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(long I) {
45 AddInteger((unsigned long)I);
47 void FoldingSetNodeID::AddInteger(unsigned long I) {
48 if (sizeof(long) == sizeof(int))
49 AddInteger(unsigned(I));
50 else if (sizeof(long) == sizeof(long long)) {
51 AddInteger((unsigned long long)I);
53 assert(0 && "unexpected sizeof(long)");
56 void FoldingSetNodeID::AddInteger(long long I) {
57 AddInteger((unsigned long long)I);
59 void FoldingSetNodeID::AddInteger(unsigned long long I) {
60 AddInteger(unsigned(I));
61 if ((uint64_t)(int)I != I)
62 Bits.push_back(unsigned(I >> 32));
64 void FoldingSetNodeID::AddFloat(float F) {
65 Bits.push_back(FloatToBits(F));
67 void FoldingSetNodeID::AddDouble(double D) {
68 AddInteger(DoubleToBits(D));
71 void FoldingSetNodeID::AddString(const char *String) {
72 unsigned Size = static_cast<unsigned>(strlen(String));
76 unsigned Units = Size / 4;
78 const unsigned *Base = (const unsigned *)String;
80 // If the string is aligned do a bulk transfer.
81 if (!((intptr_t)Base & 3)) {
82 Bits.append(Base, Base + Units);
83 Pos = (Units + 1) * 4;
85 // Otherwise do it the hard way.
86 for ( Pos += 4; Pos <= Size; Pos += 4) {
87 unsigned V = ((unsigned char)String[Pos - 4] << 24) |
88 ((unsigned char)String[Pos - 3] << 16) |
89 ((unsigned char)String[Pos - 2] << 8) |
90 (unsigned char)String[Pos - 1];
95 // With the leftover bits.
97 // Pos will have overshot size by 4 - #bytes left over.
99 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru.
100 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru.
101 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
102 default: return; // Nothing left.
108 void FoldingSetNodeID::AddString(const std::string &String) {
109 unsigned Size = static_cast<unsigned>(String.size());
110 Bits.push_back(Size);
113 unsigned Units = Size / 4;
115 const unsigned *Base = (const unsigned *)String.data();
117 // If the string is aligned do a bulk transfer.
118 if (!((intptr_t)Base & 3)) {
119 Bits.append(Base, Base + Units);
120 Pos = (Units + 1) * 4;
122 // Otherwise do it the hard way.
123 for ( Pos += 4; Pos <= Size; Pos += 4) {
124 unsigned V = ((unsigned char)String[Pos - 4] << 24) |
125 ((unsigned char)String[Pos - 3] << 16) |
126 ((unsigned char)String[Pos - 2] << 8) |
127 (unsigned char)String[Pos - 1];
132 // With the leftover bits.
134 // Pos will have overshot size by 4 - #bytes left over.
135 switch (Pos - Size) {
136 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru.
137 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru.
138 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
139 default: return; // Nothing left.
145 /// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to
146 /// lookup the node in the FoldingSetImpl.
147 unsigned FoldingSetNodeID::ComputeHash() const {
148 // This is adapted from SuperFastHash by Paul Hsieh.
149 unsigned Hash = static_cast<unsigned>(Bits.size());
150 for (const unsigned *BP = &Bits[0], *E = BP+Bits.size(); BP != E; ++BP) {
152 Hash += Data & 0xFFFF;
153 unsigned Tmp = ((Data >> 16) << 11) ^ Hash;
154 Hash = (Hash << 16) ^ Tmp;
158 // Force "avalanching" of final 127 bits.
168 /// operator== - Used to compare two nodes to each other.
170 bool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS)const{
171 if (Bits.size() != RHS.Bits.size()) return false;
172 return memcmp(&Bits[0], &RHS.Bits[0], Bits.size()*sizeof(Bits[0])) == 0;
176 //===----------------------------------------------------------------------===//
177 /// Helper functions for FoldingSetImpl.
179 /// GetNextPtr - In order to save space, each bucket is a
180 /// singly-linked-list. In order to make deletion more efficient, we make
181 /// the list circular, so we can delete a node without computing its hash.
182 /// The problem with this is that the start of the hash buckets are not
183 /// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null:
184 /// use GetBucketPtr when this happens.
185 static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr) {
186 // The low bit is set if this is the pointer back to the bucket.
187 if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1)
190 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr);
195 static void **GetBucketPtr(void *NextInBucketPtr) {
196 intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr);
197 assert((Ptr & 1) && "Not a bucket pointer");
198 return reinterpret_cast<void**>(Ptr & ~intptr_t(1));
201 /// GetBucketFor - Hash the specified node ID and return the hash bucket for
202 /// the specified ID.
203 static void **GetBucketFor(const FoldingSetNodeID &ID,
204 void **Buckets, unsigned NumBuckets) {
205 // NumBuckets is always a power of 2.
206 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1);
207 return Buckets + BucketNum;
210 //===----------------------------------------------------------------------===//
211 // FoldingSetImpl Implementation
213 FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) {
214 assert(5 < Log2InitSize && Log2InitSize < 32 &&
215 "Initial hash table size out of range");
216 NumBuckets = 1 << Log2InitSize;
217 Buckets = new void*[NumBuckets+1];
220 FoldingSetImpl::~FoldingSetImpl() {
223 void FoldingSetImpl::clear() {
224 // Set all but the last bucket to null pointers.
