FoldingSet.cpp revision 27dba671c3f158a33c8cbdf5196a6fd16489be03
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> 22using namespace llvm; 23 24//===----------------------------------------------------------------------===// 25// FoldingSetNodeID Implementation 26 27/// Add* - Add various data types to Bit data. 28/// 29void 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} 39void FoldingSetNodeID::AddInteger(signed I) { 40 Bits.push_back(I); 41} 42void FoldingSetNodeID::AddInteger(unsigned I) { 43 Bits.push_back(I); 44} 45void FoldingSetNodeID::AddInteger(long I) { 46 AddInteger((unsigned long)I); 47} 48void 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} 57void FoldingSetNodeID::AddInteger(long long I) { 58 AddInteger((unsigned long long)I); 59} 60void 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 66void FoldingSetNodeID::AddString(StringRef String) { 67 unsigned Size = String.size(); 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.data(); 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/// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to 104/// lookup the node in the FoldingSetImpl. 105unsigned FoldingSetNodeID::ComputeHash() const { 106 // This is adapted from SuperFastHash by Paul Hsieh. 107 unsigned Hash = static_cast<unsigned>(Bits.size()); 108 for (const unsigned *BP = &Bits[0], *E = BP+Bits.size(); BP != E; ++BP) { 109 unsigned Data = *BP; 110 Hash += Data & 0xFFFF; 111 unsigned Tmp = ((Data >> 16) << 11) ^ Hash; 112 Hash = (Hash << 16) ^ Tmp; 113 Hash += Hash >> 11; 114 } 115 116 // Force "avalanching" of final 127 bits. 117 Hash ^= Hash << 3; 118 Hash += Hash >> 5; 119 Hash ^= Hash << 4; 120 Hash += Hash >> 17; 121 Hash ^= Hash << 25; 122 Hash += Hash >> 6; 123 return Hash; 124} 125 126/// operator== - Used to compare two nodes to each other. 127/// 128bool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS)const{ 129 if (Bits.size() != RHS.Bits.size()) return false; 130 return memcmp(&Bits[0], &RHS.Bits[0], Bits.size()*sizeof(Bits[0])) == 0; 131} 132 133 134//===----------------------------------------------------------------------===// 135/// Helper functions for FoldingSetImpl. 136 137/// GetNextPtr - In order to save space, each bucket is a 138/// singly-linked-list. In order to make deletion more efficient, we make 139/// the list circular, so we can delete a node without computing its hash. 140/// The problem with this is that the start of the hash buckets are not 141/// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null: 142/// use GetBucketPtr when this happens. 143static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr) { 144 // The low bit is set if this is the pointer back to the bucket. 145 if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1) 146 return 0; 147 148 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr); 149} 150 151 152/// testing. 153static void **GetBucketPtr(void *NextInBucketPtr) { 154 intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr); 155 assert((Ptr & 1) && "Not a bucket pointer"); 156 return reinterpret_cast<void**>(Ptr & ~intptr_t(1)); 157} 158 159/// GetBucketFor - Hash the specified node ID and return the hash bucket for 160/// the specified ID. 161static void **GetBucketFor(const FoldingSetNodeID &ID, 162 void **Buckets, unsigned NumBuckets) { 163 // NumBuckets is always a power of 2. 164 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1); 165 return Buckets + BucketNum; 166} 167 168//===----------------------------------------------------------------------===// 169// FoldingSetImpl Implementation 170 171FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) { 172 assert(5 < Log2InitSize && Log2InitSize < 32 && 173 "Initial hash table size out of range"); 174 NumBuckets = 1 << Log2InitSize; 175 Buckets = new void*[NumBuckets+1]; 176 clear(); 177} 178FoldingSetImpl::~FoldingSetImpl() { 179 delete [] Buckets; 180} 181void FoldingSetImpl::clear() { 182 // Set all but the last bucket to null pointers. 183 memset(Buckets, 0, NumBuckets*sizeof(void*)); 184 185 // Set the very last bucket to be a non-null "pointer". 186 Buckets[NumBuckets] = reinterpret_cast<void*>(-1); 187 188 // Reset the node count to zero. 189 NumNodes = 0; 190} 191 192/// GrowHashTable - Double the size of the hash table and rehash everything. 193/// 194void FoldingSetImpl::GrowHashTable() { 195 void **OldBuckets = Buckets; 196 unsigned OldNumBuckets = NumBuckets; 197 NumBuckets <<= 1; 198 199 // Clear out new buckets. 200 Buckets = new void*[NumBuckets+1]; 201 clear(); 202 203 // Walk the old buckets, rehashing nodes into their new place. 204 FoldingSetNodeID ID; 205 for (unsigned i = 0; i != OldNumBuckets; ++i) { 206 void *Probe = OldBuckets[i]; 207 if (!Probe) continue; 208 while (Node *NodeInBucket = GetNextPtr(Probe)) { 209 // Figure out the next link, remove NodeInBucket from the old link. 210 Probe = NodeInBucket->getNextInBucket(); 211 NodeInBucket->SetNextInBucket(0); 212 213 // Insert the node into the new bucket, after recomputing the hash. 214 GetNodeProfile(ID, NodeInBucket); 215 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets)); 216 ID.clear(); 217 } 218 } 219 220 delete[] OldBuckets; 221} 222 223/// FindNodeOrInsertPos - Look up the node specified by ID. If it exists, 224/// return it. If not, return the insertion token that will make insertion 225/// faster. 