lib/Support/FoldingSet.cpp

   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.
  12 //
  13 //===----------------------------------------------------------------------===//
  14
  15 #include "llvm/ADT/FoldingSet.h"
  16 #include "llvm/ADT/Hashing.h"
  17 #include "llvm/Support/Allocator.h"
  18 #include "llvm/Support/ErrorHandling.h"
  19 #include "llvm/Support/Host.h"
  20 #include "llvm/Support/MathExtras.h"
  21 #include <cassert>
  22 #include <cstring>
  23 using namespace llvm;
  24
  25 //===----------------------------------------------------------------------===//
  26 // FoldingSetNodeIDRef Implementation
  27
  28 /// ComputeHash - Compute a strong hash value for this FoldingSetNodeIDRef,
  29 /// used to lookup the node in the FoldingSetBase.
  30 unsigned FoldingSetNodeIDRef::ComputeHash() const {
  31   return static_cast<unsigned>(hash_combine_range(Data, Data+Size));
  32 }
  33
  34 bool FoldingSetNodeIDRef::operator==(FoldingSetNodeIDRef RHS) const {
  35   if (Size != RHS.Size) return false;
  36   return memcmp(Data, RHS.Data, Size*sizeof(*Data)) == 0;
  37 }
  38
  39 /// Used to compare the "ordering" of two nodes as defined by the
  40 /// profiled bits and their ordering defined by memcmp().
  41 bool FoldingSetNodeIDRef::operator<(FoldingSetNodeIDRef RHS) const {
  42   if (Size != RHS.Size)
  43     return Size < RHS.Size;
  44   return memcmp(Data, RHS.Data, Size*sizeof(*Data)) < 0;
  45 }
  46
  47 //===----------------------------------------------------------------------===//
  48 // FoldingSetNodeID Implementation
  49
  50 /// Add* - Add various data types to Bit data.
  51 ///
  52 void FoldingSetNodeID::AddPointer(const void *Ptr) {
  53   // Note: this adds pointers to the hash using sizes and endianness that
  54   // depend on the host. It doesn't matter, however, because hashing on
  55   // pointer values is inherently unstable. Nothing should depend on the
  56   // ordering of nodes in the folding set.
  57   static_assert(sizeof(uintptr_t) <= sizeof(unsigned long long),
  58                 "unexpected pointer size");
  59   AddInteger(reinterpret_cast<uintptr_t>(Ptr));
  60 }
  61 void FoldingSetNodeID::AddInteger(signed I) {
  62   Bits.push_back(I);
  63 }
  64 void FoldingSetNodeID::AddInteger(unsigned I) {
  65   Bits.push_back(I);
  66 }
  67 void FoldingSetNodeID::AddInteger(long I) {
  68   AddInteger((unsigned long)I);
  69 }
  70 void FoldingSetNodeID::AddInteger(unsigned long I) {
  71   if (sizeof(long) == sizeof(int))
  72     AddInteger(unsigned(I));
  73   else if (sizeof(long) == sizeof(long long)) {
  74     AddInteger((unsigned long long)I);
  75   } else {
  76     llvm_unreachable("unexpected sizeof(long)");
  77   }
  78 }
  79 void FoldingSetNodeID::AddInteger(long long I) {
  80   AddInteger((unsigned long long)I);
  81 }
  82 void FoldingSetNodeID::AddInteger(unsigned long long I) {
  83   AddInteger(unsigned(I));
  84   AddInteger(unsigned(I >> 32));
  85 }
  86
  87 void FoldingSetNodeID::AddString(StringRef String) {
  88   unsigned Size =  String.size();
  89   Bits.push_back(Size);
  90   if (!Size) return;
  91
  92   unsigned Units = Size / 4;
  93   unsigned Pos = 0;
  94   const unsigned *Base = (const unsigned*) String.data();
  95
  96   // If the string is aligned do a bulk transfer.
