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/ADT/APFloat.h"
19 #include "llvm/Support/MathExtras.h"
23 //===----------------------------------------------------------------------===//
24 // FoldingSetImpl::NodeID Implementation
26 /// Add* - Add various data types to Bit data.
28 void FoldingSetImpl::NodeID::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 FoldingSetImpl::NodeID::AddInteger(signed I) {
41 void FoldingSetImpl::NodeID::AddInteger(unsigned I) {
44 void FoldingSetImpl::NodeID::AddInteger(int64_t I) {
45 AddInteger((uint64_t)I);
47 void FoldingSetImpl::NodeID::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 FoldingSetImpl::NodeID::AddFloat(float F) {
55 Bits.push_back(FloatToBits(F));
57 void FoldingSetImpl::NodeID::AddDouble(double D) {
58 AddInteger(DoubleToBits(D));
60 void FoldingSetImpl::NodeID::AddAPFloat(const APFloat& apf) {
61 APInt api = apf.convertToAPInt();
62 const uint64_t *p = api.getRawData();
63 for (unsigned i=0; i<api.getNumWords(); i++)
66 void FoldingSetImpl::NodeID::AddString(const std::string &String) {
67 unsigned Size = String.size();
71 unsigned Units = Size / 4;
73 const unsigned *Base = (const unsigned *)String.data();
75 // If the string is aligned do a bulk transfer.
76 if (!((intptr_t)Base & 3)) {
77 Bits.append(Base, Base + Units);
78 Pos = (Units + 1) * 4;
80 // Otherwise do it the hard way.
81 for ( Pos += 4; Pos <= Size; Pos += 4) {
82 unsigned V = ((unsigned char)String[Pos - 4] << 24) |
83 ((unsigned char)String[Pos - 3] << 16) |
84 ((unsigned char)String[Pos - 2] << 8) |
85 (unsigned char)String[Pos - 1];
90 // With the leftover bits.
92 // Pos will have overshot size by 4 - #bytes left over.
94 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru.
95 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru.
96 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
97 default: return; // Nothing left.
103 /// ComputeHash - Compute a strong hash value for this NodeID, used to
104 /// lookup the node in the FoldingSetImpl.
105 unsigned FoldingSetImpl::NodeID::ComputeHash() const {
106 // This is adapted from SuperFastHash by Paul Hsieh.
107 unsigned Hash = Bits.size();
108 for (const unsigned *BP = &Bits[0], *E = BP+Bits.size(); BP != E; ++BP) {
110 Hash += Data & 0xFFFF;
111 unsigned Tmp = ((Data >> 16) << 11) ^ Hash;
112 Hash = (Hash << 16) ^ Tmp;
116 // Force "avalanching" of final 127 bits.
126 /// operator== - Used to compare two nodes to each other.
128 bool FoldingSetImpl::NodeID::operator==(const FoldingSetImpl::NodeID &RHS)const{
129 if (Bits.size() != RHS.Bits.size()) return false;
130 return memcmp(&Bits[0], &RHS.Bits[0], Bits.size()*sizeof(Bits[0])) == 0;
134 //===----------------------------------------------------------------------===//
135 /// Helper functions for FoldingSetImpl.
137 /// GetNextPtr - In order to save space, each bucket is a
138 /// singly-linked-list. In order to make deletion more efficient, we make
139 /// the list circular, so we can delete a node without computing its hash.
140 /// The problem with this is that the start of the hash buckets are not
141 /// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null:
142 /// use GetBucketPtr when this happens.
143 static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr) {
144 // The low bit is set if this is the pointer back to the bucket.
145 if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1)
148 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr);
151 /// GetBucketPtr - Provides a casting of a bucket pointer for isNode
153 static void **GetBucketPtr(void *NextInBucketPtr) {
154 intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr);
155 assert((Ptr & 1) && "Not a bucket pointer");
156 return reinterpret_cast<void**>(Ptr & ~intptr_t(1));
159 /// GetBucketFor - Hash the specified node ID and return the hash bucket for
160 /// the specified ID.
161 static void **GetBucketFor(const FoldingSetImpl::NodeID &ID,
162 void **Buckets, unsigned NumBuckets) {
163 // NumBuckets is always a power of 2.
164 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1);
165 return Buckets + BucketNum;
168 //===----------------------------------------------------------------------===//
169 // FoldingSetImpl Implementation
171 FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) : NumNodes(0) {
172 assert(5 < Log2InitSize && Log2InitSize < 32 &&
173 "Initial hash table size out of range");
174 NumBuckets = 1 << Log2InitSize;
175 Buckets = new void*[NumBuckets+1];
176 memset(Buckets, 0, NumBuckets*sizeof(void*));
178 // Set the very last bucket to be a non-null "pointer".
