruction::EQ,
+		Instruction::LT,
+		Instruction::SLT,
+		Instruction::GT,
+		Instruction::SGT
+	})
+		rules.push_back({
+			{Instruction::ISZERO, {{Instruction::ISZERO, {{op, {X, Y}}}}}},
+			[=]() -> Pattern { return {op, {X, Y}}; },
+			false
+		});
+
+	rules.push_back({
+		{Instruction::ISZERO, {{Instruction::ISZERO, {{Instruction::ISZERO, {X}}}}}},
+		[=]() -> Pattern { return {Instruction::ISZERO, {X}}; },
+		false
+	});
+
+	rules.push_back({
+		{Instruction::ISZERO, {{Instruction::XOR, {X, Y}}}},
+		[=]() -> Pattern { return { Instruction::EQ, {X, Y} }; },
+		false
+	});
+
+	// Associative operations
+	for (auto const& opFun: std::vector<std::pair<Instruction,std::function<u256(u256 const&,u256 const&)>>>{
+		{Instruction::ADD, std::plus<u256>()},
+		{Instruction::MUL, std::multiplies<u256>()},
+		{Instruction::AND, std::bit_and<u256>()},
+		{Instruction::OR, std::bit_or<u256>()},
+		{Instruction::XOR, std::bit_xor<u256>()}
+	})
+	{
+		auto op = opFun.first;
+		auto fun = opFun.second;
+		// Moving constants to the outside, order matters here - we first add rules
+		// for constants and then for non-constants.
+		// xa can be (X, A) or (A, X)
+		for (auto xa: {std::vector<Pattern>{X, A}, std::vector<Pattern>{A, X}})
+		{
+			rules += std::vector<SimplificationRule<Pattern>>{{
+				// (X+A)+B -> X+(A+B)
+				{op, {{op, xa}, B}},
+				[=]() -> Pattern { return {op, {X, fun(A.d(), B.d())}}; },
+				false
+			}, {
+				// (X+A)+Y -> (X+Y)+A
+				{op, {{op, xa}, Y}},
+				[=]() -> Pattern { return {op, {{op, {X, Y}}, A}}; },
+				false
+			}, {
+				// B+(X+A) -> X+(A+B)
+				{op, {B, {op, xa}}},
+				[=]() -> Pattern { return {op, {X, fun(A.d(), B.d())}}; },
+				false
+			}, {
+				// Y+(X+A) -> (Y+X)+A
+				{op, {Y, {op, xa}}},
+				[=]() -> Pattern { return {op, {{op, {Y, X}}, A}}; },
+				false
+			}};
+		}
+	}
+
+	// move constants across subtractions
+	rules += std::vector<SimplificationRule<Pattern>>{
+		{
+			// X - A -> X + (-A)
+			{Instruction::SUB, {X, A}},
+			[=]() -> Pattern { return {Instruction::ADD, {X, 0 - A.d()}}; },
+			false
+		}, {
+			// (X + A) - Y -> (X - Y) + A
+			{Instruction::SUB, {{Instruction::ADD, {X, A}}, Y}},
+			[=]() -> Pattern { return {Instruction::ADD, {{Instruction::SUB, {X, Y}}, A}}; },
+			false
+		}, {
+			// (A + X) - Y -> (X - Y) + A
+			{Instruction::SUB, {{Instruction::ADD, {A, X}}, Y}},
+			[=]() -> Pattern { return {Instruction::ADD, {{Instruction::SUB, {X, Y}}, A}}; },
+			false
+		}, {
+			// X - (Y + A) -> (X - Y) + (-A)
+			{Instruction::SUB, {X, {Instruction::ADD, {Y, A}}}},
+			[=]() -> Pattern { return {Instruction::ADD, {{Instruction::SUB, {X, Y}}, 0 - A.d()}}; },
+			false
+		}, {
+			// X - (A + Y) -> (X - Y) + (-A)
+			{Instruction::SUB, {X, {Instruction::ADD, {A, Y}}}},
+			[=]() -> Pattern { return {Instruction::ADD, {{Instruction::SUB, {X, Y}}, 0 - A.d()}}; },
+			false
+		}
+	};
+	return rules;
+}
+
+}
+}
diff --git a/libevmasm/SimplificationRule.h b/libevmasm/SimplificationRule.h
new file mode 100644
index 00000000..7b4dea68
--- /dev/null
+++ b/libevmasm/SimplificationRule.h
@@ -0,0 +1,45 @@
+/*
+	This file is part of solidity.
+
+	solidity is free software: you can redistribute it and/or modify
+	it under the terms of the GNU General Public License as published by
+	the Free Software Foundation, either version 3 of the License, or
+	(at your option) any later version.
+
+	solidity is distributed in the hope that it will be useful,
+	but WITHOUT ANY WARRANTY; without even the implied warranty of
+	MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+	GNU General Public License for more details.
+
+	You should have received a copy of the GNU General Public License
+	along with solidity.  If not, see <http://www.gnu.org/licenses/>.
+*/
+/**
+ * Expression simplification pattern.
+ */
+
+#pragma once
+
+#include <functional>
+
+namespace dev
+{
+namespace solidity
+{
+
+/**
+ * Rule that contains a pattern, an action that can be applied
+ * after the pattern has matched and a bool that indicates
+ * whether the action would remove something from the expression
+ * than is not a constant literal.
+ */
+template <class Pattern>
+struct SimplificationRule
+{
+	Pattern pattern;
+	std::function<Pattern()> action;
+	bool removesNonConstants;
+};
+
+}
+}
diff --git a/libevmasm/SimplificationRules.cpp b/libevmasm/SimplificationRules.cpp
index e6c51f95..53a5f9fc 100644
--- a/libevmasm/SimplificationRules.cpp
+++ b/libevmasm/SimplificationRules.cpp
@@ -23,7 +23,6 @@
 
