solidity/libyul/optimiser/DataFlowAnalyzer.cpp

481 lines
16 KiB
C++
Raw Normal View History

/*(
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/>.
*/
/**
2018-06-06 04:43:58 +00:00
* Base class to perform data flow analysis during AST walks.
* Tracks assignments and is used as base class for both Rematerialiser and
* Common Subexpression Eliminator.
*/
2018-10-15 09:52:35 +00:00
#include <libyul/optimiser/DataFlowAnalyzer.h>
2018-10-15 09:52:35 +00:00
#include <libyul/optimiser/NameCollector.h>
#include <libyul/optimiser/Semantics.h>
#include <libyul/optimiser/OptimizerUtilities.h>
#include <libyul/optimiser/KnowledgeBase.h>
#include <libyul/AST.h>
#include <libyul/Dialect.h>
2018-10-15 09:52:35 +00:00
#include <libyul/Exceptions.h>
#include <libyul/Utilities.h>
#include <libsolutil/CommonData.h>
#include <libsolutil/cxx20.h>
#include <variant>
#include <range/v3/view/reverse.hpp>
2019-12-11 16:31:36 +00:00
using namespace solidity;
using namespace solidity::util;
using namespace solidity::yul;
DataFlowAnalyzer::DataFlowAnalyzer(
Dialect const& _dialect,
MemoryAndStorage _analyzeStores,
std::map<YulString, SideEffects> _functionSideEffects
):
2021-11-04 14:40:57 +00:00
m_dialect(_dialect),
m_functionSideEffects(std::move(_functionSideEffects)),
m_knowledgeBase([this](YulString _var) { return variableValue(_var); }),
m_analyzeStores(_analyzeStores == MemoryAndStorage::Analyze)
{
if (m_analyzeStores)
{
if (auto const* builtin = _dialect.memoryStoreFunction(YulString{}))
m_storeFunctionName[static_cast<unsigned>(StoreLoadLocation::Memory)] = builtin->name;
if (auto const* builtin = _dialect.memoryLoadFunction(YulString{}))
m_loadFunctionName[static_cast<unsigned>(StoreLoadLocation::Memory)] = builtin->name;
if (auto const* builtin = _dialect.storageStoreFunction(YulString{}))
m_storeFunctionName[static_cast<unsigned>(StoreLoadLocation::Storage)] = builtin->name;
if (auto const* builtin = _dialect.storageLoadFunction(YulString{}))
m_loadFunctionName[static_cast<unsigned>(StoreLoadLocation::Storage)] = builtin->name;
}
}
2019-05-21 13:52:15 +00:00
void DataFlowAnalyzer::operator()(ExpressionStatement& _statement)
{
if (m_analyzeStores)
2019-05-27 22:14:01 +00:00
{
if (auto vars = isSimpleStore(StoreLoadLocation::Storage, _statement))
{
ASTModifier::operator()(_statement);
2022-03-17 18:04:00 +00:00
cxx20::erase_if(m_state.environment.storage, mapTuple([&](auto&& key, auto&& value) {
return
!m_knowledgeBase.knownToBeDifferent(vars->first, key) &&
vars->second != value;
}));
2022-03-17 18:04:00 +00:00
m_state.environment.storage[vars->first] = vars->second;
return;
}
else if (auto vars = isSimpleStore(StoreLoadLocation::Memory, _statement))
{
ASTModifier::operator()(_statement);
2022-03-17 18:04:00 +00:00
cxx20::erase_if(m_state.environment.memory, mapTuple([&](auto&& key, auto&& /* value */) {
return !m_knowledgeBase.knownToBeDifferentByAtLeast32(vars->first, key);
}));
// TODO erase keccak knowledge, but in a more clever way
m_state.environment.keccak = {};
2022-03-17 18:04:00 +00:00
m_state.environment.memory[vars->first] = vars->second;
return;
}
2019-05-21 13:52:15 +00:00
}
clearKnowledgeIfInvalidated(_statement.expression);
ASTModifier::operator()(_statement);
2019-05-21 13:52:15 +00:00
}
void DataFlowAnalyzer::operator()(Assignment& _assignment)
{
std::set<YulString> names;
for (auto const& var: _assignment.variableNames)
names.emplace(var.name);
2018-05-09 09:43:14 +00:00
assertThrow(_assignment.value, OptimizerException, "");
