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/ADT/APFloat.h"
19 #include "llvm/ADT/APInt.h"
20 #include "llvm/Support/MathExtras.h"
24 //===----------------------------------------------------------------------===//
25 // FoldingSetNodeID Implementation
27 /// Add* - Add various data types to Bit data.
29 void FoldingSetNodeID::AddPointer(const void *Ptr) {
30 // Note: this adds pointers to the hash using sizes and endianness that
31 // depend on the host. It doesn't matter however, because hashing on
32 // pointer values in inherently unstable. Nothing should depend on the
33 // ordering of nodes in the folding set.
34 intptr_t PtrI = (intptr_t)Ptr;
35 Bits.push_back(unsigned(PtrI));
36 if (sizeof(intptr_t) > sizeof(unsigned))
37 Bits.push_back(unsigned(uint64_t(PtrI) >> 32));
39 void FoldingSetNodeID::AddInteger(signed I) {
42 void FoldingSetNodeID::AddInteger(unsigned I) {
45 void FoldingSetNodeID::AddInteger(int64_t I) {
46 AddInteger((uint64_t)I);
48 void FoldingSetNodeID::AddInteger(uint64_t I) {
49 Bits.push_back(unsigned(I));
51 // If the integer is small, encode it just as 32-bits.
52 if ((uint64_t)(int)I != I)
53 Bits.push_back(unsigned(I >> 32));
55 void FoldingSetNodeID::AddFloat(float F) {
56 Bits.push_back(FloatToBits(F));
58 void FoldingSetNodeID::AddDouble(double D) {
59 AddInteger(DoubleToBits(D));
61 void FoldingSetNodeID::AddAPFloat(const APFloat& apf) {
62 APInt api = apf.convertToAPInt();
65 void FoldingSetNodeID::AddAPInt(const APInt& api) {
66 const uint64_t *p = api.getRawData();
67 for (unsigned i=0; i<api.getNumWords(); i++)
70 void FoldingSetNodeID::AddString(const std::string &String) {
71 unsigned Size = String.size();
75 unsigned Units = Size / 4;
77 const unsigned *Base = (const unsigned *)String.data();
79 // If the string is aligned do a bulk transfer.
80 if (!((intptr_t)Base & 3)) {
81 Bits.append(Base, Base + Units);
82 Pos = (Units + 1) * 4;
84 // Otherwise do it the hard way.
85 for ( Pos += 4; Pos <= Size; Pos += 4) {
86 unsigned V = ((unsigned char)String[Pos - 4] << 24) |
87 ((unsigned char)String[Pos - 3] << 16) |
88 ((unsigned char)String[Pos - 2] << 8) |
89 (unsigned char)String[Pos - 1];
94 // With the leftover bits.
96 // Pos will have overshot size by 4 - #bytes left over.
98 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru.
99 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru.
100 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
101 default: return; // Nothing left.
107 /// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to
108 /// lookup the node in the FoldingSetImpl.
109 unsigned FoldingSetNodeID::ComputeHash() const {
110 // This is adapted from SuperFastHash by Paul Hsieh.
111 unsigned Hash = Bits.size();
112 for (const unsigned *BP = &Bits[0], *E = BP+Bits.size(); BP != E; ++BP) {
114 Hash += Data & 0xFFFF;
115 unsigned Tmp = ((Data >> 16) << 11) ^ Hash;
116 Hash = (Hash << 16) ^ Tmp;
120 // Force "avalanching" of final 127 bits.
130 /// operator== - Used to compare two nodes to each other.
132 bool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS)const{
133 if (Bits.size() != RHS.Bits.size()) return false;
134 return memcmp(&Bits[0], &RHS.Bits[0], Bits.size()*sizeof(Bits[0])) == 0;
138 //===----------------------------------------------------------------------===//
139 /// Helper functions for FoldingSetImpl.
141 /// GetNextPtr - In order to save space, each bucket is a
142 /// singly-linked-list. In order to make deletion more efficient, we make
143 /// the list circular, so we can delete a node without computing its hash.
144 /// The problem with this is that the start of the hash buckets are not
145 /// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null:
146 /// use GetBucketPtr when this happens.
147 static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr) {
148 // The low bit is set if this is the pointer back to the bucket.
149 if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1)
152 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr);
157 static void **GetBucketPtr(void *NextInBucketPtr) {
158 intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr);
159 assert((Ptr & 1) && "Not a bucket pointer");
160 return reinterpret_cast<void**>(Ptr & ~intptr_t(1));
163 /// GetBucketFor - Hash the specified node ID and return the hash bucket for
164 /// the specified ID.
165 static void **GetBucketFor(const FoldingSetNodeID &ID,
166 void **Buckets, unsigned NumBuckets) {
167 // NumBuckets is always a power of 2.
168 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1);
169 return Buckets + BucketNum;
172 //===----------------------------------------------------------------------===//
173 // FoldingSetImpl Implementation
175 FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) : NumNodes(0) {
176 assert(5 < Log2InitSize && Log2InitSize < 32 &&
177 "Initial hash table size out of range");
178 NumBuckets = 1 << Log2InitSize;
179 Buckets = new void*[NumBuckets+1];
180 memset(Buckets, 0, NumBuckets*sizeof(void*));
182 // Set the very last bucket to be a non-null "pointer".
