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