2019-05-27 22:14:01 +00:00
clearKnowledgeIfInvalidated(*_assignment.value);
visit(*_assignment.value);
2020-11-26 17:07:12 +00:00
handleAssignment(names, _assignment.value.get(), false);
}
void DataFlowAnalyzer::operator()(VariableDeclaration& _varDecl)
{
std::set<YulString> names;
for (auto const& var: _varDecl.variables)
names.emplace(var.name);
2018-02-05 17:02:32 +00:00
m_variableScopes.back().variables += names;
if (_varDecl.value)
2019-05-21 13:52:15 +00:00
{
2019-05-27 22:14:01 +00:00
clearKnowledgeIfInvalidated(*_varDecl.value);
visit(*_varDecl.value);
2019-05-21 13:52:15 +00:00
}
2020-11-26 17:07:12 +00:00
handleAssignment(names, _varDecl.value.get(), true);
}
void DataFlowAnalyzer::operator()(If& _if)
{
2019-05-27 22:14:01 +00:00
clearKnowledgeIfInvalidated(*_if.condition);
2022-03-17 18:04:00 +00:00
Environment preEnvironment = m_state.environment;
2019-05-21 13:52:15 +00:00
ASTModifier::operator()(_if);
2022-03-17 18:04:00 +00:00
joinKnowledge(preEnvironment);
2019-05-21 13:52:15 +00:00
2021-11-04 14:40:57 +00:00
clearValues(assignedVariableNames(_if.body));
}
void DataFlowAnalyzer::operator()(Switch& _switch)
{
2019-05-27 22:14:01 +00:00
clearKnowledgeIfInvalidated(*_switch.expression);
visit(*_switch.expression);
std::set<YulString> assignedVariables;
for (auto& _case: _switch.cases)
{
2022-03-17 18:04:00 +00:00
Environment preEnvironment = m_state.environment;
(*this)(_case.body);
2022-03-17 18:04:00 +00:00
joinKnowledge(preEnvironment);
2019-05-21 13:52:15 +00:00
std::set<YulString> variables = assignedVariableNames(_case.body);
2021-11-04 14:40:57 +00:00
assignedVariables += variables;
// This is a little too destructive, we could retain the old values.
2021-11-04 14:40:57 +00:00
clearValues(variables);
2019-05-27 22:14:01 +00:00
clearKnowledgeIfInvalidated(_case.body);
}
2019-05-21 13:52:15 +00:00
for (auto& _case: _switch.cases)
2019-05-27 22:14:01 +00:00
clearKnowledgeIfInvalidated(_case.body);
clearValues(assignedVariables);
}
void DataFlowAnalyzer::operator()(FunctionDefinition& _fun)
{
// Save all information. We might rather reinstantiate this class,
// but this could be difficult if it is subclassed.
2022-03-10 18:35:58 +00:00
ScopedSaveAndRestore stateResetter(m_state, {});
2021-04-13 11:20:59 +00:00
ScopedSaveAndRestore loopDepthResetter(m_loopDepth, 0u);
pushScope(true);
for (auto const& parameter: _fun.parameters)
m_variableScopes.back().variables.emplace(parameter.name);
for (auto const& var: _fun.returnVariables)
{
m_variableScopes.back().variables.emplace(var.name);
2020-11-26 17:07:12 +00:00
handleAssignment({var.name}, nullptr, true);
}
ASTModifier::operator()(_fun);
2019-10-28 14:25:02 +00:00
// Note that the contents of return variables, storage and memory at this point
// might be incorrect due to the fact that the DataFlowAnalyzer ignores the ``leave``
// statement.
popScope();
}
void DataFlowAnalyzer::operator()(ForLoop& _for)
{
// If the pre block was not empty,
// we would have to deal with more complicated scoping rules.
assertThrow(_for.pre.statements.empty(), OptimizerException, "");
2019-11-27 09:44:40 +00:00
++m_loopDepth;
AssignmentsSinceContinue assignmentsSinceCont;
assignmentsSinceCont(_for.body);
std::set<YulString> assignedVariables =
2021-11-04 14:40:57 +00:00
assignedVariableNames(_for.body) + assignedVariableNames(_for.post);
clearValues(assignedVariables);
2019-05-21 13:52:15 +00:00
// break/continue are tricky for storage and thus we almost always clear here.
