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
13 //===----------------------------------------------------------------------===//
15 #include "llvm/ADT/FoldingSet.h"
16 #include "llvm/ADT/Hashing.h"
17 #include "llvm/Support/Allocator.h"
18 #include "llvm/Support/ErrorHandling.h"
19 #include "llvm/Support/Host.h"
20 #include "llvm/Support/MathExtras.h"
25 //===----------------------------------------------------------------------===//
26 // FoldingSetNodeIDRef Implementation
28 /// ComputeHash - Compute a strong hash value for this FoldingSetNodeIDRef,
29 /// used to lookup the node in the FoldingSetBase.
30 unsigned FoldingSetNodeIDRef::ComputeHash() const {
31 return static_cast<unsigned>(hash_combine_range(Data, Data+Size));
34 bool FoldingSetNodeIDRef::operator==(FoldingSetNodeIDRef RHS) const {
35 if (Size != RHS.Size) return false;
36 return memcmp(Data, RHS.Data, Size*sizeof(*Data)) == 0;
39 /// Used to compare the "ordering" of two nodes as defined by the
40 /// profiled bits and their ordering defined by memcmp().
41 bool FoldingSetNodeIDRef::operator<(FoldingSetNodeIDRef RHS) const {
43 return Size < RHS.Size;
44 return memcmp(Data, RHS.Data, Size*sizeof(*Data)) < 0;
47 //===----------------------------------------------------------------------===//
48 // FoldingSetNodeID Implementation
50 /// Add* - Add various data types to Bit data.
52 void FoldingSetNodeID::AddPointer(const void *Ptr) {
53 // Note: this adds pointers to the hash using sizes and endianness that
54 // depend on the host. It doesn't matter, however, because hashing on
55 // pointer values is inherently unstable. Nothing should depend on the
56 // ordering of nodes in the folding set.
57 static_assert(sizeof(uintptr_t) <= sizeof(unsigned long long),
58 "unexpected pointer size");
59 AddInteger(reinterpret_cast<uintptr_t>(Ptr));
61 void FoldingSetNodeID::AddInteger(signed I) {
64 void FoldingSetNodeID::AddInteger(unsigned I) {
67 void FoldingSetNodeID::AddInteger(long I) {
68 AddInteger((unsigned long)I);
70 void FoldingSetNodeID::AddInteger(unsigned long I) {
71 if (sizeof(long) == sizeof(int))
72 AddInteger(unsigned(I));
73 else if (sizeof(long) == sizeof(long long)) {
74 AddInteger((unsigned long long)I);
76 llvm_unreachable("unexpected sizeof(long)");
79 void FoldingSetNodeID::AddInteger(long long I) {
80 AddInteger((unsigned long long)I);
82 void FoldingSetNodeID::AddInteger(unsigned long long I) {
83 AddInteger(unsigned(I));
84 AddInteger(unsigned(I >> 32));
87 void FoldingSetNodeID::AddString(StringRef String) {
88 unsigned Size = String.size();
92 unsigned Units = Size / 4;
94 const unsigned *Base = (const unsigned*) String.data();
96 // If the string is aligned do a bulk transfer.
97 if (!((intptr_t)Base & 3)) {
98 Bits.append(Base, Base + Units);
99 Pos = (Units + 1) * 4;
101 // Otherwise do it the hard way.
102 // To be compatible with above bulk transfer, we need to take endianness
104 static_assert(sys::IsBigEndianHost || sys::IsLittleEndianHost,
105 "Unexpected host endianness");
106 if (sys::IsBigEndianHost) {
107 for (Pos += 4; Pos <= Size; Pos += 4) {
108 unsigned V = ((unsigned char)String[Pos - 4] << 24) |
109 ((unsigned char)String[Pos - 3] << 16) |
110 ((unsigned char)String[Pos - 2] << 8) |
111 (unsigned char)String[Pos - 1];
114 } else { // Little-endian host
115 for (Pos += 4; Pos <= Size; Pos += 4) {
116 unsigned V = ((unsigned char)String[Pos - 1] << 24) |
117 ((unsigned char)String[Pos - 2] << 16) |
118 ((unsigned char)String[Pos - 3] << 8) |
119 (unsigned char)String[Pos - 4];
125 // With the leftover bits.
