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/Allocator.h" 19 #include "llvm/Support/ErrorHandling.h" 20 #include "llvm/Support/MathExtras.h" 21 #include "llvm/Support/Host.h" 22 #include <cassert> 23 #include <cstring> 24 using namespace llvm; 25 26 //===----------------------------------------------------------------------===// 27 // FoldingSetNodeIDRef Implementation 28 29 /// ComputeHash - Compute a strong hash value for this FoldingSetNodeIDRef, 30 /// used to lookup the node in the FoldingSetImpl. 31 unsigned FoldingSetNodeIDRef::ComputeHash() const { 32 // This is adapted from SuperFastHash by Paul Hsieh. 33 unsigned Hash = static_cast<unsigned>(Size); 34 for (const unsigned *BP = Data, *E = BP+Size; BP != E; ++BP) { 35 unsigned Data = *BP; 36 Hash += Data & 0xFFFF; 37 unsigned Tmp = ((Data >> 16) << 11) ^ Hash; 38 Hash = (Hash << 16) ^ Tmp; 39 Hash += Hash >> 11; 40 } 41 42 // Force "avalanching" of final 127 bits. 43 Hash ^= Hash << 3; 44 Hash += Hash >> 5; 45 Hash ^= Hash << 4; 46 Hash += Hash >> 17; 47 Hash ^= Hash << 25; 48 Hash += Hash >> 6; 49 return Hash; 50 } 51 52 bool FoldingSetNodeIDRef::operator==(FoldingSetNodeIDRef RHS) const { 53 if (Size != RHS.Size) return false; 54 return memcmp(Data, RHS.Data, Size*sizeof(*Data)) == 0; 55 } 56 57 //===----------------------------------------------------------------------===// 58 // FoldingSetNodeID Implementation 59 60 /// Add* - Add various data types to Bit data. 61 /// 62 void FoldingSetNodeID::AddPointer(const void *Ptr) { 63 // Note: this adds pointers to the hash using sizes and endianness that 64 // depend on the host. It doesn't matter however, because hashing on 65 // pointer values in inherently unstable. Nothing should depend on the 66 // ordering of nodes in the folding set. 67 Bits.append(reinterpret_cast<unsigned *>(&Ptr), 68 reinterpret_cast<unsigned *>(&Ptr+1)); 69 } 70 void FoldingSetNodeID::AddInteger(signed I) { 71 Bits.push_back(I); 72 } 73 void FoldingSetNodeID::AddInteger(unsigned I) { 74 Bits.push_back(I); 75 } 76 void FoldingSetNodeID::AddInteger(long I) { 77 AddInteger((unsigned long)I); 78 } 79 void FoldingSetNodeID::AddInteger(unsigned long I) { 80 if (sizeof(long) == sizeof(int)) 81 AddInteger(unsigned(I)); 82 else if (sizeof(long) == sizeof(long long)) { 83 AddInteger((unsigned long long)I); 84 } else { 85 llvm_unreachable("unexpected sizeof(long)"); 86 } 87 } 88 void FoldingSetNodeID::AddInteger(long long I) { 89 AddInteger((unsigned long long)I); 90 } 91 void FoldingSetNodeID::AddInteger(unsigned long long I) { 92 AddInteger(unsigned(I)); 93 if ((uint64_t)(unsigned)I != I) 94 Bits.push_back(unsigned(I >> 32)); 95 } 96 97 void FoldingSetNodeID::AddString(StringRef String) { 98 unsigned Size = String.size(); 99 Bits.push_back(Size); 100 if (!Size) return; 101 102 unsigned Units = Size / 4; 103 unsigned Pos = 0; 104 const unsigned *Base = (const unsigned*) String.data(); 105 106 // If the string is aligned do a bulk transfer. 107 if (!((intptr_t)Base & 3)) { 108 Bits.append(Base, Base + Units); 109 Pos = (Units + 1) * 4; 110 } else { 111 // Otherwise do it the hard way. 112 // To be compatible with above bulk transfer, we need to take endianness 113 // into account. 