FoldingSet.cpp revision 72e61b85017286418ca5266c2ff7b88782e2fe00
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> 20#include <cstring> 21using namespace llvm; 22 23//===----------------------------------------------------------------------===// 24// FoldingSetNodeID Implementation 25 26/// Add* - Add various data types to Bit data. 27/// 28void FoldingSetNodeID::AddPointer(const void *Ptr) { 29 // Note: this adds pointers to the hash using sizes and endianness that 30 // depend on the host. It doesn't matter however, because hashing on 31 // pointer values in inherently unstable. Nothing should depend on the 32 // ordering of nodes in the folding set. 33 intptr_t PtrI = (intptr_t)Ptr; 34 Bits.push_back(unsigned(PtrI)); 35 if (sizeof(intptr_t) > sizeof(unsigned)) 36 Bits.push_back(unsigned(uint64_t(PtrI) >> 32)); 37} 38void FoldingSetNodeID::AddInteger(signed I) { 39 Bits.push_back(I); 40} 41void FoldingSetNodeID::AddInteger(unsigned I) { 42 Bits.push_back(I); 43} 44void FoldingSetNodeID::AddInteger(int64_t I) { 45 AddInteger((uint64_t)I); 46} 47void FoldingSetNodeID::AddInteger(uint64_t I) { 48 Bits.push_back(unsigned(I)); 49 50 // If the integer is small, encode it just as 32-bits. 51 if ((uint64_t)(int)I != I) 52 Bits.push_back(unsigned(I >> 32)); 53} 54void FoldingSetNodeID::AddFloat(float F) { 55 Bits.push_back(FloatToBits(F)); 56} 57void FoldingSetNodeID::AddDouble(double D) { 58 AddInteger(DoubleToBits(D)); 59} 60 61void FoldingSetNodeID::AddString(const char *String) { 62 unsigned Size = static_cast<unsigned>(strlen(String)); 63 Bits.push_back(Size); 64 if (!Size) return; 65 66 unsigned Units = Size / 4; 67 unsigned Pos = 0; 68 const unsigned *Base = (const unsigned *)String; 69 70 // If the string is aligned do a bulk transfer. 71 if (!((intptr_t)Base & 3)) { 72 Bits.append(Base, Base + Units); 73 Pos = (Units + 1) * 4; 74 } else { 75 // Otherwise do it the hard way. 76 for ( Pos += 4; Pos <= Size; Pos += 4) { 77 unsigned V = ((unsigned char)String[Pos - 4] << 24) | 78 ((unsigned char)String[Pos - 3] << 16) | 79 ((unsigned char)String[Pos - 2] << 8) | 80 (unsigned char)String[Pos - 1]; 81 Bits.push_back(V); 82 } 83 } 84 85 // With the leftover bits. 86 unsigned V = 0; 87 // Pos will have overshot size by 4 - #bytes left over. 88 switch (Pos - Size) { 89 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru. 90 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru. 91 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break; 92 default: return; // Nothing left. 93 } 94 95 Bits.push_back(V); 96} 97 98void FoldingSetNodeID::AddString(const std::string &String) { 99 unsigned Size = static_cast<unsigned>(String.size()); 100 Bits.push_back(Size); 101 if (!Size) return; 102 103 unsigned Units = Size / 4; 104 unsigned Pos = 0; 105 const unsigned *Base = (const unsigned *)String.data(); 106 107 // If the string is aligned do a bulk transfer. 108 if (!((intptr_t)Base & 3)) { 109 Bits.append(Base, Base + Units); 110 Pos = (Units + 1) * 4; 111 } else { 112 // Otherwise do it the hard way. 113 for ( Pos += 4; Pos <= Size; Pos += 4) { 114 unsigned V = ((unsigned char)String[Pos - 4] << 24) | 115 ((unsigned char)String[Pos - 3] << 16) | 116 ((unsigned char)String[Pos - 2] << 8) | 117 (unsigned char)String[Pos - 1]; 118 Bits.push_back(V); 119 } 120 } 121 122 // With the leftover bits. 123 unsigned V = 0; 124 // Pos will have overshot size by 4 - #bytes left over. 125 switch (Pos - Size) { 126 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru. 127 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru. 128 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break; 129 default: return; // Nothing left. 130 } 131 132 Bits.push_back(V); 133} 134 135/// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to 136/// lookup the node in the FoldingSetImpl. 137unsigned FoldingSetNodeID::ComputeHash() const { 138 // This is adapted from SuperFastHash by Paul Hsieh. 139 unsigned Hash = static_cast<unsigned>(Bits.size()); 140 for (const unsigned *BP = &Bits[0], *E = BP+Bits.size(); BP != E; ++BP) { 141 unsigned Data = *BP; 142 Hash += Data & 0xFFFF; 143 unsigned Tmp = ((Data >> 16) << 11) ^ Hash; 144 Hash = (Hash << 16) ^ Tmp; 145 Hash += Hash >> 11; 146 } 147 148 // Force "avalanching" of final 127 bits. 149 Hash ^= Hash << 3; 150 Hash += Hash >> 5; 151 Hash ^= Hash << 4; 152 Hash += Hash >> 17; 153 Hash ^= Hash << 25; 154 Hash += Hash >> 6; 155 return Hash; 156} 157 158/// operator== - Used to compare two nodes to each other. 