1 //===-- Support/FoldingSet.cpp - Uniquing Hash Set --------------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // 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"
23 //===----------------------------------------------------------------------===//
24 // FoldingSetNodeID Implementation
26 /// Add* - Add various data types to Bit data.
28 void FoldingSetNodeID::AddPointer(const void *Ptr) {
29 // Note: this adds pointers to the hash using sizes and endianness that
30 // depend on the host. It doesn't matter however, because hashing on
31 // pointer values in inherently unstable. Nothing should depend on the
32 // ordering of nodes in the folding set.
33 intptr_t PtrI = (intptr_t)Ptr;
34 Bits.push_back(unsigned(PtrI));
35 if (sizeof(intptr_t) > sizeof(unsigned))
36 Bits.push_back(unsigned(uint64_t(PtrI) >> 32));
38 void FoldingSetNodeID::AddInteger(signed I) {
41 void FoldingSetNodeID::AddInteger(unsigned I) {
44 void FoldingSetNodeID::AddInteger(int64_t I) {
45 AddInteger((uint64_t)I);
47 void FoldingSetNodeID::AddInteger(uint64_t I) {
48 Bits.push_back(unsigned(I));
50 // If the integer is small, encode it just as 32-bits.
51 if ((uint64_t)(int)I != I)
52 Bits.push_back(unsigned(I >> 32));
54 void FoldingSetNodeID::AddFloat(float F) {
55 Bits.push_back(FloatToBits(F));
57 void FoldingSetNodeID::AddDouble(double D) {
58 AddInteger(DoubleToBits(D));
61 void FoldingSetNodeID::AddString(const char *String) {
62 unsigned Size = static_cast<unsigned>(strlen(String));
66 unsigned Units = Size / 4;
68 const unsigned *Base = (const unsigned *)String;
70 // If the string is aligned do a bulk transfer.
71 if (!((intptr_t)Base & 3)) {
72 Bits.append(Base, Base + Units);
73 Pos = (Units + 1) * 4;
75 // Otherwise do it the hard way.
76 for ( Pos += 4; Pos <= Size; Pos += 4) {
77 unsigned V = ((unsigned char)String[Pos - 4] << 24) |
78 ((unsigned char)String[Pos - 3] << 16) |
79 ((unsigned char)String[Pos - 2] << 8) |
80 (unsigned char)String[Pos - 1];
85 // With the leftover bits.
87 // Pos will have overshot size by 4 - #bytes left over.
89 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru.
90 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru.
91 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
92 default: return; // Nothing left.
98 void FoldingSetNodeID::AddString(const std::string &String) {
99 unsigned Size = static_cast<unsigned>(String.size());
100 Bits.push_back(Size);
103 unsigned Units = Size / 4;
105 const unsigned *Base = (const unsigned *)String.data();
107 // If the string is aligned do a bulk transfer.
108 if (!((intptr_t)Base & 3)) {
109 Bits.append(Base, Base + Units);
110 Pos = (Units + 1) * 4;
112 // Otherwise do it the hard way.
113 for ( Pos += 4; Pos <= Size; Pos += 4) {
114 unsigned V = ((unsigned char)String[Pos - 4] << 24) |
115 ((unsigned char)String[Pos - 3] << 16) |
116 ((unsigned char)String[Pos - 2] << 8) |
117 (unsigned char)String[Pos - 1];
122 // With the leftover bits.
124 // Pos will have overshot size by 4 - #bytes left over.
125 switch (Pos - Size) {
126 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru.
127 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru.
128 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
129 default: return; // Nothing left.
135 /// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to
136 /// lookup the node in the FoldingSetImpl.
137 unsigned FoldingSetNodeID::ComputeHash() const {
138 // This is adapted from SuperFastHash by Paul Hsieh.
139 unsigned Hash = static_cast<unsigned>(Bits.size());
140 for (const unsigned *BP = &Bits[0], *E = BP+Bits.size(); BP != E; ++BP) {
142 Hash += Data & 0xFFFF;
143 unsigned Tmp = ((Data >> 16) << 11) ^ Hash;
144 Hash = (Hash << 16) ^ Tmp;
148 // Force "avalanching" of final 127 bits.
158 /// operator== - Used to compare two nodes to each other.
160 bool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS)const{
161 if (Bits.size() != RHS.Bits.size()) return false;
162 return memcmp(&Bits[0], &RHS.Bits[0], Bits.size()*sizeof(Bits[0])) == 0;
166 //===----------------------------------------------------------------------===//
167 /// Helper functions for FoldingSetImpl.
169 /// GetNextPtr - In order to save space, each bucket is a
170 /// singly-linked-list. In order to make deletion more efficient, we make
171 /// the list circular, so we can delete a node without computing its hash.
172 /// The problem with this is that the start of the hash buckets are not
173 /// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null:
174 /// use GetBucketPtr when this happens.
175 static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr) {
176 // The low bit is set if this is the pointer back to the bucket.
177 if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1)
180 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr);
185 static void **GetBucketPtr(void *NextInBucketPtr) {
186 intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr);
187 assert((Ptr & 1) && "Not a bucket pointer");
188 return reinterpret_cast<void**>(Ptr & ~intptr_t(1));
191 /// GetBucketFor - Hash the specified node ID and return the hash bucket for
192 /// the specified ID.
193 static void **GetBucketFor(const FoldingSetNodeID &ID,
194 void **Buckets, unsigned NumBuckets) {
195 // NumBuckets is always a power of 2.