 #include <libevmasm/ExpressionClasses.h>
 #include <utility>
-#include <tuple>
 #include <functional>
 #include <boost/range/adaptor/reversed.hpp>
 #include <boost/noncopyable.hpp>
@@ -31,12 +30,14 @@
 #include <libevmasm/CommonSubexpressionEliminator.h>
 #include <libevmasm/SimplificationRules.h>
 
+#include <libevmasm/RuleList.h>
+
 using namespace std;
 using namespace dev;
 using namespace dev::eth;
 
 
-pair<Pattern, function<Pattern()> > const* Rules::findFirstMatch(
+SimplificationRule<Pattern> const* Rules::findFirstMatch(
 	Expression const& _expr,
 	ExpressionClasses const& _classes
 )
@@ -46,32 +47,22 @@ pair<Pattern, function<Pattern()> > const* Rules::findFirstMatch(
 	assertThrow(_expr.item, OptimizerException, "");
 	for (auto const& rule: m_rules[byte(_expr.item->instruction())])
 	{
-		if (rule.first.matches(_expr, _classes))
+		if (rule.pattern.matches(_expr, _classes))
 			return &rule;
 		resetMatchGroups();
 	}
 	return nullptr;
 }
 
-void Rules::addRules(std::vector<std::pair<Pattern, std::function<Pattern ()> > > const& _rules)
+void Rules::addRules(std::vector<SimplificationRule<Pattern>> const& _rules)
 {
 	for (auto const& r: _rules)
 		addRule(r);
 }
 
-void Rules::addRule(std::pair<Pattern, std::function<Pattern()> > const& _rule)
+void Rules::addRule(SimplificationRule<Pattern> const& _rule)
 {
-	m_rules[byte(_rule.first.instruction())].push_back(_rule);
-}
-
-template <class S> S divWorkaround(S const& _a, S const& _b)
-{
-	return (S)(bigint(_a) / bigint(_b));
-}
-
-template <class S> S modWorkaround(S const& _a, S const& _b)
-{
-	return (S)(bigint(_a) % bigint(_b));
+	m_rules[byte(_rule.pattern.instruction())].push_back(_rule);
 }
 
