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 #define DEBUG_TYPE "foldingset"
18 #include "llvm/ADT/FoldingSet.h"
19 #include "llvm/Support/MathExtras.h"
20 #include "llvm/Support/Debug.h"
24 //===----------------------------------------------------------------------===//
25 // FoldingSetImpl::NodeID Implementation
27 /// Add* - Add various data types to Bit data.
29 void FoldingSetImpl::NodeID::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 FoldingSetImpl::NodeID::AddInteger(signed I) {
42 void FoldingSetImpl::NodeID::AddInteger(unsigned I) {
45 void FoldingSetImpl::NodeID::AddInteger(uint64_t I) {
46 Bits.push_back(unsigned(I));
48 // If the integer is small, encode it just as 32-bits.
49 if ((uint64_t)(int)I != I)
50 Bits.push_back(unsigned(I >> 32));
52 void FoldingSetImpl::NodeID::AddFloat(float F) {
53 Bits.push_back(FloatToBits(F));
55 void FoldingSetImpl::NodeID::AddDouble(double D) {
56 AddInteger(DoubleToBits(D));
58 void FoldingSetImpl::NodeID::AddString(const std::string &String) {
59 unsigned Size = String.size();
63 unsigned Units = Size / 4;
65 const unsigned *Base = (const unsigned *)String.data();
67 // If the string is aligned do a bulk transfer.
68 if (!((intptr_t)Base & 3)) {
69 Bits.append(Base, Base + Units);
70 Pos = (Units + 1) * 4;
72 // Otherwise do it the hard way.
73 for ( Pos += 4; Pos <= Size; Pos += 4) {
74 unsigned V = ((unsigned char)String[Pos - 4] << 24) |
75 ((unsigned char)String[Pos - 3] << 16) |
76 ((unsigned char)String[Pos - 2] << 8) |
77 (unsigned char)String[Pos - 1];
82 // With the leftover bits.
84 // Pos will have overshot size by 4 - #bytes left over.
86 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru.
87 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru.
88 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
89 default: return; // Nothing left.
95 /// ComputeHash - Compute a strong hash value for this NodeID, used to
96 /// lookup the node in the FoldingSetImpl.
97 unsigned FoldingSetImpl::NodeID::ComputeHash() const {
98 // This is adapted from SuperFastHash by Paul Hsieh.
99 unsigned Hash = Bits.size();
100 for (const unsigned *BP = &Bits[0], *E = BP+Bits.size(); BP != E; ++BP) {
102 Hash += Data & 0xFFFF;
103 unsigned Tmp = ((Data >> 16) << 11) ^ Hash;
104 Hash = (Hash << 16) ^ Tmp;
108 // Force "avalanching" of final 127 bits.
118 /// operator== - Used to compare two nodes to each other.
120 bool FoldingSetImpl::NodeID::operator==(const FoldingSetImpl::NodeID &RHS)const{
121 if (Bits.size() != RHS.Bits.size()) return false;
122 return memcmp(&Bits[0], &RHS.Bits[0], Bits.size()*sizeof(Bits[0])) == 0;
126 //===----------------------------------------------------------------------===//
127 /// Helper functions for FoldingSetImpl.
129 /// GetNextPtr - In order to save space, each bucket is a
130 /// singly-linked-list. In order to make deletion more efficient, we make
131 /// the list circular, so we can delete a node without computing its hash.
132 /// The problem with this is that the start of the hash buckets are not
133 /// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null:
134 /// use GetBucketPtr when this happens.
135 static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr,
136 void **Buckets, unsigned NumBuckets) {
137 if (NextInBucketPtr >= Buckets && NextInBucketPtr < Buckets + NumBuckets)
139 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr);
142 /// GetBucketPtr - Provides a casting of a bucket pointer for isNode
144 static void **GetBucketPtr(void *NextInBucketPtr) {
145 return static_cast<void**>(NextInBucketPtr);
148 /// GetBucketFor - Hash the specified node ID and return the hash bucket for
149 /// the specified ID.
