FoldingSet.cpp revision 218893
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/Allocator.h" 19#include "llvm/Support/ErrorHandling.h" 20#include "llvm/Support/MathExtras.h" 21#include "llvm/Support/Host.h" 22#include <cassert> 23#include <cstring> 24using namespace llvm; 25 26//===----------------------------------------------------------------------===// 27// FoldingSetNodeIDRef Implementation 28 29/// ComputeHash - Compute a strong hash value for this FoldingSetNodeIDRef, 30/// used to lookup the node in the FoldingSetImpl. 31unsigned FoldingSetNodeIDRef::ComputeHash() const { 32 // This is adapted from SuperFastHash by Paul Hsieh. 33 unsigned Hash = static_cast<unsigned>(Size); 34 for (const unsigned *BP = Data, *E = BP+Size; BP != E; ++BP) { 35 unsigned Data = *BP; 36 Hash += Data & 0xFFFF; 37 unsigned Tmp = ((Data >> 16) << 11) ^ Hash; 38 Hash = (Hash << 16) ^ Tmp; 39 Hash += Hash >> 11; 40 } 41 42 // Force "avalanching" of final 127 bits. 43 Hash ^= Hash << 3; 44 Hash += Hash >> 5; 45 Hash ^= Hash << 4; 46 Hash += Hash >> 17; 47 Hash ^= Hash << 25; 48 Hash += Hash >> 6; 49 return Hash; 50} 51 52bool FoldingSetNodeIDRef::operator==(FoldingSetNodeIDRef RHS) const { 53 if (Size != RHS.Size) return false; 54 return memcmp(Data, RHS.Data, Size*sizeof(*Data)) == 0; 55} 56 57//===----------------------------------------------------------------------===// 58// FoldingSetNodeID Implementation 59 60/// Add* - Add various data types to Bit data. 61/// 62void FoldingSetNodeID::AddPointer(const void *Ptr) { 63 // Note: this adds pointers to the hash using sizes and endianness that 64 // depend on the host. It doesn't matter however, because hashing on 65 // pointer values in inherently unstable. Nothing should depend on the 66 // ordering of nodes in the folding set. 67 intptr_t PtrI = (intptr_t)Ptr; 68 Bits.push_back(unsigned(PtrI)); 69 if (sizeof(intptr_t) > sizeof(unsigned)) 70 Bits.push_back(unsigned(uint64_t(PtrI) >> 32)); 71} 72void FoldingSetNodeID::AddInteger(signed I) { 73 Bits.push_back(I); 74} 75void FoldingSetNodeID::AddInteger(unsigned I) { 76 Bits.push_back(I); 77} 78void FoldingSetNodeID::AddInteger(long I) { 79 AddInteger((unsigned long)I); 80} 81void FoldingSetNodeID::AddInteger(unsigned long I) { 82 if (sizeof(long) == sizeof(int)) 83 AddInteger(unsigned(I)); 84 else if (sizeof(long) == sizeof(long long)) { 85 AddInteger((unsigned long long)I); 86 } else { 87 llvm_unreachable("unexpected sizeof(long)"); 88 } 89} 90void FoldingSetNodeID::AddInteger(long long I) { 91 AddInteger((unsigned long long)I); 92} 93void FoldingSetNodeID::AddInteger(unsigned long long I) { 94 AddInteger(unsigned(I)); 95 if ((uint64_t)(int)I != I) 96 Bits.push_back(unsigned(I >> 32)); 97} 98 99void FoldingSetNodeID::AddString(StringRef String) { 100 unsigned Size = String.size(); 101 Bits.push_back(Size); 102 if (!Size) return; 103 104 unsigned Units = Size / 4; 105 unsigned Pos = 0; 106 const unsigned *Base = (const unsigned*) String.data(); 107 108 // If the string is aligned do a bulk transfer. 109 if (!((intptr_t)Base & 3)) { 110 Bits.append(Base, Base + Units); 111 Pos = (Units + 1) * 4; 112 } else { 113 // Otherwise do it the hard way. 114 // To be compatible with above bulk transfer, we need to take endianness 115 // into account. 