 Rules::Rules()
@@ -84,165 +75,13 @@ Rules::Rules()
 	// Anything.
 	Pattern X;
 	Pattern Y;
-	Pattern Z;
 	A.setMatchGroup(1, m_matchGroups);
 	B.setMatchGroup(2, m_matchGroups);
 	C.setMatchGroup(3, m_matchGroups);
 	X.setMatchGroup(4, m_matchGroups);
 	Y.setMatchGroup(5, m_matchGroups);
-	Z.setMatchGroup(6, m_matchGroups);
-
-	addRules(vector<pair<Pattern, function<Pattern()>>>{
-		// arithmetics on constants
-		{{Instruction::ADD, {A, B}}, [=]{ return A.d() + B.d(); }},
-		{{Instruction::MUL, {A, B}}, [=]{ return A.d() * B.d(); }},
-		{{Instruction::SUB, {A, B}}, [=]{ return A.d() - B.d(); }},
-		{{Instruction::DIV, {A, B}}, [=]{ return B.d() == 0 ? 0 : divWorkaround(A.d(), B.d()); }},
-		{{Instruction::SDIV, {A, B}}, [=]{ return B.d() == 0 ? 0 : s2u(divWorkaround(u2s(A.d()), u2s(B.d()))); }},
-		{{Instruction::MOD, {A, B}}, [=]{ return B.d() == 0 ? 0 : modWorkaround(A.d(), B.d()); }},
-		{{Instruction::SMOD, {A, B}}, [=]{ return B.d() == 0 ? 0 : s2u(modWorkaround(u2s(A.d()), u2s(B.d()))); }},
-		{{Instruction::EXP, {A, B}}, [=]{ return u256(boost::multiprecision::powm(bigint(A.d()), bigint(B.d()), bigint(1) << 256)); }},
-		{{Instruction::NOT, {A}}, [=]{ return ~A.d(); }},
-		{{Instruction::LT, {A, B}}, [=]() -> u256 { return A.d() < B.d() ? 1 : 0; }},
-		{{Instruction::GT, {A, B}}, [=]() -> u256 { return A.d() > B.d() ? 1 : 0; }},
-		{{Instruction::SLT, {A, B}}, [=]() -> u256 { return u2s(A.d()) < u2s(B.d()) ? 1 : 0; }},
-		{{Instruction::SGT, {A, B}}, [=]() -> u256 { return u2s(A.d()) > u2s(B.d()) ? 1 : 0; }},
-		{{Instruction::EQ, {A, B}}, [=]() -> u256 { return A.d() == B.d() ? 1 : 0; }},
-		{{Instruction::ISZERO, {A}}, [=]() -> u256 { return A.d() == 0 ? 1 : 0; }},
-		{{Instruction::AND, {A, B}}, [=]{ return A.d() & B.d(); }},
-		{{Instruction::OR, {A, B}}, [=]{ return A.d() | B.d(); }},
-		{{Instruction::XOR, {A, B}}, [=]{ return A.d() ^ B.d(); }},
-		{{Instruction::BYTE, {A, B}}, [=]{ return A.d() >= 32 ? 0 : (B.d() >> unsigned(8 * (31 - A.d()))) & 0xff; }},
-		{{Instruction::ADDMOD, {A, B, C}}, [=]{ return C.d() == 0 ? 0 : u256((bigint(A.d()) + bigint(B.d())) % C.d()); }},
-		{{Instruction::MULMOD, {A, B, C}}, [=]{ return C.d() == 0 ? 0 : u256((bigint(A.d()) * bigint(B.d())) % C.d()); }},
-		{{Instruction::MULMOD, {A, B, C}}, [=]{ return A.d() * B.d(); }},
-		{{Instruction::SIGNEXTEND, {A, B}}, [=]() -> u256 {
-			if (A.d() >= 31)
-				return B.d();
-			unsigned testBit = unsigned(A.d()) * 8 + 7;
-			u256 mask = (u256(1) << testBit) - 1;
-			return u256(boost::multiprecision::bit_test(B.d(), testBit) ? B.d() | ~mask : B.d() & mask);
-		}},
-
-		// invariants involving known constants (commutative instructions will be checked with swapped operants too)
-		{{Instruction::ADD, {X, 0}}, [=]{ return X; }},
-		{{Instruction::SUB, {X, 0}}, [=]{ return X; }},
-		{{Instruction::MUL, {X, 0}}, [=]{ return u256(0); }},
-		{{Instruction::MUL, {X, 1}}, [=]{ return X; }},
-		{{Instruction::DIV, {X, 0}}, [=]{ return u256(0); }},
-		{{Instruction::DIV, {0, X}}, [=]{ return u256(0); }},
-		{{Instruction::DIV, {X, 1}}, [=]{ return X; }},
-		{{Instruction::SDIV, {X, 0}}, [=]{ return u256(0); }},
-		{{Instruction::SDIV, {0, X}}, [=]{ return u256(0); }},
-		{{Instruction::SDIV, {X, 1}}, [=]{ return X; }},
-		{{Instruction::AND, {X, ~u256(0)}}, [=]{ return X; }},
-		{{Instruction::AND, {X, 0}}, [=]{ return u256(0); }},
-		{{Instruction::OR, {X, 0}}, [=]{ return X; }},
-		{{Instruction::OR, {X, ~u256(0)}}, [=]{ return ~u256(0); }},
-		{{Instruction::XOR, {X, 0}}, [=]{ return X; }},
-		{{Instruction::MOD, {X, 0}}, [=]{ return u256(0); }},
-		{{Instruction::MOD, {0, X}}, [=]{ return u256(0); }},
-		{{Instruction::EQ, {X, 0}}, [=]() -> Pattern { return {Instruction::ISZERO, {X}}; } },
-
-		// operations involving an expression and itself
-		{{Instruction::AND, {X, X}}, [=]{ return X; }},
-		{{Instruction::OR, {X, X}}, [=]{ return X; }},
-		{{Instruction::XOR, {X, X}}, [=]{ return u256(0); }},
-		{{Instruction::SUB, {X, X}}, [=]{ return u256(0); }},
-		{{Instruction::EQ, {X, X}}, [=]{ return u256(1); }},
-		{{Instruction::LT, {X, X}}, [=]{ return u256(0); }},
-		{{Instruction::SLT, {X, X}}, [=]{ return u256(0); }},
-		{{Instruction::GT, {X, X}}, [=]{ return u256(0); }},
-		{{Instruction::SGT, {X, X}}, [=]{ return u256(0); }},
-		{{Instruction::MOD, {X, X}}, [=]{ return u256(0); }},
-
-		// logical instruction combinations
-		{{Instruction::NOT, {{Instruction::NOT, {X}}}}, [=]{ return X; }},
-		{{Instruction::XOR, {{{X}, {Instruction::XOR, {X, Y}}}}}, [=]{ return Y; }},
-		{{Instruction::OR, {{{X}, {Instruction::AND, {X, Y}}}}}, [=]{ return X; }},
-		{{Instruction::AND, {{{X}, {Instruction::OR, {X, Y}}}}}, [=]{ return X; }},
-		{{Instruction::AND, {{{X}, {Instruction::NOT, {X}}}}}, [=]{ return u256(0); }},
-		{{Instruction::OR, {{{X}, {Instruction::NOT, {X}}}}}, [=]{ return ~u256(0); }},
-	});
-
-	// Double negation of opcodes with binary result
-	for (auto const& op: vector<Instruction>{
-		Instruction::EQ,
-		Instruction::LT,
-		Instruction::SLT,
-		Instruction::GT,
-		Instruction::SGT
-	})
-		addRule({
-			{Instruction::ISZERO, {{Instruction::ISZERO, {{op, {X, Y}}}}}},
-			[=]() -> Pattern { return {op, {X, Y}}; }
-		});
-
-	addRule({
-		{Instruction::ISZERO, {{Instruction::ISZERO, {{Instruction::ISZERO, {X}}}}}},
-		[=]() -> Pattern { return {Instruction::ISZERO, {X}}; }
-	});
-
-	addRule({
-		{Instruction::ISZERO, {{Instruction::XOR, {X, Y}}}},
-		[=]() -> Pattern { return { Instruction::EQ, {X, Y} }; }
-	});
-
-	// Associative operations
-	for (auto const& opFun: vector<pair<Instruction,function<u256(u256 const&,u256 const&)>>>{
-		{Instruction::ADD, plus<u256>()},
-		{Instruction::MUL, multiplies<u256>()},
-		{Instruction::AND, bit_and<u256>()},
-		{Instruction::OR, bit_or<u256>()},
-		{Instruction::XOR, bit_xor<u256>()}
-	})
-	{
-		auto op = opFun.first;
-		auto fun = opFun.second;
-		// Moving constants to the outside, order matters here!
-		// we need actions that return expressions (or patterns?) here, and we need also reversed rules
-		// (X+A)+B -> X+(A+B)
-		addRules(vector<pair<Pattern, function<Pattern()>>>{{
-			{op, {{op, {X, A}}, B}},
-			[=]() -> Pattern { return {op, {X, fun(A.d(), B.d())}}; }
-		}, {
-		// X+(Y+A) -> (X+Y)+A
-			{op, {{op, {X, A}}, Y}},
-			[=]() -> Pattern { return {op, {{op, {X, Y}}, A}}; }
-		}, {
-		// For now, we still need explicit commutativity for the inner pattern
-			{op, {{op, {A, X}}, B}},
-			[=]() -> Pattern { return {op, {X, fun(A.d(), B.d())}}; }
-		}, {
-			{op, {{op, {A, X}}, Y}},
-			[=]() -> Pattern { return {op, {{op, {X, Y}}, A}}; }
-		}});
-	}
 
