lib/Support/FoldingSet.cpp

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