150 static void **GetBucketFor(const FoldingSetImpl::NodeID &ID,
151 void **Buckets, unsigned NumBuckets) {
152 // NumBuckets is always a power of 2.
153 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1);
154 return Buckets + BucketNum;
157 //===----------------------------------------------------------------------===//
158 // FoldingSetImpl Implementation
160 FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) : NumNodes(0) {
161 assert(5 < Log2InitSize && Log2InitSize < 32 &&
162 "Initial hash table size out of range");
163 NumBuckets = 1 << Log2InitSize;
164 Buckets = new void*[NumBuckets];
165 memset(Buckets, 0, NumBuckets*sizeof(void*));
167 FoldingSetImpl::~FoldingSetImpl() {
171 /// GrowHashTable - Double the size of the hash table and rehash everything.
173 void FoldingSetImpl::GrowHashTable() {
174 void **OldBuckets = Buckets;
175 unsigned OldNumBuckets = NumBuckets;
178 // Reset the node count to zero: we're going to reinsert everything.
181 // Clear out new buckets.
182 Buckets = new void*[NumBuckets];
183 memset(Buckets, 0, NumBuckets*sizeof(void*));
185 // Walk the old buckets, rehashing nodes into their new place.
186 for (unsigned i = 0; i != OldNumBuckets; ++i) {
187 void *Probe = OldBuckets[i];
188 if (!Probe) continue;
189 while (Node *NodeInBucket = GetNextPtr(Probe, OldBuckets, OldNumBuckets)) {
190 // Figure out the next link, remove NodeInBucket from the old link.
191 Probe = NodeInBucket->getNextInBucket();
192 NodeInBucket->SetNextInBucket(0);
194 // Insert the node into the new bucket, after recomputing the hash.
196 GetNodeProfile(ID, NodeInBucket);
197 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets));
204 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists,
205 /// return it. If not, return the insertion token that will make insertion
207 FoldingSetImpl::Node *FoldingSetImpl::FindNodeOrInsertPos(const NodeID &ID,
209 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets);
210 void *Probe = *Bucket;
214 while (Node *NodeInBucket = GetNextPtr(Probe, Buckets, NumBuckets)) {
216 GetNodeProfile(OtherID, NodeInBucket);
220 Probe = NodeInBucket->getNextInBucket();
223 // Didn't find the node, return null with the bucket as the InsertPos.
228 /// InsertNode - Insert the specified node into the folding set, knowing that it
229 /// is not already in the map. InsertPos must be obtained from
230 /// FindNodeOrInsertPos.
231 void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) {
232 assert(N->getNextInBucket() == 0);
233 // Do we need to grow the hashtable?
234 DEBUG(DOUT << "INSERT: " << N << '\n');
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 DEBUG(DOUT << "REMOVE: " << N << '\n');
265 void *Ptr = N->getNextInBucket();
266 if (Ptr == 0) return false; // Not in folding set.
269 N->SetNextInBucket(0);
271 // Remember what N originally pointed to, either a bucket or another node.
272 void *NodeNextPtr = Ptr;
274 // Chase around the list until we find the node (or bucket) which points to N.
276 if (Node *NodeInBucket = GetNextPtr(Ptr, Buckets, NumBuckets)) {
278 Ptr = NodeInBucket->getNextInBucket();
280 // We found a node that points to N, change it to point to N's next node,
281 // removing N from the list.
283 NodeInBucket->SetNextInBucket(NodeNextPtr);
287 void **Bucket = GetBucketPtr(Ptr);
290 // If we found that the bucket points to N, update the bucket to point to
293 *Bucket = NodeNextPtr;
300 /// GetOrInsertNode - If there is an existing simple Node exactly
301 /// equal to the specified node, return it. Otherwise, insert 'N' and it
303 FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) {
305 GetNodeProfile(ID, N);
307 if (Node *E = FindNodeOrInsertPos(ID, IP))