eof: Support CALLF, JUMPF and RETF

Author: rodiazet

libevmasm/Assembly.cpp

@@ -407,9 +407,18 @@ void Assembly::assemblyStream(
 		f.feed(i, _debugInfoSelection);
 	f.flush();
 
-	// Implementing this requires introduction of CALLF, RETF and JUMPF
 	if (m_codeSections.size() > 1)
-		solUnimplemented("Add support for more code sections");
+	{
+		for (size_t i = 1; i < m_codeSections.size(); ++i)
+		{
+			_out << std::endl << _prefix << "code_section_" << i << ": assembly {\n";
+			Functionalizer codeSectionF(_out, _prefix + "    ", _sourceCodes, *this);
+			for (auto const& item: m_codeSections[i].items)
+				codeSectionF.feed(item, _debugInfoSelection);
+			codeSectionF.flush();
+			_out << _prefix << "}" << std::endl;
+		}
+	}
 
 	if (!m_data.empty() || !m_subs.empty())
 	{
@@ -1415,7 +1424,10 @@ LinkerObject const& Assembly::assembleEOF() const
 					item.instruction() != Instruction::RETURNCONTRACT &&
 					item.instruction() != Instruction::EOFCREATE &&
 					item.instruction() != Instruction::RJUMP &&
-					item.instruction() != Instruction::RJUMPI
+					item.instruction() != Instruction::RJUMPI &&
+					item.instruction() != Instruction::CALLF &&
+					item.instruction() != Instruction::JUMPF &&
+					item.instruction() != Instruction::RETF
 				);
 				solAssert(!(item.instruction() >= Instruction::PUSH0 && item.instruction() <= Instruction::PUSH32));
 				ret.bytecode += assembleOperation(item);
@@ -1470,6 +1482,25 @@ LinkerObject const& Assembly::assembleEOF() const
 				appendBigEndianUint16(ret.bytecode, item.data());
 				break;
 			}
+			case CallF:
+			{
+				ret.bytecode.push_back(static_cast<uint8_t>(Instruction::CALLF));
+				solAssert(item.data() <= std::numeric_limits<uint16_t>::max(), "Invalid callf index value.");
+				appendBigEndianUint16(ret.bytecode, item.data());
+				break;
+			}
+			case JumpF:
+			{
+				ret.bytecode.push_back(static_cast<uint8_t>(Instruction::JUMPF));
+				solAssert(item.data() <= std::numeric_limits<uint16_t>::max(), "Invalid jumpf index value.");
+				appendBigEndianUint16(ret.bytecode, item.data());
+				break;
+			}
+			case RetF:
+			{
+				ret.bytecode.push_back(static_cast<uint8_t>(Instruction::RETF));
+				break;
+			}
 			default:
 				solThrow(InvalidOpcode, "Unexpected opcode while assembling.");
 			}

