FoldingSet.cpp revision 205407
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 <cassert> 22#include <cstring> 23using namespace llvm; 24 25//===----------------------------------------------------------------------===// 26// FoldingSetNodeID Implementation 27 28/// Add* - Add various data types to Bit data. 29/// 30void FoldingSetNodeID::AddPointer(const void *Ptr) { 31 // Note: this adds pointers to the hash using sizes and endianness that 32 // depend on the host. It doesn't matter however, because hashing on 33 // pointer values in inherently unstable. Nothing should depend on the 34 // ordering of nodes in the folding set. 35 intptr_t PtrI = (intptr_t)Ptr; 36 Bits.push_back(unsigned(PtrI)); 37 if (sizeof(intptr_t) > sizeof(unsigned)) 38 Bits.push_back(unsigned(uint64_t(PtrI) >> 32)); 39} 40void FoldingSetNodeID::AddInteger(signed I) { 41 Bits.push_back(I); 42} 43void FoldingSetNodeID::AddInteger(unsigned I) { 44 Bits.push_back(I); 45} 46void FoldingSetNodeID::AddInteger(long I) { 47 AddInteger((unsigned long)I); 48} 49void FoldingSetNodeID::AddInteger(unsigned long I) { 50 if (sizeof(long) == sizeof(int)) 51 AddInteger(unsigned(I)); 52 else if (sizeof(long) == sizeof(long long)) { 53 AddInteger((unsigned long long)I); 54 } else { 55 llvm_unreachable("unexpected sizeof(long)"); 56 } 57} 58void FoldingSetNodeID::AddInteger(long long I) { 59 AddInteger((unsigned long long)I); 60} 61void FoldingSetNodeID::AddInteger(unsigned long long I) { 62 AddInteger(unsigned(I)); 63 if ((uint64_t)(int)I != I) 64 Bits.push_back(unsigned(I >> 32)); 65} 66 67void FoldingSetNodeID::AddString(StringRef String) { 68 unsigned Size = String.size(); 69 Bits.push_back(Size); 70 if (!Size) return; 71 72 unsigned Units = Size / 4; 73 unsigned Pos = 0; 74 const unsigned *Base = (const unsigned*) String.data(); 75 76 // If the string is aligned do a bulk transfer. 77 if (!((intptr_t)Base & 3)) { 78 Bits.append(Base, Base + Units); 79 Pos = (Units + 1) * 4; 80 } else { 81 // Otherwise do it the hard way. 82 for (Pos += 4; Pos <= Size; Pos += 4) { 83 unsigned V = ((unsigned char)String[Pos - 4] << 24) | 84 ((unsigned char)String[Pos - 3] << 16) | 85 ((unsigned char)String[Pos - 2] << 8) | 86 (unsigned char)String[Pos - 1]; 87 Bits.push_back(V); 88 } 89 } 90 91 // With the leftover bits. 92 unsigned V = 0; 93 // Pos will have overshot size by 4 - #bytes left over. 94 switch (Pos - Size) { 95 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru. 96 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru. 97 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break; 98 default: return; // Nothing left. 99 } 100 101 Bits.push_back(V); 102} 103 104/// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to 105/// lookup the node in the FoldingSetImpl. 106unsigned FoldingSetNodeID::ComputeHash() const { 107 // This is adapted from SuperFastHash by Paul Hsieh. 108 unsigned Hash = static_cast<unsigned>(Bits.size()); 109 for (const unsigned *BP = &Bits[0], *E = BP+Bits.size(); BP != E; ++BP) { 110 unsigned Data = *BP; 111 Hash += Data & 0xFFFF; 112 unsigned Tmp = ((Data >> 16) << 11) ^ Hash; 113 Hash = (Hash << 16) ^ Tmp; 114 Hash += Hash >> 11; 115 } 116 117 // Force "avalanching" of final 127 bits. 118 Hash ^= Hash << 3; 119 Hash += Hash >> 5; 120 Hash ^= Hash << 4; 121 Hash += Hash >> 17; 122 Hash ^= Hash << 25; 123 Hash += Hash >> 6; 124 return Hash; 125} 126 127/// operator== - Used to compare two nodes to each other. 128/// 129bool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS)const{ 130 if (Bits.size() != RHS.Bits.size()) return false; 131 return memcmp(&Bits[0], &RHS.Bits[0], Bits.size()*sizeof(Bits[0])) == 0; 132} 133 134/// Intern - Copy this node's data to a memory region allocated from the 135/// given allocator and return a FoldingSetNodeIDRef describing the 136/// interned data. 137FoldingSetNodeIDRef 138FoldingSetNodeID::Intern(BumpPtrAllocator &Allocator) const { 139 unsigned *New = Allocator.Allocate<unsigned>(Bits.size()); 140 std::uninitialized_copy(Bits.begin(), Bits.end(), New); 141 return FoldingSetNodeIDRef(New, Bits.size()); 142} 143 144//===----------------------------------------------------------------------===// 145/// Helper functions for FoldingSetImpl. 