FoldingSet.cpp revision 39c6d3aac1a049a9e9696b20fb9e9fdd6e906b91
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/Support/MathExtras.h" 19#include <cassert> 20using namespace llvm; 21 22//===----------------------------------------------------------------------===// 23// FoldingSetImpl::NodeID Implementation 24 25/// Add* - Add various data types to Bit data. 26/// 27void FoldingSetImpl::NodeID::AddPointer(const void *Ptr) { 28 // Note: this adds pointers to the hash using sizes and endianness that 29 // depend on the host. It doesn't matter however, because hashing on 30 // pointer values in inherently unstable. Nothing should depend on the 31 // ordering of nodes in the folding set. 32 intptr_t PtrI = (intptr_t)Ptr; 33 Bits.push_back(unsigned(PtrI)); 34 if (sizeof(intptr_t) > sizeof(unsigned)) 35 Bits.push_back(unsigned(uint64_t(PtrI) >> 32)); 36} 37void FoldingSetImpl::NodeID::AddInteger(signed I) { 38 Bits.push_back(I); 39} 40void FoldingSetImpl::NodeID::AddInteger(unsigned I) { 41 Bits.push_back(I); 42} 43void FoldingSetImpl::NodeID::AddInteger(uint64_t I) { 44 Bits.push_back(unsigned(I)); 45 Bits.push_back(unsigned(I >> 32)); 46} 47void FoldingSetImpl::NodeID::AddFloat(float F) { 48 Bits.push_back(FloatToBits(F)); 49} 50void FoldingSetImpl::NodeID::AddDouble(double D) { 51 AddInteger(DoubleToBits(D)); 52} 53void FoldingSetImpl::NodeID::AddString(const std::string &String) { 54 unsigned Size = String.size(); 55 Bits.push_back(Size); 56 if (!Size) return; 57 58 unsigned Units = Size / 4; 59 unsigned Pos = 0; 60 const unsigned *Base = (const unsigned *)String.data(); 61 62 // If the string is aligned do a bulk transfer. 63 if (!((intptr_t)Base & 3)) { 64 Bits.append(Base, Base + Units); 65 Pos = (Units + 1) * 4; 66 } else { 67 // Otherwise do it the hard way. 68 for ( Pos += 4; Pos <= Size; Pos += 4) { 69 unsigned V = ((unsigned char)String[Pos - 4] << 24) | 70 ((unsigned char)String[Pos - 3] << 16) | 71 ((unsigned char)String[Pos - 2] << 8) | 72 (unsigned char)String[Pos - 1]; 73 Bits.push_back(V); 74 } 75 } 76 77 // With the leftover bits. 78 unsigned V = 0; 79 // Pos will have overshot size by 4 - #bytes left over. 80 switch (Pos - Size) { 81 case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru. 82 case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru. 83 case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break; 84 default: return; // Nothing left. 85 } 86 87 Bits.push_back(V); 88} 89 90/// ComputeHash - Compute a strong hash value for this NodeID, used to 91/// lookup the node in the FoldingSetImpl. 92unsigned FoldingSetImpl::NodeID::ComputeHash() const { 93 // This is adapted from SuperFastHash by Paul Hsieh. 94 unsigned Hash = Bits.size(); 95 for (const unsigned *BP = &Bits[0], *E = BP+Bits.size(); BP != E; ++BP) { 96 unsigned Data = *BP; 97 Hash += Data & 0xFFFF; 98 unsigned Tmp = ((Data >> 16) << 11) ^ Hash; 99 Hash = (Hash << 16) ^ Tmp; 100 Hash += Hash >> 11; 101 } 102 103 // Force "avalanching" of final 127 bits. 104 Hash ^= Hash << 3; 105 Hash += Hash >> 5; 106 Hash ^= Hash << 4; 107 Hash += Hash >> 17; 108 Hash ^= Hash << 25; 109 Hash += Hash >> 6; 110 return Hash; 111} 112 113/// operator== - Used to compare two nodes to each other. 114/// 115bool FoldingSetImpl::NodeID::operator==(const FoldingSetImpl::NodeID &RHS)const{ 116 if (Bits.size() != RHS.Bits.size()) return false; 117 return memcmp(&Bits[0], &RHS.Bits[0], Bits.size()*sizeof(Bits[0])) == 0; 118} 119 120 121//===----------------------------------------------------------------------===// 122/// Helper functions for FoldingSetImpl. 123 124/// GetNextPtr - In order to save space, each bucket is a 125/// singly-linked-list. In order to make deletion more efficient, we make 126/// the list circular, so we can delete a node without computing its hash. 127/// The problem with this is that the start of the hash buckets are not 128/// Nodes. If NextInBucketPtr is a bucket pointer, this method returns null 129/// : use GetBucketPtr when this happens. 130static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr, 131 void **Buckets, unsigned NumBuckets) { 132 if (NextInBucketPtr >= Buckets && NextInBucketPtr < Buckets + NumBuckets) 133 return 0; 134 return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr); 135} 136 137/// GetBucketPtr - Provides a casting of a bucket pointer for isNode 138/// testing. 139static void **GetBucketPtr(void *NextInBucketPtr) { 140 return static_cast<void**>(NextInBucketPtr); 141} 142 143/// GetBucketFor - Hash the specified node ID and return the hash bucket for 144/// the specified ID. 145static void **GetBucketFor(const FoldingSetImpl::NodeID &ID, 146 void **Buckets, unsigned NumBuckets) { 147 // NumBuckets is always a power of 2. 