116 if (sys::isBigEndianHost()) { 117 for (Pos += 4; Pos <= Size; Pos += 4) { 118 unsigned V = ((unsigned char)String[Pos - 4] << 24) | 119 ((unsigned char)String[Pos - 3] << 16) | 120 ((unsigned char)String[Pos - 2] << 8) | 121 (unsigned char)String[Pos - 1]; 122 Bits.push_back(V); 123 } 124 } else { 125 assert(sys::isLittleEndianHost() && "Unexpected host endianness"); 126 for (Pos += 4; Pos <= Size; Pos += 4) { 127 unsigned V = ((unsigned char)String[Pos - 1] << 24) | 128 ((unsigned char)String[Pos - 2] << 16) | 129 ((unsigned char)String[Pos - 3] << 8) | 130 (unsigned char)String[Pos - 4]; 131 Bits.push_back(V); 132 } 133 } 134 } 135 136 // With the leftover bits. 137 unsigned V = 0; 138 // Pos will have overshot size by 4 - #bytes left over. 139 // No need to take endianness into account here - this is always executed. 140 switch (Pos - Size) { 141 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru. 142 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru. 143 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break; 144 default: return; // Nothing left. 145 } 146 147 Bits.push_back(V); 148} 149 150/// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to 151/// lookup the node in the FoldingSetImpl. 152unsigned FoldingSetNodeID::ComputeHash() const { 153 return FoldingSetNodeIDRef(Bits.data(), Bits.size()).ComputeHash(); 154} 155 156/// operator== - Used to compare two nodes to each other. 157/// 158bool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS)const{ 159 return *this == FoldingSetNodeIDRef(RHS.Bits.data(), RHS.Bits.size()); 160} 161 162/// operator== - Used to compare two nodes to each other. 163/// 164bool FoldingSetNodeID::operator==(FoldingSetNodeIDRef RHS) const { 165 return FoldingSetNodeIDRef(Bits.data(), Bits.size()) == RHS; 166} 167 168/// Intern - Copy this node's data to a memory region allocated from the 169/// given allocator and return a FoldingSetNodeIDRef describing the 170/// interned data. 171FoldingSetNodeIDRef 172FoldingSetNodeID::Intern(BumpPtrAllocator &Allocator) const { 173 unsigned *New = Allocator.Allocate<unsigned>(Bits.size()); 174 std::uninitialized_copy(Bits.begin(), Bits.end(), New); 175 return FoldingSetNodeIDRef(New, Bits.size()); 176} 177 178//===----------------------------------------------------------------------===// 179/// Helper functions for FoldingSetImpl. 180 181/// GetNextPtr - In order to save space, each bucket is a 182/// singly-linked-list. In order to make deletion more efficient, we make 183/// the list circular, so we can delete a node without computing its hash. 184/// The problem with this is that the start of the hash buckets are not 185/// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null: 186/// use GetBucketPtr when this happens. 187static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr) { 188 // The low bit is set if this is the pointer back to the bucket. 189 if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1) 190 return 0; 191 192 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr); 193} 194 195 196/// testing. 197static void **GetBucketPtr(void *NextInBucketPtr) { 198 intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr); 199 assert((Ptr & 1) && "Not a bucket pointer"); 200 return reinterpret_cast<void**>(Ptr & ~intptr_t(1)); 201} 202 203/// GetBucketFor - Hash the specified node ID and return the hash bucket for 204/// the specified ID. 205static void **GetBucketFor(unsigned Hash, void **Buckets, unsigned NumBuckets) { 206 // NumBuckets is always a power of 2. 207 unsigned BucketNum = Hash & (NumBuckets-1); 208 return Buckets + BucketNum; 209} 210 211/// AllocateBuckets - Allocated initialized bucket memory. 