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();
57 unsigned Units = Size / 4;
59 const unsigned *Base = (const unsigned *)String.data();
61 // If the string is aligned do a bulk transfer.
62 if (!((intptr_t)Base & 3)) {
63 Bits.append(Base, Base + Units);
64 Pos = (Units + 1) * 4;
66 // Otherwise do it the hard way.
67 for ( Pos += 4; Pos <= Size; Pos += 4) {
68 unsigned V = ((unsigned char)String[Pos - 4] << 24) |
69 ((unsigned char)String[Pos - 3] << 16) |
70 ((unsigned char)String[Pos - 2] << 8) |
71 (unsigned char)String[Pos - 1];
76 // With the leftover bits.
78 // Pos will have overshot size by 4 - #bytes left over.
80 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru.
81 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru.
82 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
83 default: return; // Nothing left.
89 /// ComputeHash - Compute a strong hash value for this NodeID, used to
90 /// lookup the node in the FoldingSetImpl.
91 unsigned FoldingSetImpl::NodeID::ComputeHash() const {
92 // This is adapted from SuperFastHash by Paul Hsieh.
93 unsigned Hash = Bits.size();
94 for (const unsigned *BP = &Bits[0], *E = BP+Bits.size(); BP != E; ++BP) {
96 Hash += Data & 0xFFFF;
97 unsigned Tmp = ((Data >> 16) << 11) ^ Hash;
98 Hash = (Hash << 16) ^ Tmp;
102 // Force "avalanching" of final 127 bits.
112 /// operator== - Used to compare two nodes to each other.
114 bool FoldingSetImpl::NodeID::operator==(const FoldingSetImpl::NodeID &RHS)const{
115 if (Bits.size() != RHS.Bits.size()) return false;
116 return memcmp(&Bits[0], &RHS.Bits[0], Bits.size()*sizeof(Bits[0])) == 0;
120 //===----------------------------------------------------------------------===//
121 /// Helper functions for FoldingSetImpl.
123 /// GetNextPtr - In order to save space, each bucket is a
124 /// singly-linked-list. In order to make deletion more efficient, we make
125 /// the list circular, so we can delete a node without computing its hash.
126 /// The problem with this is that the start of the hash buckets are not
127 /// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null
128 /// : use GetBucketPtr when this happens.
129 static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr,
130 void **Buckets, unsigned NumBuckets) {
131 if (NextInBucketPtr >= Buckets && NextInBucketPtr < Buckets + NumBuckets)
133 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr);
136 /// GetBucketPtr - Provides a casting of a bucket pointer for isNode
138 static void **GetBucketPtr(void *NextInBucketPtr) {
139 return static_cast<void**>(NextInBucketPtr);
142 /// GetBucketFor - Hash the specified node ID and return the hash bucket for
143 /// the specified ID.
144 static void **GetBucketFor(const FoldingSetImpl::NodeID &ID,
145 void **Buckets, unsigned NumBuckets) {
146 // NumBuckets is always a power of 2.
147 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1);
148 return Buckets + BucketNum;
151 //===----------------------------------------------------------------------===//
152 // FoldingSetImpl Implementation
154 FoldingSetImpl::FoldingSetImpl() : NumNodes(0) {
156 Buckets = new void*[NumBuckets];
157 memset(Buckets, 0, NumBuckets*sizeof(void*));
159 FoldingSetImpl::~FoldingSetImpl() {
163 /// GrowHashTable - Double the size of the hash table and rehash everything.
165 void FoldingSetImpl::GrowHashTable() {
166 void **OldBuckets = Buckets;
167 unsigned OldNumBuckets = NumBuckets;
170 // Reset the node count to zero: we're going to reinsert everything.
173 // Clear out new buckets.
174 Buckets = new void*[NumBuckets];
175 memset(Buckets, 0, NumBuckets*sizeof(void*));
177 // Walk the old buckets, rehashing nodes into their new place.
178 for (unsigned i = 0; i != OldNumBuckets; ++i) {
179 void *Probe = OldBuckets[i];
180 if (!Probe) continue;
181 while (Node *NodeInBucket = GetNextPtr(Probe, OldBuckets, OldNumBuckets)){
182 // Figure out the next link, remove NodeInBucket from the old link.
183 Probe = NodeInBucket->getNextInBucket();
184 NodeInBucket->SetNextInBucket(0);
186 // Insert the node into the new bucket, after recomputing the hash.
188 GetNodeProfile(ID, NodeInBucket);
189 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets));
196 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists,
197 /// return it. If not, return the insertion token that will make insertion
199 FoldingSetImpl::Node *FoldingSetImpl::FindNodeOrInsertPos(const NodeID &ID,
201 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets);
202 void *Probe = *Bucket;
206 while (Node *NodeInBucket = GetNextPtr(Probe, Buckets, NumBuckets)) {
208 GetNodeProfile(OtherID, NodeInBucket);
212 Probe = NodeInBucket->getNextInBucket();
215 // Didn't find the node, return null with the bucket as the InsertPos.
220 /// InsertNode - Insert the specified node into the folding set, knowing that it
221 /// is not already in the map. InsertPos must be obtained from
222 /// FindNodeOrInsertPos.
223 void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) {
225 // Do we need to grow the hashtable?
226 if (NumNodes > NumBuckets*2) {
229 GetNodeProfile(ID, N);
230 InsertPos = GetBucketFor(ID, Buckets, NumBuckets);
233 /// The insert position is actually a bucket pointer.
234 void **Bucket = static_cast<void**>(InsertPos);
236 void *Next = *Bucket;
238 // If this is the first insertion into this bucket, its next pointer will be
239 // null. Pretend as if it pointed to itself.
243 // Set the nodes next pointer, and make the bucket point to the node.
244 N->SetNextInBucket(Next);
248 /// RemoveNode - Remove a node from the folding set, returning true if one was
249 /// removed or false if the node was not in the folding set.
250 bool FoldingSetImpl::RemoveNode(Node *N) {
251 // Because each bucket is a circular list, we don't need to compute N's hash
252 // to remove it. Chase around the list until we find the node (or bucket)
253 // which points to N.
254 void *Ptr = N->getNextInBucket();
255 if (Ptr == 0) return false; // Not in folding set.
259 void *NodeNextPtr = Ptr;
260 N->SetNextInBucket(0);
262 if (Node *NodeInBucket = GetNextPtr(Ptr, Buckets, NumBuckets)) {
264 Ptr = NodeInBucket->getNextInBucket();
266 // We found a node that points to N, change it to point to N's next node,
267 // removing N from the list.
269 NodeInBucket->SetNextInBucket(NodeNextPtr);
273 void **Bucket = GetBucketPtr(Ptr);
276 // If we found that the bucket points to N, update the bucket to point to
279 *Bucket = NodeNextPtr;
286 /// GetOrInsertNode - If there is an existing simple Node exactly
287 /// equal to the specified node, return it. Otherwise, insert 'N' and it
289 FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) {
291 GetNodeProfile(ID, N);
293 if (Node *E = FindNodeOrInsertPos(ID, IP))