159/// 160bool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS)const{ 161 if (Bits.size() != RHS.Bits.size()) return false; 162 return memcmp(&Bits[0], &RHS.Bits[0], Bits.size()*sizeof(Bits[0])) == 0; 163} 164 165 166//===----------------------------------------------------------------------===// 167/// Helper functions for FoldingSetImpl. 168 169/// GetNextPtr - In order to save space, each bucket is a 170/// singly-linked-list. In order to make deletion more efficient, we make 171/// the list circular, so we can delete a node without computing its hash. 172/// The problem with this is that the start of the hash buckets are not 173/// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null: 174/// use GetBucketPtr when this happens. 175static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr) { 176 // The low bit is set if this is the pointer back to the bucket. 177 if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1) 178 return 0; 179 180 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr); 181} 182 183 184/// testing. 185static void **GetBucketPtr(void *NextInBucketPtr) { 186 intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr); 187 assert((Ptr & 1) && "Not a bucket pointer"); 188 return reinterpret_cast<void**>(Ptr & ~intptr_t(1)); 189} 190 191/// GetBucketFor - Hash the specified node ID and return the hash bucket for 192/// the specified ID. 193static void **GetBucketFor(const FoldingSetNodeID &ID, 194 void **Buckets, unsigned NumBuckets) { 195 // NumBuckets is always a power of 2. 196 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1); 197 return Buckets + BucketNum; 198} 199 200//===----------------------------------------------------------------------===// 201// FoldingSetImpl Implementation 202 203FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) : NumNodes(0) { 204 assert(5 < Log2InitSize && Log2InitSize < 32 && 205 "Initial hash table size out of range"); 206 NumBuckets = 1 << Log2InitSize; 207 Buckets = new void*[NumBuckets+1]; 208 memset(Buckets, 0, NumBuckets*sizeof(void*)); 209 210 // Set the very last bucket to be a non-null "pointer". 211 Buckets[NumBuckets] = reinterpret_cast<void*>(-1); 212} 213FoldingSetImpl::~FoldingSetImpl() { 214 delete [] Buckets; 215} 216 217/// GrowHashTable - Double the size of the hash table and rehash everything. 218/// 219void FoldingSetImpl::GrowHashTable() { 220 void **OldBuckets = Buckets; 221 unsigned OldNumBuckets = NumBuckets; 222 NumBuckets <<= 1; 223 224 // Reset the node count to zero: we're going to reinsert everything. 225 NumNodes = 0; 226 227 // Clear out new buckets. 228 Buckets = new void*[NumBuckets+1]; 229 memset(Buckets, 0, NumBuckets*sizeof(void*)); 230 231 // Set the very last bucket to be a non-null "pointer". 232 Buckets[NumBuckets] = reinterpret_cast<void*>(-1); 233 234 // Walk the old buckets, rehashing nodes into their new place. 235 for (unsigned i = 0; i != OldNumBuckets; ++i) { 236 void *Probe = OldBuckets[i]; 237 if (!Probe) continue; 238 while (Node *NodeInBucket = GetNextPtr(Probe)) { 239 // Figure out the next link, remove NodeInBucket from the old link. 240 Probe = NodeInBucket->getNextInBucket(); 241 NodeInBucket->SetNextInBucket(0); 242 243 // Insert the node into the new bucket, after recomputing the hash. 244 FoldingSetNodeID ID; 245 GetNodeProfile(ID, NodeInBucket); 246 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets)); 247 } 248 } 249 250 delete[] OldBuckets; 251} 252 253/// FindNodeOrInsertPos - Look up the node specified by ID. If it exists, 254/// return it. If not, return the insertion token that will make insertion 255/// faster. 256FoldingSetImpl::Node 257*FoldingSetImpl::FindNodeOrInsertPos(const FoldingSetNodeID &ID, 258 void *&InsertPos) { 259 260 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets); 261 void *Probe = *Bucket; 262 263 InsertPos = 0; 264 265 while (Node *NodeInBucket = GetNextPtr(Probe)) { 266 FoldingSetNodeID OtherID; 267 GetNodeProfile(OtherID, NodeInBucket); 268 if (OtherID == ID) 269 return NodeInBucket; 270 271 Probe = NodeInBucket->getNextInBucket(); 272 } 273 274 // Didn't find the node, return null with the bucket as the InsertPos. 