2019-05-27 22:14:01 +00:00
clearKnowledgeIfInvalidated(*_for.condition);
clearKnowledgeIfInvalidated(_for.post);
clearKnowledgeIfInvalidated(_for.body);
2019-05-21 13:52:15 +00:00
visit(*_for.condition);
(*this)(_for.body);
clearValues(assignmentsSinceCont.names());
2019-05-27 22:14:01 +00:00
clearKnowledgeIfInvalidated(_for.body);
(*this)(_for.post);
2021-11-04 14:40:57 +00:00
clearValues(assignedVariables);
2019-05-27 22:14:01 +00:00
clearKnowledgeIfInvalidated(*_for.condition);
clearKnowledgeIfInvalidated(_for.post);
clearKnowledgeIfInvalidated(_for.body);
2019-11-27 09:44:40 +00:00
--m_loopDepth;
}
void DataFlowAnalyzer::operator()(Block& _block)
{
size_t numScopes = m_variableScopes.size();
pushScope(false);
ASTModifier::operator()(_block);
popScope();
2018-05-09 09:43:14 +00:00
assertThrow(numScopes == m_variableScopes.size(), OptimizerException, "");
}
std::optional<YulString> DataFlowAnalyzer::storageValue(YulString _key) const
{
if (YulString const* value = valueOrNullptr(m_state.environment.storage, _key))
return *value;
else
return std::nullopt;
}
std::optional<YulString> DataFlowAnalyzer::memoryValue(YulString _key) const
{
if (YulString const* value = valueOrNullptr(m_state.environment.memory, _key))
return *value;
else
return std::nullopt;
}
std::optional<YulString> DataFlowAnalyzer::keccakValue(YulString _start, YulString _length) const
{
if (YulString const* value = valueOrNullptr(m_state.environment.keccak, std::make_pair(_start, _length)))
return *value;
else
return std::nullopt;
}
void DataFlowAnalyzer::handleAssignment(std::set<YulString> const& _variables, Expression* _value, bool _isDeclaration)
{
2020-11-26 17:07:12 +00:00
if (!_isDeclaration)
clearValues(_variables);
MovableChecker movableChecker{m_dialect, &m_functionSideEffects};
if (_value)
movableChecker.visit(*_value);
else
for (auto const& var: _variables)
2019-11-27 09:44:40 +00:00
assignValue(var, &m_zero);
if (_value && _variables.size() == 1)
{
YulString name = *_variables.begin();
// Expression has to be movable and cannot contain a reference
// to the variable that will be assigned to.
if (movableChecker.movable() && !movableChecker.referencedVariables().count(name))
2019-11-27 09:44:40 +00:00
assignValue(name, _value);
}
auto const& referencedVariables = movableChecker.referencedVariables();
for (auto const& name: _variables)
2019-05-21 13:52:15 +00:00
{
2022-03-10 18:35:58 +00:00
m_state.references[name] = referencedVariables;
2020-11-26 17:07:12 +00:00
if (!_isDeclaration)
{
// assignment to slot denoted by "name"
2022-03-17 18:04:00 +00:00
m_state.environment.storage.erase(name);
2020-11-26 17:07:12 +00:00
// assignment to slot contents denoted by "name"
2022-03-17 18:04:00 +00:00
cxx20::erase_if(m_state.environment.storage, mapTuple([&name](auto&& /* key */, auto&& value) { return value == name; }));
2020-11-26 17:07:12 +00:00
// assignment to slot denoted by "name"
2022-03-17 18:04:00 +00:00
m_state.environment.memory.erase(name);
2020-11-26 17:07:12 +00:00
// assignment to slot contents denoted by "name"
cxx20::erase_if(m_state.environment.keccak, [&name](auto&& _item) {
return _item.first.first == name || _item.first.second == name || _item.second == name;
});
2022-03-17 18:04:00 +00:00
cxx20::erase_if(m_state.environment.memory, mapTuple([&name](auto&& /* key */, auto&& value) { return value == name; }));
2020-11-26 17:07:12 +00:00
}
2019-05-21 13:52:15 +00:00
}
if (_value && _variables.size() == 1)
{
YulString variable = *_variables.begin();
if (!movableChecker.referencedVariables().count(variable))
{
// This might erase additional knowledge about the slot.
