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/Allocator.h"
19 #include "llvm/Support/ErrorHandling.h"
20 #include "llvm/Support/MathExtras.h"
25 //===----------------------------------------------------------------------===//
26 // FoldingSetNodeID Implementation
28 /// Add* - Add various data types to Bit data.
30 void FoldingSetNodeID::AddPointer(const void *Ptr) {
31 // Note: this adds pointers to the hash using sizes and endianness that
32 // depend on the host. It doesn't matter however, because hashing on
33 // pointer values in inherently unstable. Nothing should depend on the
34 // ordering of nodes in the folding set.
35 intptr_t PtrI = (intptr_t)Ptr;
36 Bits.push_back(unsigned(PtrI));
37 if (sizeof(intptr_t) > sizeof(unsigned))
38 Bits.push_back(unsigned(uint64_t(PtrI) >> 32));
40 void FoldingSetNodeID::AddInteger(signed I) {
43 void FoldingSetNodeID::AddInteger(unsigned I) {
46 void FoldingSetNodeID::AddInteger(long I) {
47 AddInteger((unsigned long)I);
49 void FoldingSetNodeID::AddInteger(unsigned long I) {
50 if (sizeof(long) == sizeof(int))
51 AddInteger(unsigned(I));
52 else if (sizeof(long) == sizeof(long long)) {
53 AddInteger((unsigned long long)I);
55 llvm_unreachable("unexpected sizeof(long)");
58 void FoldingSetNodeID::AddInteger(long long I) {
59 AddInteger((unsigned long long)I);
61 void FoldingSetNodeID::AddInteger(unsigned long long I) {
62 AddInteger(unsigned(I));
63 if ((uint64_t)(int)I != I)
64 Bits.push_back(unsigned(I >> 32));
67 void FoldingSetNodeID::AddString(StringRef String) {
68 unsigned Size = String.size();
72 unsigned Units = Size / 4;
74 const unsigned *Base = (const unsigned*) String.data();
76 // If the string is aligned do a bulk transfer.
77 if (!((intptr_t)Base & 3)) {
78 Bits.append(Base, Base + Units);
79 Pos = (Units + 1) * 4;
81 // Otherwise do it the hard way.
82 for (Pos += 4; Pos <= Size; Pos += 4) {
83 unsigned V = ((unsigned char)String[Pos - 4] << 24) |
84 ((unsigned char)String[Pos - 3] << 16) |
85 ((unsigned char)String[Pos - 2] << 8) |
86 (unsigned char)String[Pos - 1];
91 // With the leftover bits.
93 // Pos will have overshot size by 4 - #bytes left over.
95 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru.
96 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru.
97 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
98 default: return; // Nothing left.
104 /// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to
105 /// lookup the node in the FoldingSetImpl.
106 unsigned FoldingSetNodeID::ComputeHash() const {
107 // This is adapted from SuperFastHash by Paul Hsieh.
108 unsigned Hash = static_cast<unsigned>(Bits.size());
109 for (const unsigned *BP = &Bits[0], *E = BP+Bits.size(); BP != E; ++BP) {
111 Hash += Data & 0xFFFF;
112 unsigned Tmp = ((Data >> 16) << 11) ^ Hash;
113 Hash = (Hash << 16) ^ Tmp;
117 // Force "avalanching" of final 127 bits.
127 /// operator== - Used to compare two nodes to each other.
129 bool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS)const{
130 if (Bits.size() != RHS.Bits.size()) return false;
131 return memcmp(&Bits[0], &RHS.Bits[0], Bits.size()*sizeof(Bits[0])) == 0;
134 /// Intern - Copy this node's data to a memory region allocated from the
135 /// given allocator and return a FoldingSetNodeIDRef describing the
138 FoldingSetNodeID::Intern(BumpPtrAllocator &Allocator) const {
139 unsigned *New = Allocator.Allocate<unsigned>(Bits.size());
140 std::uninitialized_copy(Bits.begin(), Bits.end(), New);
141 return FoldingSetNodeIDRef(New, Bits.size());
144 //===----------------------------------------------------------------------===//
145 /// Helper functions for FoldingSetImpl.
147 /// GetNextPtr - In order to save space, each bucket is a
148 /// singly-linked-list. In order to make deletion more efficient, we make
149 /// the list circular, so we can delete a node without computing its hash.
150 /// The problem with this is that the start of the hash buckets are not
151 /// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null:
152 /// use GetBucketPtr when this happens.
153 static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr) {
154 // The low bit is set if this is the pointer back to the bucket.
155 if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1)
158 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr);
163 static void **GetBucketPtr(void *NextInBucketPtr) {
164 intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr);
165 assert((Ptr & 1) && "Not a bucket pointer");
166 return reinterpret_cast<void**>(Ptr & ~intptr_t(1));
169 /// GetBucketFor - Hash the specified node ID and return the hash bucket for
170 /// the specified ID.
171 static void **GetBucketFor(const FoldingSetNodeID &ID,
172 void **Buckets, unsigned NumBuckets) {
173 // NumBuckets is always a power of 2.
174 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1);
175 return Buckets + BucketNum;
178 /// AllocateBuckets - Allocated initialized bucket memory.
179 static void **AllocateBuckets(unsigned NumBuckets) {
180 void **Buckets = static_cast<void**>(calloc(NumBuckets+1, sizeof(void*)));
181 // Set the very last bucket to be a non-null "pointer".