libevmasm/Assembly.h

@@ -65,9 +65,50 @@ class Assembly
 	}
 
 	std::optional<uint8_t> eofVersion() const { return m_eofVersion; }
+	bool supportsFunctions() const { return m_eofVersion.has_value(); }
 	bool supportsRelativeJumps() const { return m_eofVersion.has_value(); }
 	AssemblyItem newTag() { assertThrow(m_usedTags < 0xffffffff, AssemblyException, ""); return AssemblyItem(Tag, m_usedTags++); }
 	AssemblyItem newPushTag() { assertThrow(m_usedTags < 0xffffffff, AssemblyException, ""); return AssemblyItem(PushTag, m_usedTags++); }
+	AssemblyItem newFunctionCall(uint16_t _functionID)
+	{
+		solAssert(_functionID < m_codeSections.size(), "Call to undeclared function.");
+		solAssert(_functionID > 0, "Cannot call call section 0");
+		auto const& section = m_codeSections.at(_functionID);
+		if (section.outputs != 0x80)
+			return AssemblyItem::functionCall(_functionID, section.inputs, section.outputs);
+		else
+			return AssemblyItem::jumpF(_functionID, section.inputs);
+	}
+
+	AssemblyItem newFunctionReturn()
+	{
+		return AssemblyItem::functionReturn(m_codeSections.at(m_currentCodeSection).outputs);
+	}
+
+	uint16_t createFunction(uint8_t _args, uint8_t _rets)
+	{
+		size_t functionID = m_codeSections.size();
+		solAssert(functionID < 1024, "Too many functions.");
+		solAssert(m_currentCodeSection == 0, "Functions need to be declared from the main block.");
+		solAssert(_rets <= 0x80, "Too many function returns.");
+		m_codeSections.emplace_back(CodeSection{_args, _rets, {}});
+		return static_cast<uint16_t>(functionID);
+	}
+
+	void beginFunction(uint16_t _functionID)
+	{
+		solAssert(m_currentCodeSection == 0, "Atempted to begin a function before ending the last one.");
+		solAssert(_functionID < m_codeSections.size(), "Attempt to begin an undeclared function.");
+		auto& section = m_codeSections.at(_functionID);
+		solAssert(section.items.empty(), "Function already defined.");
+		m_currentCodeSection = _functionID;
+	}
+	void endFunction()
+	{
+		solAssert(m_currentCodeSection != 0, "End function without begin function.");
+		m_currentCodeSection = 0;
+	}
+
 	/// Returns a tag identified by the given name. Creates it if it does not yet exist.
 	AssemblyItem namedTag(std::string const& _name, size_t _params, size_t _returns, std::optional<uint64_t> _sourceID);
 	AssemblyItem newData(bytes const& _data) { util::h256 h(util::keccak256(util::asString(_data))); m_data[h] = _data; return AssemblyItem(PushData, h); }
@@ -111,6 +152,16 @@ class Assembly
 		return append(AssemblyItem::returnContract(_containerId));
 	}
 
+	AssemblyItem appendFunctionCall(uint16_t _functionID)
+	{
+		return append(newFunctionCall(_functionID));
+	}
+
+	AssemblyItem appendFunctionReturn()
+	{
+		return append(newFunctionReturn());
+	}
+
 	AssemblyItem appendJump() { auto ret = append(newPushTag()); append(Instruction::JUMP); return ret; }
 	AssemblyItem appendJumpI() { auto ret = append(newPushTag()); append(Instruction::JUMPI); return ret; }
 	AssemblyItem appendJump(AssemblyItem const& _tag) { auto ret = append(_tag.pushTag()); append(Instruction::JUMP); return ret; }

libevmasm/AssemblyItem.cpp

@@ -78,6 +78,9 @@ std::pair<std::string, std::string> AssemblyItem::nameAndData(langutil::EVMVersi
 	case ReturnContract:
 	case RelativeJump:
 	case ConditionalRelativeJump:
+	case CallF:
+	case JumpF:
+	case RetF:
 		return {instructionInfo(instruction(), _evmVersion).name, m_data != nullptr ? toStringInHex(*m_data) : ""};
 	case Push:
 		return {"PUSH", toStringInHex(data())};
@@ -130,6 +133,7 @@ size_t AssemblyItem::bytesRequired(size_t _addressLength, langutil::EVMVersion _
 	{
 	case Operation:
 	case Tag: // 1 byte for the JUMPDEST
+	case RetF:
 		return 1;
 	case Push:
 		return
@@ -172,6 +176,8 @@ size_t AssemblyItem::bytesRequired(size_t _addressLength, langutil::EVMVersion _
 	case RelativeJump:
 	case ConditionalRelativeJump:
 	case AuxDataLoadN:
+	case JumpF:
+	case CallF:
 		return 1 + 2;
 	case EOFCreate:
 		return 2;
@@ -186,14 +192,21 @@ size_t AssemblyItem::bytesRequired(size_t _addressLength, langutil::EVMVersion _
 
