From 5fe9171b38404bdb2296b849182769083c9d07b3 Mon Sep 17 00:00:00 2001 From: "Vikram S. Adve" Date: Mon, 4 Nov 2002 14:15:57 +0000 Subject: Generic graph iterator to enumerate the SCCs of a graph in linear time using Tarjan's DFS algorithm. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@4531 91177308-0d34-0410-b5e6-96231b3b80d8 --- include/Support/SCCIterator.h | 221 +++++++++++++++++++++++++++++++++++++++++ include/llvm/ADT/SCCIterator.h | 221 +++++++++++++++++++++++++++++++++++++++++ 2 files changed, 442 insertions(+) create mode 100644 include/Support/SCCIterator.h create mode 100644 include/llvm/ADT/SCCIterator.h diff --git a/include/Support/SCCIterator.h b/include/Support/SCCIterator.h new file mode 100644 index 0000000000..e737f7c990 --- /dev/null +++ b/include/Support/SCCIterator.h @@ -0,0 +1,221 @@ +//===-- Support/TarjanSCCIterator.h -Generic Tarjan SCC iterator -*- C++ -*--=// +// +// This builds on the Support/GraphTraits.h file to find the strongly +// connected components (SCCs) of a graph in O(N+E) time using +// Tarjan's DFS algorithm. +// +// The SCC iterator has the important property that if a node in SCC S1 +// has an edge to a node in SCC S2, then it visits S1 *after* S2. +// +// To visit S1 *before* S2, use the TarjanSCCIterator on the Inverse graph. +// (NOTE: This requires some simple wrappers and is not supported yet.) +//===----------------------------------------------------------------------===// + +#ifndef LLVM_SUPPORT_TARJANSCC_ITERATOR_H +#define LLVM_SUPPORT_TARJANSCC_ITERATOR_H + +#include "Support/GraphTraits.h" +#include +#include +#include +#include +#include + + +//-------------------------------------------------------------------------- +// class SCC : A simple representation of an SCC in a generic Graph. +//-------------------------------------------------------------------------- + +template > +struct SCC: public std::vector { + + typedef typename GT::NodeType NodeType; + typedef typename GT::ChildIteratorType ChildItTy; + + typedef std::vector super; + typedef typename super::iterator iterator; + typedef typename super::const_iterator const_iterator; + typedef typename super::reverse_iterator reverse_iterator; + typedef typename super::const_reverse_iterator const_reverse_iterator; + + // HasLoop() -- Test if this SCC has a loop. If it has more than one + // node, this is trivially true. If not, it may still contain a loop + // if the node has an edge back to itself. + bool HasLoop() const { + if (size() > 1) return true; + NodeType* N = front(); + for (ChildItTy CI=GT::child_begin(N), CE=GT::child_end(N); CI != CE; ++CI) + if (*CI == N) + return true; + return false; + } +}; + +//-------------------------------------------------------------------------- +// class TarjanSCC_iterator: Enumerate the SCCs of a directed graph, in +// reverse topological order of the SCC DAG. +//-------------------------------------------------------------------------- + +const unsigned long MAXLONG = (1 << (8 * sizeof(unsigned long) - 1)); + +namespace { + Statistic<> NumSCCs("NumSCCs", "Number of Strongly Connected Components"); + Statistic<> MaxSCCSize("MaxSCCSize", "Size of largest Strongly Connected Component"); +} + +template > +class TarjanSCC_iterator : public forward_iterator, ptrdiff_t> +{ + typedef SCC SccTy; + typedef forward_iterator super; + typedef typename super::reference reference; + typedef typename super::pointer pointer; + typedef typename GT::NodeType NodeType; + typedef typename GT::ChildIteratorType ChildItTy; + + // The visit counters used to detect when a complete SCC is on the stack. + // visitNum is the global counter. + // nodeVisitNumbers are per-node visit numbers, also used as DFS flags. + unsigned long visitNum; + std::map nodeVisitNumbers; + + // SCCNodeStack - Stack holding nodes of the SCC. + std::stack SCCNodeStack; + + // CurrentSCC - The current SCC, retrieved using operator*(). + SccTy CurrentSCC; + + // VisitStack - Used to