  97   if (!((intptr_t)Base & 3)) {
  98     Bits.append(Base, Base + Units);
  99     Pos = (Units + 1) * 4;
 100   } else {
 101     // Otherwise do it the hard way.
 102     // To be compatible with above bulk transfer, we need to take endianness
 103     // into account.
 104     static_assert(sys::IsBigEndianHost || sys::IsLittleEndianHost,
 105                   "Unexpected host endianness");
 106     if (sys::IsBigEndianHost) {
 107       for (Pos += 4; Pos <= Size; Pos += 4) {
 108         unsigned V = ((unsigned char)String[Pos - 4] << 24) |
 109                      ((unsigned char)String[Pos - 3] << 16) |
 110                      ((unsigned char)String[Pos - 2] << 8) |
 111                       (unsigned char)String[Pos - 1];
 112         Bits.push_back(V);
 113       }
 114     } else {  // Little-endian host
 115       for (Pos += 4; Pos <= Size; Pos += 4) {
 116         unsigned V = ((unsigned char)String[Pos - 1] << 24) |
 117                      ((unsigned char)String[Pos - 2] << 16) |
 118                      ((unsigned char)String[Pos - 3] << 8) |
 119                       (unsigned char)String[Pos - 4];
 120         Bits.push_back(V);
 121       }
 122     }
 123   }
 124
 125   // With the leftover bits.
 126   unsigned V = 0;
 127   // Pos will have overshot size by 4 - #bytes left over.
 128   // No need to take endianness into account here - this is always executed.
 129   switch (Pos - Size) {
 130   case 1: V = (V << 8) | (unsigned char)String[Size - 3]; LLVM_FALLTHROUGH;
 131   case 2: V = (V << 8) | (unsigned char)String[Size - 2]; LLVM_FALLTHROUGH;
 132   case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
 133   default: return; // Nothing left.
 134   }
 135
 136   Bits.push_back(V);
 137 }
 138
 139 // AddNodeID - Adds the Bit data of another ID to *this.
 140 void FoldingSetNodeID::AddNodeID(const FoldingSetNodeID &ID) {
 141   Bits.append(ID.Bits.begin(), ID.Bits.end());
 142 }
 143
 144 /// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to
 145 /// lookup the node in the FoldingSetBase.
 146 unsigned FoldingSetNodeID::ComputeHash() const {
 147   return FoldingSetNodeIDRef(Bits.data(), Bits.size()).ComputeHash();
 148 }
 149
 150 /// operator== - Used to compare two nodes to each other.
 151 ///
 152 bool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS) const {
 153   return *this == FoldingSetNodeIDRef(RHS.Bits.data(), RHS.Bits.size());
 154 }
 155
 156 /// operator== - Used to compare two nodes to each other.
 157 ///
 158 bool FoldingSetNodeID::operator==(FoldingSetNodeIDRef RHS) const {
 159   return FoldingSetNodeIDRef(Bits.data(), Bits.size()) == RHS;
 160 }
 161
 162 /// Used to compare the "ordering" of two nodes as defined by the
 163 /// profiled bits and their ordering defined by memcmp().
 164 bool FoldingSetNodeID::operator<(const FoldingSetNodeID &RHS) const {
 165   return *this < FoldingSetNodeIDRef(RHS.Bits.data(), RHS.Bits.size());
 166 }
 167
 168 bool FoldingSetNodeID::operator<(FoldingSetNodeIDRef RHS) const {
 169   return FoldingSetNodeIDRef(Bits.data(), Bits.size()) < RHS;
 170 }
 171
 172 /// Intern - Copy this node's data to a memory region allocated from the
 173 /// given allocator and return a FoldingSetNodeIDRef describing the
 174 /// interned data.
 175 FoldingSetNodeIDRef
 176 FoldingSetNodeID::Intern(BumpPtrAllocator &Allocator) const {
 177   unsigned *New = Allocator.Allocate<unsigned>(Bits.size());
 178   std::uninitialized_copy(Bits.begin(), Bits.end(), New);
 179   return FoldingSetNodeIDRef(New, Bits.size());
 180 }
 181
 182 //===----------------------------------------------------------------------===//
 183 /// Helper functions for FoldingSetBase.