182 Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
186 //===----------------------------------------------------------------------===//
187 // FoldingSetImpl Implementation
189 FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) {
190 assert(5 < Log2InitSize && Log2InitSize < 32 &&
191 "Initial hash table size out of range");
192 NumBuckets = 1 << Log2InitSize;
193 Buckets = AllocateBuckets(NumBuckets);
196 FoldingSetImpl::~FoldingSetImpl() {
199 void FoldingSetImpl::clear() {
200 // Set all but the last bucket to null pointers.
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 // Reset the node count to zero.
210 /// GrowHashTable - Double the size of the hash table and rehash everything.
212 void FoldingSetImpl::GrowHashTable() {
213 void **OldBuckets = Buckets;
214 unsigned OldNumBuckets = NumBuckets;
217 // Clear out new buckets.
218 Buckets = AllocateBuckets(NumBuckets);
221 // Walk the old buckets, rehashing nodes into their new place.
223 for (unsigned i = 0; i != OldNumBuckets; ++i) {
224 void *Probe = OldBuckets[i];
225 if (!Probe) continue;
226 while (Node *NodeInBucket = GetNextPtr(Probe)) {
227 // Figure out the next link, remove NodeInBucket from the old link.
228 Probe = NodeInBucket->getNextInBucket();
229 NodeInBucket->SetNextInBucket(0);
231 // Insert the node into the new bucket, after recomputing the hash.
232 GetNodeProfile(ID, NodeInBucket);
233 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets));
241 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists,
242 /// return it. If not, return the insertion token that will make insertion
245 *FoldingSetImpl::FindNodeOrInsertPos(const FoldingSetNodeID &ID,
248 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets);
249 void *Probe = *Bucket;
253 FoldingSetNodeID OtherID;
254 while (Node *NodeInBucket = GetNextPtr(Probe)) {
255 GetNodeProfile(OtherID, NodeInBucket);
259 Probe = NodeInBucket->getNextInBucket();
263 // Didn't find the node, return null with the bucket as the InsertPos.
268 /// InsertNode - Insert the specified node into the folding set, knowing that it
269 /// is not already in the map. InsertPos must be obtained from
270 /// FindNodeOrInsertPos.
271 void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) {
272 assert(N->getNextInBucket() == 0);
273 // Do we need to grow the hashtable?
274 if (NumNodes+1 > NumBuckets*2) {
277 GetNodeProfile(ID, N);
278 InsertPos = GetBucketFor(ID, Buckets, NumBuckets);
283 /// The insert position is actually a bucket pointer.
284 void **Bucket = static_cast<void**>(InsertPos);
286 void *Next = *Bucket;
288 // If this is the first insertion into this bucket, its next pointer will be
289 // null. Pretend as if it pointed to itself, setting the low bit to indicate
290 // that it is a pointer to the bucket.
292 Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1);
294 // Set the node's next pointer, and make the bucket point to the node.
295 N->SetNextInBucket(Next);
299 /// RemoveNode - Remove a node from the folding set, returning true if one was
300 /// removed or false if the node was not in the folding set.
301 bool FoldingSetImpl::RemoveNode(Node *N) {
302 // Because each bucket is a circular list, we don't need to compute N's hash
304 void *Ptr = N->getNextInBucket();
305 if (Ptr == 0) return false; // Not in folding set.
308 N->SetNextInBucket(0);
310 // Remember what N originally pointed to, either a bucket or another node.
311 void *NodeNextPtr = Ptr;
313 // Chase around the list until we find the node (or bucket) which points to N.
315 if (Node *NodeInBucket = GetNextPtr(Ptr)) {
317 Ptr = NodeInBucket->getNextInBucket();
319 // We found a node that points to N, change it to point to N's next node,
320 // removing N from the list.
322 NodeInBucket->SetNextInBucket(NodeNextPtr);
326 void **Bucket = GetBucketPtr(Ptr);
329 // If we found that the bucket points to N, update the bucket to point to
332 *Bucket = NodeNextPtr;
339 /// GetOrInsertNode - If there is an existing simple Node exactly
340 /// equal to the specified node, return it. Otherwise, insert 'N' and it
342 FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) {
344 GetNodeProfile(ID, N);
346 if (Node *E = FindNodeOrInsertPos(ID, IP))
352 //===----------------------------------------------------------------------===//
353 // FoldingSetIteratorImpl Implementation
355 FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) {
356 // Skip to the first non-null non-self-cycle bucket.
357 while (*Bucket != reinterpret_cast<void*>(-1) &&
358 (*Bucket == 0 || GetNextPtr(*Bucket) == 0))
361 NodePtr = static_cast<FoldingSetNode*>(*Bucket);
364 void FoldingSetIteratorImpl::advance() {
365 // If there is another link within this bucket, go to it.
366 void *Probe = NodePtr->getNextInBucket();
368 if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe))
369 NodePtr = NextNodeInBucket;
371 // Otherwise, this is the last link in this bucket.
372 void **Bucket = GetBucketPtr(Probe);
374 // Skip to the next non-null non-self-cycle bucket.
377 } while (*Bucket != reinterpret_cast<void*>(-1) &&
378 (*Bucket == 0 || GetNextPtr(*Bucket) == 0));
380 NodePtr = static_cast<FoldingSetNode*>(*Bucket);
384 //===----------------------------------------------------------------------===//
385 // FoldingSetBucketIteratorImpl Implementation
387 FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) {
388 Ptr = (*Bucket == 0 || GetNextPtr(*Bucket) == 0) ? (void*) Bucket : *Bucket;