183 Buckets[NumBuckets] = reinterpret_cast<void*>(-2);
185 FoldingSetImpl::~FoldingSetImpl() {
189 /// GrowHashTable - Double the size of the hash table and rehash everything.
191 void FoldingSetImpl::GrowHashTable() {
192 void **OldBuckets = Buckets;
193 unsigned OldNumBuckets = NumBuckets;
196 // Reset the node count to zero: we're going to reinsert everything.
199 // Clear out new buckets.
200 Buckets = new void*[NumBuckets+1];
201 memset(Buckets, 0, NumBuckets*sizeof(void*));
203 // Set the very last bucket to be a non-null "pointer".
204 Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
206 // Walk the old buckets, rehashing nodes into their new place.
207 for (unsigned i = 0; i != OldNumBuckets; ++i) {
208 void *Probe = OldBuckets[i];
209 if (!Probe) continue;
210 while (Node *NodeInBucket = GetNextPtr(Probe)) {
211 // Figure out the next link, remove NodeInBucket from the old link.
212 Probe = NodeInBucket->getNextInBucket();
213 NodeInBucket->SetNextInBucket(0);
215 // Insert the node into the new bucket, after recomputing the hash.
217 GetNodeProfile(ID, NodeInBucket);
218 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets));
225 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists,
226 /// return it. If not, return the insertion token that will make insertion
229 *FoldingSetImpl::FindNodeOrInsertPos(const FoldingSetNodeID &ID,
232 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets);
233 void *Probe = *Bucket;
237 while (Node *NodeInBucket = GetNextPtr(Probe)) {
238 FoldingSetNodeID OtherID;
239 GetNodeProfile(OtherID, NodeInBucket);
243 Probe = NodeInBucket->getNextInBucket();
246 // Didn't find the node, return null with the bucket as the InsertPos.
251 /// InsertNode - Insert the specified node into the folding set, knowing that it
252 /// is not already in the map. InsertPos must be obtained from
253 /// FindNodeOrInsertPos.
254 void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) {
255 assert(N->getNextInBucket() == 0);
256 // Do we need to grow the hashtable?
257 if (NumNodes+1 > NumBuckets*2) {
260 GetNodeProfile(ID, N);
261 InsertPos = GetBucketFor(ID, Buckets, NumBuckets);
266 /// The insert position is actually a bucket pointer.
267 void **Bucket = static_cast<void**>(InsertPos);
269 void *Next = *Bucket;
271 // If this is the first insertion into this bucket, its next pointer will be
272 // null. Pretend as if it pointed to itself, setting the low bit to indicate
273 // that it is a pointer to the bucket.
275 Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1);
277 // Set the node's next pointer, and make the bucket point to the node.
278 N->SetNextInBucket(Next);
282 /// RemoveNode - Remove a node from the folding set, returning true if one was
283 /// removed or false if the node was not in the folding set.
284 bool FoldingSetImpl::RemoveNode(Node *N) {
285 // Because each bucket is a circular list, we don't need to compute N's hash
287 void *Ptr = N->getNextInBucket();
288 if (Ptr == 0) return false; // Not in folding set.
291 N->SetNextInBucket(0);
293 // Remember what N originally pointed to, either a bucket or another node.
294 void *NodeNextPtr = Ptr;
296 // Chase around the list until we find the node (or bucket) which points to N.
298 if (Node *NodeInBucket = GetNextPtr(Ptr)) {
300 Ptr = NodeInBucket->getNextInBucket();
302 // We found a node that points to N, change it to point to N's next node,
303 // removing N from the list.
305 NodeInBucket->SetNextInBucket(NodeNextPtr);
309 void **Bucket = GetBucketPtr(Ptr);
312 // If we found that the bucket points to N, update the bucket to point to
315 *Bucket = NodeNextPtr;
322 /// GetOrInsertNode - If there is an existing simple Node exactly
323 /// equal to the specified node, return it. Otherwise, insert 'N' and it
325 FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) {
327 GetNodeProfile(ID, N);
329 if (Node *E = FindNodeOrInsertPos(ID, IP))
335 //===----------------------------------------------------------------------===//
336 // FoldingSetIteratorImpl Implementation
338 FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) {
339 // Skip to the first non-null non-self-cycle bucket.
340 while (*Bucket == 0 || GetNextPtr(*Bucket) == 0)
343 NodePtr = static_cast<FoldingSetNode*>(*Bucket);
346 void FoldingSetIteratorImpl::advance() {
347 // If there is another link within this bucket, go to it.
348 void *Probe = NodePtr->getNextInBucket();
350 if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe))
351 NodePtr = NextNodeInBucket;
353 // Otherwise, this is the last link in this bucket.
354 void **Bucket = GetBucketPtr(Probe);
356 // Skip to the next non-null non-self-cycle bucket.
359 } while (*Bucket == 0 || GetNextPtr(*Bucket) == 0);
361 NodePtr = static_cast<FoldingSetNode*>(*Bucket);
365 //===----------------------------------------------------------------------===//
366 // FoldingSetBucketIteratorImpl Implementation
368 FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) {
369 Ptr = (*Bucket == 0 || GetNextPtr(*Bucket) == 0) ? (void*) Bucket : *Bucket;