114 if (sys::isBigEndianHost()) { 115 for (Pos += 4; Pos <= Size; Pos += 4) { 116 unsigned V = ((unsigned char)String[Pos - 4] << 24) | 117 ((unsigned char)String[Pos - 3] << 16) | 118 ((unsigned char)String[Pos - 2] << 8) | 119 (unsigned char)String[Pos - 1]; 120 Bits.push_back(V); 121 } 122 } else { 123 assert(sys::isLittleEndianHost() && "Unexpected host endianness"); 124 for (Pos += 4; Pos <= Size; Pos += 4) { 125 unsigned V = ((unsigned char)String[Pos - 1] << 24) | 126 ((unsigned char)String[Pos - 2] << 16) | 127 ((unsigned char)String[Pos - 3] << 8) | 128 (unsigned char)String[Pos - 4]; 129 Bits.push_back(V); 130 } 131 } 132 } 133 134 // With the leftover bits. 135 unsigned V = 0; 136 // Pos will have overshot size by 4 - #bytes left over. 137 // No need to take endianness into account here - this is always executed. 138 switch (Pos - Size) { 139 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru. 140 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru. 141 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break; 142 default: return; // Nothing left. 143 } 144 145 Bits.push_back(V); 146 } 147 148 // AddNodeID - Adds the Bit data of another ID to *this. 149 void FoldingSetNodeID::AddNodeID(const FoldingSetNodeID &ID) { 150 Bits.append(ID.Bits.begin(), ID.Bits.end()); 151 } 152 153 /// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to 154 /// lookup the node in the FoldingSetImpl. 155 unsigned FoldingSetNodeID::ComputeHash() const { 156 return FoldingSetNodeIDRef(Bits.data(), Bits.size()).ComputeHash(); 157 } 158 159 /// operator== - Used to compare two nodes to each other. 160 /// 161 bool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS)const{ 162 return *this == FoldingSetNodeIDRef(RHS.Bits.data(), RHS.Bits.size()); 163 } 164 165 /// operator== - Used to compare two nodes to each other. 166 /// 167 bool FoldingSetNodeID::operator==(FoldingSetNodeIDRef RHS) const { 168 return FoldingSetNodeIDRef(Bits.data(), Bits.size()) == RHS; 169 } 170 171 /// Intern - Copy this node's data to a memory region allocated from the 172 /// given allocator and return a FoldingSetNodeIDRef describing the 173 /// interned data. 174 FoldingSetNodeIDRef 175 FoldingSetNodeID::Intern(BumpPtrAllocator &Allocator) const { 176 unsigned *New = Allocator.Allocate<unsigned>(Bits.size()); 177 std::uninitialized_copy(Bits.begin(), Bits.end(), New); 178 return FoldingSetNodeIDRef(New, Bits.size()); 179 } 180 181 //===----------------------------------------------------------------------===// 182 /// Helper functions for FoldingSetImpl. 183 184 /// GetNextPtr - In order to save space, each bucket is a 185 /// singly-linked-list. In order to make deletion more efficient, we make 186 /// the list circular, so we can delete a node without computing its hash. 187 /// The problem with this is that the start of the hash buckets are not 188 /// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null: 189 /// use GetBucketPtr when this happens. 190 static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr) { 191 // The low bit is set if this is the pointer back to the bucket. 192 if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1) 193 return 0; 194 195 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr); 196 } 197 198 199 /// testing. 200 static void **GetBucketPtr(void *NextInBucketPtr) { 201 intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr); 202 assert((Ptr & 1) && "Not a bucket pointer"); 203 return reinterpret_cast<void**>(Ptr & ~intptr_t(1)); 204 } 205 206 /// GetBucketFor - Hash the specified node ID and return the hash bucket for 207 /// the specified ID. 208 static void **GetBucketFor(unsigned Hash, void **Buckets, unsigned NumBuckets) { 209 // NumBuckets is always a power of 2. 