maintain the ordering. Top = current block + // First element is basic block pointer, second is the 'next child' to visit + std::stack > VisitStack; + + // MinVistNumStack - Stack holding the "min" values for each node in the DFS. + // This is used to track the minimum uplink values for all children of + // the corresponding node on the VisitStack. + std::stack MinVisitNumStack; + + // A single "visit" within the non-recursive DFS traversal. + void DFSVisitOne(NodeType* N) { + ++visitNum; // Global counter for the visit order + nodeVisitNumbers[N] = visitNum; + SCCNodeStack.push(N); + MinVisitNumStack.push(visitNum); + VisitStack.push(make_pair(N, GT::child_begin(N))); + DEBUG(std::cerr << "TarjanSCC: Node " << N << + " : visitNum = " << visitNum << "\n"); + } + + // The stack-based DFS traversal; defined below. + void DFSVisitChildren() { + assert(!VisitStack.empty()); + while (VisitStack.top().second != GT::child_end(VisitStack.top().first)) + { // TOS has at least one more child so continue DFS + NodeType *childN = *VisitStack.top().second++; + if (nodeVisitNumbers.find(childN) == nodeVisitNumbers.end()) + { // this node has never been seen + DFSVisitOne(childN); + } + else + { + unsigned long childNum = nodeVisitNumbers[childN]; + if (MinVisitNumStack.top() > childNum) + MinVisitNumStack.top() = childNum; + } + } + } + + // Compute the next SCC using the DFS traversal. + void GetNextSCC() { + assert(VisitStack.size() == MinVisitNumStack.size()); + CurrentSCC.clear(); // Prepare to compute the next SCC + while (! VisitStack.empty()) + { + DFSVisitChildren(); + + assert(VisitStack.top().second==GT::child_end(VisitStack.top().first)); + NodeType* visitingN = VisitStack.top().first; + unsigned long minVisitNum = MinVisitNumStack.top(); + VisitStack.pop(); + MinVisitNumStack.pop(); + if (! MinVisitNumStack.empty() && MinVisitNumStack.top() > minVisitNum) + MinVisitNumStack.top() = minVisitNum; + + DEBUG(std::cerr << "TarjanSCC: Popped node " << visitingN << + " : minVisitNum = " << minVisitNum << "; Node visit num = " << + nodeVisitNumbers[visitingN] << "\n"); + + if (minVisitNum == nodeVisitNumbers[visitingN]) + { // A full SCC is on the SCCNodeStack! It includes all nodes below + // visitingN on the stack. Copy those nodes to CurrentSCC, + // reset their minVisit values, and return (this suspends + // the DFS traversal till the next ++). + do { + CurrentSCC.push_back(SCCNodeStack.top()); + SCCNodeStack.pop(); + nodeVisitNumbers[CurrentSCC.back()] = MAXLONG; + } while (CurrentSCC.back() != visitingN); + + ++NumSCCs; + if (CurrentSCC.size() > MaxSCCSize) MaxSCCSize = CurrentSCC.size(); + + return; + } + } + } + + inline TarjanSCC_iterator(NodeType *entryN) : visitNum(0) { + DFSVisitOne(entryN); + GetNextSCC(); + } + inline TarjanSCC_iterator() { /* End is when DFS stack is empty */ } + +public: + typedef TarjanSCC_iterator _Self; + + // Provide static "constructors"... + static inline _Self begin(GraphT& G) { return _Self(GT::getEntryNode(G)); } + static inline _Self end (GraphT& G) { return _Self(); } + + // Direct loop termination test (I.fini() is more efficient than I == end()) + inline bool fini() const { + return VisitStack.empty(); + } + + inline bool operator==(const _Self& x) const { + return VisitStack == x.VisitStack; + } + inline bool operator!=(const _Self& x) const { return !operator==(x); } + + // Iterator traversal: forward iteration only + inline _Self& operator++() { // Preincrement + GetNextSCC(); + return *this; + } + inline _Self operator++(int) { // Postincrement + _Self tmp = *this; ++*this; return tmp; + } + + // Retrieve a pointer to the current SCC. Returns NULL when done. + inline const SccTy* operator*() const { + assert(!CurrentSCC.empty() || fini()); + return CurrentSCC.empty()? NULL : &CurrentSCC; + } + inline SccTy* operator*() { + assert(!CurrentSCC.empty() || fini()); + return CurrentSCC.empty()? NULL : &CurrentSCC; + } +}; + + +// Global constructor for the