lib/Support/FoldingSet.cpp

   1 //===-- Support/FoldingSet.cpp - Uniquing Hash Set --------------*- C++ -*-===//
   2 //
   3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
   4 // See https://llvm.org/LICENSE.txt for license information.
   5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
   6 //
   7 //===----------------------------------------------------------------------===//
   8 //
   9 // This file implements a hash set that can be used to remove duplication of
  10 // nodes in a graph.
  11 //
  12 //===----------------------------------------------------------------------===//
  13
  14 #include "llvm/ADT/FoldingSet.h"
  15 #include "llvm/ADT/Hashing.h"
  16 #include "llvm/Support/Allocator.h"
  17 #include "llvm/Support/ErrorHandling.h"
  18 #include "llvm/Support/Host.h"
  19 #include "llvm/Support/MathExtras.h"
  20 #include <cassert>
  21 #include <cstring>
  22 using namespace llvm;
  23
  24 //===----------------------------------------------------------------------===//
  25 // FoldingSetNodeIDRef Implementation
  26
  27 /// ComputeHash - Compute a strong hash value for this FoldingSetNodeIDRef,
  28 /// used to lookup the node in the FoldingSetBase.
  29 unsigned FoldingSetNodeIDRef::ComputeHash() const {
  30   return static_cast<unsigned>(hash_combine_range(Data, Data+Size));
  31 }
  32
  33 bool FoldingSetNodeIDRef::operator==(FoldingSetNodeIDRef RHS) const {
  34   if (Size != RHS.Size) return false;
  35   return memcmp(Data, RHS.Data, Size*sizeof(*Data)) == 0;
  36 }
  37
  38 /// Used to compare the "ordering" of two nodes as defined by the
  39 /// profiled bits and their ordering defined by memcmp().
  40 bool FoldingSetNodeIDRef::operator<(FoldingSetNodeIDRef RHS) const {
  41   if (Size != RHS.Size)
  42     return Size < RHS.Size;
  43   return memcmp(Data, RHS.Data, Size*sizeof(*Data)) < 0;
  44 }
  45
  46 //===----------------------------------------------------------------------===//
  47 // FoldingSetNodeID Implementation
  48
  49 /// Add* - Add various data types to Bit data.
  50 ///
  51 void FoldingSetNodeID::AddPointer(const void *Ptr) {
  52   // Note: this adds pointers to the hash using sizes and endianness that
  53   // depend on the host. It doesn't matter, however, because hashing on
  54   // pointer values is inherently unstable. Nothing should depend on the
  55   // ordering of nodes in the folding set.
  56   static_assert(sizeof(uintptr_t) <= sizeof(unsigned long long),
  57                 "unexpected pointer size");
  58   AddInteger(reinterpret_cast<uintptr_t>(Ptr));
  59 }
  60 void FoldingSetNodeID::AddInteger(signed I) {
  61   Bits.push_back(I);
  62 }
  63 void FoldingSetNodeID::AddInteger(unsigned I) {
  64   Bits.push_back(I);
  65 }
  66 void FoldingSetNodeID::AddInteger(long I) {
  67   AddInteger((unsigned long)I);
  68 }
  69 void FoldingSetNodeID::AddInteger(unsigned long I) {
  70   if (sizeof(long) == sizeof(int))
  71     AddInteger(unsigned(I));
  72   else if (sizeof(long) == sizeof(long long)) {
  73     AddInteger((unsigned long long)I);
  74   } else {
  75     llvm_unreachable("unexpected sizeof(long)");
  76   }
  77 }
  78 void FoldingSetNodeID::AddInteger(long long I) {
  79   AddInteger((unsigned long long)I);
  80 }
  81 void FoldingSetNodeID::AddInteger(unsigned long long I) {
  82   AddInteger(unsigned(I));
  83   AddInteger(unsigned(I >> 32));
  84 }
  85
  86 void FoldingSetNodeID::AddString(StringRef String) {
  87   unsigned Size =  String.size();
  88   Bits.push_back(Size);
  89   if (!Size) return;
  90
  91   unsigned Units = Size / 4;
  92   unsigned Pos = 0;
  93   const unsigned *Base = (const unsigned*) String.data();
  94
  95   // If the string is aligned do a bulk transfer.
  96   if (!((intptr_t)Base & 3)) {
  97     Bits.append(Base, Base + Units);
  98     Pos = (Units + 1) * 4;
  99   } else {
 100     // Otherwise do it the hard way.
