FoldingSet.cpp revision 20cd13f54f2b1a6307016c47330045de13b140e2
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(long I) { 45 AddInteger((unsigned long)I); 46} 47void FoldingSetNodeID::AddInteger(unsigned long I) { 48 if (sizeof(long) == sizeof(int)) 49 AddInteger(unsigned(I)); 50 else if (sizeof(long) == sizeof(long long)) { 51 AddInteger((unsigned long long)I); 52 } else { 53 assert(0 && "unexpected sizeof(long)"); 54 } 55} 56void FoldingSetNodeID::AddInteger(long long I) { 57 AddInteger((unsigned long long)I); 58} 59void FoldingSetNodeID::AddInteger(unsigned long long I) { 60 AddInteger(unsigned(I)); 61 if ((uint64_t)(int)I != I) 62 Bits.push_back(unsigned(I >> 32)); 63} 64void FoldingSetNodeID::AddFloat(float F) { 65 Bits.push_back(FloatToBits(F)); 66} 67void FoldingSetNodeID::AddDouble(double D) { 68 AddInteger(DoubleToBits(D)); 69} 70 71void FoldingSetNodeID::AddString(const char *String) { 72 unsigned Size = static_cast<unsigned>(strlen(String)); 73 Bits.push_back(Size); 74 if (!Size) return; 75 76 unsigned Units = Size / 4; 77 unsigned Pos = 0; 78 const unsigned *Base = (const unsigned *)String; 79 80 // If the string is aligned do a bulk transfer. 81 if (!((intptr_t)Base & 3)) { 82 Bits.append(Base, Base + Units); 83 Pos = (Units + 1) * 4; 84 } else { 85 // Otherwise do it the hard way. 86 for ( Pos += 4; Pos <= Size; Pos += 4) { 87 unsigned V = ((unsigned char)String[Pos - 4] << 24) | 88 ((unsigned char)String[Pos - 3] << 16) | 89 ((unsigned char)String[Pos - 2] << 8) | 90 (unsigned char)String[Pos - 1]; 91 Bits.push_back(V); 92 } 93 } 94 95 // With the leftover bits. 96 unsigned V = 0; 97 // Pos will have overshot size by 4 - #bytes left over. 98 switch (Pos - Size) { 99 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru. 100 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru. 101 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break; 102 default: return; // Nothing left. 103 } 104 105 Bits.push_back(V); 106} 107 108void FoldingSetNodeID::AddString(const std::string &String) { 109 unsigned Size = static_cast<unsigned>(String.size()); 110 Bits.push_back(Size); 111 if (!Size) return; 112 113 unsigned Units = Size / 4; 114 unsigned Pos = 0; 115 const unsigned *Base = (const unsigned *)String.data(); 116 117 // If the string is aligned do a bulk transfer. 118 if (!((intptr_t)Base & 3)) { 119 Bits.append(Base, Base + Units); 120 Pos = (Units + 1) * 4; 121 } else { 122 // Otherwise do it the hard way. 123 for ( Pos += 4; Pos <= Size; Pos += 4) { 124 unsigned V = ((unsigned char)String[Pos - 4] << 24) | 125 ((unsigned char)String[Pos - 3] << 16) | 126 ((unsigned char)String[Pos - 2] << 8) | 127 (unsigned char)String[Pos - 1]; 128 Bits.push_back(V); 129 } 130 } 131 132 // With the leftover bits. 133 unsigned V = 0; 134 // Pos will have overshot size by 4 - #bytes left over. 135 switch (Pos - Size) { 136 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru. 137 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru. 138 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break; 139 default: return; // Nothing left. 140 } 141 142 Bits.push_back(V); 143} 144 145/// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to 146/// lookup the node in the FoldingSetImpl. 147unsigned FoldingSetNodeID::ComputeHash() const { 148 // This is adapted from SuperFastHash by Paul Hsieh. 149 unsigned Hash = static_cast<unsigned>(Bits.size()); 150 for (const unsigned *BP = &Bits[0], *E = BP+Bits.size(); BP != E; ++BP) { 151 unsigned Data = *BP; 152 Hash += Data & 0xFFFF; 153 unsigned Tmp = ((Data >> 16) << 11) ^ Hash; 154 Hash = (Hash << 16) ^ Tmp; 155 Hash += Hash >> 11; 156 } 157 158 // Force "avalanching" of final 127 bits. 159 Hash ^= Hash << 3; 160 Hash += Hash >> 5; 161 Hash ^= Hash << 4; 162 Hash += Hash >> 17; 163 Hash ^= Hash << 25; 164 Hash += Hash >> 6; 165 return Hash; 166} 167 168/// operator== - Used to compare two nodes to each other. 