1 //===-- Support/FoldingSet.cpp - Uniquing Hash Set --------------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by James M. Laskey and is distributed under
6 // the University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements a hash set that can be used to remove duplication of
11 // nodes in a graph. This code was originally created by Chris Lattner for use
12 // with SelectionDAGCSEMap, but was isolated to provide use across the llvm code
15 //===----------------------------------------------------------------------===//
17 #include "llvm/ADT/FoldingSet.h"
18 #include "llvm/Support/MathExtras.h"
22 //===----------------------------------------------------------------------===//
23 // FoldingSetImpl::NodeID Implementation
25 /// Add* - Add various data types to Bit data.
27 void FoldingSetImpl::NodeID::AddPointer(const void *Ptr) {
28 // Note: this adds pointers to the hash using sizes and endianness that
29 // depend on the host. It doesn't matter however, because hashing on
30 // pointer values in inherently unstable. Nothing should depend on the
31 // ordering of nodes in the folding set.
32 intptr_t PtrI = (intptr_t)Ptr;
33 Bits.push_back(unsigned(PtrI));
34 if (sizeof(intptr_t) > sizeof(unsigned))
35 Bits.push_back(unsigned(uint64_t(PtrI) >> 32));
37 void FoldingSetImpl::NodeID::AddInteger(signed I) {
40 void FoldingSetImpl::NodeID::AddInteger(unsigned I) {
43 void FoldingSetImpl::NodeID::AddInteger(uint64_t I) {
44 Bits.push_back(unsigned(I));
45 Bits.push_back(unsigned(I >> 32));
47 void FoldingSetImpl::NodeID::AddFloat(float F) {
48 Bits.push_back(FloatToBits(F));
50 void FoldingSetImpl::NodeID::AddDouble(double D) {
51 AddInteger(DoubleToBits(D));
53 void FoldingSetImpl::NodeID::AddString(const std::string &String) {
54 unsigned Size = String.size();
58 unsigned Units = Size / 4;
60 const unsigned *Base = (const unsigned *)String.data();
62 // If the string is aligned do a bulk transfer.
63 if (!((intptr_t)Base & 3)) {
64 Bits.append(Base, Base + Units);
65 Pos = (Units + 1) * 4;
67 // Otherwise do it the hard way.
68 for ( Pos += 4; Pos <= Size; Pos += 4) {
69 unsigned V = ((unsigned char)String[Pos - 4] << 24) |
70 ((unsigned char)String[Pos - 3] << 16) |
71 ((unsigned char)String[Pos - 2] << 8) |
72 (unsigned char)String[Pos - 1];
77 // With the leftover bits.
79 // Pos will have overshot size by 4 - #bytes left over.
81 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru.
82 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru.
83 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
84 default: return; // Nothing left.
90 /// ComputeHash - Compute a strong hash value for this NodeID, used to
91 /// lookup the node in the FoldingSetImpl.
92 unsigned FoldingSetImpl::NodeID::ComputeHash() const {
93 // This is adapted from SuperFastHash by Paul Hsieh.
94 unsigned Hash = Bits.size();
95 for (const unsigned *BP = &Bits[0], *E = BP+Bits.size(); BP != E; ++BP) {
97 Hash += Data & 0xFFFF;
98 unsigned Tmp = ((Data >> 16) << 11) ^ Hash;
99 Hash = (Hash << 16) ^ Tmp;
103 // Force "avalanching" of final 127 bits.
113 /// operator== - Used to compare two nodes to each other.
115 bool FoldingSetImpl::NodeID::operator==(const FoldingSetImpl::NodeID &RHS)const{
116 if (Bits.size() != RHS.Bits.size()) return false;
117 return memcmp(&Bits[0], &RHS.Bits[0], Bits.size()*sizeof(Bits[0])) == 0;
121 //===----------------------------------------------------------------------===//
122 /// Helper functions for FoldingSetImpl.
124 /// GetNextPtr - In order to save space, each bucket is a
125 /// singly-linked-list. In order to make deletion more efficient, we make
126 /// the list circular, so we can delete a node without computing its hash.
127 /// The problem with this is that the start of the hash buckets are not
128 /// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null:
129 /// use GetBucketPtr when this happens.
130 static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr,
131 void **Buckets, unsigned NumBuckets) {
132 if (NextInBucketPtr >= Buckets && NextInBucketPtr < Buckets + NumBuckets)
134 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr);
137 /// GetBucketPtr - Provides a casting of a bucket pointer for isNode
139 static void **GetBucketPtr(void *NextInBucketPtr) {
140 return static_cast<void**>(NextInBucketPtr);
143 /// GetBucketFor - Hash the specified node ID and return the hash bucket for
144 /// the specified ID.