212static void **AllocateBuckets(unsigned NumBuckets) { 213 void **Buckets = static_cast<void**>(calloc(NumBuckets+1, sizeof(void*))); 214 // Set the very last bucket to be a non-null "pointer". 215 Buckets[NumBuckets] = reinterpret_cast<void*>(-1); 216 return Buckets; 217} 218 219//===----------------------------------------------------------------------===// 220// FoldingSetImpl Implementation 221 222FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) { 223 assert(5 < Log2InitSize && Log2InitSize < 32 && 224 "Initial hash table size out of range"); 225 NumBuckets = 1 << Log2InitSize; 226 Buckets = AllocateBuckets(NumBuckets); 227 NumNodes = 0; 228} 229FoldingSetImpl::~FoldingSetImpl() { 230 free(Buckets); 231} 232void FoldingSetImpl::clear() { 233 // Set all but the last bucket to null pointers. 234 memset(Buckets, 0, NumBuckets*sizeof(void*)); 235 236 // Set the very last bucket to be a non-null "pointer". 237 Buckets[NumBuckets] = reinterpret_cast<void*>(-1); 238 239 // Reset the node count to zero. 240 NumNodes = 0; 241} 242 243/// GrowHashTable - Double the size of the hash table and rehash everything. 244/// 245void FoldingSetImpl::GrowHashTable() { 246 void **OldBuckets = Buckets; 247 unsigned OldNumBuckets = NumBuckets; 248 NumBuckets <<= 1; 249 250 // Clear out new buckets. 251 Buckets = AllocateBuckets(NumBuckets); 252 NumNodes = 0; 253 254 // Walk the old buckets, rehashing nodes into their new place. 255 FoldingSetNodeID TempID; 256 for (unsigned i = 0; i != OldNumBuckets; ++i) { 257 void *Probe = OldBuckets[i]; 258 if (!Probe) continue; 259 while (Node *NodeInBucket = GetNextPtr(Probe)) { 260 // Figure out the next link, remove NodeInBucket from the old link. 261 Probe = NodeInBucket->getNextInBucket(); 262 NodeInBucket->SetNextInBucket(0); 263 264 // Insert the node into the new bucket, after recomputing the hash. 265 InsertNode(NodeInBucket, 266 GetBucketFor(ComputeNodeHash(NodeInBucket, TempID), 267 Buckets, NumBuckets)); 268 TempID.clear(); 269 } 270 } 271 272 free(OldBuckets); 273} 274 275/// FindNodeOrInsertPos - Look up the node specified by ID. If it exists, 276/// return it. If not, return the insertion token that will make insertion 277/// faster. 278FoldingSetImpl::Node 279*FoldingSetImpl::FindNodeOrInsertPos(const FoldingSetNodeID &ID, 280 void *&InsertPos) { 281 282 void **Bucket = GetBucketFor(ID.ComputeHash(), Buckets, NumBuckets); 283 void *Probe = *Bucket; 284 285 InsertPos = 0; 286 287 FoldingSetNodeID TempID; 288 while (Node *NodeInBucket = GetNextPtr(Probe)) { 289 if (NodeEquals(NodeInBucket, ID, TempID)) 290 return NodeInBucket; 291 TempID.clear(); 292 293 Probe = NodeInBucket->getNextInBucket(); 294 } 295 296 // Didn't find the node, return null with the bucket as the InsertPos. 297 InsertPos = Bucket; 298 return 0; 299} 300 301/// InsertNode - Insert the specified node into the folding set, knowing that it 302/// is not already in the map. InsertPos must be obtained from 303/// FindNodeOrInsertPos. 304void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) { 305 assert(N->getNextInBucket() == 0); 306 // Do we need to grow the hashtable? 307 if (NumNodes+1 > NumBuckets*2) { 308 GrowHashTable(); 309 FoldingSetNodeID TempID; 310 InsertPos = GetBucketFor(ComputeNodeHash(N, TempID), Buckets, NumBuckets); 311 } 312 313 ++NumNodes; 314 315 /// The insert position is actually a bucket pointer. 