169/// 170bool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS)const{ 171 if (Bits.size() != RHS.Bits.size()) return false; 172 return memcmp(&Bits[0], &RHS.Bits[0], Bits.size()*sizeof(Bits[0])) == 0; 173} 174 175 176//===----------------------------------------------------------------------===// 177/// Helper functions for FoldingSetImpl. 178 179/// GetNextPtr - In order to save space, each bucket is a 180/// singly-linked-list. In order to make deletion more efficient, we make 181/// the list circular, so we can delete a node without computing its hash. 182/// The problem with this is that the start of the hash buckets are not 183/// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null: 184/// use GetBucketPtr when this happens. 185static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr) { 186 // The low bit is set if this is the pointer back to the bucket. 187 if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1) 188 return 0; 189 190 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr); 191} 192 193 194/// testing. 195static void **GetBucketPtr(void *NextInBucketPtr) { 196 intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr); 197 assert((Ptr & 1) && "Not a bucket pointer"); 198 return reinterpret_cast<void**>(Ptr & ~intptr_t(1)); 199} 200 201/// GetBucketFor - Hash the specified node ID and return the hash bucket for 202/// the specified ID. 203static void **GetBucketFor(const FoldingSetNodeID &ID, 204 void **Buckets, unsigned NumBuckets) { 205 // NumBuckets is always a power of 2. 206 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1); 207 return Buckets + BucketNum; 208} 209 210//===----------------------------------------------------------------------===// 211// FoldingSetImpl Implementation 212 213FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) { 214 assert(5 < Log2InitSize && Log2InitSize < 32 && 215 "Initial hash table size out of range"); 216 NumBuckets = 1 << Log2InitSize; 217 Buckets = new void*[NumBuckets+1]; 218 clear(); 219} 220FoldingSetImpl::~FoldingSetImpl() { 221 delete [] Buckets; 222} 223void FoldingSetImpl::clear() { 224 // Set all but the last bucket to null pointers. 225 memset(Buckets, 0, NumBuckets*sizeof(void*)); 226 227 // Set the very last bucket to be a non-null "pointer". 228 Buckets[NumBuckets] = reinterpret_cast<void*>(-1); 229 230 // Reset the node count to zero. 231 NumNodes = 0; 232} 233 234/// GrowHashTable - Double the size of the hash table and rehash everything. 235/// 236void FoldingSetImpl::GrowHashTable() { 237 void **OldBuckets = Buckets; 238 unsigned OldNumBuckets = NumBuckets; 239 NumBuckets <<= 1; 240 241 // Clear out new buckets. 242 Buckets = new void*[NumBuckets+1]; 243 clear(); 244 245 // Walk the old buckets, rehashing nodes into their new place. 246 FoldingSetNodeID ID; 247 for (unsigned i = 0; i != OldNumBuckets; ++i) { 248 void *Probe = OldBuckets[i]; 249 if (!Probe) continue; 250 while (Node *NodeInBucket = GetNextPtr(Probe)) { 251 // Figure out the next link, remove NodeInBucket from the old link. 252 Probe = NodeInBucket->getNextInBucket(); 253 NodeInBucket->SetNextInBucket(0); 254 255 // Insert the node into the new bucket, after recomputing the hash. 256 GetNodeProfile(ID, NodeInBucket); 257 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets)); 258 ID.clear(); 259 } 260 } 261 262 delete[] OldBuckets; 263} 264 265/// FindNodeOrInsertPos - Look up the node specified by ID. If it exists, 266/// return it. If not, return the insertion token that will make insertion 267/// faster. 268FoldingSetImpl::Node 269*FoldingSetImpl::FindNodeOrInsertPos(const FoldingSetNodeID &ID, 270 void *&InsertPos) { 271 272 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets); 273 void *Probe = *Bucket; 274 275 InsertPos = 0; 276 277 FoldingSetNodeID OtherID; 278 while (Node *NodeInBucket = GetNextPtr(Probe)) { 279 GetNodeProfile(OtherID, NodeInBucket); 280 if (OtherID == ID) 281 return NodeInBucket; 282 283 Probe = NodeInBucket->getNextInBucket(); 284 OtherID.clear(); 285 } 286 287 // Didn't find the node, return null with the bucket as the InsertPos. 