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 //===----------------------------------------------------------------------===//
179 // FoldingSetImpl Implementation
181 FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) {
182 assert(5 < Log2InitSize && Log2InitSize < 32 &&
183 "Initial hash table size out of range");
184 NumBuckets = 1 << Log2InitSize;
185 Buckets = new void*[NumBuckets+1];
188 FoldingSetImpl::~FoldingSetImpl() {
191 void FoldingSetImpl::clear() {
192 // Set all but the last bucket to null pointers.
193 memset(Buckets, 0, NumBuckets*sizeof(void*));
195 // Set the very last bucket to be a non-null "pointer".
196 Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
198 // Reset the node count to zero.
202 /// GrowHashTable - Double the size of the hash table and rehash everything.
204 void FoldingSetImpl::GrowHashTable() {
205 void **OldBuckets = Buckets;
206 unsigned OldNumBuckets = NumBuckets;
209 // Clear out new buckets.
210 Buckets = new void*[NumBuckets+1];
213 // Walk the old buckets, rehashing nodes into their new place.
215 for (unsigned i = 0; i != OldNumBuckets; ++i) {
216 void *Probe = OldBuckets[i];
217 if (!Probe) continue;
218 while (Node *NodeInBucket = GetNextPtr(Probe)) {
219 // Figure out the next link, remove NodeInBucket from the old link.
220 Probe = NodeInBucket->getNextInBucket();
221 NodeInBucket->SetNextInBucket(0);
223 // Insert the node into the new bucket, after recomputing the hash.
224 GetNodeProfile(ID, NodeInBucket);
225 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets));
233 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists,
234 /// return it. If not, return the insertion token that will make insertion
237 *FoldingSetImpl::FindNodeOrInsertPos(const FoldingSetNodeID &ID,
240 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets);
241 void *Probe = *Bucket;
245 FoldingSetNodeID OtherID;
246 while (Node *NodeInBucket = GetNextPtr(Probe)) {
247 GetNodeProfile(OtherID, NodeInBucket);
251 Probe = NodeInBucket->getNextInBucket();
255 // Didn't find the node, return null with the bucket as the InsertPos.
260 /// InsertNode - Insert the specified node into the folding set, knowing that it
261 /// is not already in the map. InsertPos must be obtained from
262 /// FindNodeOrInsertPos.
263 void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) {
264 assert(N->getNextInBucket() == 0);
265 // Do we need to grow the hashtable?
266 if (NumNodes+1 > NumBuckets*2) {
269 GetNodeProfile(ID, N);
270 InsertPos = GetBucketFor(ID, Buckets, NumBuckets);
275 /// The insert position is actually a bucket pointer.
276 void **Bucket = static_cast<void**>(InsertPos);
278 void *Next = *Bucket;
280 // If this is the first insertion into this bucket, its next pointer will be
281 // null. Pretend as if it pointed to itself, setting the low bit to indicate
282 // that it is a pointer to the bucket.
284 Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1);
286 // Set the node's next pointer, and make the bucket point to the node.
287 N->SetNextInBucket(Next);
291 /// RemoveNode - Remove a node from the folding set, returning true if one was
292 /// removed or false if the node was not in the folding set.
293 bool FoldingSetImpl::RemoveNode(Node *N) {
294 // Because each bucket is a circular list, we don't need to compute N's hash
296 void *Ptr = N->getNextInBucket();
297 if (Ptr == 0) return false; // Not in folding set.
300 N->SetNextInBucket(0);
302 // Remember what N originally pointed to, either a bucket or another node.
303 void *NodeNextPtr = Ptr;
305 // Chase around the list until we find the node (or bucket) which points to N.
307 if (Node *NodeInBucket = GetNextPtr(Ptr)) {
309 Ptr = NodeInBucket->getNextInBucket();
311 // We found a node that points to N, change it to point to N's next node,
312 // removing N from the list.
314 NodeInBucket->SetNextInBucket(NodeNextPtr);
318 void **Bucket = GetBucketPtr(Ptr);
321 // If we found that the bucket points to N, update the bucket to point to
324 *Bucket = NodeNextPtr;
331 /// GetOrInsertNode - If there is an existing simple Node exactly
332 /// equal to the specified node, return it. Otherwise, insert 'N' and it
334 FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) {
336 GetNodeProfile(ID, N);
338 if (Node *E = FindNodeOrInsertPos(ID, IP))
344 //===----------------------------------------------------------------------===//
345 // FoldingSetIteratorImpl Implementation
347 FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) {
348 // Skip to the first non-null non-self-cycle bucket.
349 while (*Bucket != reinterpret_cast<void*>(-1) &&
350 (*Bucket == 0 || GetNextPtr(*Bucket) == 0))
353 NodePtr = static_cast<FoldingSetNode*>(*Bucket);
356 void FoldingSetIteratorImpl::advance() {
357 // If there is another link within this bucket, go to it.
358 void *Probe = NodePtr->getNextInBucket();
360 if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe))
361 NodePtr = NextNodeInBucket;
363 // Otherwise, this is the last link in this bucket.
364 void **Bucket = GetBucketPtr(Probe);
366 // Skip to the next non-null non-self-cycle bucket.
369 } while (*Bucket != reinterpret_cast<void*>(-1) &&
370 (*Bucket == 0 || GetNextPtr(*Bucket) == 0));
372 NodePtr = static_cast<FoldingSetNode*>(*Bucket);
376 //===----------------------------------------------------------------------===//
377 // FoldingSetBucketIteratorImpl Implementation
379 FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) {
380 Ptr = (*Bucket == 0 || GetNextPtr(*Bucket) == 0) ? (void*) Bucket : *Bucket;