226FoldingSetImpl::Node 227*FoldingSetImpl::FindNodeOrInsertPos(const FoldingSetNodeID &ID, 228 void *&InsertPos) { 229 230 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets); 231 void *Probe = *Bucket; 232 233 InsertPos = 0; 234 235 FoldingSetNodeID OtherID; 236 while (Node *NodeInBucket = GetNextPtr(Probe)) { 237 GetNodeProfile(OtherID, NodeInBucket); 238 if (OtherID == ID) 239 return NodeInBucket; 240 241 Probe = NodeInBucket->getNextInBucket(); 242 OtherID.clear(); 243 } 244 245 // Didn't find the node, return null with the bucket as the InsertPos. 246 InsertPos = Bucket; 247 return 0; 248} 249 250/// InsertNode - Insert the specified node into the folding set, knowing that it 251/// is not already in the map. InsertPos must be obtained from 252/// FindNodeOrInsertPos. 253void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) { 254 assert(N->getNextInBucket() == 0); 255 // Do we need to grow the hashtable? 256 if (NumNodes+1 > NumBuckets*2) { 257 GrowHashTable(); 258 FoldingSetNodeID ID; 259 GetNodeProfile(ID, N); 260 InsertPos = GetBucketFor(ID, Buckets, NumBuckets); 261 } 262 263 ++NumNodes; 264 265 /// The insert position is actually a bucket pointer. 266 void **Bucket = static_cast<void**>(InsertPos); 267 268 void *Next = *Bucket; 269 270 // If this is the first insertion into this bucket, its next pointer will be 271 // null. Pretend as if it pointed to itself, setting the low bit to indicate 272 // that it is a pointer to the bucket. 273 if (Next == 0) 274 Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1); 275 276 // Set the node's next pointer, and make the bucket point to the node. 277 N->SetNextInBucket(Next); 278 *Bucket = N; 279} 280 281/// RemoveNode - Remove a node from the folding set, returning true if one was 282/// removed or false if the node was not in the folding set. 283bool FoldingSetImpl::RemoveNode(Node *N) { 284 // Because each bucket is a circular list, we don't need to compute N's hash 285 // to remove it. 286 void *Ptr = N->getNextInBucket(); 287 if (Ptr == 0) return false; // Not in folding set. 288 289 --NumNodes; 290 N->SetNextInBucket(0); 291 292 // Remember what N originally pointed to, either a bucket or another node. 293 void *NodeNextPtr = Ptr; 294 295 // Chase around the list until we find the node (or bucket) which points to N. 296 while (true) { 297 if (Node *NodeInBucket = GetNextPtr(Ptr)) { 298 // Advance pointer. 299 Ptr = NodeInBucket->getNextInBucket(); 300 301 // We found a node that points to N, change it to point to N's next node, 302 // removing N from the list. 303 if (Ptr == N) { 304 NodeInBucket->SetNextInBucket(NodeNextPtr); 305 return true; 306 } 307 } else { 308 void **Bucket = GetBucketPtr(Ptr); 309 Ptr = *Bucket; 310 311 // If we found that the bucket points to N, update the bucket to point to 312 // whatever is next. 313 if (Ptr == N) { 314 *Bucket = NodeNextPtr; 315 return true; 316 } 317 } 318 } 319} 320 321/// GetOrInsertNode - If there is an existing simple Node exactly 322/// equal to the specified node, return it. Otherwise, insert 'N' and it 323/// instead. 324FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) { 325 FoldingSetNodeID ID; 326 GetNodeProfile(ID, N); 327 void *IP; 328 if (Node *E = FindNodeOrInsertPos(ID, IP)) 329 return E; 330 InsertNode(N, IP); 331 return N; 332} 333 334//===----------------------------------------------------------------------===// 335// FoldingSetIteratorImpl Implementation 336 337FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) { 338 // Skip to the first non-null non-self-cycle bucket. 339 while (*Bucket != reinterpret_cast<void*>(-1) && 340 (*Bucket == 0 || GetNextPtr(*Bucket) == 0)) 341 ++Bucket; 342 343 NodePtr = static_cast<FoldingSetNode*>(*Bucket); 344} 345 346void FoldingSetIteratorImpl::advance() { 347 // If there is another link within this bucket, go to it. 348 void *Probe = NodePtr->getNextInBucket(); 349 350 if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe)) 351 NodePtr = NextNodeInBucket; 352 else { 353 // Otherwise, this is the last link in this bucket. 354 void **Bucket = GetBucketPtr(Probe); 355 356 // Skip to the next non-null non-self-cycle bucket. 357 do { 358 ++Bucket; 359 } while (*Bucket != reinterpret_cast<void*>(-1) && 360 (*Bucket == 0 || GetNextPtr(*Bucket) == 0)); 361 362 NodePtr = static_cast<FoldingSetNode*>(*Bucket); 363 } 364} 365 366//===----------------------------------------------------------------------===// 367// FoldingSetBucketIteratorImpl Implementation 368 369FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) { 370 Ptr = (*Bucket == 0 || GetNextPtr(*Bucket) == 0) ? (void*) Bucket : *Bucket; 371} 372