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