146 147/// GetNextPtr - In order to save space, each bucket is a 148/// singly-linked-list. In order to make deletion more efficient, we make 149/// the list circular, so we can delete a node without computing its hash. 150/// The problem with this is that the start of the hash buckets are not 151/// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null: 152/// use GetBucketPtr when this happens. 153static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr) { 154 // The low bit is set if this is the pointer back to the bucket. 155 if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1) 156 return 0; 157 158 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr); 159} 160 161 162/// testing. 163static void **GetBucketPtr(void *NextInBucketPtr) { 164 intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr); 165 assert((Ptr & 1) && "Not a bucket pointer"); 166 return reinterpret_cast<void**>(Ptr & ~intptr_t(1)); 167} 168 169/// GetBucketFor - Hash the specified node ID and return the hash bucket for 170/// the specified ID. 171static void **GetBucketFor(const FoldingSetNodeID &ID, 172 void **Buckets, unsigned NumBuckets) { 173 // NumBuckets is always a power of 2. 174 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1); 175 return Buckets + BucketNum; 176} 177 178//===----------------------------------------------------------------------===// 179// FoldingSetImpl Implementation 180 181FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) { 182 assert(5 < Log2InitSize && Log2InitSize < 32 && 183 "Initial hash table size out of range"); 184 NumBuckets = 1 << Log2InitSize; 185 Buckets = new void*[NumBuckets+1]; 186 clear(); 187} 188FoldingSetImpl::~FoldingSetImpl() { 189 delete [] Buckets; 190} 191void FoldingSetImpl::clear() { 192 // Set all but the last bucket to null pointers. 193 memset(Buckets, 0, NumBuckets*sizeof(void*)); 194 195 // Set the very last bucket to be a non-null "pointer". 196 Buckets[NumBuckets] = reinterpret_cast<void*>(-1); 197 198 // Reset the node count to zero. 199 NumNodes = 0; 200} 201 202/// GrowHashTable - Double the size of the hash table and rehash everything. 203/// 204void FoldingSetImpl::GrowHashTable() { 205 void **OldBuckets = Buckets; 206 unsigned OldNumBuckets = NumBuckets; 207 NumBuckets <<= 1; 208 209 // Clear out new buckets. 210 Buckets = new void*[NumBuckets+1]; 211 clear(); 212 213 // Walk the old buckets, rehashing nodes into their new place. 214 FoldingSetNodeID ID; 215 for (unsigned i = 0; i != OldNumBuckets; ++i) { 216 void *Probe = OldBuckets[i]; 217 if (!Probe) continue; 218 while (Node *NodeInBucket = GetNextPtr(Probe)) { 219 // Figure out the next link, remove NodeInBucket from the old link. 220 Probe = NodeInBucket->getNextInBucket(); 221 NodeInBucket->SetNextInBucket(0); 222 223 // Insert the node into the new bucket, after recomputing the hash. 224 GetNodeProfile(ID, NodeInBucket); 225 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets)); 226 ID.clear(); 227 } 228 } 229 230 delete[] OldBuckets; 231} 232 233/// FindNodeOrInsertPos - Look up the node specified by ID. If it exists, 234/// return it. If not, return the insertion token that will make insertion 235/// faster. 236FoldingSetImpl::Node 237*FoldingSetImpl::FindNodeOrInsertPos(const FoldingSetNodeID &ID, 238 void *&InsertPos) { 239 240 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets); 241 void *Probe = *Bucket; 242 243 InsertPos = 0; 244 245 FoldingSetNodeID OtherID; 246 while (Node *NodeInBucket = GetNextPtr(Probe)) { 247 GetNodeProfile(OtherID, NodeInBucket); 248 if (OtherID == ID) 249 return NodeInBucket; 250 251 Probe = NodeInBucket->getNextInBucket(); 252 OtherID.clear(); 253 } 254 255 // Didn't find the node, return null with the bucket as the InsertPos. 