FoldingSet.cpp revision 251662
174462Salfred//===-- Support/FoldingSet.cpp - Uniquing Hash Set --------------*- C++ -*-===// 274462Salfred// 3259118Shrs// The LLVM Compiler Infrastructure 4259118Shrs// 58870Srgrimes// This file is distributed under the University of Illinois Open Source 6259118Shrs// License. See LICENSE.TXT for details. 7259118Shrs// 8259118Shrs//===----------------------------------------------------------------------===// 98870Srgrimes// 10259118Shrs// This file implements a hash set that can be used to remove duplication of 11259118Shrs// nodes in a graph. 12259118Shrs// 13259118Shrs//===----------------------------------------------------------------------===// 14259118Shrs 15259118Shrs#include "llvm/ADT/FoldingSet.h" 16259118Shrs#include "llvm/ADT/Hashing.h" 17259118Shrs#include "llvm/Support/Allocator.h" 18259118Shrs#include "llvm/Support/ErrorHandling.h" 198870Srgrimes#include "llvm/Support/Host.h" 20259118Shrs#include "llvm/Support/MathExtras.h" 21259118Shrs#include <cassert> 22259118Shrs#include <cstring> 23259118Shrsusing namespace llvm; 24259118Shrs 25259118Shrs//===----------------------------------------------------------------------===// 26259118Shrs// FoldingSetNodeIDRef Implementation 27259118Shrs 28259118Shrs/// ComputeHash - Compute a strong hash value for this FoldingSetNodeIDRef, 29259118Shrs/// used to lookup the node in the FoldingSetImpl. 30259118Shrsunsigned FoldingSetNodeIDRef::ComputeHash() const { 31259118Shrs return static_cast<unsigned>(hash_combine_range(Data, Data+Size)); 321902Swollman} 331902Swollman 341902Swollmanbool FoldingSetNodeIDRef::operator==(FoldingSetNodeIDRef RHS) const { 35136582Sobrien if (Size != RHS.Size) return false; 3692986Sobrien return memcmp(Data, RHS.Data, Size*sizeof(*Data)) == 0; 371902Swollman} 3892986Sobrien 3992986Sobrien/// Used to compare the "ordering" of two nodes as defined by the 401902Swollman/// profiled bits and their ordering defined by memcmp(). 411902Swollmanbool FoldingSetNodeIDRef::operator<(FoldingSetNodeIDRef RHS) const { 421902Swollman if (Size != RHS.Size) 431902Swollman return Size < RHS.Size; 441902Swollman return memcmp(Data, RHS.Data, Size*sizeof(*Data)) < 0; 451902Swollman} 461902Swollman 471902Swollman//===----------------------------------------------------------------------===// 481902Swollman// FoldingSetNodeID Implementation 491902Swollman 5074462Salfred/// Add* - Add various data types to Bit data. 5174462Salfred/// 5274462Salfredvoid FoldingSetNodeID::AddPointer(const void *Ptr) { 5374462Salfred // Note: this adds pointers to the hash using sizes and endianness that 5474462Salfred // depend on the host. It doesn't matter however, because hashing on 5511669Sphk // pointer values in inherently unstable. Nothing should depend on the 5674462Salfred // ordering of nodes in the folding set. 571902Swollman Bits.append(reinterpret_cast<unsigned *>(&Ptr), 581902Swollman reinterpret_cast<unsigned *>(&Ptr+1)); 5974462Salfred} 601902Swollmanvoid FoldingSetNodeID::AddInteger(signed I) { 6192905Sobrien Bits.push_back(I); 6292905Sobrien} 6392905Sobrienvoid FoldingSetNodeID::AddInteger(unsigned I) { 6492905Sobrien Bits.push_back(I); 6592905Sobrien} 6692905Sobrienvoid FoldingSetNodeID::AddInteger(long I) { 6792905Sobrien AddInteger((unsigned long)I); 6874462Salfred} 6992905Sobrienvoid FoldingSetNodeID::AddInteger(unsigned long I) { 7092905Sobrien if (sizeof(long) == sizeof(int)) 7192905Sobrien AddInteger(unsigned(I)); 7292905Sobrien else if (sizeof(long) == sizeof(long