// 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))
2022-03-17 18:04:00 +00:00
m_state.environment.memory[*key] = variable;
else if (auto key = isSimpleLoad(StoreLoadLocation::Storage, *_value))
2022-03-17 18:04:00 +00:00
m_state.environment.storage[*key] = variable;
else if (auto arguments = isKeccak(*_value))
m_state.environment.keccak[*arguments] = variable;
}
}
}
void DataFlowAnalyzer::pushScope(bool _functionScope)
{
m_variableScopes.emplace_back(_functionScope);
}
void DataFlowAnalyzer::popScope()
{
for (auto const& name: m_variableScopes.back().variables)
{
2022-03-10 18:35:58 +00:00
m_state.value.erase(name);
m_state.references.erase(name);
}
m_variableScopes.pop_back();
}
void DataFlowAnalyzer::clearValues(std::set<YulString> _variables)
{
// All variables that reference variables to be cleared also have to be
// cleared, but not recursively, since only the value of the original
// variables changes. Example:
// let a := 1
// let b := a
// let c := b
// a := 2
// add(b, c)
// In the last line, we can replace c by b, but not b by a.
//
// This cannot be easily tested since the substitutions will be done
// one by one on the fly, and the last line will just be add(1, 1)
2019-05-21 13:52:15 +00:00
// First clear storage knowledge, because we do not have to clear
// storage knowledge of variables whose expression has changed,
// since the value is still unchanged.
2021-01-13 16:23:27 +00:00
auto eraseCondition = mapTuple([&_variables](auto&& key, auto&& value) {
return _variables.count(key) || _variables.count(value);
});
2022-03-17 18:04:00 +00:00
cxx20::erase_if(m_state.environment.storage, eraseCondition);
cxx20::erase_if(m_state.environment.memory, eraseCondition);
cxx20::erase_if(m_state.environment.keccak, [&_variables](auto&& _item) {
return
_variables.count(_item.first.first) ||
_variables.count(_item.first.second) ||
_variables.count(_item.second);
});
2019-05-21 13:52:15 +00:00
// Also clear variables that reference variables to be cleared.
2021-01-13 11:43:38 +00:00
for (auto const& variableToClear: _variables)
2022-03-10 18:35:58 +00:00
for (auto const& [ref, names]: m_state.references)
2021-01-13 11:43:38 +00:00
if (names.count(variableToClear))
_variables.emplace(ref);
// Clear the value and update the reference relation.
for (auto const& name: _variables)
{
2022-03-10 18:35:58 +00:00
m_state.value.erase(name);
m_state.references.erase(name);
}
}
2019-11-27 09:44:40 +00:00
void DataFlowAnalyzer::assignValue(YulString _variable, Expression const* _value)
{
2022-03-10 18:35:58 +00:00
m_state.value[_variable] = {_value, m_loopDepth};
2019-11-27 09:44:40 +00:00
}
2019-05-27 22:14:01 +00:00
void DataFlowAnalyzer::clearKnowledgeIfInvalidated(Block const& _block)
2019-05-21 13:52:15 +00:00
{
if (!m_analyzeStores)
return;
SideEffectsCollector sideEffects(m_dialect, _block, &m_functionSideEffects);
2019-05-27 22:14:01 +00:00
if (sideEffects.invalidatesStorage())
2022-03-17 18:04:00 +00:00
m_state.environment.storage.clear();
2019-05-27 22:14:01 +00:00
if (sideEffects.invalidatesMemory())
{
2022-03-17 18:04:00 +00:00
m_state.environment.memory.clear();
m_state.environment.keccak.clear();
}
2019-05-21 13:52:15 +00:00
}
2019-05-27 22:14:01 +00:00
void DataFlowAnalyzer::clearKnowledgeIfInvalidated(Expression const& _expr)
2019-05-21 13:52:15 +00:00
{
if (!m_analyzeStores)
return;
SideEffectsCollector sideEffects(m_dialect, _expr, &m_functionSideEffects);
2019-05-27 22:14:01 +00:00
if (sideEffects.invalidatesStorage())