 184
 185 /// GetNextPtr - In order to save space, each bucket is a
 186 /// singly-linked-list. In order to make deletion more efficient, we make
 187 /// the list circular, so we can delete a node without computing its hash.
 188 /// The problem with this is that the start of the hash buckets are not
 189 /// Nodes.  If NextInBucketPtr is a bucket pointer, this method returns null:
 190 /// use GetBucketPtr when this happens.
 191 static FoldingSetBase::Node *GetNextPtr(void *NextInBucketPtr) {
 192   // The low bit is set if this is the pointer back to the bucket.
 193   if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1)
 194     return nullptr;
 195
 196   return static_cast<FoldingSetBase::Node*>(NextInBucketPtr);
 197 }
 198
 199
 200 /// testing.
 201 static void **GetBucketPtr(void *NextInBucketPtr) {
 202   intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr);
 203   assert((Ptr & 1) && "Not a bucket pointer");
 204   return reinterpret_cast<void**>(Ptr & ~intptr_t(1));
 205 }
 206
 207 /// GetBucketFor - Hash the specified node ID and return the hash bucket for
 208 /// the specified ID.
 209 static void **GetBucketFor(unsigned Hash, void **Buckets, unsigned NumBuckets) {
 210   // NumBuckets is always a power of 2.
 211   unsigned BucketNum = Hash & (NumBuckets-1);
 212   return Buckets + BucketNum;
 213 }
 214
 215 /// AllocateBuckets - Allocated initialized bucket memory.
 216 static void **AllocateBuckets(unsigned NumBuckets) {
 217   void **Buckets = static_cast<void**>(safe_calloc(NumBuckets + 1,
 218                                                    sizeof(void*)));
 219   // Set the very last bucket to be a non-null "pointer".
 220   Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
 221   return Buckets;
 222 }
 223
 224 //===----------------------------------------------------------------------===//
 225 // FoldingSetBase Implementation
 226
 227 void FoldingSetBase::anchor() {}
 228
 229 FoldingSetBase::FoldingSetBase(unsigned Log2InitSize) {
 230   assert(5 < Log2InitSize && Log2InitSize < 32 &&
 231          "Initial hash table size out of range");
 232   NumBuckets = 1 << Log2InitSize;
 233   Buckets = AllocateBuckets(NumBuckets);
 234   NumNodes = 0;
 235 }
 236
 237 FoldingSetBase::FoldingSetBase(FoldingSetBase &&Arg)
 238     : Buckets(Arg.Buckets), NumBuckets(Arg.NumBuckets), NumNodes(Arg.NumNodes) {
 239   Arg.Buckets = nullptr;
 240   Arg.NumBuckets = 0;
 241   Arg.NumNodes = 0;
 242 }
 243
 244 FoldingSetBase &FoldingSetBase::operator=(FoldingSetBase &&RHS) {
 245   free(Buckets); // This may be null if the set is in a moved-from state.
 246   Buckets = RHS.Buckets;
 247   NumBuckets = RHS.NumBuckets;
 248   NumNodes = RHS.NumNodes;
 249   RHS.Buckets = nullptr;
 250   RHS.NumBuckets = 0;
 251   RHS.NumNodes = 0;
 252   return *this;
 253 }
 254
 255 FoldingSetBase::~FoldingSetBase() {
 256   free(Buckets);
 257 }
 258
 259 void FoldingSetBase::clear() {
 260   // Set all but the last bucket to null pointers.
 261   memset(Buckets, 0, NumBuckets*sizeof(void*));
 262
 263   // Set the very last bucket to be a non-null "pointer".
 264   Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
 265
 266   // Reset the node count to zero.