Tarjan SCC iterator. Use *I == NULL or I.fini() +// to test termination efficiently, instead of I == the "end" iterator. +template +TarjanSCC_iterator tarj_begin(T G) +{ + return TarjanSCC_iterator::begin(G); +} + + +//===----------------------------------------------------------------------===// + +#endif diff --git a/include/llvm/ADT/SCCIterator.h b/include/llvm/ADT/SCCIterator.h new file mode 100644 index 0000000000..e737f7c990 --- /dev/null +++ b/include/llvm/ADT/SCCIterator.h @@ -0,0 +1,221 @@ +//===-- Support/TarjanSCCIterator.h -Generic Tarjan SCC iterator -*- C++ -*--=// +// +// This builds on the Support/GraphTraits.h file to find the strongly +// connected components (SCCs) of a graph in O(N+E) time using +// Tarjan's DFS algorithm. +// +// The SCC iterator has the important property that if a node in SCC S1 +// has an edge to a node in SCC S2, then it visits S1 *after* S2. +// +// To visit S1 *before* S2, use the TarjanSCCIterator on the Inverse graph. +// (NOTE: This requires some simple wrappers and is not supported yet.) +//===----------------------------------------------------------------------===// + +#ifndef LLVM_SUPPORT_TARJANSCC_ITERATOR_H +#define LLVM_SUPPORT_TARJANSCC_ITERATOR_H + +#include "Support/GraphTraits.h" +#include +#include +#include +#include +#include + + +//-------------------------------------------------------------------------- +// class SCC : A simple representation of an SCC in a generic Graph. +//-------------------------------------------------------------------------- + +template > +struct SCC: public std::vector { + + typedef typename GT::NodeType NodeType; + typedef typename GT::ChildIteratorType ChildItTy; + + typedef std::vector super; + typedef typename super::iterator iterator; + typedef typename super::const_iterator const_iterator; + typedef typename super::reverse_iterator reverse_iterator; + typedef typename super::const_reverse_iterator const_reverse_iterator; + + // HasLoop() -- Test if this SCC has a loop. If it has more than one + // node, this is trivially true. If not, it may still contain a loop + // if the node has an edge back to itself. + bool HasLoop() const { + if (size() > 1) return true; + NodeType* N = front(); + for (ChildItTy CI=GT::child_begin(N), CE=GT::child_end(N); CI != CE; ++CI) + if (*CI == N) + return true; + return false; + } +}; + +//-------------------------------------------------------------------------- +// class TarjanSCC_iterator: Enumerate the SCCs of a directed graph, in +// reverse topological order of the SCC DAG. +//-------------------------------------------------------------------------- + +const unsigned long MAXLONG = (1 << (8 * sizeof(unsigned long) - 1)); + +namespace { + Statistic<> NumSCCs("NumSCCs", "Number of Strongly Connected Components"); + Statistic<> MaxSCCSize("MaxSCCSize", "Size of largest Strongly Connected Component"); +} + +template > +class TarjanSCC_iterator : public forward_iterator, ptrdiff_t> +{ + typedef SCC SccTy; + typedef forward_iterator super; + typedef typename super::reference reference; + typedef typename super::pointer pointer; + typedef typename GT::NodeType NodeType; + typedef typename GT::ChildIteratorType ChildItTy; + + // The visit counters used to detect when a complete SCC is on the stack. + // visitNum is the global counter. + // nodeVisitNumbers are per-node visit numbers, also used as DFS flags. + unsigned long visitNum; + std::map nodeVisitNumbers; + + // SCCNodeStack - Stack holding nodes of the SCC. + std::stack SCCNodeStack; + + // CurrentSCC - The current SCC, retrieved using operator*(). + SccTy CurrentSCC; + + // VisitStack - Used to maintain the ordering. Top = current block + // First element is basic block pointer, second is the 'next child' to visit + std::stack > VisitStack; + + // MinVistNumStack - Stack holding the "min" values for each node in the DFS. + // This is used to track the minimum uplink values for all children of + // the corresponding