 101     // To be compatible with above bulk transfer, we need to take endianness
 102     // into account.
 103     static_assert(sys::IsBigEndianHost || sys::IsLittleEndianHost,
 104                   "Unexpected host endianness");
 105     if (sys::IsBigEndianHost) {
 106       for (Pos += 4; Pos <= Size; Pos += 4) {
 107         unsigned V = ((unsigned char)String[Pos - 4] << 24) |
 108                      ((unsigned char)String[Pos - 3] << 16) |
 109                      ((unsigned char)String[Pos - 2] << 8) |
 110                       (unsigned char)String[Pos - 1];
 111         Bits.push_back(V);
 112       }
 113     } else {  // Little-endian host
 114       for (Pos += 4; Pos <= Size; Pos += 4) {
 115         unsigned V = ((unsigned char)String[Pos - 1] << 24) |
 116                      ((unsigned char)String[Pos - 2] << 16) |
 117                      ((unsigned char)String[Pos - 3] << 8) |
 118                       (unsigned char)String[Pos - 4];
 119         Bits.push_back(V);
 120       }
 121     }
 122   }
 123
 124   // With the leftover bits.
 125   unsigned V = 0;
 126   // Pos will have overshot size by 4 - #bytes left over.
 127   // No need to take endianness into account here - this is always executed.
 128   switch (Pos - Size) {
 129   case 1: V = (V << 8) | (unsigned char)String[Size - 3]; LLVM_FALLTHROUGH;
 130   case 2: V = (V << 8) | (unsigned char)String[Size - 2]; LLVM_FALLTHROUGH;
 131   case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
 132   default: return; // Nothing left.
 133   }
 134
 135   Bits.push_back(V);
 136 }
 137
 138 // AddNodeID - Adds the Bit data of another ID to *this.
 139 void FoldingSetNodeID::AddNodeID(const FoldingSetNodeID &ID) {
 140   Bits.append(ID.Bits.begin(), ID.Bits.end());
 141 }
 142
 143 /// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to
 144 /// lookup the node in the FoldingSetBase.
 145 unsigned FoldingSetNodeID::ComputeHash() const {
 146   return FoldingSetNodeIDRef(Bits.data(), Bits.size()).ComputeHash();
 147 }
 148
 149 /// operator== - Used to compare two nodes to each other.
 150 ///
 151 bool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS) const {
 152   return *this == FoldingSetNodeIDRef(RHS.Bits.data(), RHS.Bits.size());
 153 }
 154
 155 /// operator== - Used to compare two nodes to each other.
 156 ///
 157 bool FoldingSetNodeID::operator==(FoldingSetNodeIDRef RHS) const {
 158   return FoldingSetNodeIDRef(Bits.data(), Bits.size()) == RHS;
 159 }
 160
 161 /// Used to compare the "ordering" of two nodes as defined by the
 162 /// profiled bits and their ordering defined by memcmp().
 163 bool FoldingSetNodeID::operator<(const FoldingSetNodeID &RHS) const {
 164   return *this < FoldingSetNodeIDRef(RHS.Bits.data(), RHS.Bits.size());
 165 }
 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 FoldingSetBase.
 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 FoldingSetBase::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 nullptr;
 194
 195   return static_cast<FoldingSetBase::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**>(safe_calloc(NumBuckets + 1,
 217                                                    sizeof(void*)));
 218   // Set the very last bucket to be a non-null "pointer".
 219   Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
 220   return Buckets;
 221 }
 222
 223 //===----------------------------------------------------------------------===//
 224 // FoldingSetBase Implementation
 225
 226 void FoldingSetBase::anchor() {}
 227
 228 FoldingSetBase::FoldingSetBase(unsigned Log2InitSize) {
 229   assert(5 < Log2InitSize && Log2InitSize < 32 &&
 230          "Initial hash table size out of range");
 231   NumBuckets = 1 << Log2InitSize;
 232   Buckets = AllocateBuckets(NumBuckets);
 233   NumNodes = 0;
 234 }
 235
 236 FoldingSetBase::FoldingSetBase(FoldingSetBase &&Arg)
 237     : Buckets(Arg.Buckets), NumBuckets(Arg.NumBuckets), NumNodes(Arg.NumNodes) {
 238   Arg.Buckets = nullptr;
 239   Arg.NumBuckets = 0;
 240   Arg.NumNodes = 0;
 241 }
 242
 243 FoldingSetBase &FoldingSetBase::operator=(FoldingSetBase &&RHS) {
 244   free(Buckets); // This may be null if the set is in a moved-from state.