210 unsigned BucketNum = Hash & (NumBuckets-1); 211 return Buckets + BucketNum; 212 } 213 214 /// AllocateBuckets - Allocated initialized bucket memory. 215 static void **AllocateBuckets(unsigned NumBuckets) { 216 void **Buckets = static_cast<void**>(calloc(NumBuckets+1, sizeof(void*))); 217 // Set the very last bucket to be a non-null "pointer". 218 Buckets[NumBuckets] = reinterpret_cast<void*>(-1); 219 return Buckets; 220 } 221 222 //===----------------------------------------------------------------------===// 223 // FoldingSetImpl Implementation 224 225 FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) { 226 assert(5 < Log2InitSize && Log2InitSize < 32 && 227 "Initial hash table size out of range"); 228 NumBuckets = 1 << Log2InitSize; 229 Buckets = AllocateBuckets(NumBuckets); 230 NumNodes = 0; 231 } 232 FoldingSetImpl::~FoldingSetImpl() { 233 free(Buckets); 234 } 235 void FoldingSetImpl::clear() { 236 // Set all but the last bucket to null pointers. 237 memset(Buckets, 0, NumBuckets*sizeof(void*)); 238 239 // Set the very last bucket to be a non-null "pointer". 240 Buckets[NumBuckets] = reinterpret_cast<void*>(-1); 241 242 // Reset the node count to zero. 243 NumNodes = 0; 244 } 245 246 /// GrowHashTable - Double the size of the hash table and rehash everything. 247 /// 248 void FoldingSetImpl::GrowHashTable() { 249 void **OldBuckets = Buckets; 250 unsigned OldNumBuckets = NumBuckets; 251 NumBuckets <<= 1; 252 253 // Clear out new buckets. 254 Buckets = AllocateBuckets(NumBuckets); 255 NumNodes = 0; 256 257 // Walk the old buckets, rehashing nodes into their new place. 258 FoldingSetNodeID TempID; 259 for (unsigned i = 0; i != OldNumBuckets; ++i) { 260 void *Probe = OldBuckets[i]; 261 if (!Probe) continue; 262 while (Node *NodeInBucket = GetNextPtr(Probe)) { 263 // Figure out the next link, remove NodeInBucket from the old link. 264 Probe = NodeInBucket->getNextInBucket(); 265 NodeInBucket->SetNextInBucket(0); 266 267 // Insert the node into the new bucket, after recomputing the hash. 268 InsertNode(NodeInBucket, 269 GetBucketFor(ComputeNodeHash(NodeInBucket, TempID), 270 Buckets, NumBuckets)); 271 TempID.clear(); 272 } 273 } 274 275 free(OldBuckets); 276 } 277 278 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists, 279 /// return it. If not, return the insertion token that will make insertion 280 /// faster. 281 FoldingSetImpl::Node 282 *FoldingSetImpl::FindNodeOrInsertPos(const FoldingSetNodeID &ID, 283 void *&InsertPos) { 284 285 void **Bucket = GetBucketFor(ID.ComputeHash(), Buckets, NumBuckets); 286 void *Probe = *Bucket; 287 288 InsertPos = 0; 289 290 FoldingSetNodeID TempID; 291 while (Node *NodeInBucket = GetNextPtr(Probe)) { 292 if (NodeEquals(NodeInBucket, ID, TempID)) 293 return NodeInBucket; 294 TempID.clear(); 295 296 Probe = NodeInBucket->getNextInBucket(); 297 } 298 299 // Didn't find the node, return null with the bucket as the InsertPos. 300 InsertPos = Bucket; 301 return 0; 302 } 303 304 /// InsertNode - Insert the specified node into the folding set, knowing that it 305 /// is not already in the map. InsertPos must be obtained from 306 /// FindNodeOrInsertPos. 307 void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) { 308 assert(N->getNextInBucket() == 0); 309 // Do we need to grow the hashtable? 310 if (NumNodes+1 > NumBuckets*2) { 311 GrowHashTable(); 312 FoldingSetNodeID TempID; 313 InsertPos = GetBucketFor(ComputeNodeHash(N, TempID), Buckets, NumBuckets); 314 } 315 316 ++NumNodes; 317 318 /// The insert position is actually a bucket pointer. 