node on the VisitStack. + std::stack MinVisitNumStack; + + // A single "visit" within the non-recursive DFS traversal. + void DFSVisitOne(NodeType* N) { + ++visitNum; // Global counter for the visit order + nodeVisitNumbers[N] = visitNum; + SCCNodeStack.push(N); + MinVisitNumStack.push(visitNum); + VisitStack.push(make_pair(N, GT::child_begin(N))); + DEBUG(std::cerr << "TarjanSCC: Node " << N << + " : visitNum = " << visitNum << "\n"); + } + + // The stack-based DFS traversal; defined below. + void DFSVisitChildren() { + assert(!VisitStack.empty()); + while (VisitStack.top().second != GT::child_end(VisitStack.top().first)) + { // TOS has at least one more child so continue DFS + NodeType *childN = *VisitStack.top().second++; + if (nodeVisitNumbers.find(childN) == nodeVisitNumbers.end()) + { // this node has never been seen + DFSVisitOne(childN); + } + else + { + unsigned long childNum = nodeVisitNumbers[childN]; + if (MinVisitNumStack.top() > childNum) + MinVisitNumStack.top() = childNum; + } + } + } + + // Compute the next SCC using the DFS traversal. + void GetNextSCC() { + assert(VisitStack.size() == MinVisitNumStack.size()); + CurrentSCC.clear(); // Prepare to compute the next SCC + while (! VisitStack.empty()) + { + DFSVisitChildren(); + + assert(VisitStack.top().second==GT::child_end(VisitStack.top().first)); + NodeType* visitingN = VisitStack.top().first; + unsigned long minVisitNum = MinVisitNumStack.top(); + VisitStack.pop(); + MinVisitNumStack.pop(); + if (! MinVisitNumStack.empty() && MinVisitNumStack.top() > minVisitNum) + MinVisitNumStack.top() = minVisitNum; + + DEBUG(std::cerr << "TarjanSCC: Popped node " << visitingN << + " : minVisitNum = " << minVisitNum << "; Node visit num = " << + nodeVisitNumbers[visitingN] << "\n"); + + if (minVisitNum == nodeVisitNumbers[visitingN]) + { // A full SCC is on the SCCNodeStack! It includes all nodes below + // visitingN on the stack. Copy those nodes to CurrentSCC, + // reset their minVisit values, and return (this suspends + // the DFS traversal till the next ++). + do { + CurrentSCC.push_back(SCCNodeStack.top()); + SCCNodeStack.pop(); + nodeVisitNumbers[CurrentSCC.back()] = MAXLONG; + } while (CurrentSCC.back() != visitingN); + + ++NumSCCs; + if (CurrentSCC.size() > MaxSCCSize) MaxSCCSize = CurrentSCC.size(); + + return; + } + } + } + + inline TarjanSCC_iterator(NodeType *entryN) : visitNum(0) { + DFSVisitOne(entryN); + GetNextSCC(); + } + inline TarjanSCC_iterator() { /* End is when DFS stack is empty */ } + +public: + typedef TarjanSCC_iterator _Self; + + // Provide static "constructors"... + static inline _Self begin(GraphT& G) { return _Self(GT::getEntryNode(G)); } + static inline _Self end (GraphT& G) { return _Self(); } + + // Direct loop termination test (I.fini() is more efficient than I == end()) + inline bool fini() const { + return VisitStack.empty(); + } + + inline bool operator==(const _Self& x) const { + return VisitStack == x.VisitStack; + } + inline bool operator!=(const _Self& x) const { return !operator==(x); } + + // Iterator traversal: forward iteration only + inline _Self& operator++() { // Preincrement + GetNextSCC(); + return *this; + } + inline _Self operator++(int) { // Postincrement + _Self tmp = *this; ++*this; return tmp; + } + + // Retrieve a pointer to the current SCC. Returns NULL when done. + inline const SccTy* operator*() const { + assert(!CurrentSCC.empty() || fini()); + return CurrentSCC.empty()? NULL : &CurrentSCC; + } + inline SccTy* operator*() { + assert(!CurrentSCC.empty() || fini()); + return CurrentSCC.empty()? NULL : &CurrentSCC; + } +}; + + +// Global constructor for the Tarjan SCC iterator. Use *I == NULL or I.fini() +// to test termination efficiently, instead of I == the "end" iterator. +template +TarjanSCC_iterator tarj_begin(T G) +{ + return TarjanSCC_iterator::begin(G); +} + + +//===----------------------------------------------------------------------===// + +#endif -- cgit v1.2.3