127 // Pos will have overshot size by 4 - #bytes left over.
128 // No need to take endianness into account here - this is always executed.
129 switch (Pos - Size) {
130 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; LLVM_FALLTHROUGH;
131 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; LLVM_FALLTHROUGH;
132 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
133 default: return; // Nothing left.
139 // AddNodeID - Adds the Bit data of another ID to *this.
140 void FoldingSetNodeID::AddNodeID(const FoldingSetNodeID &ID) {
141 Bits.append(ID.Bits.begin(), ID.Bits.end());
144 /// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to
145 /// lookup the node in the FoldingSetBase.
146 unsigned FoldingSetNodeID::ComputeHash() const {
147 return FoldingSetNodeIDRef(Bits.data(), Bits.size()).ComputeHash();
150 /// operator== - Used to compare two nodes to each other.
152 bool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS) const {
153 return *this == FoldingSetNodeIDRef(RHS.Bits.data(), RHS.Bits.size());
156 /// operator== - Used to compare two nodes to each other.
158 bool FoldingSetNodeID::operator==(FoldingSetNodeIDRef RHS) const {
159 return FoldingSetNodeIDRef(Bits.data(), Bits.size()) == RHS;
162 /// Used to compare the "ordering" of two nodes as defined by the
163 /// profiled bits and their ordering defined by memcmp().
164 bool FoldingSetNodeID::operator<(const FoldingSetNodeID &RHS) const {
165 return *this < FoldingSetNodeIDRef(RHS.Bits.data(), RHS.Bits.size());
168 bool FoldingSetNodeID::operator<(FoldingSetNodeIDRef RHS) const {
169 return FoldingSetNodeIDRef(Bits.data(), Bits.size()) < RHS;
172 /// Intern - Copy this node's data to a memory region allocated from the
173 /// given allocator and return a FoldingSetNodeIDRef describing the
176 FoldingSetNodeID::Intern(BumpPtrAllocator &Allocator) const {
177 unsigned *New = Allocator.Allocate<unsigned>(Bits.size());
178 std::uninitialized_copy(Bits.begin(), Bits.end(), New);
179 return FoldingSetNodeIDRef(New, Bits.size());
182 //===----------------------------------------------------------------------===//
183 /// Helper functions for FoldingSetBase.
185 /// GetNextPtr - In order to save space, each bucket is a
186 /// singly-linked-list. In order to make deletion more efficient, we make
187 /// the list circular, so we can delete a node without computing its hash.
188 /// The problem with this is that the start of the hash buckets are not
189 /// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null:
190 /// use GetBucketPtr when this happens.
191 static FoldingSetBase::Node *GetNextPtr(void *NextInBucketPtr) {
192 // The low bit is set if this is the pointer back to the bucket.
193 if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1)
196 return static_cast<FoldingSetBase::Node*>(NextInBucketPtr);
201 static void **GetBucketPtr(void *NextInBucketPtr) {
202 intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr);
203 assert((Ptr & 1) && "Not a bucket pointer");
204 return reinterpret_cast<void**>(Ptr & ~intptr_t(1));
207 /// GetBucketFor - Hash the specified node ID and return the hash bucket for
208 /// the specified ID.
209 static void **GetBucketFor(unsigned Hash, void **Buckets, unsigned NumBuckets) {
210 // NumBuckets is always a power of 2.
211 unsigned BucketNum = Hash & (NumBuckets-1);
212 return Buckets + BucketNum;
215 /// AllocateBuckets - Allocated initialized bucket memory.
