Add flag to indicate whether it can be applied to expressions with side-effects.

4 files changed, 104 insertions, 87 deletions

diff --git a/libevmasm/ExpressionClasses.cpp b/libevmasm/ExpressionClasses.cpp
index fc283b0b..58c254b1 100644
--- a/libevmasm/ExpressionClasses.cpp
+++ b/libevmasm/ExpressionClasses.cpp
@@ -202,7 +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()));
+		return rebuildExpression(ExpressionTemplate(std::get<1>(*match)(), _expr.item->location()));
 	}
 
 	if (!_secondRun && _expr.arguments.size() == 2 && SemanticInformation::isCommutativeOperation(*_expr.item))
diff --git a/libevmasm/RuleList.h b/libevmasm/RuleList.h
index d95b014d..70a3ef71 100644
--- a/libevmasm/RuleList.h
+++ b/libevmasm/RuleList.h
@@ -26,6 +26,8 @@
 
 #include <libevmasm/Instruction.h>
 
+#include <libdevcore/CommonData.h>
+
 namespace dev
 {
 namespace solidity
@@ -43,11 +45,12 @@ template <class S> S modWorkaround(S const& _a, S const& _b)
 
 /// @returns a list of simplification rules given certain match placeholders.
 /// A, B and C should represent constants, X and Y arbitrary expressions.
-/// As the simplification can remove instructions, care has to be taken if multiple
-/// non-constant expressions are used. The simplifications should not change the
-/// order of operations, though.
+/// The third element in the tuple is a boolean flag that indicates whether
+/// any non-constant elements in the pattern are removed by applying it.
+/// The simplifications should neven change the order of evaluation of
+/// arbitrary operations, though.
 template <class Pattern>
-std::vector<std::pair<Pattern, std::function<Pattern()>>> simplificationRuleList(
+std::vector<std::tuple<Pattern, std::function<Pattern()>, bool>> simplificationRuleList(
 	Pattern A,
 	Pattern B,
 	Pattern C,
@@ -55,78 +58,78 @@ std::vector<std::pair<Pattern, std::function<Pattern()>>> simplificationRuleList
 	Pattern Y
 )
 {
-	std::vector<std::pair<Pattern, std::function<Pattern()>>> rules;
-	rules += std::vector<std::pair<Pattern, std::function<Pattern()>>>{
+	std::vector<std::tuple<Pattern, std::function<Pattern()>, bool>> rules;
+	rules += std::vector<std::tuple<Pattern, std::function<Pattern()>, bool>>{
 		// 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::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 (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}}; } },
+		{{Instruction::ADD, {X, 0}}, [=]{ return X; }, false},
+		{{Instruction::SUB, {X, 0}}, [=]{ return X; }, false},
+		{{Instruction::MUL, {X, 0}}, [=]{ return u256(0); }, true},
+		{{Instruction::MUL, {X, 1}}, [=]{ return 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, {X, 0}}, [=]{ return u256(0); }, true},
+		{{Instruction::OR, {X, 0}}, [=]{ return X; }, false},
+		{{Instruction::OR, {X, ~u256(0)}}, [=]{ return ~u256(0); }, true},
+		{{Instruction::XOR, {X, 0}}, [=]{ 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 },
 
 		// 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); }},
+		{{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; }},
-		{{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); }},
+		{{Instruction::NOT, {{Instruction::NOT, {X}}}}, [=]{ return X; }, false},
+		{{Instruction::XOR, {{{X}, {Instruction::XOR, {X, Y}}}}}, [=]{ return Y; }, true},
+		{{Instruction::OR, {{{X}, {Instruction::AND, {X, Y}}}}}, [=]{ return X; }, true},
+		{{Instruction::AND, {{{X}, {Instruction::OR, {X, Y}}}}}, [=]{ return X; }, true},
+		{{Instruction::AND, {{{X}, {Instruction::NOT, {X}}}}}, [=]{ return u256(0); }, true},
+		{{Instruction::OR, {{{X}, {Instruction::NOT, {X}}}}}, [=]{ return ~u256(0); }, true},
 	};
 
 	// Double negation of opcodes with binary result
@@ -139,17 +142,20 @@ std::vector<std::pair<Pattern, std::function<Pattern()>>> simplificationRuleList
 	})
 		rules.push_back({
 			{Instruction::ISZERO, {{Instruction::ISZERO, {{op, {X, Y}}}}}},
-			[=]() -> Pattern { return {op, {X, Y}}; }
+			[=]() -> Pattern { return {op, {X, Y}}; },
+			false
 		});
 
 	rules.push_back({
 		{Instruction::ISZERO, {{Instruction::ISZERO, {{Instruction::ISZERO, {X}}}}}},
-		[=]() -> Pattern { return {Instruction::ISZERO, {X}}; }
+		[=]() -> Pattern { return {Instruction::ISZERO, {X}}; },
+		false
 	});
 
 	rules.push_back({
 		{Instruction::ISZERO, {{Instruction::XOR, {X, Y}}}},
-		[=]() -> Pattern { return { Instruction::EQ, {X, Y} }; }
+		[=]() -> Pattern { return { Instruction::EQ, {X, Y} }; },
+		false
 	});
 
