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"
21 //===----------------------------------------------------------------------===//
22 // FoldingSetImpl::NodeID Implementation
24 /// Add* - Add various data types to Bit data.
26 void FoldingSetImpl::NodeID::AddPointer(const void *Ptr) {
27 // Note: this adds pointers to the hash using sizes and endianness that
28 // depend on the host. It doesn't matter however, because hashing on
29 // pointer values in inherently unstable. Nothing should depend on the
30 // ordering of nodes in the folding set.
31 intptr_t PtrI = (intptr_t)Ptr;
32 Bits.push_back(unsigned(PtrI));
33 if (sizeof(intptr_t) > sizeof(unsigned))
34 Bits.push_back(unsigned(uint64_t(PtrI) >> 32));
36 void FoldingSetImpl::NodeID::AddInteger(signed I) {
39 void FoldingSetImpl::NodeID::AddInteger(unsigned I) {
42 void FoldingSetImpl::NodeID::AddInteger(uint64_t I) {
43 Bits.push_back(unsigned(I));
44 Bits.push_back(unsigned(I >> 32));
46 void FoldingSetImpl::NodeID::AddFloat(float F) {
47 Bits.push_back(FloatToBits(F));
49 void FoldingSetImpl::NodeID::AddDouble(double D) {
50 AddInteger(DoubleToBits(D));
52 void FoldingSetImpl::NodeID::AddString(const std::string &String) {
53 unsigned Size = String.size();
54 unsigned Units = Size / 4;
56 const unsigned *Base = (const unsigned *)String.data();
58 // If the string is aligned do a bulk transfer.
59 if (!((intptr_t)Base & 3)) {
60 Bits.append(Base, Base + Units);
61 Pos = Units * sizeof(unsigned);
63 // Otherwise do it the hard way.
64 for ( Pos += 4; Pos < Size; Pos += 4) {
65 unsigned V = ((unsigned char)String[Pos - 4] << 24) |
66 ((unsigned char)String[Pos - 3] << 16) |
67 ((unsigned char)String[Pos - 2] << 8) |
68 (unsigned char)String[Pos - 1];
73 // With the leftover bits.
75 // Pos will have overshot size by 4 - #bytes left over.
77 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru.
78 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru.
79 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
80 case 0: return; // Nothing left.
86 /// ComputeHash - Compute a strong hash value for this NodeID, used to
87 /// lookup the node in the FoldingSetImpl.
88 unsigned FoldingSetImpl::NodeID::ComputeHash() const {
89 // This is adapted from SuperFastHash by Paul Hsieh.
90 unsigned Hash = Bits.size();
91 for (const unsigned *BP = &Bits[0], *E = BP+Bits.size(); BP != E; ++BP) {
93 Hash += Data & 0xFFFF;
94 unsigned Tmp = ((Data >> 16) << 11) ^ Hash;
95 Hash = (Hash << 16) ^ Tmp;
99 // Force "avalanching" of final 127 bits.
109 /// operator== - Used to compare two nodes to each other.
111 bool FoldingSetImpl::NodeID::operator==(const FoldingSetImpl::NodeID &RHS)const{
112 if (Bits.size() != RHS.Bits.size()) return false;
113 return memcmp(&Bits[0], &RHS.Bits[0], Bits.size()*sizeof(Bits[0])) == 0;
117 //===----------------------------------------------------------------------===//
118 /// Helper functions for FoldingSetImpl.
120 /// GetNextPtr - In order to save space, each bucket is a
121 /// singly-linked-list. In order to make deletion more efficient, we make
122 /// the list circular, so we can delete a node without computing its hash.
123 /// The problem with this is that the start of the hash buckets are not
124 /// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null
125 /// : use GetBucketPtr when this happens.
126 static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr,
127 void **Buckets, unsigned NumBuckets) {
128 if (NextInBucketPtr >= Buckets && NextInBucketPtr < Buckets + NumBuckets)
130 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr);
133 /// GetBucketPtr - Provides a casting of a bucket pointer for isNode
135 static void **GetBucketPtr(void *NextInBucketPtr) {
136 return static_cast<void**>(NextInBucketPtr);
139 /// GetBucketFor - Hash the specified node ID and return the hash bucket for
140 /// the specified ID.
