lib/Support/StringMap.cpp

   1 //===--- StringMap.cpp - String Hash table map implementation -------------===//
   2 //
   3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
   4 // See https://llvm.org/LICENSE.txt for license information.
   5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
   6 //
   7 //===----------------------------------------------------------------------===//
   8 //
   9 // This file implements the StringMap class.
  10 //
  11 //===----------------------------------------------------------------------===//
  12
  13 #include "llvm/ADT/StringMap.h"
  14 #include "llvm/ADT/StringExtras.h"
  15 #include "llvm/Support/Compiler.h"
  16 #include "llvm/Support/DJB.h"
  17 #include "llvm/Support/MathExtras.h"
  18 #include <cassert>
  19
  20 using namespace llvm;
  21
  22 /// Returns the number of buckets to allocate to ensure that the DenseMap can
  23 /// accommodate \p NumEntries without need to grow().
  24 static unsigned getMinBucketToReserveForEntries(unsigned NumEntries) {
  25   // Ensure that "NumEntries * 4 < NumBuckets * 3"
  26   if (NumEntries == 0)
  27     return 0;
  28   // +1 is required because of the strict equality.
  29   // For example if NumEntries is 48, we need to return 401.
  30   return NextPowerOf2(NumEntries * 4 / 3 + 1);
  31 }
  32
  33 StringMapImpl::StringMapImpl(unsigned InitSize, unsigned itemSize) {
  34   ItemSize = itemSize;
  35
  36   // If a size is specified, initialize the table with that many buckets.
  37   if (InitSize) {
  38     // The table will grow when the number of entries reach 3/4 of the number of
  39     // buckets. To guarantee that "InitSize" number of entries can be inserted
  40     // in the table without growing, we allocate just what is needed here.
  41     init(getMinBucketToReserveForEntries(InitSize));
  42     return;
  43   }
  44
  45   // Otherwise, initialize it with zero buckets to avoid the allocation.
  46   TheTable = nullptr;
  47   NumBuckets = 0;
  48   NumItems = 0;
  49   NumTombstones = 0;
  50 }
  51
  52 void StringMapImpl::init(unsigned InitSize) {
  53   assert((InitSize & (InitSize-1)) == 0 &&
  54          "Init Size must be a power of 2 or zero!");
  55
  56   unsigned NewNumBuckets = InitSize ? InitSize : 16;
  57   NumItems = 0;
  58   NumTombstones = 0;
  59
  60   TheTable = static_cast<StringMapEntryBase **>(
  61       safe_calloc(NewNumBuckets+1,
  62                   sizeof(StringMapEntryBase **) + sizeof(unsigned)));
  63
  64   // Set the member only if TheTable was successfully allocated
  65   NumBuckets = NewNumBuckets;
  66
  67   // Allocate one extra bucket, set it to look filled so the iterators stop at
  68   // end.
  69   TheTable[NumBuckets] = (StringMapEntryBase*)2;
  70 }
  71
  72 /// LookupBucketFor - Look up the bucket that the specified string should end
  73 /// up in.  If it already exists as a key in the map, the Item pointer for the
  74 /// specified bucket will be non-null.  Otherwise, it will be null.  In either
  75 /// case, the FullHashValue field of the bucket will be set to the hash value
  76 /// of the string.
  77 unsigned StringMapImpl::LookupBucketFor(StringRef Name) {
  78   unsigned HTSize = NumBuckets;
  79   if (HTSize == 0) {  // Hash table unallocated so far?
  80     init(16);
  81     HTSize = NumBuckets;
  82   }
  83   unsigned FullHashValue = djbHash(Name, 0);
  84   unsigned BucketNo = FullHashValue & (HTSize-1);
  85   unsigned *HashTable = (unsigned *)(TheTable + NumBuckets + 1);
  86
  87   unsigned ProbeAmt = 1;
  88   int FirstTombstone = -1;
  89   while (true) {
  90     StringMapEntryBase *BucketItem = TheTable[BucketNo];
  91     // If we found an empty bucket, this key isn't in the table yet, return it.
  92     if (LLVM_LIKELY(!BucketItem)) {
  93       // If we found a tombstone, we want to reuse the tombstone instead of an
  94       // empty bucket.  This reduces probing.