225 memset(Buckets, 0, NumBuckets*sizeof(void*));
227 // Set the very last bucket to be a non-null "pointer".
228 Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
230 // Reset the node count to zero.
234 /// GrowHashTable - Double the size of the hash table and rehash everything.
236 void FoldingSetImpl::GrowHashTable() {
237 void **OldBuckets = Buckets;
238 unsigned OldNumBuckets = NumBuckets;
241 // Clear out new buckets.
242 Buckets = new void*[NumBuckets+1];
245 // Walk the old buckets, rehashing nodes into their new place.
247 for (unsigned i = 0; i != OldNumBuckets; ++i) {
248 void *Probe = OldBuckets[i];
249 if (!Probe) continue;
250 while (Node *NodeInBucket = GetNextPtr(Probe)) {
251 // Figure out the next link, remove NodeInBucket from the old link.
252 Probe = NodeInBucket->getNextInBucket();
253 NodeInBucket->SetNextInBucket(0);
255 // Insert the node into the new bucket, after recomputing the hash.
256 GetNodeProfile(ID, NodeInBucket);
257 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets));
265 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists,
266 /// return it. If not, return the insertion token that will make insertion
269 *FoldingSetImpl::FindNodeOrInsertPos(const FoldingSetNodeID &ID,
272 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets);
273 void *Probe = *Bucket;
277 FoldingSetNodeID OtherID;
278 while (Node *NodeInBucket = GetNextPtr(Probe)) {
279 GetNodeProfile(OtherID, NodeInBucket);
283 Probe = NodeInBucket->getNextInBucket();
287 // Didn't find the node, return null with the bucket as the InsertPos.
292 /// InsertNode - Insert the specified node into the folding set, knowing that it
293 /// is not already in the map. InsertPos must be obtained from
294 /// FindNodeOrInsertPos.
295 void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) {
296 assert(N->getNextInBucket() == 0);
297 // Do we need to grow the hashtable?
298 if (NumNodes+1 > NumBuckets*2) {
301 GetNodeProfile(ID, N);
302 InsertPos = GetBucketFor(ID, Buckets, NumBuckets);
307 /// The insert position is actually a bucket pointer.
308 void **Bucket = static_cast<void**>(InsertPos);
310 void *Next = *Bucket;
312 // If this is the first insertion into this bucket, its next pointer will be
313 // null. Pretend as if it pointed to itself, setting the low bit to indicate
314 // that it is a pointer to the bucket.
316 Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1);
318 // Set the node's next pointer, and make the bucket point to the node.
319 N->SetNextInBucket(Next);
323 /// RemoveNode - Remove a node from the folding set, returning true if one was
324 /// removed or false if the node was not in the folding set.
325 bool FoldingSetImpl::RemoveNode(Node *N) {
326 // Because each bucket is a circular list, we don't need to compute N's hash
328 void *Ptr = N->getNextInBucket();
329 if (Ptr == 0) return false; // Not in folding set.
332 N->SetNextInBucket(0);
334 // Remember what N originally pointed to, either a bucket or another node.
335 void *NodeNextPtr = Ptr;
337 // Chase around the list until we find the node (or bucket) which points to N.
339 if (Node *NodeInBucket = GetNextPtr(Ptr)) {
341 Ptr = NodeInBucket->getNextInBucket();
343 // We found a node that points to N, change it to point to N's next node,
344 // removing N from the list.
346 NodeInBucket->SetNextInBucket(NodeNextPtr);
350 void **Bucket = GetBucketPtr(Ptr);
353 // If we found that the bucket points to N, update the bucket to point to
356 *Bucket = NodeNextPtr;
363 /// GetOrInsertNode - If there is an existing simple Node exactly
364 /// equal to the specified node, return it. Otherwise, insert 'N' and it
366 FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) {
368 GetNodeProfile(ID, N);
370 if (Node *E = FindNodeOrInsertPos(ID, IP))
376 //===----------------------------------------------------------------------===//
377 // FoldingSetIteratorImpl Implementation
379 FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) {
380 // Skip to the first non-null non-self-cycle bucket.
381 while (*Bucket != reinterpret_cast<void*>(-1) &&
382 (*Bucket == 0 || GetNextPtr(*Bucket) == 0))
385 NodePtr = static_cast<FoldingSetNode*>(*Bucket);
388 void FoldingSetIteratorImpl::advance() {
389 // If there is another link within this bucket, go to it.
390 void *Probe = NodePtr->getNextInBucket();
392 if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe))
393 NodePtr = NextNodeInBucket;
395 // Otherwise, this is the last link in this bucket.
396 void **Bucket = GetBucketPtr(Probe);
398 // Skip to the next non-null non-self-cycle bucket.
401 } while (*Bucket != reinterpret_cast<void*>(-1) &&
402 (*Bucket == 0 || GetNextPtr(*Bucket) == 0));
404 NodePtr = static_cast<FoldingSetNode*>(*Bucket);
408 //===----------------------------------------------------------------------===//
409 // FoldingSetBucketIteratorImpl Implementation
411 FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) {
412 Ptr = (*Bucket == 0 || GetNextPtr(*Bucket) == 0) ? (void*) Bucket : *Bucket;