cerc-io/solidity
11
0
Fork 0
mirror of https://github.com/ethereum/solidity synced 2023-10-03 13:03:40 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Disable redundent assign eliminator for deeply nested loops.

Browse Source
This commit is contained in:
chriseth
2019-04-23 09:50:25 +02:00
parent 4312f6ec23
commit aad4373da4
4 changed files with 202 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
libyul/optimiser/RedundantAssignEliminator.cpp
+32 -10
View File
@@ -117,13 +117,13 @@ void RedundantAssignEliminator::operator()(ForLoop const& _forLoop)
{
ForLoopInfo outerForLoopInfo;
swap(outerForLoopInfo, m_forLoopInfo);
++m_forLoopNestingDepth;
// If the pre block was not empty,
// we would have to deal with more complicated scoping rules.
assertThrow(_forLoop.pre.statements.empty(), OptimizerException, "");
// We just run the loop twice to account for the
// back edge.
// We just run the loop twice to account for the back edge.
// There need not be more runs because we only have three different states.
visit(*_forLoop.condition);
@@ -137,24 +137,46 @@ void RedundantAssignEliminator::operator()(ForLoop const& _forLoop)
visit(*_forLoop.condition);
TrackedAssignments oneRun{m_assignments};
if (m_forLoopNestingDepth < 6)
{
// Do the second run only for small nesting depths to avoid horrible runtime.
TrackedAssignments oneRun{m_assignments};
(*this)(_forLoop.body);
(*this)(_forLoop.body);
merge(m_assignments, move(m_forLoopInfo.pendingContinueStmts));
m_forLoopInfo.pendingContinueStmts.clear();
(*this)(_forLoop.post);
merge(m_assignments, move(m_forLoopInfo.pendingContinueStmts));
m_forLoopInfo.pendingContinueStmts.clear();
(*this)(_forLoop.post);
visit(*_forLoop.condition);
visit(*_forLoop.condition);
// Order of merging does not matter because "max" is commutative and associative.
merge(m_assignments, move(oneRun));
}
else
{
// Shortcut to avoid horrible runtime:
// Change all assignments that were newly introduced in the for loop to "used".
// We do not have to do that with the "break" or "continue" paths, because
// they will be joined later anyway.
// TODO parallel traversal might be more efficient here.
for (auto& var: m_assignments)
for (auto& assignment: var.second)
{
auto zeroIt = zeroRuns.find(var.first);
if (zeroIt != zeroRuns.end() && zeroIt->second.count(assignment.first))
continue;
assignment.second = State::Value::Used;
}
}
// Order does not matter because "max" is commutative and associative.
merge(m_assignments, move(oneRun));
// Order of merging does not matter because "max" is commutative and associative.
merge(m_assignments, move(zeroRuns));
merge(m_assignments, move(m_forLoopInfo.pendingBreakStmts));
m_forLoopInfo.pendingBreakStmts.clear();
// Restore potential outer for-loop states.
swap(m_forLoopInfo, outerForLoopInfo);
--m_forLoopNestingDepth;
}
void RedundantAssignEliminator::operator()(Break const&)