216 static void **AllocateBuckets(unsigned NumBuckets) {
217 void **Buckets = static_cast<void**>(calloc(NumBuckets+1, sizeof(void*)));
218 // Set the very last bucket to be a non-null "pointer".
219 Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
223 //===----------------------------------------------------------------------===//
224 // FoldingSetBase Implementation
226 void FoldingSetBase::anchor() {}
228 FoldingSetBase::FoldingSetBase(unsigned Log2InitSize) {
229 assert(5 < Log2InitSize && Log2InitSize < 32 &&
230 "Initial hash table size out of range");
231 NumBuckets = 1 << Log2InitSize;
232 Buckets = AllocateBuckets(NumBuckets);
236 FoldingSetBase::FoldingSetBase(FoldingSetBase &&Arg)
237 : Buckets(Arg.Buckets), NumBuckets(Arg.NumBuckets), NumNodes(Arg.NumNodes) {
238 Arg.Buckets = nullptr;
243 FoldingSetBase &FoldingSetBase::operator=(FoldingSetBase &&RHS) {
244 free(Buckets); // This may be null if the set is in a moved-from state.
245 Buckets = RHS.Buckets;
246 NumBuckets = RHS.NumBuckets;
247 NumNodes = RHS.NumNodes;
248 RHS.Buckets = nullptr;
254 FoldingSetBase::~FoldingSetBase() {
258 void FoldingSetBase::clear() {
259 // Set all but the last bucket to null pointers.
260 memset(Buckets, 0, NumBuckets*sizeof(void*));
262 // Set the very last bucket to be a non-null "pointer".
263 Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
265 // Reset the node count to zero.
269 void FoldingSetBase::GrowBucketCount(unsigned NewBucketCount) {
270 assert((NewBucketCount > NumBuckets) && "Can't shrink a folding set with GrowBucketCount");
271 assert(isPowerOf2_32(NewBucketCount) && "Bad bucket count!");
272 void **OldBuckets = Buckets;
273 unsigned OldNumBuckets = NumBuckets;
274 NumBuckets = NewBucketCount;
276 // Clear out new buckets.
277 Buckets = AllocateBuckets(NumBuckets);
280 // Walk the old buckets, rehashing nodes into their new place.
281 FoldingSetNodeID TempID;
282 for (unsigned i = 0; i != OldNumBuckets; ++i) {
283 void *Probe = OldBuckets[i];
284 if (!Probe) continue;
285 while (Node *NodeInBucket = GetNextPtr(Probe)) {
286 // Figure out the next link, remove NodeInBucket from the old link.
287 Probe = NodeInBucket->getNextInBucket();
288 NodeInBucket->SetNextInBucket(nullptr);
290 // Insert the node into the new bucket, after recomputing the hash.
291 InsertNode(NodeInBucket,
292 GetBucketFor(ComputeNodeHash(NodeInBucket, TempID),
293 Buckets, NumBuckets));
301 /// GrowHashTable - Double the size of the hash table and rehash everything.
303 void FoldingSetBase::GrowHashTable() {
304 GrowBucketCount(NumBuckets * 2);
307 void FoldingSetBase::reserve(unsigned EltCount) {
308 // This will give us somewhere between EltCount / 2 and
309 // EltCount buckets. This puts us in the load factor
310 // range of 1.0 - 2.0.
311 if(EltCount < capacity())
313 GrowBucketCount(PowerOf2Floor(EltCount));
316 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists,
317 /// return it. If not, return the insertion token that will make insertion
319 FoldingSetBase::Node *
320 FoldingSetBase::FindNodeOrInsertPos(const FoldingSetNodeID &ID,
322 unsigned IDHash = ID.ComputeHash();
323 void **Bucket = GetBucketFor(IDHash, Buckets, NumBuckets);
324 void *Probe = *Bucket;
328 FoldingSetNodeID TempID;
329 while (Node *NodeInBucket = GetNextPtr(Probe)) {
330 if (NodeEquals(NodeInBucket, ID, IDHash, TempID))
334 Probe = NodeInBucket->getNextInBucket();
337 // Didn't find the node, return null with the bucket as the InsertPos.
