FoldingSet.cpp revision 1f801fa5ada9cb40fb97ae755c282e91af54a1bc
1//===-- Support/FoldingSet.cpp - Uniquing Hash Set --------------*- C++ -*-===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 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 13// set. 14// 15//===----------------------------------------------------------------------===// 16 17#include "llvm/ADT/FoldingSet.h" 18#include "llvm/Support/MathExtras.h" 19#include <cassert> 20using namespace llvm; 21 22//===----------------------------------------------------------------------===// 23// FoldingSetNodeID Implementation 24 25/// Add* - Add various data types to Bit data. 26/// 27void FoldingSetNodeID::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)); 36} 37void FoldingSetNodeID::AddInteger(signed I) { 38 Bits.push_back(I); 39} 40void FoldingSetNodeID::AddInteger(unsigned I) { 41 Bits.push_back(I); 42} 43void FoldingSetNodeID::AddInteger(int64_t I) { 44 AddInteger((uint64_t)I); 45} 46void FoldingSetNodeID::AddInteger(uint64_t I) { 47 Bits.push_back(unsigned(I)); 48 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)); 52} 53void FoldingSetNodeID::AddFloat(float F) { 54 Bits.push_back(FloatToBits(F)); 55} 56void FoldingSetNodeID::AddDouble(double D) { 57 AddInteger(DoubleToBits(D)); 58} 59void FoldingSetNodeID::AddString(const std::string &String) { 60 unsigned Size = String.size(); 61 Bits.push_back(Size); 62 if (!Size) return; 63 64 unsigned Units = Size / 4; 65 unsigned Pos = 0; 66 const unsigned *Base = (const unsigned *)String.data(); 67 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; 72 } else { 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]; 79 Bits.push_back(V); 80 } 81 } 82 83 // With the leftover bits. 84 unsigned V = 0; 85 // Pos will have overshot size by 4 - #bytes left over. 86 switch (Pos - Size) { 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. 91 } 92 93 Bits.push_back(V); 94} 95 96/// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to 97/// lookup the node in the FoldingSetImpl. 98unsigned FoldingSetNodeID::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) { 102 unsigned Data = *BP; 103 Hash += Data & 0xFFFF; 104 unsigned Tmp = ((Data >> 16) << 11) ^ Hash; 105 Hash = (Hash << 16) ^ Tmp; 106 Hash += Hash >> 11; 107 } 108 109 // Force "avalanching" of final 127 bits. 110 Hash ^= Hash << 3; 111 Hash += Hash >> 5; 112 Hash ^= Hash << 4; 113 Hash += Hash >> 17; 114 Hash ^= Hash << 25; 115 Hash += Hash >> 6; 116 return Hash; 117} 118 119/// operator== - Used to compare two nodes to each other. 120/// 121bool FoldingSetNodeID::operator==(const FoldingSetNodeID &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; 124} 125 126 127//===----------------------------------------------------------------------===// 128/// Helper functions for FoldingSetImpl. 129 130/// 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. 136static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr) { 137 // The low bit is set if this is the pointer back to the bucket. 138 if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1) 139 return 0; 140 141 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr); 142} 143 144 145/// testing. 146static void **GetBucketPtr(void *NextInBucketPtr) { 147 intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr); 148 assert((Ptr & 1) && "Not a bucket pointer"); 149 return reinterpret_cast<void**>(Ptr & ~intptr_t(1)); 150} 151 152/// GetBucketFor - Hash the specified node ID and return the hash bucket for 153/// the specified ID. 154static void **GetBucketFor(const FoldingSetNodeID &ID, 155 void **Buckets, unsigned NumBuckets) { 156 // NumBuckets is always a power of 2. 157 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1); 158 return Buckets + BucketNum; 159} 160 161//===----------------------------------------------------------------------===// 162// FoldingSetImpl Implementation 163 164FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) : NumNodes(0) { 165 assert(5 < Log2InitSize && Log2InitSize < 32 && 166 "Initial hash table size out of range"); 167 NumBuckets = 1 << Log2InitSize; 168 Buckets = new void*[NumBuckets+1]; 169 memset(Buckets, 0, NumBuckets*sizeof(void*)); 170 171 // Set the very last bucket to be a non-null "pointer". 172 Buckets[NumBuckets] = reinterpret_cast<void*>(-2); 173} 174FoldingSetImpl::~FoldingSetImpl() { 175 delete [] Buckets; 176} 177 178/// GrowHashTable - Double the size of the hash table and rehash everything. 179/// 180void FoldingSetImpl::GrowHashTable() { 181 void **OldBuckets = Buckets; 182 unsigned OldNumBuckets = NumBuckets; 183 NumBuckets <<= 1; 184 185 // Reset the node count to zero: we're going to reinsert everything. 186 NumNodes = 0; 187 188 // Clear out new buckets. 189 Buckets = new void*[NumBuckets+1]; 190 memset(Buckets, 0, NumBuckets*sizeof(void*)); 191 192 // Set the very last bucket to be a non-null "pointer". 193 Buckets[NumBuckets] = reinterpret_cast<void*>(-1); 194 195 // Walk the old buckets, rehashing nodes into their new place. 