 size_t AssemblyItem::arguments() const
 {
-	if (hasInstruction())
+	if (hasInstruction() && type() != CallF && type() != JumpF && type() != RetF)
 		// The latest EVMVersion is used here, since the InstructionInfo is assumed to be
 		// the same across all EVM versions except for the instruction name.
 		return static_cast<size_t>(instructionInfo(instruction(), EVMVersion()).args);
 	else if (type() == VerbatimBytecode)
 		return std::get<0>(*m_verbatimBytecode);
 	else if (type() == AssignImmutable)
 		return 2;
+	else if (type() == CallF || type() == JumpF)
+	{
+		solAssert(m_functionSignature.has_value());
+		return (*m_functionSignature).argsNum;
+	}
+	else if (type() == RetF)
+		return static_cast<size_t>(data());
 	else
 		return 0;
 }
@@ -226,6 +239,12 @@ size_t AssemblyItem::returnValues() const
 		return std::get<1>(*m_verbatimBytecode);
 	case AuxDataLoadN:
 		return 1;
+	case CallF:
+		solAssert(m_functionSignature.has_value());
+		return (*m_functionSignature).retsNum;
+	case JumpF:
+	case RetF:
+		return 0;
 	case AssignImmutable:
 	case UndefinedItem:
 		break;
@@ -244,6 +263,9 @@ bool AssemblyItem::canBeFunctional() const
 	case ReturnContract:
 	case RelativeJump:
 	case ConditionalRelativeJump:
+	case CallF:
+	case JumpF:
+	case RetF:
 		return !isDupInstruction(instruction()) && !isSwapInstruction(instruction());
 	case Push:
 	case PushTag:
@@ -374,6 +396,15 @@ std::string AssemblyItem::toAssemblyText(Assembly const& _assembly) const
 	case ConditionalRelativeJump:
 		text = "rjumpi{" + std::string("tag_") + std::to_string(static_cast<size_t>(data())) + "}";
 		break;
+	case CallF:
+		text = "callf{" +  std::to_string(static_cast<size_t>(data())) + "}";
+		break;
+	case JumpF:
+		text = "jumpf{" +  std::to_string(static_cast<size_t>(data())) + "}";
+		break;
+	case RetF:
+		text = "retf";
+		break;
 	}
 	if (m_jumpType == JumpType::IntoFunction || m_jumpType == JumpType::OutOfFunction)
 	{
@@ -396,6 +427,9 @@ std::ostream& solidity::evmasm::operator<<(std::ostream& _out, AssemblyItem cons
 	case ReturnContract:
 	case RelativeJump:
 	case ConditionalRelativeJump:
+	case CallF:
+	case JumpF:
+	case RetF:
 		_out << " " << instructionInfo(_item.instruction(), EVMVersion()).name;
 		if (_item.instruction() == Instruction::JUMP || _item.instruction() == Instruction::JUMPI)
 			_out << "\t" << _item.getJumpTypeAsString();

libevmasm/AssemblyItem.h

@@ -59,6 +59,9 @@ enum AssemblyItemType
 	ReturnContract, ///< Returns new container (with auxiliary data filled in) to be deployed
 	RelativeJump, ///< Jumps to relative position accordingly to its argument
 	ConditionalRelativeJump, ///< Same as RelativeJump but takes condition from the stack
+	CallF, ///< Calls a function under function ID. Function is returning.
+	JumpF, ///< Calls a function under function ID. Function is non-returning.
+	RetF, ///< Returns from returning function to the caller.
 	VerbatimBytecode ///< Contains data that is inserted into the bytecode code section without modification.
 };
 
@@ -105,6 +108,25 @@ class AssemblyItem
 		m_debugData{langutil::DebugData::create()}
 	{}
 