342 /// InsertNode - Insert the specified node into the folding set, knowing that it
343 /// is not already in the map. InsertPos must be obtained from
344 /// FindNodeOrInsertPos.
345 void FoldingSetBase::InsertNode(Node *N, void *InsertPos) {
346 assert(!N->getNextInBucket());
347 // Do we need to grow the hashtable?
348 if (NumNodes+1 > capacity()) {
350 FoldingSetNodeID TempID;
351 InsertPos = GetBucketFor(ComputeNodeHash(N, TempID), Buckets, NumBuckets);
356 /// The insert position is actually a bucket pointer.
357 void **Bucket = static_cast<void**>(InsertPos);
359 void *Next = *Bucket;
361 // If this is the first insertion into this bucket, its next pointer will be
362 // null. Pretend as if it pointed to itself, setting the low bit to indicate
363 // that it is a pointer to the bucket.
365 Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1);
367 // Set the node's next pointer, and make the bucket point to the node.
368 N->SetNextInBucket(Next);
372 /// RemoveNode - Remove a node from the folding set, returning true if one was
373 /// removed or false if the node was not in the folding set.
374 bool FoldingSetBase::RemoveNode(Node *N) {
375 // Because each bucket is a circular list, we don't need to compute N's hash
377 void *Ptr = N->getNextInBucket();
378 if (!Ptr) return false; // Not in folding set.
381 N->SetNextInBucket(nullptr);
383 // Remember what N originally pointed to, either a bucket or another node.
384 void *NodeNextPtr = Ptr;
386 // Chase around the list until we find the node (or bucket) which points to N.
388 if (Node *NodeInBucket = GetNextPtr(Ptr)) {
390 Ptr = NodeInBucket->getNextInBucket();
392 // We found a node that points to N, change it to point to N's next node,
393 // removing N from the list.
395 NodeInBucket->SetNextInBucket(NodeNextPtr);
399 void **Bucket = GetBucketPtr(Ptr);
402 // If we found that the bucket points to N, update the bucket to point to
405 *Bucket = NodeNextPtr;
412 /// GetOrInsertNode - If there is an existing simple Node exactly
413 /// equal to the specified node, return it. Otherwise, insert 'N' and it
415 FoldingSetBase::Node *FoldingSetBase::GetOrInsertNode(FoldingSetBase::Node *N) {
417 GetNodeProfile(N, ID);
419 if (Node *E = FindNodeOrInsertPos(ID, IP))
425 //===----------------------------------------------------------------------===//
426 // FoldingSetIteratorImpl Implementation
428 FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) {
429 // Skip to the first non-null non-self-cycle bucket.
430 while (*Bucket != reinterpret_cast<void*>(-1) &&
431 (!*Bucket || !GetNextPtr(*Bucket)))
434 NodePtr = static_cast<FoldingSetNode*>(*Bucket);
437 void FoldingSetIteratorImpl::advance() {
438 // If there is another link within this bucket, go to it.
439 void *Probe = NodePtr->getNextInBucket();
441 if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe))
442 NodePtr = NextNodeInBucket;
444 // Otherwise, this is the last link in this bucket.
445 void **Bucket = GetBucketPtr(Probe);
447 // Skip to the next non-null non-self-cycle bucket.
450 } while (*Bucket != reinterpret_cast<void*>(-1) &&
451 (!*Bucket || !GetNextPtr(*Bucket)));
453 NodePtr = static_cast<FoldingSetNode*>(*Bucket);
457 //===----------------------------------------------------------------------===//
458 // FoldingSetBucketIteratorImpl Implementation
460 FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) {
461 Ptr = (!*Bucket || !GetNextPtr(*Bucket)) ? (void*) Bucket : *Bucket;