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));
47 Bits.push_back(unsigned(I >> 32));
49 void FoldingSetImpl::NodeID::AddFloat(float F) {
50 Bits.push_back(FloatToBits(F));
52 void FoldingSetImpl::NodeID::AddDouble(double D) {
53 AddInteger(DoubleToBits(D));
55 void FoldingSetImpl::NodeID::AddString(const std::string &String) {
56 unsigned Size = String.size();
60 unsigned Units = Size / 4;
62 const unsigned *Base = (const unsigned *)String.data();
64 // If the string is aligned do a bulk transfer.
65 if (!((intptr_t)Base & 3)) {
66 Bits.append(Base, Base + Units);
67 Pos = (Units + 1) * 4;
69 // Otherwise do it the hard way.
70 for ( Pos += 4; Pos <= Size; Pos += 4) {
71 unsigned V = ((unsigned char)String[Pos - 4] << 24) |
72 ((unsigned char)String[Pos - 3] << 16) |
73 ((unsigned char)String[Pos - 2] << 8) |
74 (unsigned char)String[Pos - 1];
79 // With the leftover bits.
81 // Pos will have overshot size by 4 - #bytes left over.
83 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru.
84 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru.
85 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
86 default: return; // Nothing left.
92 /// ComputeHash - Compute a strong hash value for this NodeID, used to
93 /// lookup the node in the FoldingSetImpl.
94 unsigned FoldingSetImpl::NodeID::ComputeHash() const {
95 // This is adapted from SuperFastHash by Paul Hsieh.
96 unsigned Hash = Bits.size();
97 for (const unsigned *BP = &Bits[0], *E = BP+Bits.size(); BP != E; ++BP) {
99 Hash += Data & 0xFFFF;
100 unsigned Tmp = ((Data >> 16) << 11) ^ Hash;
101 Hash = (Hash << 16) ^ Tmp;
105 // Force "avalanching" of final 127 bits.
115 /// operator== - Used to compare two nodes to each other.
117 bool FoldingSetImpl::NodeID::operator==(const FoldingSetImpl::NodeID &RHS)const{
118 if (Bits.size() != RHS.Bits.size()) return false;
119 return memcmp(&Bits[0], &RHS.Bits[0], Bits.size()*sizeof(Bits[0])) == 0;
123 //===----------------------------------------------------------------------===//
124 /// Helper functions for FoldingSetImpl.
126 /// GetNextPtr - In order to save space, each bucket is a
127 /// singly-linked-list. In order to make deletion more efficient, we make
128 /// the list circular, so we can delete a node without computing its hash.
129 /// The problem with this is that the start of the hash buckets are not
130 /// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null:
131 /// use GetBucketPtr when this happens.
132 static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr,
133 void **Buckets, unsigned NumBuckets) {
134 if (NextInBucketPtr >= Buckets && NextInBucketPtr < Buckets + NumBuckets)
136 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr);
139 /// GetBucketPtr - Provides a casting of a bucket pointer for isNode
141 static void **GetBucketPtr(void *NextInBucketPtr) {
142 return static_cast<void**>(NextInBucketPtr);
145 /// GetBucketFor - Hash the specified node ID and return the hash bucket for
146 /// the specified ID.
147 static void **GetBucketFor(const FoldingSetImpl::NodeID &ID,
148 void **Buckets, unsigned NumBuckets) {
149 // NumBuckets is always a power of 2.
150 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1);
151 return Buckets + BucketNum;
154 //===----------------------------------------------------------------------===//
155 // FoldingSetImpl Implementation
157 FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) : NumNodes(0) {
158 assert(5 < Log2InitSize && Log2InitSize < 32 &&
159 "Initial hash table size out of range");
160 NumBuckets = 1 << Log2InitSize;
161 Buckets = new void*[NumBuckets];
162 memset(Buckets, 0, NumBuckets*sizeof(void*));
164 FoldingSetImpl::~FoldingSetImpl() {
168 /// GrowHashTable - Double the size of the hash table and rehash everything.