196 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1);
197 return Buckets + BucketNum;
200 //===----------------------------------------------------------------------===//
201 // FoldingSetImpl Implementation
203 FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) : NumNodes(0) {
204 assert(5 < Log2InitSize && Log2InitSize < 32 &&
205 "Initial hash table size out of range");
206 NumBuckets = 1 << Log2InitSize;
207 Buckets = new void*[NumBuckets+1];
208 memset(Buckets, 0, NumBuckets*sizeof(void*));
210 // Set the very last bucket to be a non-null "pointer".
211 Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
213 FoldingSetImpl::~FoldingSetImpl() {
217 /// GrowHashTable - Double the size of the hash table and rehash everything.
219 void FoldingSetImpl::GrowHashTable() {
220 void **OldBuckets = Buckets;
221 unsigned OldNumBuckets = NumBuckets;
224 // Reset the node count to zero: we're going to reinsert everything.
227 // Clear out new buckets.
228 Buckets = new void*[NumBuckets+1];
229 memset(Buckets, 0, NumBuckets*sizeof(void*));
231 // Set the very last bucket to be a non-null "pointer".
232 Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
234 // Walk the old buckets, rehashing nodes into their new place.
235 for (unsigned i = 0; i != OldNumBuckets; ++i) {
236 void *Probe = OldBuckets[i];
237 if (!Probe) continue;
238 while (Node *NodeInBucket = GetNextPtr(Probe)) {
239 // Figure out the next link, remove NodeInBucket from the old link.
240 Probe = NodeInBucket->getNextInBucket();
241 NodeInBucket->SetNextInBucket(0);
243 // Insert the node into the new bucket, after recomputing the hash.
245 GetNodeProfile(ID, NodeInBucket);
246 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets));
253 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists,
254 /// return it. If not, return the insertion token that will make insertion
257 *FoldingSetImpl::FindNodeOrInsertPos(const FoldingSetNodeID &ID,
260 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets);
261 void *Probe = *Bucket;
265 while (Node *NodeInBucket = GetNextPtr(Probe)) {
266 FoldingSetNodeID OtherID;
267 GetNodeProfile(OtherID, NodeInBucket);
271 Probe = NodeInBucket->getNextInBucket();
274 // Didn't find the node, return null with the bucket as the InsertPos.
279 /// InsertNode - Insert the specified node into the folding set, knowing that it
280 /// is not already in the map. InsertPos must be obtained from
281 /// FindNodeOrInsertPos.
282 void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) {
283 assert(N->getNextInBucket() == 0);
284 // Do we need to grow the hashtable?
285 if (NumNodes+1 > NumBuckets*2) {
288 GetNodeProfile(ID, N);
289 InsertPos = GetBucketFor(ID, Buckets, NumBuckets);
294 /// The insert position is actually a bucket pointer.
295 void **Bucket = static_cast<void**>(InsertPos);
297 void *Next = *Bucket;
299 // If this is the first insertion into this bucket, its next pointer will be
300 // null. Pretend as if it pointed to itself, setting the low bit to indicate
301 // that it is a pointer to the bucket.
303 Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1);
305 // Set the node's next pointer, and make the bucket point to the node.
306 N->SetNextInBucket(Next);
310 /// RemoveNode - Remove a node from the folding set, returning true if one was
311 /// removed or false if the node was not in the folding set.
312 bool FoldingSetImpl::RemoveNode(Node *N) {
313 // Because each bucket is a circular list, we don't need to compute N's hash
315 void *Ptr = N->getNextInBucket();
316 if (Ptr == 0) return false; // Not in folding set.
319 N->SetNextInBucket(0);
321 // Remember what N originally pointed to, either a bucket or another node.
322 void *NodeNextPtr = Ptr;
324 // Chase around the list until we find the node (or bucket) which points to N.
326 if (Node *NodeInBucket = GetNextPtr(Ptr)) {
328 Ptr = NodeInBucket->getNextInBucket();
330 // We found a node that points to N, change it to point to N's next node,
331 // removing N from the list.
333 NodeInBucket->SetNextInBucket(NodeNextPtr);
337 void **Bucket = GetBucketPtr(Ptr);
340 // If we found that the bucket points to N, update the bucket to point to
343 *Bucket = NodeNextPtr;
350 /// GetOrInsertNode - If there is an existing simple Node exactly
351 /// equal to the specified node, return it. Otherwise, insert 'N' and it
353 FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) {
355 GetNodeProfile(ID, N);
357 if (Node *E = FindNodeOrInsertPos(ID, IP))
363 //===----------------------------------------------------------------------===//
364 // FoldingSetIteratorImpl Implementation
366 FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) {
367 // Skip to the first non-null non-self-cycle bucket.
368 while (*Bucket != reinterpret_cast<void*>(-1) &&
369 (*Bucket == 0 || GetNextPtr(*Bucket) == 0))
372 NodePtr = static_cast<FoldingSetNode*>(*Bucket);
375 void FoldingSetIteratorImpl::advance() {
376 // If there is another link within this bucket, go to it.
377 void *Probe = NodePtr->getNextInBucket();
379 if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe))
380 NodePtr = NextNodeInBucket;
382 // Otherwise, this is the last link in this bucket.
383 void **Bucket = GetBucketPtr(Probe);
385 // Skip to the next non-null non-self-cycle bucket.
388 } while (*Bucket != reinterpret_cast<void*>(-1) &&
389 (*Bucket == 0 || GetNextPtr(*Bucket) == 0));
391 NodePtr = static_cast<FoldingSetNode*>(*Bucket);
395 //===----------------------------------------------------------------------===//
396 // FoldingSetBucketIteratorImpl Implementation
398 FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) {
399 Ptr = (*Bucket == 0 || GetNextPtr(*Bucket) == 0) ? (void*) Bucket : *Bucket;