 245   Buckets = RHS.Buckets;
 246   NumBuckets = RHS.NumBuckets;
 247   NumNodes = RHS.NumNodes;
 248   RHS.Buckets = nullptr;
 249   RHS.NumBuckets = 0;
 250   RHS.NumNodes = 0;
 251   return *this;
 252 }
 253
 254 FoldingSetBase::~FoldingSetBase() {
 255   free(Buckets);
 256 }
 257
 258 void FoldingSetBase::clear() {
 259   // Set all but the last bucket to null pointers.
 260   memset(Buckets, 0, NumBuckets*sizeof(void*));
 261
 262   // Set the very last bucket to be a non-null "pointer".
 263   Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
 264
 265   // Reset the node count to zero.
 266   NumNodes = 0;
 267 }
 268
 269 void FoldingSetBase::GrowBucketCount(unsigned NewBucketCount) {
 270   assert((NewBucketCount > NumBuckets) && "Can't shrink a folding set with GrowBucketCount");
 271   assert(isPowerOf2_32(NewBucketCount) && "Bad bucket count!");
 272   void **OldBuckets = Buckets;
 273   unsigned OldNumBuckets = NumBuckets;
 274
 275   // Clear out new buckets.
 276   Buckets = AllocateBuckets(NewBucketCount);
 277   // Set NumBuckets only if allocation of new buckets was successful.
 278   NumBuckets = NewBucketCount;
 279   NumNodes = 0;
 280
 281   // Walk the old buckets, rehashing nodes into their new place.
 282   FoldingSetNodeID TempID;
 283   for (unsigned i = 0; i != OldNumBuckets; ++i) {
 284     void *Probe = OldBuckets[i];
 285     if (!Probe) continue;
 286     while (Node *NodeInBucket = GetNextPtr(Probe)) {
 287       // Figure out the next link, remove NodeInBucket from the old link.
 288       Probe = NodeInBucket->getNextInBucket();
 289       NodeInBucket->SetNextInBucket(nullptr);
 290
 291       // Insert the node into the new bucket, after recomputing the hash.
 292       InsertNode(NodeInBucket,
 293                  GetBucketFor(ComputeNodeHash(NodeInBucket, TempID),
 294                               Buckets, NumBuckets));
 295       TempID.clear();
 296     }
 297   }
 298
 299   free(OldBuckets);
 300 }
 301
 302 /// GrowHashTable - Double the size of the hash table and rehash everything.
 303 ///
 304 void FoldingSetBase::GrowHashTable() {
 305   GrowBucketCount(NumBuckets * 2);
 306 }
 307
 308 void FoldingSetBase::reserve(unsigned EltCount) {
 309   // This will give us somewhere between EltCount / 2 and
 310   // EltCount buckets.  This puts us in the load factor
 311   // range of 1.0 - 2.0.
 312   if(EltCount < capacity())
 313     return;
 314   GrowBucketCount(PowerOf2Floor(EltCount));
 315 }
 316
 317 /// FindNodeOrInsertPos - Look up the node specified by ID.  If it exists,
 318 /// return it.  If not, return the insertion token that will make insertion
 319 /// faster.
 320 FoldingSetBase::Node *
 321 FoldingSetBase::FindNodeOrInsertPos(const FoldingSetNodeID &ID,
 322                                     void *&InsertPos) {
 323   unsigned IDHash = ID.ComputeHash();
 324   void **Bucket = GetBucketFor(IDHash, Buckets, NumBuckets);
 325   void *Probe = *Bucket;
 326
 327   InsertPos = nullptr;
 328
 329   FoldingSetNodeID TempID;
 330   while (Node *NodeInBucket = GetNextPtr(Probe)) {
 331     if (NodeEquals(NodeInBucket, ID, IDHash, TempID))
 332       return NodeInBucket;
 333     TempID.clear();
 334
 335     Probe = NodeInBucket->getNextInBucket();
 336   }
 337
 338   // Didn't find the node, return null with the bucket as the InsertPos.
 339   InsertPos = Bucket;
 340   return nullptr;
 341 }
 342
 343 /// InsertNode - Insert the specified node into the folding set, knowing that it
 344 /// is not already in the map.  InsertPos must be obtained from
 345 /// FindNodeOrInsertPos.
 346 void FoldingSetBase::InsertNode(Node *N, void *InsertPos) {
 347   assert(!N->getNextInBucket());
 348   // Do we need to grow the hashtable?
