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"
22 //===----------------------------------------------------------------------===//
23 // FoldingSetNodeID Implementation
25 /// Add* - Add various data types to Bit data.
27 void FoldingSetNodeID::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 FoldingSetNodeID::AddInteger(signed I) {
40 void FoldingSetNodeID::AddInteger(unsigned I) {
43 void FoldingSetNodeID::AddInteger(int64_t I) {
44 AddInteger((uint64_t)I);
46 void FoldingSetNodeID::AddInteger(uint64_t I) {
47 Bits.push_back(unsigned(I));
49 // If the integer is small, encode it just as 32-bits.
50 if ((uint64_t)(int)I != I)
51 Bits.push_back(unsigned(I >> 32));
53 void FoldingSetNodeID::AddFloat(float F) {
54 Bits.push_back(FloatToBits(F));
56 void FoldingSetNodeID::AddDouble(double D) {
57 AddInteger(DoubleToBits(D));
59 void FoldingSetNodeID::AddString(const std::string &String) {
60 unsigned Size = String.size();
64 unsigned Units = Size / 4;
66 const unsigned *Base = (const unsigned *)String.data();
68 // If the string is aligned do a bulk transfer.
69 if (!((intptr_t)Base & 3)) {
70 Bits.append(Base, Base + Units);
71 Pos = (Units + 1) * 4;
73 // Otherwise do it the hard way.
74 for ( Pos += 4; Pos <= Size; Pos += 4) {
75 unsigned V = ((unsigned char)String[Pos - 4] << 24) |
76 ((unsigned char)String[Pos - 3] << 16) |
77 ((unsigned char)String[Pos - 2] << 8) |
78 (unsigned char)String[Pos - 1];
83 // With the leftover bits.
85 // Pos will have overshot size by 4 - #bytes left over.
87 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru.
88 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru.
89 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
90 default: return; // Nothing left.
96 /// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to
97 /// lookup the node in the FoldingSetImpl.
98 unsigned FoldingSetNodeID::ComputeHash() const {
99 // This is adapted from SuperFastHash by Paul Hsieh.
100 unsigned Hash = Bits.size();
101 for (const unsigned *BP = &Bits[0], *E = BP+Bits.size(); BP != E; ++BP) {
103 Hash += Data & 0xFFFF;
104 unsigned Tmp = ((Data >> 16) << 11) ^ Hash;
105 Hash = (Hash << 16) ^ Tmp;
109 // Force "avalanching" of final 127 bits.
119 /// operator== - Used to compare two nodes to each other.
121 bool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS)const{
122 if (Bits.size() != RHS.Bits.size()) return false;
123 return memcmp(&Bits[0], &RHS.Bits[0], Bits.size()*sizeof(Bits[0])) == 0;
127 //===----------------------------------------------------------------------===//
128 /// Helper functions for FoldingSetImpl.
130 /// GetNextPtr - In order to save space, each bucket is a
131 /// singly-linked-list. In order to make deletion more efficient, we make
132 /// the list circular, so we can delete a node without computing its hash.
133 /// The problem with this is that the start of the hash buckets are not
134 /// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null:
135 /// use GetBucketPtr when this happens.
136 static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr) {
137 // The low bit is set if this is the pointer back to the bucket.
138 if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1)
141 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr);
146 static void **GetBucketPtr(void *NextInBucketPtr) {
147 intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr);
148 assert((Ptr & 1) && "Not a bucket pointer");
149 return reinterpret_cast<void**>(Ptr & ~intptr_t(1));
152 /// GetBucketFor - Hash the specified node ID and return the hash bucket for
153 /// the specified ID.
154 static void **GetBucketFor(const FoldingSetNodeID &ID,
155 void **Buckets, unsigned NumBuckets) {
156 // NumBuckets is always a power of 2.
157 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1);
158 return Buckets + BucketNum;
161 //===----------------------------------------------------------------------===//
162 // FoldingSetImpl Implementation
164 FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) : NumNodes(0) {
165 assert(5 < Log2InitSize && Log2InitSize < 32 &&
166 "Initial hash table size out of range");
167 NumBuckets = 1 << Log2InitSize;
168 Buckets = new void*[NumBuckets+1];
169 memset(Buckets, 0, NumBuckets*sizeof(void*));
171 // Set the very last bucket to be a non-null "pointer".
172 Buckets[NumBuckets] = reinterpret_cast<void*>(-2);
174 FoldingSetImpl::~FoldingSetImpl() {
178 /// GrowHashTable - Double the size of the hash table and rehash everything.
180 void FoldingSetImpl::GrowHashTable() {
181 void **OldBuckets = Buckets;
182 unsigned OldNumBuckets = NumBuckets;
185 // Reset the node count to zero: we're going to reinsert everything.
188 // Clear out new buckets.