148 unsigned BucketNum = ID.ComputeHash() & (NumBuckets-1); 149 return Buckets + BucketNum; 150} 151 152//===----------------------------------------------------------------------===// 153// FoldingSetImpl Implementation 154 155FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) : NumNodes(0) { 156 assert(5 < Log2InitSize && Log2InitSize < 32 && 157 "Initial hash table size out of range"); 158 NumBuckets = 1 << Log2InitSize; 159 Buckets = new void*[NumBuckets]; 160 memset(Buckets, 0, NumBuckets*sizeof(void*)); 161} 162FoldingSetImpl::~FoldingSetImpl() { 163 delete [] Buckets; 164} 165 166/// GrowHashTable - Double the size of the hash table and rehash everything. 167/// 168void FoldingSetImpl::GrowHashTable() { 169 void **OldBuckets = Buckets; 170 unsigned OldNumBuckets = NumBuckets; 171 NumBuckets <<= 1; 172 173 // Reset the node count to zero: we're going to reinsert everything. 174 NumNodes = 0; 175 176 // Clear out new buckets. 177 Buckets = new void*[NumBuckets]; 178 memset(Buckets, 0, NumBuckets*sizeof(void*)); 179 180 // Walk the old buckets, rehashing nodes into their new place. 181 for (unsigned i = 0; i != OldNumBuckets; ++i) { 182 void *Probe = OldBuckets[i]; 183 if (!Probe) continue; 184 while (Node *NodeInBucket = GetNextPtr(Probe, OldBuckets, OldNumBuckets)){ 185 // Figure out the next link, remove NodeInBucket from the old link. 186 Probe = NodeInBucket->getNextInBucket(); 187 NodeInBucket->SetNextInBucket(0); 188 189 // Insert the node into the new bucket, after recomputing the hash. 190 NodeID ID; 191 GetNodeProfile(ID, NodeInBucket); 192 InsertNode(NodeInBucket, GetBucketFor(ID, Buckets, NumBuckets)); 193 } 194 } 195 196 delete[] OldBuckets; 197} 198 199/// FindNodeOrInsertPos - Look up the node specified by ID. If it exists, 200/// return it. If not, return the insertion token that will make insertion 201/// faster. 202FoldingSetImpl::Node *FoldingSetImpl::FindNodeOrInsertPos(const NodeID &ID, 203 void *&InsertPos) { 204 void **Bucket = GetBucketFor(ID, Buckets, NumBuckets); 205 void *Probe = *Bucket; 206 207 InsertPos = 0; 208 209 while (Node *NodeInBucket = GetNextPtr(Probe, Buckets, NumBuckets)) { 210 NodeID OtherID; 211 GetNodeProfile(OtherID, NodeInBucket); 212 if (OtherID == ID) 213 return NodeInBucket; 214 215 Probe = NodeInBucket->getNextInBucket(); 216 } 217 218 // Didn't find the node, return null with the bucket as the InsertPos. 219 InsertPos = Bucket; 220 return 0; 221} 222 223/// InsertNode - Insert the specified node into the folding set, knowing that it 224/// is not already in the map. InsertPos must be obtained from 225/// FindNodeOrInsertPos. 226void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) { 227 ++NumNodes; 228 // Do we need to grow the hashtable? 229 if (NumNodes > NumBuckets*2) { 230 GrowHashTable(); 231 NodeID ID; 232 GetNodeProfile(ID, N); 233 InsertPos = GetBucketFor(ID, Buckets, NumBuckets); 234 } 235 236 /// The insert position is actually a bucket pointer. 237 void **Bucket = static_cast<void**>(InsertPos); 238 239 void *Next = *Bucket; 240 241 // If this is the first insertion into this bucket, its next pointer will be 242 // null. Pretend as if it pointed to itself. 243 if (Next == 0) 244 Next = Bucket; 245 246 // Set the nodes next pointer, and make the bucket point to the node. 247 N->SetNextInBucket(Next); 248 *Bucket = N; 249} 250 251/// RemoveNode - Remove a node from the folding set, returning true if one was 252/// removed or false if the node was not in the folding set. 253bool FoldingSetImpl::RemoveNode(Node *N) { 254 // Because each bucket is a circular list, we don't need to compute N's hash 255 // to remove it. Chase around the list until we find the node (or bucket) 256 // which points to N. 257 void *Ptr = N->getNextInBucket(); 258 if (Ptr == 0) return false; // Not in folding set. 259 260 --NumNodes; 261 262 void *NodeNextPtr = Ptr; 263 N->SetNextInBucket(0); 264 while (true) { 265 if (Node *NodeInBucket = GetNextPtr(Ptr, Buckets, NumBuckets)) { 266 // Advance pointer. 267 Ptr = NodeInBucket->getNextInBucket(); 268 269 // We found a node that points to N, change it to point to N's next node, 270 // removing N from the list. 271 if (Ptr == N) { 272 NodeInBucket->SetNextInBucket(NodeNextPtr); 273 return true; 274 } 275 } else { 276 void **Bucket = GetBucketPtr(Ptr); 277 Ptr = *Bucket; 278 279 // If we found that the bucket points to N, update the bucket to point to 280 // whatever is next. 281 if (Ptr == N) { 282 *Bucket = NodeNextPtr; 283 return true; 284 } 285 } 286 } 287} 288 289/// GetOrInsertNode - If there is an existing simple Node exactly 290/// equal to the specified node, return it. Otherwise, insert 'N' and it 291/// instead. 292FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) { 293 NodeID ID; 294 GetNodeProfile(ID, N); 295 void *IP; 296 if (Node *E = FindNodeOrInsertPos(ID, IP)) 297 return E; 298 InsertNode(N, IP); 299 return N; 300} 301