170 void FoldingSetImpl::GrowHashTable() {
171 void **OldBuckets = Buckets;
172 unsigned OldNumBuckets = NumBuckets;
175 // Reset the node count to zero: we're going to reinsert everything.
178 // Clear out new buckets.
179 Buckets = new void*[NumBuckets];
180 memset(Buckets, 0, NumBuckets*sizeof(void*));
182 // Walk the old buckets, rehashing nodes into their new place.
183 for (unsigned i = 0; i != OldNumBuckets; ++i) {
184 void *Probe = OldBuckets[i];
185 if (!Probe) continue;
186 while (Node *NodeInBucket = GetNextPtr(Probe, OldBuckets, OldNumBuckets)) {
187 // Figure out the next link, remove NodeInBucket from the old link.
188 Probe = NodeInBucket->getNextInBucket();
189 NodeInBucket->SetNextInBucket(0);
191 // Insert the node into the new bucket, after recomputing the hash.
193 GetNodeProfile(ID, NodeInBucket);
194 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets));
201 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists,
202 /// return it. If not, return the insertion token that will make insertion
204 FoldingSetImpl::Node *FoldingSetImpl::FindNodeOrInsertPos(const NodeID &ID,
206 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets);
207 void *Probe = *Bucket;
211 while (Node *NodeInBucket = GetNextPtr(Probe, Buckets, NumBuckets)) {
213 GetNodeProfile(OtherID, NodeInBucket);
217 Probe = NodeInBucket->getNextInBucket();
220 // Didn't find the node, return null with the bucket as the InsertPos.
225 /// InsertNode - Insert the specified node into the folding set, knowing that it
226 /// is not already in the map. InsertPos must be obtained from
227 /// FindNodeOrInsertPos.
228 void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) {
229 // Do we need to grow the hashtable?
230 DEBUG(DOUT << "INSERT: " << N << '\n');
231 if (NumNodes+1 > NumBuckets*2) {
234 GetNodeProfile(ID, N);
235 InsertPos = GetBucketFor(ID, Buckets, NumBuckets);
240 /// The insert position is actually a bucket pointer.
241 void **Bucket = static_cast<void**>(InsertPos);
243 void *Next = *Bucket;
245 // If this is the first insertion into this bucket, its next pointer will be
246 // null. Pretend as if it pointed to itself.
250 // Set the node's next pointer, and make the bucket point to the node.
251 N->SetNextInBucket(Next);
255 /// RemoveNode - Remove a node from the folding set, returning true if one was
256 /// removed or false if the node was not in the folding set.
257 bool FoldingSetImpl::RemoveNode(Node *N) {
258 // Because each bucket is a circular list, we don't need to compute N's hash
259 // to remove it. Chase around the list until we find the node (or bucket)
260 // which points to N.
261 DEBUG(DOUT << "REMOVE: " << N << '\n');
262 void *Ptr = N->getNextInBucket();
263 if (Ptr == 0) return false; // Not in folding set.
267 void *NodeNextPtr = Ptr;
268 N->SetNextInBucket(0);
270 if (Node *NodeInBucket = GetNextPtr(Ptr, Buckets, NumBuckets)) {
272 Ptr = NodeInBucket->getNextInBucket();
274 // We found a node that points to N, change it to point to N's next node,
275 // removing N from the list.
277 NodeInBucket->SetNextInBucket(NodeNextPtr);
281 void **Bucket = GetBucketPtr(Ptr);
284 // If we found that the bucket points to N, update the bucket to point to
287 *Bucket = NodeNextPtr;
294 /// GetOrInsertNode - If there is an existing simple Node exactly
295 /// equal to the specified node, return it. Otherwise, insert 'N' and it
297 FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) {
299 GetNodeProfile(ID, N);
301 if (Node *E = FindNodeOrInsertPos(ID, IP))