+	static AssemblyItem functionCall(uint16_t _functionID, uint8_t _args, uint8_t _rets, langutil::DebugData::ConstPtr _debugData = langutil::DebugData::create())
+	{
+		// TODO: Make this constructor this way that it's impossible to create it without setting functions signature.
+		// It can be done by template constructor with Instruction as template parameter i.e. Same for JumpF below.
+		AssemblyItem result(CallF, Instruction::CALLF, _functionID, _debugData);
+		result.m_functionSignature = {_args, _rets};
+		return result;
+	}
+	static AssemblyItem jumpF(uint16_t _functionID, uint8_t _args, langutil::DebugData::ConstPtr _debugData = langutil::DebugData::create())
+	{
+		AssemblyItem result(JumpF, Instruction::JUMPF, _functionID, _debugData);
+		result.m_functionSignature = {_args, 0};
+		return result;
+	}
+	static AssemblyItem functionReturn(uint8_t _rets, langutil::DebugData::ConstPtr _debugData = langutil::DebugData::create())
+	{
+		return AssemblyItem(RetF, Instruction::RETF, _rets, _debugData);
+	}
+
 	static AssemblyItem eofCreate(ContainerID _containerID, langutil::DebugData::ConstPtr _debugData = langutil::DebugData::create())
 	{
 		return AssemblyItem(EOFCreate, Instruction::EOFCREATE, _containerID, std::move(_debugData));
@@ -162,7 +184,10 @@ class AssemblyItem
 			m_type == EOFCreate ||
 			m_type == ReturnContract ||
 			m_type == RelativeJump ||
-			m_type == ConditionalRelativeJump;
+			m_type == ConditionalRelativeJump ||
+			m_type == CallF ||
+			m_type == JumpF ||
+			m_type == RetF;
 	}
 	/// @returns the instruction of this item (only valid if hasInstruction returns true)
 	Instruction instruction() const
@@ -267,6 +292,13 @@ class AssemblyItem
 	AssemblyItemType m_type;
 	Instruction m_instruction; ///< Only valid if m_type == Operation
 	std::shared_ptr<u256> m_data; ///< Only valid if m_type != Operation
+
+	struct FunctionSignature
+	{
+		uint8_t argsNum;
+		uint8_t retsNum;
+	};
+	std::optional<FunctionSignature> m_functionSignature; ///< Only valid if m_type == CallF or JumpF
 	/// If m_type == VerbatimBytecode, this holds number of arguments, number of
 	/// return variables and verbatim bytecode.
 	std::optional<std::tuple<size_t, size_t, bytes>> m_verbatimBytecode;

libevmasm/Instruction.cpp

@@ -169,6 +169,9 @@ std::map<std::string, Instruction> const solidity::evmasm::c_instructions =
 	{ "LOG3", Instruction::LOG3 },
 	{ "LOG4", Instruction::LOG4 },
 	{ "DATALOADN", Instruction::DATALOADN },
+	{ "CALLF", Instruction::CALLF },
+	{ "RETF", Instruction::RETF },
+	{ "JUMPF", Instruction::JUMPF },
 	{ "RJUMP", Instruction::RJUMP },
 	{ "RJUMPI", Instruction::RJUMPI },
 	{ "EOFCREATE", Instruction::EOFCREATE },
@@ -330,6 +333,9 @@ static std::map<Instruction, InstructionInfo> const c_instructionInfo =
 	{Instruction::LOG2,           {"LOG2",            0,  4,   0,  true,       Tier::Special}},
 	{Instruction::LOG3,           {"LOG3",            0,  5,   0,  true,       Tier::Special}},
 	{Instruction::LOG4,           {"LOG4",            0,  6,   0,  true,       Tier::Special}},
+	{Instruction::RETF,           {"RETF",            0,  0,   0,  true,       Tier::Special}},
+	{Instruction::CALLF,          {"CALLF",           2,  0,   0,  true,       Tier::Special}},
+	{Instruction::JUMPF,          {"JUMPF",           2,  0,   0,  true,       Tier::Special}},
 	{Instruction::EOFCREATE,      {"EOFCREATE",       1,  4,   1,  true,       Tier::Special}},
 	{Instruction::RETURNCONTRACT, {"RETURNCONTRACT",  1,  2,   0,  true,       Tier::Special}},
 	{Instruction::CREATE,         {"CREATE",          0,  3,   1,  true,       Tier::Special}},