-	// move constants across subtractions
-	addRules(vector<pair<Pattern, function<Pattern()>>>{
-		{
-			// X - A -> X + (-A)
-			{Instruction::SUB, {X, A}},
-			[=]() -> Pattern { return {Instruction::ADD, {X, 0 - A.d()}}; }
-		}, {
-			// (X + A) - Y -> (X - Y) + A
-			{Instruction::SUB, {{Instruction::ADD, {X, A}}, Y}},
-			[=]() -> Pattern { return {Instruction::ADD, {{Instruction::SUB, {X, Y}}, A}}; }
-		}, {
-			// (A + X) - Y -> (X - Y) + A
-			{Instruction::SUB, {{Instruction::ADD, {A, X}}, Y}},
-			[=]() -> Pattern { return {Instruction::ADD, {{Instruction::SUB, {X, Y}}, A}}; }
-		}, {
-			// X - (Y + A) -> (X - Y) + (-A)
-			{Instruction::SUB, {X, {Instruction::ADD, {Y, A}}}},
-			[=]() -> Pattern { return {Instruction::ADD, {{Instruction::SUB, {X, Y}}, 0 - A.d()}}; }
-		}, {
-			// X - (A + Y) -> (X - Y) + (-A)
-			{Instruction::SUB, {X, {Instruction::ADD, {A, Y}}}},
-			[=]() -> Pattern { return {Instruction::ADD, {{Instruction::SUB, {X, Y}}, 0 - A.d()}}; }
-		}
-	});
+	addRules(simplificationRuleList(A, B, C, X, Y));
 }
 
