FoldingSet.cpp revision 18529f35151f18345519e38496ac72350ee15f38
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/ADT/MathExtras.h" 19using namespace llvm; 20 21//===----------------------------------------------------------------------===// 22// FoldingSetImpl::NodeID Implementation 23 24/// Add* - Add various data types to Bit data. 25/// 26void FoldingSetImpl::NodeID::AddPointer(const void *Ptr) { 27 // Note: this adds pointers to the hash using sizes and endianness that 28 // depend on the host. It doesn't matter however, because hashing on 29 // pointer values in inherently unstable. Nothing should depend on the 30 // ordering of nodes in the folding set. 31 intptr_t PtrI = (intptr_t)Ptr; 32 Bits.push_back(unsigned(PtrI)); 33 if (sizeof(intptr_t) > sizeof(unsigned)) 34 Bits.push_back(unsigned(uint64_t(PtrI) >> 32)); 35} 36void FoldingSetImpl::NodeID::AddInteger(signed I) { 37 Bits.push_back(I); 38} 39void FoldingSetImpl::NodeID::AddInteger(unsigned I) { 40 Bits.push_back(I); 41} 42void FoldingSetImpl::NodeID::AddInteger(uint64_t I) { 43 Bits.push_back(unsigned(I)); 44 Bits.push_back(unsigned(I >> 32)); 45} 46void FoldingSetImpl::NodeID::AddFloat(float F) { 47 Bits.push_back(FloatToBits(F)); 48} 49void FoldingSetImpl::NodeID::AddDouble(double D) { 50 AddInteger(DoubleToBits(D)); 51} 52void FoldingSetImpl::NodeID::AddString(const std::string &String) { 53 unsigned Size = String.size(); 54 unsigned Units = Size / 4; 55 unsigned Pos = 0; 56 const unsigned *Base = (const unsigned *)String.data(); 57 58 // If the string is aligned do a bulk transfer. 59 if (!((intptr_t)Base & 3)) { 60 Bits.insert(Bits.end(), Base, Base + Units); 61 Pos = Units * sizeof(unsigned); 62 } else { 63 // Otherwise do it the hard way. 64 for ( Pos += 4; Pos < Size; Pos += 4) { 65 unsigned V = ((unsigned char)String[Pos - 4] << 24) | 66 ((unsigned char)String[Pos - 3] << 16) | 67 ((unsigned char)String[Pos - 2] << 8) | 68 (unsigned char)String[Pos - 1]; 69 Bits.push_back(V); 70 } 71 } 72 73 // With the leftover bits. 74 unsigned V = 0; 75 // Pos will have overshot size by 4 - #bytes left over. 76 switch (Pos - Size) { 77 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru. 78 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru. 79 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break; 80 case 0: return; // Nothing left. 81 } 82 83 Bits.push_back(V); 84} 85 86/// ComputeHash - Compute a strong hash value for this NodeID, used to 87/// lookup the node in the FoldingSetImpl. 88unsigned FoldingSetImpl::NodeID::ComputeHash() const { 89 // This is adapted from SuperFastHash by Paul Hsieh. 90 unsigned Hash = Bits.size(); 91 for (const unsigned *BP = &Bits[0], *E = BP+Bits.size(); BP != E; ++BP) { 92 unsigned Data = *BP; 93 Hash += Data & 0xFFFF; 94 unsigned Tmp = ((Data >> 16) << 11) ^ Hash; 95 Hash = (Hash << 16) ^ Tmp; 96 Hash += Hash >> 11; 97 } 98 99 // Force "avalanching" of final 127 bits. 100 Hash ^= Hash << 3; 101 Hash += Hash >> 5; 102 Hash ^= Hash << 4; 103 Hash += Hash >> 17; 104 Hash ^= Hash << 25; 105 Hash += Hash >> 6; 106 return Hash; 107} 108 109/// operator== - Used to compare two nodes to each other. 110/// 111bool FoldingSetImpl::NodeID::operator==(const FoldingSetImpl::NodeID &RHS)const{ 112 if (Bits.size() != RHS.Bits.size()) return false; 113 return memcmp(&Bits[0], &RHS.Bits[0], Bits.size()*sizeof(Bits[0])) == 0; 114} 115 116 117//===----------------------------------------------------------------------===// 118/// Helper functions for FoldingSetImpl. 119 120/// GetNextPtr - In order to save space, each bucket is a 121/// singly-linked-list. In order to make deletion more efficient, we make 122/// the list circular, so we can delete a node without computing its hash. 123/// The problem with this is that the start of the hash buckets are not 124/// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null 125/// : use GetBucketPtr when this happens. 126static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr, 127 void **Buckets, unsigned NumBuckets) { 128 if (NextInBucketPtr >= Buckets && NextInBucketPtr < Buckets + NumBuckets) 129 return 0; 130 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr); 131} 132 133/// GetBucketPtr - Provides a casting of a bucket pointer for isNode 134/// testing. 135static void **GetBucketPtr(void *NextInBucketPtr) { 136 return static_cast<void**>(NextInBucketPtr); 137} 138 139/// GetBucketFor - Hash the specified node ID and return the hash bucket for 140/// the specified ID. 141static void **GetBucketFor(const FoldingSetImpl::NodeID &ID, 142 void **Buckets, unsigned NumBuckets) { 143 // NumBuckets is always a power of 2. 