 267   NumNodes = 0;
 268 }
 269
 270 void FoldingSetBase::GrowBucketCount(unsigned NewBucketCount) {
 271   assert((NewBucketCount > NumBuckets) && "Can't shrink a folding set with GrowBucketCount");
 272   assert(isPowerOf2_32(NewBucketCount) && "Bad bucket count!");
 273   void **OldBuckets = Buckets;
 274   unsigned OldNumBuckets = NumBuckets;
 275
 276   // Clear out new buckets.
 277   Buckets = AllocateBuckets(NewBucketCount);
 278   // Set NumBuckets only if allocation of new buckets was successful.
 279   NumBuckets = NewBucketCount;
 280   NumNodes = 0;
 281
 282   // Walk the old buckets, rehashing nodes into their new place.
 283   FoldingSetNodeID TempID;
 284   for (unsigned i = 0; i != OldNumBuckets; ++i) {
 285     void *Probe = OldBuckets[i];
 286     if (!Probe) continue;
 287     while (Node *NodeInBucket = GetNextPtr(Probe)) {
 288       // Figure out the next link, remove NodeInBucket from the old link.
 289       Probe = NodeInBucket->getNextInBucket();
 290       NodeInBucket->SetNextInBucket(nullptr);
 291
 292       // Insert the node into the new bucket, after recomputing the hash.
 293       InsertNode(NodeInBucket,
 294                  GetBucketFor(ComputeNodeHash(NodeInBucket, TempID),
 295                               Buckets, NumBuckets));
 296       TempID.clear();
 297     }
 298   }
 299
 300   free(OldBuckets);
 301 }
 302
 303 /// GrowHashTable - Double the size of the hash table and rehash everything.
 304 ///
 305 void FoldingSetBase::GrowHashTable() {
 306   GrowBucketCount(NumBuckets * 2);
 307 }
 308
 309 void FoldingSetBase::reserve(unsigned EltCount) {
 310   // This will give us somewhere between EltCount / 2 and
 311   // EltCount buckets.  This puts us in the load factor
 312   // range of 1.0 - 2.0.
 313   if(EltCount < capacity())
 314     return;
 315   GrowBucketCount(PowerOf2Floor(EltCount));
 316 }
 317
 318 /// FindNodeOrInsertPos - Look up the node specified by ID.  If it exists,
 319 /// return it.  If not, return the insertion token that will make insertion
 320 /// faster.
 321 FoldingSetBase::Node *
 322 FoldingSetBase::FindNodeOrInsertPos(const FoldingSetNodeID &ID,
 323                                     void *&InsertPos) {
 324   unsigned IDHash = ID.ComputeHash();
 325   void **Bucket = GetBucketFor(IDHash, Buckets, NumBuckets);
 326   void *Probe = *Bucket;
 327
 328   InsertPos = nullptr;
 329
 330   FoldingSetNodeID TempID;
 331   while (Node *NodeInBucket = GetNextPtr(Probe)) {
 332     if (NodeEquals(NodeInBucket, ID, IDHash, TempID))
 333       return NodeInBucket;
 334     TempID.clear();
 335
 336     Probe = NodeInBucket->getNextInBucket();
 337   }
 338
 339   // Didn't find the node, return null with the bucket as the InsertPos.
 340   InsertPos = Bucket;
 341   return nullptr;
 342 }
 343
 344 /// InsertNode - Insert the specified node into the folding set, knowing that it
 345 /// is not already in the map.  InsertPos must be obtained from
 346 /// FindNodeOrInsertPos.
 347 void FoldingSetBase::InsertNode(Node *N, void *InsertPos) {
 348   assert(!N->getNextInBucket());
 349   // Do we need to grow the hashtable?
 350   if (NumNodes+1 > capacity()) {
 351     GrowHashTable();
 352     FoldingSetNodeID TempID;
 353     InsertPos = GetBucketFor(ComputeNodeHash(N, TempID), Buckets, NumBuckets);
 354   }
 355
 356   ++NumNodes;
 357
 358   /// The insert position is actually a bucket pointer.
