1 //===-- Support/FoldingSet.cpp - Uniquing Hash Set --------------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by James M. Laskey and is distributed under
6 // the University of Illinois Open Source 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 // FoldingSetImpl::NodeID Implementation
25 /// Add* - Add various data types to Bit data.
27 void FoldingSetImpl::NodeID::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 FoldingSetImpl::NodeID::AddInteger(signed I) {
40 void FoldingSetImpl::NodeID::AddInteger(unsigned I) {
43 void FoldingSetImpl::NodeID::AddInteger(int64_t I) {
44 AddInteger((uint64_t)I);
46 void FoldingSetImpl::NodeID::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 FoldingSetImpl::NodeID::AddFloat(float F) {
54 Bits.push_back(FloatToBits(F));
56 void FoldingSetImpl::NodeID::AddDouble(double D) {
57 AddInteger(DoubleToBits(D));
59 void FoldingSetImpl::NodeID::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 NodeID, used to
97 /// lookup the node in the FoldingSetImpl.
98 unsigned FoldingSetImpl::NodeID::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 FoldingSetImpl::NodeID::operator==(const FoldingSetImpl::NodeID &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 void **Buckets, unsigned NumBuckets) {
138 if (NextInBucketPtr >= Buckets && NextInBucketPtr < Buckets + NumBuckets)
140 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr);
143 /// GetBucketPtr - Provides a casting of a bucket pointer for isNode
145 static void **GetBucketPtr(void *NextInBucketPtr) {
146 return static_cast<void**>(NextInBucketPtr);
149 /// GetBucketFor - Hash the specified node ID and return the hash bucket for
150 /// the specified ID.
151 static void **GetBucketFor(const FoldingSetImpl::NodeID &ID,
152 void **Buckets, unsigned NumBuckets) {
153 // NumBuckets is always a power of 2.
154 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1);
155 return Buckets + BucketNum;
158 //===----------------------------------------------------------------------===//
159 // FoldingSetImpl Implementation
161 FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) : NumNodes(0) {
162 assert(5 < Log2InitSize && Log2InitSize < 32 &&
163 "Initial hash table size out of range");
164 NumBuckets = 1 << Log2InitSize;
165 Buckets = new void*[NumBuckets];
166 memset(Buckets, 0, NumBuckets*sizeof(void*));
168 FoldingSetImpl::~FoldingSetImpl() {
172 /// GrowHashTable - Double the size of the hash table and rehash everything.
174 void FoldingSetImpl::GrowHashTable() {
175 void **OldBuckets = Buckets;
176 unsigned OldNumBuckets = NumBuckets;
179 // Reset the node count to zero: we're going to reinsert everything.
182 // Clear out new buckets.
183 Buckets = new void*[NumBuckets];
184 memset(Buckets, 0, NumBuckets*sizeof(void*));
186 // Walk the old buckets, rehashing nodes into their new place.
187 for (unsigned i = 0; i != OldNumBuckets; ++i) {
188 void *Probe = OldBuckets[i];
189 if (!Probe) continue;
190 while (Node *NodeInBucket = GetNextPtr(Probe, OldBuckets, OldNumBuckets)) {
191 // Figure out the next link, remove NodeInBucket from the old link.
192 Probe = NodeInBucket->getNextInBucket();
193 NodeInBucket->SetNextInBucket(0);
195 // Insert the node into the new bucket, after recomputing the hash.
197 GetNodeProfile(ID, NodeInBucket);
198 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets));
205 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists,
206 /// return it. If not, return the insertion token that will make insertion
208 FoldingSetImpl::Node *FoldingSetImpl::FindNodeOrInsertPos(const NodeID &ID,
210 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets);
211 void *Probe = *Bucket;
215 while (Node *NodeInBucket = GetNextPtr(Probe, Buckets, NumBuckets)) {
217 GetNodeProfile(OtherID, NodeInBucket);
221 Probe = NodeInBucket->getNextInBucket();
224 // Didn't find the node, return null with the bucket as the InsertPos.
229 /// InsertNode - Insert the specified node into the folding set, knowing that it
230 /// is not already in the map. InsertPos must be obtained from
231 /// FindNodeOrInsertPos.
232 void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) {
233 assert(N->getNextInBucket() == 0);
234 // Do we need to grow the hashtable?
235 if (NumNodes+1 > NumBuckets*2) {
238 GetNodeProfile(ID, N);
239 InsertPos = GetBucketFor(ID, Buckets, NumBuckets);
244 /// The insert position is actually a bucket pointer.
245 void **Bucket = static_cast<void**>(InsertPos);
247 void *Next = *Bucket;
249 // If this is the first insertion into this bucket, its next pointer will be
250 // null. Pretend as if it pointed to itself.
254 // Set the node's next pointer, and make the bucket point to the node.
255 N->SetNextInBucket(Next);
259 /// RemoveNode - Remove a node from the folding set, returning true if one was
260 /// removed or false if the node was not in the folding set.
261 bool FoldingSetImpl::RemoveNode(Node *N) {
262 // Because each bucket is a circular list, we don't need to compute N's hash
264 void *Ptr = N->getNextInBucket();
265 if (Ptr == 0) return false; // Not in folding set.
268 N->SetNextInBucket(0);
270 // Remember what N originally pointed to, either a bucket or another node.
271 void *NodeNextPtr = Ptr;
273 // Chase around the list until we find the node (or bucket) which points to N.
275 if (Node *NodeInBucket = GetNextPtr(Ptr, Buckets, NumBuckets)) {
277 Ptr = NodeInBucket->getNextInBucket();
279 // We found a node that points to N, change it to point to N's next node,
280 // removing N from the list.
282 NodeInBucket->SetNextInBucket(NodeNextPtr);
286 void **Bucket = GetBucketPtr(Ptr);
289 // If we found that the bucket points to N, update the bucket to point to
292 *Bucket = NodeNextPtr;
299 /// GetOrInsertNode - If there is an existing simple Node exactly
300 /// equal to the specified node, return it. Otherwise, insert 'N' and it
302 FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) {
304 GetNodeProfile(ID, N);
306 if (Node *E = FindNodeOrInsertPos(ID, IP))