 Pattern::Pattern(Instruction _instruction, std::vector<Pattern> const& _arguments):
diff --git a/libevmasm/SimplificationRules.h b/libevmasm/SimplificationRules.h
index a4da5849..53f7e595 100644
--- a/libevmasm/SimplificationRules.h
+++ b/libevmasm/SimplificationRules.h
@@ -24,6 +24,7 @@
 #pragma once
 
 #include <libevmasm/ExpressionClasses.h>
+#include <libevmasm/SimplificationRule.h>
 
 #include <functional>
 #include <vector>
@@ -47,19 +48,21 @@ public:
 
 	/// @returns a pointer to the first matching pattern and sets the match
 	/// groups accordingly.
-	std::pair<Pattern, std::function<Pattern()>> const* findFirstMatch(
+	SimplificationRule<Pattern> const* findFirstMatch(
 		Expression const& _expr,
 		ExpressionClasses const& _classes
 	);
 
 private:
-	void addRules(std::vector<std::pair<Pattern, std::function<Pattern()>>> const& _rules);
-	void addRule(std::pair<Pattern, std::function<Pattern()>> const& _rule);
+	void addRules(std::vector<SimplificationRule<Pattern>> const& _rules);
+	void addRule(SimplificationRule<Pattern> const& _rule);
 
 	void resetMatchGroups() { m_matchGroups.clear(); }
 
 	std::map<unsigned, Expression const*> m_matchGroups;
-	std::vector<std::pair<Pattern, std::function<Pattern()>>> m_rules[256];
+	/// Pattern to match, replacement to be applied and flag indicating whether
+	/// the replacement might remove some elements (except constants).
+	std::vector<SimplificationRule<Pattern>> m_rules[256];
 };
 
 /**