141 static void **GetBucketFor(const FoldingSetImpl::NodeID &ID,
142 void **Buckets, unsigned NumBuckets) {
143 // NumBuckets is always a power of 2.
144 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1);
145 return Buckets + BucketNum;
148 //===----------------------------------------------------------------------===//
149 // FoldingSetImpl Implementation
151 FoldingSetImpl::FoldingSetImpl() : NumNodes(0) {
153 Buckets = new void*[NumBuckets];
154 memset(Buckets, 0, NumBuckets*sizeof(void*));
156 FoldingSetImpl::~FoldingSetImpl() {
160 /// GrowHashTable - Double the size of the hash table and rehash everything.
162 void FoldingSetImpl::GrowHashTable() {
163 void **OldBuckets = Buckets;
164 unsigned OldNumBuckets = NumBuckets;
167 // Reset the node count to zero: we're going to reinsert everything.
170 // Clear out new buckets.
171 Buckets = new void*[NumBuckets];
172 memset(Buckets, 0, NumBuckets*sizeof(void*));
174 // Walk the old buckets, rehashing nodes into their new place.
175 for (unsigned i = 0; i != OldNumBuckets; ++i) {
176 void *Probe = OldBuckets[i];
177 if (!Probe) continue;
178 while (Node *NodeInBucket = GetNextPtr(Probe, OldBuckets, OldNumBuckets)){
179 // Figure out the next link, remove NodeInBucket from the old link.
180 Probe = NodeInBucket->getNextInBucket();
181 NodeInBucket->SetNextInBucket(0);
183 // Insert the node into the new bucket, after recomputing the hash.
185 GetNodeProfile(ID, NodeInBucket);
186 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets));
193 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists,
194 /// return it. If not, return the insertion token that will make insertion
196 FoldingSetImpl::Node *FoldingSetImpl::FindNodeOrInsertPos(const NodeID &ID,
198 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets);
199 void *Probe = *Bucket;
203 while (Node *NodeInBucket = GetNextPtr(Probe, Buckets, NumBuckets)) {
205 GetNodeProfile(OtherID, NodeInBucket);
209 Probe = NodeInBucket->getNextInBucket();
212 // Didn't find the node, return null with the bucket as the InsertPos.
217 /// InsertNode - Insert the specified node into the folding set, knowing that it
218 /// is not already in the map. InsertPos must be obtained from
219 /// FindNodeOrInsertPos.
220 void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) {
222 // Do we need to grow the hashtable?
223 if (NumNodes > NumBuckets*2) {
226 GetNodeProfile(ID, N);
227 InsertPos = GetBucketFor(ID, Buckets, NumBuckets);
230 /// The insert position is actually a bucket pointer.
231 void **Bucket = static_cast<void**>(InsertPos);
233 void *Next = *Bucket;
235 // If this is the first insertion into this bucket, its next pointer will be
236 // null. Pretend as if it pointed to itself.
240 // Set the nodes next pointer, and make the bucket point to the node.
241 N->SetNextInBucket(Next);
245 /// RemoveNode - Remove a node from the folding set, returning true if one was
246 /// removed or false if the node was not in the folding set.
247 bool FoldingSetImpl::RemoveNode(Node *N) {
248 // Because each bucket is a circular list, we don't need to compute N's hash
249 // to remove it. Chase around the list until we find the node (or bucket)
250 // which points to N.
251 void *Ptr = N->getNextInBucket();
252 if (Ptr == 0) return false; // Not in folding set.
256 void *NodeNextPtr = Ptr;
257 N->SetNextInBucket(0);
259 if (Node *NodeInBucket = GetNextPtr(Ptr, Buckets, NumBuckets)) {
261 Ptr = NodeInBucket->getNextInBucket();
263 // We found a node that points to N, change it to point to N's next node,
264 // removing N from the list.
266 NodeInBucket->SetNextInBucket(NodeNextPtr);
270 void **Bucket = GetBucketPtr(Ptr);
273 // If we found that the bucket points to N, update the bucket to point to
276 *Bucket = NodeNextPtr;
283 /// GetOrInsertNode - If there is an existing simple Node exactly
284 /// equal to the specified node, return it. Otherwise, insert 'N' and it
286 FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) {
288 GetNodeProfile(ID, N);
290 if (Node *E = FindNodeOrInsertPos(ID, IP))