319 void **Bucket = static_cast<void**>(InsertPos); 320 321 void *Next = *Bucket; 322 323 // If this is the first insertion into this bucket, its next pointer will be 324 // null. Pretend as if it pointed to itself, setting the low bit to indicate 325 // that it is a pointer to the bucket. 326 if (Next == 0) 327 Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1); 328 329 // Set the node's next pointer, and make the bucket point to the node. 330 N->SetNextInBucket(Next); 331 *Bucket = N; 332 } 333 334 /// RemoveNode - Remove a node from the folding set, returning true if one was 335 /// removed or false if the node was not in the folding set. 336 bool FoldingSetImpl::RemoveNode(Node *N) { 337 // Because each bucket is a circular list, we don't need to compute N's hash 338 // to remove it. 339 void *Ptr = N->getNextInBucket(); 340 if (Ptr == 0) return false; // Not in folding set. 341 342 --NumNodes; 343 N->SetNextInBucket(0); 344 345 // Remember what N originally pointed to, either a bucket or another node. 346 void *NodeNextPtr = Ptr; 347 348 // Chase around the list until we find the node (or bucket) which points to N. 349 while (true) { 350 if (Node *NodeInBucket = GetNextPtr(Ptr)) { 351 // Advance pointer. 352 Ptr = NodeInBucket->getNextInBucket(); 353 354 // We found a node that points to N, change it to point to N's next node, 355 // removing N from the list. 356 if (Ptr == N) { 357 NodeInBucket->SetNextInBucket(NodeNextPtr); 358 return true; 359 } 360 } else { 361 void **Bucket = GetBucketPtr(Ptr); 362 Ptr = *Bucket; 363 364 // If we found that the bucket points to N, update the bucket to point to 365 // whatever is next. 366 if (Ptr == N) { 367 *Bucket = NodeNextPtr; 368 return true; 369 } 370 } 371 } 372 } 373 374 /// GetOrInsertNode - If there is an existing simple Node exactly 375 /// equal to the specified node, return it. Otherwise, insert 'N' and it 376 /// instead. 377 FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) { 378 FoldingSetNodeID ID; 379 GetNodeProfile(N, ID); 380 void *IP; 381 if (Node *E = FindNodeOrInsertPos(ID, IP)) 382 return E; 383 InsertNode(N, IP); 384 return N; 385 } 386 387 //===----------------------------------------------------------------------===// 388 // FoldingSetIteratorImpl Implementation 389 390 FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) { 391 // Skip to the first non-null non-self-cycle bucket. 392 while (*Bucket != reinterpret_cast<void*>(-1) && 393 (*Bucket == 0 || GetNextPtr(*Bucket) == 0)) 394 ++Bucket; 395 396 NodePtr = static_cast<FoldingSetNode*>(*Bucket); 397 } 398 399 void FoldingSetIteratorImpl::advance() { 400 // If there is another link within this bucket, go to it. 401 void *Probe = NodePtr->getNextInBucket(); 402 403 if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe)) 404 NodePtr = NextNodeInBucket; 405 else { 406 // Otherwise, this is the last link in this bucket. 407 void **Bucket = GetBucketPtr(Probe); 408 409 // Skip to the next non-null non-self-cycle bucket. 410 do { 411 ++Bucket; 412 } while (*Bucket != reinterpret_cast<void*>(-1) && 413 (*Bucket == 0 || GetNextPtr(*Bucket) == 0)); 414 415 NodePtr = static_cast<FoldingSetNode*>(*Bucket); 416 } 417 } 418 419 //===----------------------------------------------------------------------===// 420 // FoldingSetBucketIteratorImpl Implementation 421 422 FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) { 423 Ptr = (*Bucket == 0 || GetNextPtr(*Bucket) == 0) ? (void*) Bucket : *Bucket; 424 } 425