145 static void **GetBucketFor(const FoldingSetImpl::NodeID &ID,
146 void **Buckets, unsigned NumBuckets) {
147 // NumBuckets is always a power of 2.
148 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1);
149 return Buckets + BucketNum;
152 //===----------------------------------------------------------------------===//
153 // FoldingSetImpl Implementation
155 FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) : NumNodes(0) {
156 assert(5 < Log2InitSize && Log2InitSize < 32 &&
157 "Initial hash table size out of range");
158 NumBuckets = 1 << Log2InitSize;
159 Buckets = new void*[NumBuckets];
160 memset(Buckets, 0, NumBuckets*sizeof(void*));
162 FoldingSetImpl::~FoldingSetImpl() {
166 /// GrowHashTable - Double the size of the hash table and rehash everything.
168 void FoldingSetImpl::GrowHashTable() {
169 void **OldBuckets = Buckets;
170 unsigned OldNumBuckets = NumBuckets;
173 // Reset the node count to zero: we're going to reinsert everything.
176 // Clear out new buckets.
177 Buckets = new void*[NumBuckets];
178 memset(Buckets, 0, NumBuckets*sizeof(void*));
180 // Walk the old buckets, rehashing nodes into their new place.
181 for (unsigned i = 0; i != OldNumBuckets; ++i) {
182 void *Probe = OldBuckets[i];
183 if (!Probe) continue;
184 while (Node *NodeInBucket = GetNextPtr(Probe, OldBuckets, OldNumBuckets)) {
185 // Figure out the next link, remove NodeInBucket from the old link.
186 Probe = NodeInBucket->getNextInBucket();
187 NodeInBucket->SetNextInBucket(0);
189 // Insert the node into the new bucket, after recomputing the hash.
191 GetNodeProfile(ID, NodeInBucket);
192 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets));
199 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists,
200 /// return it. If not, return the insertion token that will make insertion
202 FoldingSetImpl::Node *FoldingSetImpl::FindNodeOrInsertPos(const NodeID &ID,
204 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets);
205 void *Probe = *Bucket;
209 while (Node *NodeInBucket = GetNextPtr(Probe, Buckets, NumBuckets)) {
211 GetNodeProfile(OtherID, NodeInBucket);
215 Probe = NodeInBucket->getNextInBucket();
218 // Didn't find the node, return null with the bucket as the InsertPos.
223 /// InsertNode - Insert the specified node into the folding set, knowing that it
224 /// is not already in the map. InsertPos must be obtained from
225 /// FindNodeOrInsertPos.
226 void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) {
227 // Do we need to grow the hashtable?
228 if (NumNodes+1 > NumBuckets*2) {
231 GetNodeProfile(ID, N);
232 InsertPos = GetBucketFor(ID, Buckets, NumBuckets);
237 /// The insert position is actually a bucket pointer.
238 void **Bucket = static_cast<void**>(InsertPos);
240 void *Next = *Bucket;
242 // If this is the first insertion into this bucket, its next pointer will be
243 // null. Pretend as if it pointed to itself.
247 // Set the node's next pointer, and make the bucket point to the node.
248 N->SetNextInBucket(Next);
252 /// RemoveNode - Remove a node from the folding set, returning true if one was
253 /// removed or false if the node was not in the folding set.
254 bool FoldingSetImpl::RemoveNode(Node *N) {
255 // Because each bucket is a circular list, we don't need to compute N's hash
256 // to remove it. Chase around the list until we find the node (or bucket)
257 // which points to N.
258 void *Ptr = N->getNextInBucket();
259 if (Ptr == 0) return false; // Not in folding set.
263 void *NodeNextPtr = Ptr;
264 N->SetNextInBucket(0);
266 if (Node *NodeInBucket = GetNextPtr(Ptr, Buckets, NumBuckets)) {
268 Ptr = NodeInBucket->getNextInBucket();
270 // We found a node that points to N, change it to point to N's next node,
271 // removing N from the list.
273 NodeInBucket->SetNextInBucket(NodeNextPtr);
277 void **Bucket = GetBucketPtr(Ptr);
280 // If we found that the bucket points to N, update the bucket to point to
283 *Bucket = NodeNextPtr;
290 /// GetOrInsertNode - If there is an existing simple Node exactly
291 /// equal to the specified node, return it. Otherwise, insert 'N' and it
293 FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) {
295 GetNodeProfile(ID, N);
297 if (Node *E = FindNodeOrInsertPos(ID, IP))