libevmasm/Instruction.h

@@ -186,6 +186,9 @@ enum class Instruction: uint8_t
 
 	RJUMP = 0xe0,             ///< relative jump
 	RJUMPI = 0xe1,            ///< conditional relative jump
+	CALLF = 0xe3,             ///< call function in a EOF code section
+	RETF = 0xe4,              ///< return to caller from the code section of EOF continer
+	JUMPF = 0xe5,             ///< jump to a code section of EOF contaner. No stack cleaning.
 	EOFCREATE = 0xec,         ///< create a new account with associated container code.
 	RETURNCONTRACT = 0xee,    ///< return container to be deployed with axiliary data filled in.
 	CREATE = 0xf0,            ///< create a new account with associated code

libevmasm/SemanticInformation.cpp

@@ -231,6 +231,9 @@ bool SemanticInformation::breaksCSEAnalysisBlock(AssemblyItem const& _item, bool
 	case PushDeployTimeAddress:
 	case AssignImmutable:
 	case VerbatimBytecode:
+	case CallF:
+	case JumpF:
+	case RetF:
 		return true;
 	case Push:
 	case PushTag:
@@ -331,6 +334,9 @@ bool SemanticInformation::altersControlFlow(AssemblyItem const& _item)
 	case Instruction::INVALID:
 	case Instruction::REVERT:
 	case Instruction::RETURNCONTRACT:
+	case Instruction::CALLF:
+	case Instruction::JUMPF:
+	case Instruction::RETF:
 		return true;
 	default:
 		return false;
@@ -397,6 +403,7 @@ bool SemanticInformation::isDeterministic(AssemblyItem const& _item)
 	case Instruction::RETURNDATACOPY: // depends on previous calls
 	case Instruction::RETURNDATASIZE:
 	case Instruction::EOFCREATE:
+	case Instruction::CALLF:
 		return false;
 	default:
 		return true;

liblangutil/EVMVersion.cpp

@@ -82,6 +82,9 @@ bool EVMVersion::hasOpcode(Instruction _opcode, std::optional<uint8_t> _eofVersi
 	case Instruction::DATALOADN:
 	case Instruction::RJUMP:
 	case Instruction::RJUMPI:
+	case Instruction::CALLF:
+	case Instruction::JUMPF:
+	case Instruction::RETF:
 		return _eofVersion.has_value();
 	default:
 		return true;

libyul/backends/evm/AbstractAssembly.h

@@ -57,6 +57,7 @@ class AbstractAssembly
 	using LabelID = size_t;
 	using SubID = size_t;
 	using ContainerID = uint8_t;
+	using FunctionID = uint16_t;
 	enum class JumpType { Ordinary, IntoFunction, OutOfFunction };
 
 	virtual ~AbstractAssembly() = default;
@@ -101,6 +102,19 @@ class AbstractAssembly
 	virtual void appendAssemblySize() = 0;
 	/// Creates a new sub-assembly, which can be referenced using dataSize and dataOffset.
 	virtual std::pair<std::shared_ptr<AbstractAssembly>, SubID> createSubAssembly(bool _creation, std::string _name = "") = 0;
+
+	/// Creates new function with given signature and returns newly created function ID
+	virtual FunctionID createFunction(uint8_t _args, uint8_t _rets) = 0;
+	/// Starts filling function body under given function ID
+	virtual void beginFunction(FunctionID _functionID) = 0;
+	/// Ends currently being filled function
+	virtual void endFunction() = 0;
+
+	/// Appends function call to a function under given ID
+	virtual void appendFunctionCall(FunctionID _functionID) = 0;
+	/// Appends function return from currently being filled function.
+	virtual void appendFunctionReturn() = 0;
+
 	/// Appends the offset of the given sub-assembly or data.
 	virtual void appendDataOffset(std::vector<SubID> const& _subPath) = 0;
 	/// Appends the size of the given sub-assembly or data.