196 for (unsigned i = 0; i != OldNumBuckets; ++i) { 197 void *Probe = OldBuckets[i]; 198 if (!Probe) continue; 199 while (Node *NodeInBucket = GetNextPtr(Probe)) { 200 // Figure out the next link, remove NodeInBucket from the old link. 201 Probe = NodeInBucket->getNextInBucket(); 202 NodeInBucket->SetNextInBucket(0); 203 204 // Insert the node into the new bucket, after recomputing the hash. 205 FoldingSetNodeID ID; 206 GetNodeProfile(ID, NodeInBucket); 207 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets)); 208 } 209 } 210 211 delete[] OldBuckets; 212} 213 214/// FindNodeOrInsertPos - Look up the node specified by ID. If it exists, 215/// return it. If not, return the insertion token that will make insertion 216/// faster. 217FoldingSetImpl::Node 218*FoldingSetImpl::FindNodeOrInsertPos(const FoldingSetNodeID &ID, 219 void *&InsertPos) { 220 221 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets); 222 void *Probe = *Bucket; 223 224 InsertPos = 0; 225 226 while (Node *NodeInBucket = GetNextPtr(Probe)) { 227 FoldingSetNodeID OtherID; 228 GetNodeProfile(OtherID, NodeInBucket); 229 if (OtherID == ID) 230 return NodeInBucket; 231 232 Probe = NodeInBucket->getNextInBucket(); 233 } 234 235 // Didn't find the node, return null with the bucket as the InsertPos. 236 InsertPos = Bucket; 237 return 0; 238} 239 240/// InsertNode - Insert the specified node into the folding set, knowing that it 241/// is not already in the map. InsertPos must be obtained from 242/// FindNodeOrInsertPos. 243void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) { 244 assert(N->getNextInBucket() == 0); 245 // Do we need to grow the hashtable? 246 if (NumNodes+1 > NumBuckets*2) { 247 GrowHashTable(); 248 FoldingSetNodeID ID; 249 GetNodeProfile(ID, N); 250 InsertPos = GetBucketFor(ID, Buckets, NumBuckets); 251 } 252 253 ++NumNodes; 254 255 /// The insert position is actually a bucket pointer. 256 void **Bucket = static_cast<void**>(InsertPos); 257 258 void *Next = *Bucket; 259 260 // If this is the first insertion into this bucket, its next pointer will be 261 // null. Pretend as if it pointed to itself, setting the low bit to indicate 262 // that it is a pointer to the bucket. 263 if (Next == 0) 264 Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1); 265 266 // Set the node's next pointer, and make the bucket point to the node. 267 N->SetNextInBucket(Next); 268 *Bucket = N; 269} 270 271/// RemoveNode - Remove a node from the folding set, returning true if one was 272/// removed or false if the node was not in the folding set. 273bool FoldingSetImpl::RemoveNode(Node *N) { 274 // Because each bucket is a circular list, we don't need to compute N's hash 275 // to remove it. 276 void *Ptr = N->getNextInBucket(); 277 if (Ptr == 0) return false; // Not in folding set. 278 279 --NumNodes; 280 N->SetNextInBucket(0); 281 282 // Remember what N originally pointed to, either a bucket or another node. 283 void *NodeNextPtr = Ptr; 284 285 // Chase around the list until we find the node (or bucket) which points to N. 286 while (true) { 287 if (Node *NodeInBucket = GetNextPtr(Ptr)) { 288 // Advance pointer. 289 Ptr = NodeInBucket->getNextInBucket(); 290 291 // We found a node that points to N, change it to point to N's next node, 292 // removing N from the list. 293 if (Ptr == N) { 294 NodeInBucket->SetNextInBucket(NodeNextPtr); 295 return true; 296 } 297 } else { 298 void **Bucket = GetBucketPtr(Ptr); 299 Ptr = *Bucket; 300 301 // If we found that the bucket points to N, update the bucket to point to 302 // whatever is next. 303 if (Ptr == N) { 304 *Bucket = NodeNextPtr; 305 return true; 306 } 307 } 308 } 309} 310 311/// GetOrInsertNode - If there is an existing simple Node exactly 312/// equal to the specified node, return it. Otherwise, insert 'N' and it 313/// instead. 314FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) { 315 FoldingSetNodeID ID; 316 GetNodeProfile(ID, N); 317 void *IP; 318 if (Node *E = FindNodeOrInsertPos(ID, IP)) 319 return E; 320 InsertNode(N, IP); 321 return N; 322} 323 324//===----------------------------------------------------------------------===// 325// FoldingSetIteratorImpl Implementation 326 327FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) { 328 // Skip to the first non-null non-self-cycle bucket. 329 while (*Bucket == 0 || GetNextPtr(*Bucket) == 0) 330 ++Bucket; 331 332 NodePtr = static_cast<FoldingSetNode*>(*Bucket); 333} 334 335void FoldingSetIteratorImpl::advance() { 336 // If there is another link within this bucket, go to it. 337 void *Probe = NodePtr->getNextInBucket(); 338 339 if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe)) 340 NodePtr = NextNodeInBucket; 341 else { 342 // Otherwise, this is the last link in this bucket. 343 void **Bucket = GetBucketPtr(Probe); 344 345 // Skip to the next non-null non-self-cycle bucket. 346 do { 347 ++Bucket; 348 } while (*Bucket == 0 || GetNextPtr(*Bucket) == 0); 349 350 NodePtr = static_cast<FoldingSetNode*>(*Bucket); 351 } 352} 353 354//===----------------------------------------------------------------------===// 355// FoldingSetBucketIteratorImpl Implementation 356 357FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) { 358 Ptr = (*Bucket == 0 || GetNextPtr(*Bucket) == 0) ? (void*) Bucket : *Bucket; 359} 360