long)) { 731902Swollman AddInteger((unsigned long long)I); 7474462Salfred } else { 7521062Speter llvm_unreachable("unexpected sizeof(long)"); 7621062Speter } 771902Swollman} 781902Swollmanvoid FoldingSetNodeID::AddInteger(long long I) { 791902Swollman AddInteger((unsigned long long)I); 801902Swollman} 8121062Spetervoid FoldingSetNodeID::AddInteger(unsigned long long I) { 821902Swollman AddInteger(unsigned(I)); 831902Swollman if ((uint64_t)(unsigned)I != I) 841902Swollman Bits.push_back(unsigned(I >> 32)); 8574462Salfred} 8621062Speter 8721062Spetervoid FoldingSetNodeID::AddString(StringRef String) { 8821062Speter unsigned Size = String.size(); 8921062Speter Bits.push_back(Size); 9021062Speter if (!Size) return; 9121062Speter 9221062Speter unsigned Units = Size / 4; 9321062Speter unsigned Pos = 0; 9421062Speter const unsigned *Base = (const unsigned*) String.data(); 9521062Speter 961902Swollman // If the string is aligned do a bulk transfer. 971902Swollman if (!((intptr_t)Base & 3)) { 988870Srgrimes Bits.append(Base, Base + Units); 991902Swollman Pos = (Units + 1) * 4; 1001902Swollman } else { 101283833Srodrigc // Otherwise do it the hard way. 1021902Swollman // To be compatible with above bulk transfer, we need to take endianness 1031902Swollman // into account. 1041902Swollman if (sys::IsBigEndianHost) { 10574462Salfred for (Pos += 4; Pos <= Size; Pos += 4) { 10674462Salfred unsigned V = ((unsigned char)String[Pos - 4] << 24) | 1071902Swollman ((unsigned char)String[Pos - 3] << 16) | 1081902Swollman ((unsigned char)String[Pos - 2] << 8) | 1091902Swollman (unsigned char)String[Pos - 1]; 1101902Swollman Bits.push_back(V); 11174462Salfred } 1121902Swollman } else { 113283833Srodrigc assert(sys::IsLittleEndianHost && "Unexpected host endianness"); 1141902Swollman for (Pos += 4; Pos <= Size; Pos += 4) { 11521062Speter unsigned V = ((unsigned char)String[Pos - 1] << 24) | 1161902Swollman ((unsigned char)String[Pos - 2] << 16) | 1171902Swollman ((unsigned char)String[Pos - 3] << 8) | 1181902Swollman (unsigned char)String[Pos - 4]; 119283833Srodrigc Bits.push_back(V); 1201902Swollman } 1211902Swollman } 122111962Snectar } 1231902Swollman 124111962Snectar // With the leftover bits. 12574462Salfred unsigned V = 0; 12674462Salfred // Pos will have overshot size by 4 - #bytes left over. 1271902Swollman // No need to take endianness into account here - this is always executed. 1281902Swollman switch (Pos - Size) { 1291902Swollman case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru. 1301902Swollman case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru. 131283833Srodrigc case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break; 1321902Swollman default: return; // Nothing left. 1331902Swollman } 134111962Snectar 1351902Swollman Bits.push_back(V); 136111962Snectar} 13774462Salfred 13874462Salfred// AddNodeID - Adds the Bit data of another ID to *this. 1391902Swollmanvoid FoldingSetNodeID::AddNodeID(const FoldingSetNodeID &ID) { 1401902Swollman Bits.append(ID.Bits.begin(), ID.Bits.end()); 1411902Swollman} 1421902Swollman 143283833Srodrigc/// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to 14421062Speter/// lookup the node in the FoldingSetImpl. 14574462Salfredunsigned FoldingSetNodeID::ComputeHash() const { 14621062Speter return FoldingSetNodeIDRef(Bits.data(), Bits.size()).ComputeHash(); 147111962Snectar} 14821062Speter 149111962Snectar/// operator== - Used to compare two nodes to each other. 