 349   if (NumNodes+1 > capacity()) {
 350     GrowHashTable();
 351     FoldingSetNodeID TempID;
 352     InsertPos = GetBucketFor(ComputeNodeHash(N, TempID), Buckets, NumBuckets);
 353   }
 354
 355   ++NumNodes;
 356
 357   /// The insert position is actually a bucket pointer.
 358   void **Bucket = static_cast<void**>(InsertPos);
 359
 360   void *Next = *Bucket;
 361
 362   // If this is the first insertion into this bucket, its next pointer will be
 363   // null.  Pretend as if it pointed to itself, setting the low bit to indicate
 364   // that it is a pointer to the bucket.
 365   if (!Next)
 366     Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1);
 367
 368   // Set the node's next pointer, and make the bucket point to the node.
 369   N->SetNextInBucket(Next);
 370   *Bucket = N;
 371 }
 372
 373 /// RemoveNode - Remove a node from the folding set, returning true if one was
 374 /// removed or false if the node was not in the folding set.
 375 bool FoldingSetBase::RemoveNode(Node *N) {
 376   // Because each bucket is a circular list, we don't need to compute N's hash
 377   // to remove it.
 378   void *Ptr = N->getNextInBucket();
 379   if (!Ptr) return false;  // Not in folding set.
 380
 381   --NumNodes;
 382   N->SetNextInBucket(nullptr);
 383
 384   // Remember what N originally pointed to, either a bucket or another node.
 385   void *NodeNextPtr = Ptr;
 386
 387   // Chase around the list until we find the node (or bucket) which points to N.
 388   while (true) {
 389     if (Node *NodeInBucket = GetNextPtr(Ptr)) {
 390       // Advance pointer.
 391       Ptr = NodeInBucket->getNextInBucket();
 392
 393       // We found a node that points to N, change it to point to N's next node,
 394       // removing N from the list.
 395       if (Ptr == N) {
 396         NodeInBucket->SetNextInBucket(NodeNextPtr);
 397         return true;
 398       }
 399     } else {
 400       void **Bucket = GetBucketPtr(Ptr);
 401       Ptr = *Bucket;
 402
 403       // If we found that the bucket points to N, update the bucket to point to
 404       // whatever is next.
 405       if (Ptr == N) {
 406         *Bucket = NodeNextPtr;
 407         return true;
 408       }
 409     }
 410   }
 411 }
 412
 413 /// GetOrInsertNode - If there is an existing simple Node exactly
 414 /// equal to the specified node, return it.  Otherwise, insert 'N' and it
 415 /// instead.
 416 FoldingSetBase::Node *FoldingSetBase::GetOrInsertNode(FoldingSetBase::Node *N) {
 417   FoldingSetNodeID ID;
 418   GetNodeProfile(N, ID);
 419   void *IP;
 420   if (Node *E = FindNodeOrInsertPos(ID, IP))
 421     return E;
 422   InsertNode(N, IP);
 423   return N;
 424 }
 425
 426 //===----------------------------------------------------------------------===//
 427 // FoldingSetIteratorImpl Implementation
 428
 429 FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) {
 430   // Skip to the first non-null non-self-cycle bucket.
 431   while (*Bucket != reinterpret_cast<void*>(-1) &&
 432          (!*Bucket || !GetNextPtr(*Bucket)))
 433     ++Bucket;
 434
 435   NodePtr = static_cast<FoldingSetNode*>(*Bucket);
 436 }
 437
 438 void FoldingSetIteratorImpl::advance() {
 439   // If there is another link within this bucket, go to it.
 440   void *Probe = NodePtr->getNextInBucket();
 441
 442   if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe))
 443     NodePtr = NextNodeInBucket;
 444   else {
 445     // Otherwise, this is the last link in this bucket.
 446     void **Bucket = GetBucketPtr(Probe);
 447
 448     // Skip to the next non-null non-self-cycle bucket.
 449     do {
 450       ++Bucket;
 451     } while (*Bucket != reinterpret_cast<void*>(-1) &&
 452              (!*Bucket || !GetNextPtr(*Bucket)));
 453
 454     NodePtr = static_cast<FoldingSetNode*>(*Bucket);
 455   }
 456 }
 457
 458 //===----------------------------------------------------------------------===//
 459 // FoldingSetBucketIteratorImpl Implementation
 460
 461 FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) {
 462   Ptr = (!*Bucket || !GetNextPtr(*Bucket)) ? (void*) Bucket : *Bucket;
 463 }