256 InsertPos = Bucket; 257 return 0; 258} 259 260/// InsertNode - Insert the specified node into the folding set, knowing that it 261/// is not already in the map. InsertPos must be obtained from 262/// FindNodeOrInsertPos. 263void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) { 264 assert(N->getNextInBucket() == 0); 265 // Do we need to grow the hashtable? 266 if (NumNodes+1 > NumBuckets*2) { 267 GrowHashTable(); 268 FoldingSetNodeID ID; 269 GetNodeProfile(ID, N); 270 InsertPos = GetBucketFor(ID, Buckets, NumBuckets); 271 } 272 273 ++NumNodes; 274 275 /// The insert position is actually a bucket pointer. 276 void **Bucket = static_cast<void**>(InsertPos); 277 278 void *Next = *Bucket; 279 280 // If this is the first insertion into this bucket, its next pointer will be 281 // null. Pretend as if it pointed to itself, setting the low bit to indicate 282 // that it is a pointer to the bucket. 283 if (Next == 0) 284 Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1); 285 286 // Set the node's next pointer, and make the bucket point to the node. 287 N->SetNextInBucket(Next); 288 *Bucket = N; 289} 290 291/// RemoveNode - Remove a node from the folding set, returning true if one was 292/// removed or false if the node was not in the folding set. 293bool FoldingSetImpl::RemoveNode(Node *N) { 294 // Because each bucket is a circular list, we don't need to compute N's hash 295 // to remove it. 296 void *Ptr = N->getNextInBucket(); 297 if (Ptr == 0) return false; // Not in folding set. 298 299 --NumNodes; 300 N->SetNextInBucket(0); 301 302 // Remember what N originally pointed to, either a bucket or another node. 303 void *NodeNextPtr = Ptr; 304 305 // Chase around the list until we find the node (or bucket) which points to N. 306 while (true) { 307 if (Node *NodeInBucket = GetNextPtr(Ptr)) { 308 // Advance pointer. 309 Ptr = NodeInBucket->getNextInBucket(); 310 311 // We found a node that points to N, change it to point to N's next node, 312 // removing N from the list. 313 if (Ptr == N) { 314 NodeInBucket->SetNextInBucket(NodeNextPtr); 315 return true; 316 } 317 } else { 318 void **Bucket = GetBucketPtr(Ptr); 319 Ptr = *Bucket; 320 321 // If we found that the bucket points to N, update the bucket to point to 322 // whatever is next. 323 if (Ptr == N) { 324 *Bucket = NodeNextPtr; 325 return true; 326 } 327 } 328 } 329} 330 331/// GetOrInsertNode - If there is an existing simple Node exactly 332/// equal to the specified node, return it. Otherwise, insert 'N' and it 333/// instead. 334FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) { 335 FoldingSetNodeID ID; 336 GetNodeProfile(ID, N); 337 void *IP; 338 if (Node *E = FindNodeOrInsertPos(ID, IP)) 339 return E; 340 InsertNode(N, IP); 341 return N; 342} 343 344//===----------------------------------------------------------------------===// 345// FoldingSetIteratorImpl Implementation 346 347FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) { 348 // Skip to the first non-null non-self-cycle bucket. 349 while (*Bucket != reinterpret_cast<void*>(-1) && 350 (*Bucket == 0 || GetNextPtr(*Bucket) == 0)) 351 ++Bucket; 352 353 NodePtr = static_cast<FoldingSetNode*>(*Bucket); 354} 355 356void FoldingSetIteratorImpl::advance() { 357 // If there is another link within this bucket, go to it. 358 void *Probe = NodePtr->getNextInBucket(); 359 360 if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe)) 361 NodePtr = NextNodeInBucket; 362 else { 363 // Otherwise, this is the last link in this bucket. 364 void **Bucket = GetBucketPtr(Probe); 365 366 // Skip to the next non-null non-self-cycle bucket. 367 do { 368 ++Bucket; 369 } while (*Bucket != reinterpret_cast<void*>(-1) && 370 (*Bucket == 0 || GetNextPtr(*Bucket) == 0)); 371 372 NodePtr = static_cast<FoldingSetNode*>(*Bucket); 373 } 374} 375 376//===----------------------------------------------------------------------===// 377// FoldingSetBucketIteratorImpl Implementation 378 379FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) { 380 Ptr = (*Bucket == 0 || GetNextPtr(*Bucket) == 0) ? (void*) Bucket : *Bucket; 381} 382