15074462Salfred/// 15121062Speterbool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS) const { 15274462Salfred return *this == FoldingSetNodeIDRef(RHS.Bits.data(), RHS.Bits.size()); 15321062Speter} 15421062Speter 15521062Speter/// operator== - Used to compare two nodes to each other. 15621062Speter/// 157283833Srodrigcbool FoldingSetNodeID::operator==(FoldingSetNodeIDRef RHS) const { 15821062Speter return FoldingSetNodeIDRef(Bits.data(), Bits.size()) == RHS; 15974462Salfred} 16021062Speter 161111962Snectar/// Used to compare the "ordering" of two nodes as defined by the 16221062Speter/// profiled bits and their ordering defined by memcmp(). 163111962Snectarbool FoldingSetNodeID::operator<(const FoldingSetNodeID &RHS) const { 16474462Salfred return *this < FoldingSetNodeIDRef(RHS.Bits.data(), RHS.Bits.size()); 16574462Salfred} 16674462Salfred 16721062Speterbool FoldingSetNodeID::operator<(FoldingSetNodeIDRef RHS) const { 16821062Speter return FoldingSetNodeIDRef(Bits.data(), Bits.size()) < RHS; 16921062Speter} 17021062Speter 171283833Srodrigc/// Intern - Copy this node's data to a memory region allocated from the 1721902Swollman/// given allocator and return a FoldingSetNodeIDRef describing the 1731902Swollman/// interned data. 174111962SnectarFoldingSetNodeIDRef 1751902SwollmanFoldingSetNodeID::Intern(BumpPtrAllocator &Allocator) const { 176111962Snectar unsigned *New = Allocator.Allocate<unsigned>(Bits.size()); 17774462Salfred std::uninitialized_copy(Bits.begin(), Bits.end(), New); 17874462Salfred return FoldingSetNodeIDRef(New, Bits.size()); 1791902Swollman} 1801902Swollman 1811902Swollman//===----------------------------------------------------------------------===// 1821902Swollman/// Helper functions for FoldingSetImpl. 183283833Srodrigc 1841902Swollman/// GetNextPtr - In order to save space, each bucket is a 1851902Swollman/// singly-linked-list. In order to make deletion more efficient, we make 186111962Snectar/// the list circular, so we can delete a node without computing its hash. 1871902Swollman/// The problem with this is that the start of the hash buckets are not 188111962Snectar/// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null: 18974462Salfred/// use GetBucketPtr when this happens. 19074462Salfredstatic FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr) { 1911902Swollman // The low bit is set if this is the pointer back to the bucket. 1921902Swollman if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1) 1931902Swollman return 0; 1941902Swollman 195283833Srodrigc return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr); 1961902Swollman} 1971902Swollman 19821062Speter 19974462Salfred/// testing. 2001902Swollmanstatic void **GetBucketPtr(void *NextInBucketPtr) { 2011902Swollman intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr); 2021902Swollman assert((Ptr & 1) && "Not a bucket pointer"); 203283833Srodrigc return reinterpret_cast<void**>(Ptr & ~intptr_t(1)); 2041902Swollman} 20574462Salfred 20674462Salfred/// GetBucketFor - Hash the specified node ID and return the hash bucket for 2071902Swollman/// the specified ID. 208111962Snectarstatic void **GetBucketFor(unsigned Hash, void **Buckets, unsigned NumBuckets) { 2091902Swollman // NumBuckets is always a power of 2. 2101902Swollman unsigned BucketNum = Hash & (NumBuckets-1); 211111962Snectar return Buckets + BucketNum; 2121902Swollman} 2131902Swollman 2141902Swollman/// AllocateBuckets - Allocated initialized bucket memory. 