288 InsertPos = Bucket; 289 return 0; 290} 291 292/// InsertNode - Insert the specified node into the folding set, knowing that it 293/// is not already in the map. InsertPos must be obtained from 294/// FindNodeOrInsertPos. 295void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) { 296 assert(N->getNextInBucket() == 0); 297 // Do we need to grow the hashtable? 298 if (NumNodes+1 > NumBuckets*2) { 299 GrowHashTable(); 300 FoldingSetNodeID ID; 301 GetNodeProfile(ID, N); 302 InsertPos = GetBucketFor(ID, Buckets, NumBuckets); 303 } 304 305 ++NumNodes; 306 307 /// The insert position is actually a bucket pointer. 308 void **Bucket = static_cast<void**>(InsertPos); 309 310 void *Next = *Bucket; 311 312 // If this is the first insertion into this bucket, its next pointer will be 313 // null. Pretend as if it pointed to itself, setting the low bit to indicate 314 // that it is a pointer to the bucket. 315 if (Next == 0) 316 Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1); 317 318 // Set the node's next pointer, and make the bucket point to the node. 319 N->SetNextInBucket(Next); 320 *Bucket = N; 321} 322 323/// RemoveNode - Remove a node from the folding set, returning true if one was 324/// removed or false if the node was not in the folding set. 325bool FoldingSetImpl::RemoveNode(Node *N) { 326 // Because each bucket is a circular list, we don't need to compute N's hash 327 // to remove it. 328 void *Ptr = N->getNextInBucket(); 329 if (Ptr == 0) return false; // Not in folding set. 330 331 --NumNodes; 332 N->SetNextInBucket(0); 333 334 // Remember what N originally pointed to, either a bucket or another node. 335 void *NodeNextPtr = Ptr; 336 337 // Chase around the list until we find the node (or bucket) which points to N. 338 while (true) { 339 if (Node *NodeInBucket = GetNextPtr(Ptr)) { 340 // Advance pointer. 341 Ptr = NodeInBucket->getNextInBucket(); 342 343 // We found a node that points to N, change it to point to N's next node, 344 // removing N from the list. 345 if (Ptr == N) { 346 NodeInBucket->SetNextInBucket(NodeNextPtr); 347 return true; 348 } 349 } else { 350 void **Bucket = GetBucketPtr(Ptr); 351 Ptr = *Bucket; 352 353 // If we found that the bucket points to N, update the bucket to point to 354 // whatever is next. 355 if (Ptr == N) { 356 *Bucket = NodeNextPtr; 357 return true; 358 } 359 } 360 } 361} 362 363/// GetOrInsertNode - If there is an existing simple Node exactly 364/// equal to the specified node, return it. Otherwise, insert 'N' and it 365/// instead. 366FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) { 367 FoldingSetNodeID ID; 368 GetNodeProfile(ID, N); 369 void *IP; 370 if (Node *E = FindNodeOrInsertPos(ID, IP)) 371 return E; 372 InsertNode(N, IP); 373 return N; 374} 375 376//===----------------------------------------------------------------------===// 377// FoldingSetIteratorImpl Implementation 378 379FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) { 380 // Skip to the first non-null non-self-cycle bucket. 381 while (*Bucket != reinterpret_cast<void*>(-1) && 382 (*Bucket == 0 || GetNextPtr(*Bucket) == 0)) 383 ++Bucket; 384 385 NodePtr = static_cast<FoldingSetNode*>(*Bucket); 386} 387 388void FoldingSetIteratorImpl::advance() { 389 // If there is another link within this bucket, go to it. 390 void *Probe = NodePtr->getNextInBucket(); 391 392 if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe)) 393 NodePtr = NextNodeInBucket; 394 else { 395 // Otherwise, this is the last link in this bucket. 396 void **Bucket = GetBucketPtr(Probe); 397 398 // Skip to the next non-null non-self-cycle bucket. 399 do { 400 ++Bucket; 401 } while (*Bucket != reinterpret_cast<void*>(-1) && 402 (*Bucket == 0 || GetNextPtr(*Bucket) == 0)); 403 404 NodePtr = static_cast<FoldingSetNode*>(*Bucket); 405 } 406} 407 408//===----------------------------------------------------------------------===// 409// FoldingSetBucketIteratorImpl Implementation 410 411FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) { 412 Ptr = (*Bucket == 0 || GetNextPtr(*Bucket) == 0) ? (void*) Bucket : *Bucket; 413} 414