189 Buckets = new void*[NumBuckets+1];
190 memset(Buckets, 0, NumBuckets*sizeof(void*));
192 // Set the very last bucket to be a non-null "pointer".
193 Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
195 // Walk the old buckets, rehashing nodes into their new place.
196 for (unsigned i = 0; i != OldNumBuckets; ++i) {
197 void *Probe = OldBuckets[i];
198 if (!Probe) continue;
199 while (Node *NodeInBucket = GetNextPtr(Probe)) {
200 // Figure out the next link, remove NodeInBucket from the old link.
201 Probe = NodeInBucket->getNextInBucket();
202 NodeInBucket->SetNextInBucket(0);
204 // Insert the node into the new bucket, after recomputing the hash.
206 GetNodeProfile(ID, NodeInBucket);
207 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets));
214 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists,
215 /// return it. If not, return the insertion token that will make insertion
218 *FoldingSetImpl::FindNodeOrInsertPos(const FoldingSetNodeID &ID,
221 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets);
222 void *Probe = *Bucket;
226 while (Node *NodeInBucket = GetNextPtr(Probe)) {
227 FoldingSetNodeID OtherID;
228 GetNodeProfile(OtherID, NodeInBucket);
232 Probe = NodeInBucket->getNextInBucket();
235 // Didn't find the node, return null with the bucket as the InsertPos.
240 /// InsertNode - Insert the specified node into the folding set, knowing that it
241 /// is not already in the map. InsertPos must be obtained from
242 /// FindNodeOrInsertPos.
243 void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) {
244 assert(N->getNextInBucket() == 0);
245 // Do we need to grow the hashtable?
246 if (NumNodes+1 > NumBuckets*2) {
249 GetNodeProfile(ID, N);
250 InsertPos = GetBucketFor(ID, Buckets, NumBuckets);
255 /// The insert position is actually a bucket pointer.
256 void **Bucket = static_cast<void**>(InsertPos);
258 void *Next = *Bucket;
260 // If this is the first insertion into this bucket, its next pointer will be
261 // null. Pretend as if it pointed to itself, setting the low bit to indicate
262 // that it is a pointer to the bucket.
264 Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1);
266 // Set the node's next pointer, and make the bucket point to the node.
267 N->SetNextInBucket(Next);
271 /// RemoveNode - Remove a node from the folding set, returning true if one was
272 /// removed or false if the node was not in the folding set.
273 bool FoldingSetImpl::RemoveNode(Node *N) {
274 // Because each bucket is a circular list, we don't need to compute N's hash
276 void *Ptr = N->getNextInBucket();
277 if (Ptr == 0) return false; // Not in folding set.
280 N->SetNextInBucket(0);
282 // Remember what N originally pointed to, either a bucket or another node.
283 void *NodeNextPtr = Ptr;
285 // Chase around the list until we find the node (or bucket) which points to N.
287 if (Node *NodeInBucket = GetNextPtr(Ptr)) {
289 Ptr = NodeInBucket->getNextInBucket();
291 // We found a node that points to N, change it to point to N's next node,
292 // removing N from the list.
294 NodeInBucket->SetNextInBucket(NodeNextPtr);
298 void **Bucket = GetBucketPtr(Ptr);
301 // If we found that the bucket points to N, update the bucket to point to
304 *Bucket = NodeNextPtr;
311 /// GetOrInsertNode - If there is an existing simple Node exactly
312 /// equal to the specified node, return it. Otherwise, insert 'N' and it
314 FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) {
316 GetNodeProfile(ID, N);
318 if (Node *E = FindNodeOrInsertPos(ID, IP))
324 //===----------------------------------------------------------------------===//
325 // FoldingSetIteratorImpl Implementation
327 FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) {
328 // Skip to the first non-null non-self-cycle bucket.
329 while (*Bucket == 0 || GetNextPtr(*Bucket) == 0)
332 NodePtr = static_cast<FoldingSetNode*>(*Bucket);
335 void FoldingSetIteratorImpl::advance() {
336 // If there is another link within this bucket, go to it.
337 void *Probe = NodePtr->getNextInBucket();
339 if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe))
340 NodePtr = NextNodeInBucket;
342 // Otherwise, this is the last link in this bucket.
343 void **Bucket = GetBucketPtr(Probe);
345 // Skip to the next non-null non-self-cycle bucket.
348 } while (*Bucket == 0 || GetNextPtr(*Bucket) == 0);
350 NodePtr = static_cast<FoldingSetNode*>(*Bucket);
354 //===----------------------------------------------------------------------===//
355 // FoldingSetBucketIteratorImpl Implementation
357 FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) {
358 Ptr = (*Bucket == 0 || GetNextPtr(*Bucket) == 0) ? (void*) Bucket : *Bucket;