21521062Speterstatic void **AllocateBuckets(unsigned NumBuckets) { 216283833Srodrigc void **Buckets = static_cast<void**>(calloc(NumBuckets+1, sizeof(void*))); 2171902Swollman // Set the very last bucket to be a non-null "pointer". 218297790Spfg Buckets[NumBuckets] = reinterpret_cast<void*>(-1); 2191902Swollman return Buckets; 2201902Swollman} 2211902Swollman 22274462Salfred//===----------------------------------------------------------------------===// 22374462Salfred// FoldingSetImpl Implementation 2241902Swollman 2251902SwollmanFoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) { 2261902Swollman assert(5 < Log2InitSize && Log2InitSize < 32 && 22721062Speter "Initial hash table size out of range"); 22874462Salfred NumBuckets = 1 << Log2InitSize; 22921062Speter Buckets = AllocateBuckets(NumBuckets); 230283833Srodrigc NumNodes = 0; 23121062Speter} 23274462SalfredFoldingSetImpl::~FoldingSetImpl() { 23321062Speter free(Buckets); 23421062Speter} 235void FoldingSetImpl::clear() { 236 // Set all but the last bucket to null pointers. 237 memset(Buckets, 0, NumBuckets*sizeof(void*)); 238 239 // Set the very last bucket to be a non-null "pointer". 240 Buckets[NumBuckets] = reinterpret_cast<void*>(-1); 241 242 // Reset the node count to zero. 243 NumNodes = 0; 244} 245 246/// GrowHashTable - Double the size of the hash table and rehash everything. 247/// 248void FoldingSetImpl::GrowHashTable() { 249 void **OldBuckets = Buckets; 250 unsigned OldNumBuckets = NumBuckets; 251 NumBuckets <<= 1; 252 253 // Clear out new buckets. 254 Buckets = AllocateBuckets(NumBuckets); 255 NumNodes = 0; 256 257 // Walk the old buckets, rehashing nodes into their new place. 258 FoldingSetNodeID TempID; 259 for (unsigned i = 0; i != OldNumBuckets; ++i) { 260 void *Probe = OldBuckets[i]; 261 if (!Probe) continue; 262 while (Node *NodeInBucket = GetNextPtr(Probe)) { 263 // Figure out the next link, remove NodeInBucket from the old link. 264 Probe = NodeInBucket->getNextInBucket(); 265 NodeInBucket->SetNextInBucket(0); 266 267 // Insert the node into the new bucket, after recomputing the hash. 268 InsertNode(NodeInBucket, 269 GetBucketFor(ComputeNodeHash(NodeInBucket, TempID), 270 Buckets, NumBuckets)); 271 TempID.clear(); 272 } 273 } 274 275 free(OldBuckets); 276} 277 278/// FindNodeOrInsertPos - Look up the node specified by ID. If it exists, 279/// return it. If not, return the insertion token that will make insertion 280/// faster. 281FoldingSetImpl::Node 282*FoldingSetImpl::FindNodeOrInsertPos(const FoldingSetNodeID &ID, 283 void *&InsertPos) { 284 unsigned IDHash = ID.ComputeHash(); 285 void **Bucket = GetBucketFor(IDHash, Buckets, NumBuckets); 286 void *Probe = *Bucket; 287 288 InsertPos = 0; 289 290 FoldingSetNodeID TempID; 291 while (Node *NodeInBucket = GetNextPtr(Probe)) { 292 if (NodeEquals(NodeInBucket, ID, IDHash, TempID)) 293 return NodeInBucket; 294 TempID.clear(); 295 296 Probe = NodeInBucket->getNextInBucket(); 297 } 298 299 // Didn't find the node, return null with the bucket as the InsertPos. 300 InsertPos = Bucket; 301 return 0; 302} 303 304/// InsertNode - Insert the specified node into the folding set, knowing that it 305/// is not already in the map. InsertPos must be obtained from 306/// FindNodeOrInsertPos. 307void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) { 308 assert(N->getNextInBucket() == 0); 309 // Do we need to grow the hashtable? 310 if (NumNodes+1 > NumBuckets*2) { 311 GrowHashTable(); 312 FoldingSetNodeID TempID; 313 InsertPos = GetBucketFor(ComputeNodeHash(N, TempID), Buckets, NumBuckets); 314 } 315 316 ++NumNodes; 317 318 /// The insert position is actually a bucket pointer. 