275 InsertPos = Bucket; 276 return 0; 277} 278 279/// InsertNode - Insert the specified node into the folding set, knowing that it 280/// is not already in the map. InsertPos must be obtained from 281/// FindNodeOrInsertPos. 282void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) { 283 assert(N->getNextInBucket() == 0); 284 // Do we need to grow the hashtable? 285 if (NumNodes+1 > NumBuckets*2) { 286 GrowHashTable(); 287 FoldingSetNodeID ID; 288 GetNodeProfile(ID, N); 289 InsertPos = GetBucketFor(ID, Buckets, NumBuckets); 290 } 291 292 ++NumNodes; 293 294 /// The insert position is actually a bucket pointer. 295 void **Bucket = static_cast<void**>(InsertPos); 296 297 void *Next = *Bucket; 298 299 // If this is the first insertion into this bucket, its next pointer will be 300 // null. Pretend as if it pointed to itself, setting the low bit to indicate 301 // that it is a pointer to the bucket. 302 if (Next == 0) 303 Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1); 304 305 // Set the node's next pointer, and make the bucket point to the node. 306 N->SetNextInBucket(Next); 307 *Bucket = N; 308} 309 310/// RemoveNode - Remove a node from the folding set, returning true if one was 311/// removed or false if the node was not in the folding set. 312bool FoldingSetImpl::RemoveNode(Node *N) { 313 // Because each bucket is a circular list, we don't need to compute N's hash 314 // to remove it. 315 void *Ptr = N->getNextInBucket(); 316 if (Ptr == 0) return false; // Not in folding set. 317 318 --NumNodes; 319 N->SetNextInBucket(0); 320 321 // Remember what N originally pointed to, either a bucket or another node. 322 void *NodeNextPtr = Ptr; 323 324 // Chase around the list until we find the node (or bucket) which points to N. 325 while (true) { 326 if (Node *NodeInBucket = GetNextPtr(Ptr)) { 327 // Advance pointer. 328 Ptr = NodeInBucket->getNextInBucket(); 329 330 // We found a node that points to N, change it to point to N's next node, 331 // removing N from the list. 332 if (Ptr == N) { 333 NodeInBucket->SetNextInBucket(NodeNextPtr); 334 return true; 335 } 336 } else { 337 void **Bucket = GetBucketPtr(Ptr); 338 Ptr = *Bucket; 339 340 // If we found that the bucket points to N, update the bucket to point to 341 // whatever is next. 342 if (Ptr == N) { 343 *Bucket = NodeNextPtr; 344 return true; 345 } 346 } 347 } 348} 349 350/// GetOrInsertNode - If there is an existing simple Node exactly 351/// equal to the specified node, return it. Otherwise, insert 'N' and it 352/// instead. 353FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) { 354 FoldingSetNodeID ID; 355 GetNodeProfile(ID, N); 356 void *IP; 357 if (Node *E = FindNodeOrInsertPos(ID, IP)) 358 return E; 359 InsertNode(N, IP); 360 return N; 361} 362 363//===----------------------------------------------------------------------===// 364// FoldingSetIteratorImpl Implementation 365 366FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) { 367 // Skip to the first non-null non-self-cycle bucket. 368 while (*Bucket != reinterpret_cast<void*>(-1) && 369 (*Bucket == 0 || GetNextPtr(*Bucket) == 0)) 370 ++Bucket; 371 372 NodePtr = static_cast<FoldingSetNode*>(*Bucket); 373} 374 375void FoldingSetIteratorImpl::advance() { 376 // If there is another link within this bucket, go to it. 377 void *Probe = NodePtr->getNextInBucket(); 378 379 if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe)) 380 NodePtr = NextNodeInBucket; 381 else { 382 // Otherwise, this is the last link in this bucket. 383 void **Bucket = GetBucketPtr(Probe); 384 385 // Skip to the next non-null non-self-cycle bucket. 386 do { 387 ++Bucket; 388 } while (*Bucket != reinterpret_cast<void*>(-1) && 389 (*Bucket == 0 || GetNextPtr(*Bucket) == 0)); 390 391 NodePtr = static_cast<FoldingSetNode*>(*Bucket); 392 } 393} 394 395//===----------------------------------------------------------------------===// 396// FoldingSetBucketIteratorImpl Implementation 397 398FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) { 399 Ptr = (*Bucket == 0 || GetNextPtr(*Bucket) == 0) ? (void*) Bucket : *Bucket; 400} 401