2022-03-17 18:04:00 +00:00
m_state.environment.storage.clear();
2019-05-27 22:14:01 +00:00
if (sideEffects.invalidatesMemory())
{
2022-03-17 18:04:00 +00:00
m_state.environment.memory.clear();
m_state.environment.keccak.clear();
}
2019-05-21 13:52:15 +00:00
}
bool DataFlowAnalyzer::inScope(YulString _variableName) const
{
for (auto const& scope: m_variableScopes | ranges::views::reverse)
{
2018-02-05 17:02:32 +00:00
if (scope.variables.count(_variableName))
return true;
2018-02-05 17:02:32 +00:00
if (scope.isFunction)
return false;
}
return false;
}
2019-05-21 13:52:15 +00:00
std::optional<u256> DataFlowAnalyzer::valueOfIdentifier(YulString const& _name) const
{
if (AssignedValue const* value = variableValue(_name))
if (Literal const* literal = std::get_if<Literal>(value->value))
return valueOfLiteral(*literal);
return std::nullopt;
}
std::optional<std::pair<YulString, YulString>> DataFlowAnalyzer::isSimpleStore(
StoreLoadLocation _location,
2019-05-21 13:52:15 +00:00
ExpressionStatement const& _statement
) const
{
if (FunctionCall const* funCall = std::get_if<FunctionCall>(&_statement.expression))
if (funCall->functionName.name == m_storeFunctionName[static_cast<unsigned>(_location)])
if (Identifier const* key = std::get_if<Identifier>(&funCall->arguments.front()))
if (Identifier const* value = std::get_if<Identifier>(&funCall->arguments.back()))
return std::make_pair(key->name, value->name);
2019-05-21 13:52:15 +00:00
return {};
}
std::optional<YulString> DataFlowAnalyzer::isSimpleLoad(
StoreLoadLocation _location,
Expression const& _expression
) const
{
if (FunctionCall const* funCall = std::get_if<FunctionCall>(&_expression))
if (funCall->functionName.name == m_loadFunctionName[static_cast<unsigned>(_location)])
if (Identifier const* key = std::get_if<Identifier>(&funCall->arguments.front()))
return key->name;
return {};
}
2022-03-17 18:04:00 +00:00
std::optional<std::pair<YulString, YulString>> DataFlowAnalyzer::isKeccak(Expression const& _expression) const
{
if (FunctionCall const* funCall = std::get_if<FunctionCall>(&_expression))
if (funCall->functionName.name == m_dialect.hashFunction({}))
if (Identifier const* start = std::get_if<Identifier>(&funCall->arguments.at(0)))
if (Identifier const* length = std::get_if<Identifier>(&funCall->arguments.at(1)))
return std::make_pair(start->name, length->name);
return std::nullopt;
}
2022-03-17 18:04:00 +00:00
void DataFlowAnalyzer::joinKnowledge(Environment const& _olderEnvironment)
{
if (!m_analyzeStores)
return;
joinKnowledgeHelper(m_state.environment.storage, _olderEnvironment.storage);
joinKnowledgeHelper(m_state.environment.memory, _olderEnvironment.memory);
cxx20::erase_if(m_state.environment.keccak, mapTuple([&_olderEnvironment](auto&& key, auto&& currentValue) {
YulString const* oldValue = valueOrNullptr(_olderEnvironment.keccak, key);
return !oldValue || *oldValue != currentValue;
}));
2022-03-17 18:04:00 +00:00
}
void DataFlowAnalyzer::joinKnowledgeHelper(
std::unordered_map<YulString, YulString>& _this,
std::unordered_map<YulString, YulString> const& _older
)
{
// 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_state.environment.memory and thus any overlapping write would have cleared the keys
// that are not known to be different inside m_state.environment.memory already.
cxx20::erase_if(_this, mapTuple([&_older](auto&& key, auto&& currentValue){
YulString const* oldValue = valueOrNullptr(_older, key);
2022-03-17 18:04:00 +00:00
return !oldValue || *oldValue != currentValue;
}));
}