179 Buckets[NumBuckets] = reinterpret_cast<void*>(-2);
181 FoldingSetImpl::~FoldingSetImpl() {
185 /// GrowHashTable - Double the size of the hash table and rehash everything.
187 void FoldingSetImpl::GrowHashTable() {
188 void **OldBuckets = Buckets;
189 unsigned OldNumBuckets = NumBuckets;
192 // Reset the node count to zero: we're going to reinsert everything.
195 // Clear out new buckets.
196 Buckets = new void*[NumBuckets+1];
197 memset(Buckets, 0, NumBuckets*sizeof(void*));
199 // Set the very last bucket to be a non-null "pointer".
200 Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
202 // Walk the old buckets, rehashing nodes into their new place.
203 for (unsigned i = 0; i != OldNumBuckets; ++i) {
204 void *Probe = OldBuckets[i];
205 if (!Probe) continue;
206 while (Node *NodeInBucket = GetNextPtr(Probe)) {
207 // Figure out the next link, remove NodeInBucket from the old link.
208 Probe = NodeInBucket->getNextInBucket();
209 NodeInBucket->SetNextInBucket(0);
211 // Insert the node into the new bucket, after recomputing the hash.
213 GetNodeProfile(ID, NodeInBucket);
214 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets));
221 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists,
222 /// return it. If not, return the insertion token that will make insertion
224 FoldingSetImpl::Node *FoldingSetImpl::FindNodeOrInsertPos(const NodeID &ID,
226 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets);
227 void *Probe = *Bucket;
231 while (Node *NodeInBucket = GetNextPtr(Probe)) {
233 GetNodeProfile(OtherID, NodeInBucket);
237 Probe = NodeInBucket->getNextInBucket();
240 // Didn't find the node, return null with the bucket as the InsertPos.
245 /// InsertNode - Insert the specified node into the folding set, knowing that it
246 /// is not already in the map. InsertPos must be obtained from
247 /// FindNodeOrInsertPos.
248 void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) {
249 assert(N->getNextInBucket() == 0);
250 // Do we need to grow the hashtable?
251 if (NumNodes+1 > NumBuckets*2) {
254 GetNodeProfile(ID, N);
255 InsertPos = GetBucketFor(ID, Buckets, NumBuckets);
260 /// The insert position is actually a bucket pointer.
261 void **Bucket = static_cast<void**>(InsertPos);
263 void *Next = *Bucket;
265 // If this is the first insertion into this bucket, its next pointer will be
266 // null. Pretend as if it pointed to itself, setting the low bit to indicate
267 // that it is a pointer to the bucket.
269 Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1);
271 // Set the node's next pointer, and make the bucket point to the node.
272 N->SetNextInBucket(Next);
276 /// RemoveNode - Remove a node from the folding set, returning true if one was
277 /// removed or false if the node was not in the folding set.
278 bool FoldingSetImpl::RemoveNode(Node *N) {
279 // Because each bucket is a circular list, we don't need to compute N's hash
281 void *Ptr = N->getNextInBucket();
282 if (Ptr == 0) return false; // Not in folding set.
285 N->SetNextInBucket(0);
287 // Remember what N originally pointed to, either a bucket or another node.
288 void *NodeNextPtr = Ptr;
290 // Chase around the list until we find the node (or bucket) which points to N.
292 if (Node *NodeInBucket = GetNextPtr(Ptr)) {
294 Ptr = NodeInBucket->getNextInBucket();
296 // We found a node that points to N, change it to point to N's next node,
297 // removing N from the list.
299 NodeInBucket->SetNextInBucket(NodeNextPtr);
303 void **Bucket = GetBucketPtr(Ptr);
306 // If we found that the bucket points to N, update the bucket to point to
309 *Bucket = NodeNextPtr;
316 /// GetOrInsertNode - If there is an existing simple Node exactly
317 /// equal to the specified node, return it. Otherwise, insert 'N' and it
319 FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) {
321 GetNodeProfile(ID, N);
323 if (Node *E = FindNodeOrInsertPos(ID, IP))
329 //===----------------------------------------------------------------------===//
330 // FoldingSetIteratorImpl Implementation
332 FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) {
333 // Skip to the first non-null non-self-cycle bucket.
334 while (*Bucket == 0 || GetNextPtr(*Bucket) == 0)
337 NodePtr = static_cast<FoldingSetNode*>(*Bucket);
340 void FoldingSetIteratorImpl::advance() {
341 // If there is another link within this bucket, go to it.
342 void *Probe = NodePtr->getNextInBucket();
344 if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe))
345 NodePtr = NextNodeInBucket;
347 // Otherwise, this is the last link in this bucket.
348 void **Bucket = GetBucketPtr(Probe);
350 // Skip to the next non-null non-self-cycle bucket.
353 } while (*Bucket == 0 || GetNextPtr(*Bucket) == 0);
355 NodePtr = static_cast<FoldingSetNode*>(*Bucket);