144 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1); 145 return Buckets + BucketNum; 146} 147 148//===----------------------------------------------------------------------===// 149// FoldingSetImpl Implementation 150 151FoldingSetImpl::FoldingSetImpl() : NumNodes(0) { 152 NumBuckets = 64; 153 Buckets = new void*[NumBuckets]; 154 memset(Buckets, 0, NumBuckets*sizeof(void*)); 155} 156FoldingSetImpl::~FoldingSetImpl() { 157 delete [] Buckets; 158} 159 160/// GrowHashTable - Double the size of the hash table and rehash everything. 161/// 162void FoldingSetImpl::GrowHashTable() { 163 void **OldBuckets = Buckets; 164 unsigned OldNumBuckets = NumBuckets; 165 NumBuckets <<= 1; 166 167 // Reset the node count to zero: we're going to reinsert everything. 168 NumNodes = 0; 169 170 // Clear out new buckets. 171 Buckets = new void*[NumBuckets]; 172 memset(Buckets, 0, NumBuckets*sizeof(void*)); 173 174 // Walk the old buckets, rehashing nodes into their new place. 175 for (unsigned i = 0; i != OldNumBuckets; ++i) { 176 void *Probe = OldBuckets[i]; 177 if (!Probe) continue; 178 while (Node *NodeInBucket = GetNextPtr(Probe, OldBuckets, OldNumBuckets)){ 179 // Figure out the next link, remove NodeInBucket from the old link. 180 Probe = NodeInBucket->getNextInBucket(); 181 NodeInBucket->SetNextInBucket(0); 182 183 // Insert the node into the new bucket, after recomputing the hash. 184 NodeID ID; 185 GetNodeProfile(ID, NodeInBucket); 186 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets)); 187 } 188 } 189 190 delete[] OldBuckets; 191} 192 193/// FindNodeOrInsertPos - Look up the node specified by ID. If it exists, 194/// return it. If not, return the insertion token that will make insertion 195/// faster. 196FoldingSetImpl::Node *FoldingSetImpl::FindNodeOrInsertPos(const NodeID &ID, 197 void *&InsertPos) { 198 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets); 199 void *Probe = *Bucket; 200 201 InsertPos = 0; 202 203 while (Node *NodeInBucket = GetNextPtr(Probe, Buckets, NumBuckets)) { 204 NodeID OtherID; 205 GetNodeProfile(OtherID, NodeInBucket); 206 if (OtherID == ID) 207 return NodeInBucket; 208 209 Probe = NodeInBucket->getNextInBucket(); 210 } 211 212 // Didn't find the node, return null with the bucket as the InsertPos. 213 InsertPos = Bucket; 214 return 0; 215} 216 217/// InsertNode - Insert the specified node into the folding set, knowing that it 218/// is not already in the map. InsertPos must be obtained from 219/// FindNodeOrInsertPos. 220void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) { 221 ++NumNodes; 222 // Do we need to grow the hashtable? 223 if (NumNodes > NumBuckets*2) { 224 GrowHashTable(); 225 NodeID ID; 226 GetNodeProfile(ID, N); 227 InsertPos = GetBucketFor(ID, Buckets, NumBuckets); 228 } 229 230 /// The insert position is actually a bucket pointer. 231 void **Bucket = static_cast<void**>(InsertPos); 232 233 void *Next = *Bucket; 234 235 // If this is the first insertion into this bucket, its next pointer will be 236 // null. Pretend as if it pointed to itself. 237 if (Next == 0) 238 Next = Bucket; 239 240 // Set the nodes next pointer, and make the bucket point to the node. 241 N->SetNextInBucket(Next); 242 *Bucket = N; 243} 244 245/// RemoveNode - Remove a node from the folding set, returning true if one was 246/// removed or false if the node was not in the folding set. 247bool FoldingSetImpl::RemoveNode(Node *N) { 248 // Because each bucket is a circular list, we don't need to compute N's hash 249 // to remove it. Chase around the list until we find the node (or bucket) 250 // which points to N. 251 void *Ptr = N->getNextInBucket(); 252 if (Ptr == 0) return false; // Not in folding set. 253 254 --NumNodes; 255 256 void *NodeNextPtr = Ptr; 257 N->SetNextInBucket(0); 258 while (true) { 259 if (Node *NodeInBucket = GetNextPtr(Ptr, Buckets, NumBuckets)) { 260 // Advance pointer. 261 Ptr = NodeInBucket->getNextInBucket(); 262 263 // We found a node that points to N, change it to point to N's next node, 264 // removing N from the list. 265 if (Ptr == N) { 266 NodeInBucket->SetNextInBucket(NodeNextPtr); 267 return true; 268 } 269 } else { 270 void **Bucket = GetBucketPtr(Ptr); 271 Ptr = *Bucket; 272 273 // If we found that the bucket points to N, update the bucket to point to 274 // whatever is next. 275 if (Ptr == N) { 276 *Bucket = NodeNextPtr; 277 return true; 278 } 279 } 280 } 281} 282 283/// GetOrInsertNode - If there is an existing simple Node exactly 284/// equal to the specified node, return it. Otherwise, insert 'N' and it 285/// instead. 286FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) { 287 NodeID ID; 288 GetNodeProfile(ID, N); 289 void *IP; 290 if (Node *E = FindNodeOrInsertPos(ID, IP)) 291 return E; 292 InsertNode(N, IP); 293 return N; 294} 295