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/ErrorHandling.h"
19 #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(long I) {
46 AddInteger((unsigned long)I);
48 void FoldingSetNodeID::AddInteger(unsigned long I) {
49 if (sizeof(long) == sizeof(int))
50 AddInteger(unsigned(I));
51 else if (sizeof(long) == sizeof(long long)) {
52 AddInteger((unsigned long long)I);
54 llvm_unreachable("unexpected sizeof(long)");
57 void FoldingSetNodeID::AddInteger(long long I) {
58 AddInteger((unsigned long long)I);
60 void FoldingSetNodeID::AddInteger(unsigned long long I) {
61 AddInteger(unsigned(I));
62 if ((uint64_t)(int)I != I)
63 Bits.push_back(unsigned(I >> 32));
66 void FoldingSetNodeID::AddString(StringRef 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 FoldingSetNodeID, used to
104 /// lookup the node in the FoldingSetImpl.
105 unsigned FoldingSetNodeID::ComputeHash() const {
106 // This is adapted from SuperFastHash by Paul Hsieh.
107 unsigned Hash = static_cast<unsigned>(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 FoldingSetNodeID::operator==(const FoldingSetNodeID &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);
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 FoldingSetNodeID &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) {
172 assert(5 < Log2InitSize && Log2InitSize < 32 &&
173 "Initial hash table size out of range");
174 NumBuckets = 1 << Log2InitSize;
175 Buckets = new void*[NumBuckets+1];
178 FoldingSetImpl::~FoldingSetImpl() {
181 void FoldingSetImpl::clear() {
182 // Set all but the last bucket to null pointers.
183 memset(Buckets, 0, NumBuckets*sizeof(void*));
185 // Set the very last bucket to be a non-null "pointer".
186 Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
188 // Reset the node count to zero.
192 /// GrowHashTable - Double the size of the hash table and rehash everything.
194 void FoldingSetImpl::GrowHashTable() {
195 void **OldBuckets = Buckets;
196 unsigned OldNumBuckets = NumBuckets;
199 // Clear out new buckets.
200 Buckets = new void*[NumBuckets+1];
203 // Walk the old buckets, rehashing nodes into their new place.
205 for (unsigned i = 0; i != OldNumBuckets; ++i) {
206 void *Probe = OldBuckets[i];
207 if (!Probe) continue;
208 while (Node *NodeInBucket = GetNextPtr(Probe)) {
209 // Figure out the next link, remove NodeInBucket from the old link.
210 Probe = NodeInBucket->getNextInBucket();
211 NodeInBucket->SetNextInBucket(0);
213 // Insert the node into the new bucket, after recomputing the hash.
214 GetNodeProfile(ID, NodeInBucket);
215 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets));
223 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists,
224 /// return it. If not, return the insertion token that will make insertion
227 *FoldingSetImpl::FindNodeOrInsertPos(const FoldingSetNodeID &ID,
230 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets);
231 void *Probe = *Bucket;
235 FoldingSetNodeID OtherID;
236 while (Node *NodeInBucket = GetNextPtr(Probe)) {
237 GetNodeProfile(OtherID, NodeInBucket);
241 Probe = NodeInBucket->getNextInBucket();
245 // Didn't find the node, return null with the bucket as the InsertPos.
250 /// InsertNode - Insert the specified node into the folding set, knowing that it
251 /// is not already in the map. InsertPos must be obtained from
252 /// FindNodeOrInsertPos.
253 void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) {
254 assert(N->getNextInBucket() == 0);
255 // Do we need to grow the hashtable?
256 if (NumNodes+1 > NumBuckets*2) {
259 GetNodeProfile(ID, N);
260 InsertPos = GetBucketFor(ID, Buckets, NumBuckets);
265 /// The insert position is actually a bucket pointer.
266 void **Bucket = static_cast<void**>(InsertPos);
268 void *Next = *Bucket;
270 // If this is the first insertion into this bucket, its next pointer will be
271 // null. Pretend as if it pointed to itself, setting the low bit to indicate
272 // that it is a pointer to the bucket.
274 Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1);
276 // Set the node's next pointer, and make the bucket point to the node.
277 N->SetNextInBucket(Next);
281 /// RemoveNode - Remove a node from the folding set, returning true if one was
282 /// removed or false if the node was not in the folding set.
283 bool FoldingSetImpl::RemoveNode(Node *N) {
284 // Because each bucket is a circular list, we don't need to compute N's hash
286 void *Ptr = N->getNextInBucket();
287 if (Ptr == 0) return false; // Not in folding set.
290 N->SetNextInBucket(0);
292 // Remember what N originally pointed to, either a bucket or another node.
293 void *NodeNextPtr = Ptr;
295 // Chase around the list until we find the node (or bucket) which points to N.
297 if (Node *NodeInBucket = GetNextPtr(Ptr)) {
299 Ptr = NodeInBucket->getNextInBucket();
301 // We found a node that points to N, change it to point to N's next node,
302 // removing N from the list.
304 NodeInBucket->SetNextInBucket(NodeNextPtr);
308 void **Bucket = GetBucketPtr(Ptr);
311 // If we found that the bucket points to N, update the bucket to point to
314 *Bucket = NodeNextPtr;
321 /// GetOrInsertNode - If there is an existing simple Node exactly
322 /// equal to the specified node, return it. Otherwise, insert 'N' and it
324 FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) {
326 GetNodeProfile(ID, N);
328 if (Node *E = FindNodeOrInsertPos(ID, IP))
334 //===----------------------------------------------------------------------===//
335 // FoldingSetIteratorImpl Implementation
337 FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) {
338 // Skip to the first non-null non-self-cycle bucket.
339 while (*Bucket != reinterpret_cast<void*>(-1) &&
340 (*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 != reinterpret_cast<void*>(-1) &&
360 (*Bucket == 0 || GetNextPtr(*Bucket) == 0));
362 NodePtr = static_cast<FoldingSetNode*>(*Bucket);
366 //===----------------------------------------------------------------------===//
367 // FoldingSetBucketIteratorImpl Implementation
369 FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) {
370 Ptr = (*Bucket == 0 || GetNextPtr(*Bucket) == 0) ? (void*) Bucket : *Bucket;