 	// Associative operations
@@ -166,45 +172,54 @@ std::vector<std::pair<Pattern, std::function<Pattern()>>> simplificationRuleList
 		// 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)
-		rules += std::vector<std::pair<Pattern, std::function<Pattern()>>>{{
+		rules += std::vector<std::tuple<Pattern, std::function<Pattern()>, bool>>{{
 			{op, {{op, {X, A}}, B}},
-			[=]() -> Pattern { return {op, {X, fun(A.d(), B.d())}}; }
+			[=]() -> Pattern { return {op, {X, fun(A.d(), B.d())}}; },
+			false
 		}, {
 		// X+(Y+A) -> (X+Y)+A
 			{op, {{op, {X, A}}, Y}},
-			[=]() -> Pattern { return {op, {{op, {X, Y}}, A}}; }
+			[=]() -> Pattern { return {op, {{op, {X, Y}}, A}}; },
+			false
 		}, {
 		// 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())}}; }
+			[=]() -> Pattern { return {op, {X, fun(A.d(), B.d())}}; },
+			false
 		}, {
 			{op, {{op, {A, X}}, Y}},
-			[=]() -> Pattern { return {op, {{op, {X, Y}}, A}}; }
+			[=]() -> Pattern { return {op, {{op, {X, Y}}, A}}; },
+			false
 		}};
 	}
 
 	// move constants across subtractions
-	rules += std::vector<std::pair<Pattern, std::function<Pattern()>>>{
+	rules += std::vector<std::tuple<Pattern, std::function<Pattern()>, bool>>{
 		{
 			// X - A -> X + (-A)
 			{Instruction::SUB, {X, A}},
-			[=]() -> Pattern { return {Instruction::ADD, {X, 0 - A.d()}}; }
+			[=]() -> 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}}; }
+			[=]() -> 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}}; }
+			[=]() -> 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()}}; }
+			[=]() -> 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()}}; }
+			[=]() -> Pattern { return {Instruction::ADD, {{Instruction::SUB, {X, Y}}, 0 - A.d()}}; },
+			false
 		}
 	};
 	return rules;
diff --git a/libevmasm/SimplificationRules.cpp b/libevmasm/SimplificationRules.cpp
index 01cad949..d81a993f 100644
--- a/libevmasm/SimplificationRules.cpp
+++ b/libevmasm/SimplificationRules.cpp
@@ -38,7 +38,7 @@ using namespace dev;
 using namespace dev::eth;
 
 
-pair<Pattern, function<Pattern()> > const* Rules::findFirstMatch(
+tuple<Pattern, function<Pattern()>, bool> const* Rules::findFirstMatch(
 	Expression const& _expr,
 	ExpressionClasses const& _classes
 )
@@ -48,22 +48,22 @@ 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 (std::get<0>(rule).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<std::tuple<Pattern, std::function<Pattern ()>, bool>> const& _rules)
 {
 	for (auto const& r: _rules)
 		addRule(r);
 }
 
-void Rules::addRule(std::pair<Pattern, std::function<Pattern()> > const& _rule)
+void Rules::addRule(std::tuple<Pattern, std::function<Pattern()>, bool> const& _rule)
 {
-	m_rules[byte(_rule.first.instruction())].push_back(_rule);
+	m_rules[byte(std::get<0>(_rule).instruction())].push_back(_rule);
 }
 
 Rules::Rules()
diff --git a/libevmasm/SimplificationRules.h b/libevmasm/SimplificationRules.h
index a4da5849..6040fd76 100644
--- a/libevmasm/SimplificationRules.h
+++ b/libevmasm/SimplificationRules.h
@@ -47,19 +47,21 @@ public:
 
 	/// @returns a pointer to the first matching pattern and sets the match
 	/// groups accordingly.
-	std::pair<Pattern, std::function<Pattern()>> const* findFirstMatch(
+	std::tuple<Pattern, std::function<Pattern()>, bool> 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<std::tuple<Pattern, std::function<Pattern()>, bool>> const& _rules);
+	void addRule(std::tuple<Pattern, std::function<Pattern()>, bool> 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<std::tuple<Pattern, std::function<Pattern()>, bool>> m_rules[256];
 };
 
 /**