 359   void **Bucket = static_cast<void**>(InsertPos);
 360
 361   void *Next = *Bucket;
 362
 363   // If this is the first insertion into this bucket, its next pointer will be
 364   // null.  Pretend as if it pointed to itself, setting the low bit to indicate
 365   // that it is a pointer to the bucket.
 366   if (!Next)
 367     Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1);
 368
 369   // Set the node's next pointer, and make the bucket point to the node.
 370   N->SetNextInBucket(Next);
 371   *Bucket = N;
 372 }
 373
 374 /// RemoveNode - Remove a node from the folding set, returning true if one was
 375 /// removed or false if the node was not in the folding set.
 376 bool FoldingSetBase::RemoveNode(Node *N) {
 377   // Because each bucket is a circular list, we don't need to compute N's hash
 378   // to remove it.
 379   void *Ptr = N->getNextInBucket();
 380   if (!Ptr) return false;  // Not in folding set.
 381
 382   --NumNodes;
 383   N->SetNextInBucket(nullptr);
 384
 385   // Remember what N originally pointed to, either a bucket or another node.
 386   void *NodeNextPtr = Ptr;
 387
 388   // Chase around the list until we find the node (or bucket) which points to N.
 389   while (true) {
 390     if (Node *NodeInBucket = GetNextPtr(Ptr)) {
 391       // Advance pointer.
 392       Ptr = NodeInBucket->getNextInBucket();
 393
 394       // We found a node that points to N, change it to point to N's next node,
 395       // removing N from the list.
 396       if (Ptr == N) {
 397         NodeInBucket->SetNextInBucket(NodeNextPtr);
 398         return true;
 399       }
 400     } else {
 401       void **Bucket = GetBucketPtr(Ptr);
 402       Ptr = *Bucket;
 403
 404       // If we found that the bucket points to N, update the bucket to point to
 405       // whatever is next.
 406       if (Ptr == N) {
 407         *Bucket = NodeNextPtr;
 408         return true;
 409       }
 410     }
 411   }
 412 }
 413
 414 /// GetOrInsertNode - If there is an existing simple Node exactly
 415 /// equal to the specified node, return it.  Otherwise, insert 'N' and it
 416 /// instead.
 417 FoldingSetBase::Node *FoldingSetBase::GetOrInsertNode(FoldingSetBase::Node *N) {
 418   FoldingSetNodeID ID;
 419   GetNodeProfile(N, ID);
 420   void *IP;
 421   if (Node *E = FindNodeOrInsertPos(ID, IP))
 422     return E;
 423   InsertNode(N, IP);
 424   return N;
 425 }
 426
 427 //===----------------------------------------------------------------------===//
 428 // FoldingSetIteratorImpl Implementation
 429
 430 FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) {
 431   // Skip to the first non-null non-self-cycle bucket.
 432   while (*Bucket != reinterpret_cast<void*>(-1) &&
 433          (!*Bucket || !GetNextPtr(*Bucket)))
 434     ++Bucket;
 435
 436   NodePtr = static_cast<FoldingSetNode*>(*Bucket);
 437 }
 438
 439 void FoldingSetIteratorImpl::advance() {
 440   // If there is another link within this bucket, go to it.
 441   void *Probe = NodePtr->getNextInBucket();
 442
 443   if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe))
 444     NodePtr = NextNodeInBucket;
 445   else {
 446     // Otherwise, this is the last link in this bucket.
 447     void **Bucket = GetBucketPtr(Probe);
 448
 449     // Skip to the next non-null non-self-cycle bucket.
 450     do {
 451       ++Bucket;
 452     } while (*Bucket != reinterpret_cast<void*>(-1) &&
 453              (!*Bucket || !GetNextPtr(*Bucket)));
 454
 455     NodePtr = static_cast<FoldingSetNode*>(*Bucket);
 456   }
 457 }
 458
 459 //===----------------------------------------------------------------------===//
 460 // FoldingSetBucketIteratorImpl Implementation
 461
 462 FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) {
 463   Ptr = (!*Bucket || !GetNextPtr(*Bucket)) ? (void*) Bucket : *Bucket;
 464 }