319 void **Bucket = static_cast<void**>(InsertPos); 320 321 void *Next = *Bucket; 322 323 // If this is the first insertion into this bucket, its next pointer will be 324 // null. Pretend as if it pointed to itself, setting the low bit to indicate 325 // that it is a pointer to the bucket. 326 if (Next == 0) 327 Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1); 328 329 // Set the node's next pointer, and make the bucket point to the node. 330 N->SetNextInBucket(Next); 331 *Bucket = N; 332} 333 334/// RemoveNode - Remove a node from the folding set, returning true if one was 335/// removed or false if the node was not in the folding set. 336bool FoldingSetImpl::RemoveNode(Node *N) { 337 // Because each bucket is a circular list, we don't need to compute N's hash 338 // to remove it. 339 void *Ptr = N->getNextInBucket(); 340 if (Ptr == 0) return false; // Not in folding set. 341 342 --NumNodes; 343 N->SetNextInBucket(0); 344 345 // Remember what N originally pointed to, either a bucket or another node. 346 void *NodeNextPtr = Ptr; 347 348 // Chase around the list until we find the node (or bucket) which points to N. 349 while (true) { 350 if (Node *NodeInBucket = GetNextPtr(Ptr)) { 351 // Advance pointer. 352 Ptr = NodeInBucket->getNextInBucket(); 353 354 // We found a node that points to N, change it to point to N's next node, 355 // removing N from the list. 356 if (Ptr == N) { 357 NodeInBucket->SetNextInBucket(NodeNextPtr); 358 return true; 359 } 360 } else { 361 void **Bucket = GetBucketPtr(Ptr); 362 Ptr = *Bucket; 363 364 // If we found that the bucket points to N, update the bucket to point to 365 // whatever is next. 366 if (Ptr == N) { 367 *Bucket = NodeNextPtr; 368 return true; 369 } 370 } 371 } 372} 373 374/// GetOrInsertNode - If there is an existing simple Node exactly 375/// equal to the specified node, return it. Otherwise, insert 'N' and it 376/// instead. 377FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) { 378 FoldingSetNodeID ID; 379 GetNodeProfile(N, ID); 380 void *IP; 381 if (Node *E = FindNodeOrInsertPos(ID, IP)) 382 return E; 383 InsertNode(N, IP); 384 return N; 385} 386 387//===----------------------------------------------------------------------===// 388// FoldingSetIteratorImpl Implementation 389 390FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) { 391 // Skip to the first non-null non-self-cycle bucket. 392 while (*Bucket != reinterpret_cast<void*>(-1) && 393 (*Bucket == 0 || GetNextPtr(*Bucket) == 0)) 394 ++Bucket; 395 396 NodePtr = static_cast<FoldingSetNode*>(*Bucket); 397} 398 399void FoldingSetIteratorImpl::advance() { 400 // If there is another link within this bucket, go to it. 401 void *Probe = NodePtr->getNextInBucket(); 402 403 if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe)) 404 NodePtr = NextNodeInBucket; 405 else { 406 // Otherwise, this is the last link in this bucket. 407 void **Bucket = GetBucketPtr(Probe); 408 409 // Skip to the next non-null non-self-cycle bucket. 410 do { 411 ++Bucket; 412 } while (*Bucket != reinterpret_cast<void*>(-1) && 413 (*Bucket == 0 || GetNextPtr(*Bucket) == 0)); 414 415 NodePtr = static_cast<FoldingSetNode*>(*Bucket); 416 } 417} 418 419//===----------------------------------------------------------------------===// 420// FoldingSetBucketIteratorImpl Implementation 421 422FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) { 423 Ptr = (*Bucket == 0 || GetNextPtr(*Bucket) == 0) ? (void*) Bucket : *Bucket; 424} 425