316 void **Bucket = static_cast<void**>(InsertPos); 317 318 void *Next = *Bucket; 319 320 // If this is the first insertion into this bucket, its next pointer will be 321 // null. Pretend as if it pointed to itself, setting the low bit to indicate 322 // that it is a pointer to the bucket. 323 if (Next == 0) 324 Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1); 325 326 // Set the node's next pointer, and make the bucket point to the node. 327 N->SetNextInBucket(Next); 328 *Bucket = N; 329} 330 331/// RemoveNode - Remove a node from the folding set, returning true if one was 332/// removed or false if the node was not in the folding set. 333bool FoldingSetImpl::RemoveNode(Node *N) { 334 // Because each bucket is a circular list, we don't need to compute N's hash 335 // to remove it. 336 void *Ptr = N->getNextInBucket(); 337 if (Ptr == 0) return false; // Not in folding set. 338 339 --NumNodes; 340 N->SetNextInBucket(0); 341 342 // Remember what N originally pointed to, either a bucket or another node. 343 void *NodeNextPtr = Ptr; 344 345 // Chase around the list until we find the node (or bucket) which points to N. 346 while (true) { 347 if (Node *NodeInBucket = GetNextPtr(Ptr)) { 348 // Advance pointer. 349 Ptr = NodeInBucket->getNextInBucket(); 350 351 // We found a node that points to N, change it to point to N's next node, 352 // removing N from the list. 353 if (Ptr == N) { 354 NodeInBucket->SetNextInBucket(NodeNextPtr); 355 return true; 356 } 357 } else { 358 void **Bucket = GetBucketPtr(Ptr); 359 Ptr = *Bucket; 360 361 // If we found that the bucket points to N, update the bucket to point to 362 // whatever is next. 363 if (Ptr == N) { 364 *Bucket = NodeNextPtr; 365 return true; 366 } 367 } 368 } 369} 370 371/// GetOrInsertNode - If there is an existing simple Node exactly 372/// equal to the specified node, return it. Otherwise, insert 'N' and it 373/// instead. 374FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) { 375 FoldingSetNodeID ID; 376 GetNodeProfile(N, ID); 377 void *IP; 378 if (Node *E = FindNodeOrInsertPos(ID, IP)) 379 return E; 380 InsertNode(N, IP); 381 return N; 382} 383 384//===----------------------------------------------------------------------===// 385// FoldingSetIteratorImpl Implementation 386 387FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) { 388 // Skip to the first non-null non-self-cycle bucket. 389 while (*Bucket != reinterpret_cast<void*>(-1) && 390 (*Bucket == 0 || GetNextPtr(*Bucket) == 0)) 391 ++Bucket; 392 393 NodePtr = static_cast<FoldingSetNode*>(*Bucket); 394} 395 396void FoldingSetIteratorImpl::advance() { 397 // If there is another link within this bucket, go to it. 398 void *Probe = NodePtr->getNextInBucket(); 399 400 if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe)) 401 NodePtr = NextNodeInBucket; 402 else { 403 // Otherwise, this is the last link in this bucket. 404 void **Bucket = GetBucketPtr(Probe); 405 406 // Skip to the next non-null non-self-cycle bucket. 407 do { 408 ++Bucket; 409 } while (*Bucket != reinterpret_cast<void*>(-1) && 410 (*Bucket == 0 || GetNextPtr(*Bucket) == 0)); 411 412 NodePtr = static_cast<FoldingSetNode*>(*Bucket); 413 } 414} 415 416//===----------------------------------------------------------------------===// 417// FoldingSetBucketIteratorImpl Implementation 418 419FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) { 420 Ptr = (*Bucket == 0 || GetNextPtr(*Bucket) == 0) ? (void*) Bucket : *Bucket; 421} 422