  95       if (FirstTombstone != -1) {
  96         HashTable[FirstTombstone] = FullHashValue;
  97         return FirstTombstone;
  98       }
  99
 100       HashTable[BucketNo] = FullHashValue;
 101       return BucketNo;
 102     }
 103
 104     if (BucketItem == getTombstoneVal()) {
 105       // Skip over tombstones.  However, remember the first one we see.
 106       if (FirstTombstone == -1) FirstTombstone = BucketNo;
 107     } else if (LLVM_LIKELY(HashTable[BucketNo] == FullHashValue)) {
 108       // If the full hash value matches, check deeply for a match.  The common
 109       // case here is that we are only looking at the buckets (for item info
 110       // being non-null and for the full hash value) not at the items.  This
 111       // is important for cache locality.
 112
 113       // Do the comparison like this because Name isn't necessarily
 114       // null-terminated!
 115       char *ItemStr = (char*)BucketItem+ItemSize;
 116       if (Name == StringRef(ItemStr, BucketItem->getKeyLength())) {
 117         // We found a match!
 118         return BucketNo;
 119       }
 120     }
 121
 122     // Okay, we didn't find the item.  Probe to the next bucket.
 123     BucketNo = (BucketNo+ProbeAmt) & (HTSize-1);
 124
 125     // Use quadratic probing, it has fewer clumping artifacts than linear
 126     // probing and has good cache behavior in the common case.
 127     ++ProbeAmt;
 128   }
 129 }
 130
 131 /// FindKey - Look up the bucket that contains the specified key. If it exists
 132 /// in the map, return the bucket number of the key.  Otherwise return -1.
 133 /// This does not modify the map.
 134 int StringMapImpl::FindKey(StringRef Key) const {
 135   unsigned HTSize = NumBuckets;
 136   if (HTSize == 0) return -1;  // Really empty table?
 137   unsigned FullHashValue = djbHash(Key, 0);
 138   unsigned BucketNo = FullHashValue & (HTSize-1);
 139   unsigned *HashTable = (unsigned *)(TheTable + NumBuckets + 1);
 140
 141   unsigned ProbeAmt = 1;
 142   while (true) {
 143     StringMapEntryBase *BucketItem = TheTable[BucketNo];
 144     // If we found an empty bucket, this key isn't in the table yet, return.
 145     if (LLVM_LIKELY(!BucketItem))
 146       return -1;
 147
 148     if (BucketItem == getTombstoneVal()) {
 149       // Ignore tombstones.
 150     } else if (LLVM_LIKELY(HashTable[BucketNo] == FullHashValue)) {
 151       // If the full hash value matches, check deeply for a match.  The common
 152       // case here is that we are only looking at the buckets (for item info
 153       // being non-null and for the full hash value) not at the items.  This
 154       // is important for cache locality.
 155
 156       // Do the comparison like this because NameStart isn't necessarily
 157       // null-terminated!
 158       char *ItemStr = (char*)BucketItem+ItemSize;
 159       if (Key == StringRef(ItemStr, BucketItem->getKeyLength())) {
 160         // We found a match!
 161         return BucketNo;
 162       }
 163     }
 164
 165     // Okay, we didn't find the item.  Probe to the next bucket.
 166     BucketNo = (BucketNo+ProbeAmt) & (HTSize-1);
 167
 168     // Use quadratic probing, it has fewer clumping artifacts than linear
 169     // probing and has good cache behavior in the common case.
 170     ++ProbeAmt;
 171   }
 172 }
 173
 174 /// RemoveKey - Remove the specified StringMapEntry from the table, but do not
 175 /// delete it.  This aborts if the value isn't in the table.
 176 void StringMapImpl::RemoveKey(StringMapEntryBase *V) {
 177   const char *VStr = (char*)V + ItemSize;
 178   StringMapEntryBase *V2 = RemoveKey(StringRef(VStr, V->getKeyLength()));
 179   (void)V2;
 180   assert(V == V2 && "Didn't find key?");
 181 }
 182
 183 /// RemoveKey - Remove the StringMapEntry for the specified key from the
 184 /// table, returning it.  If the key is not in the table, this returns null.
 185 StringMapEntryBase *StringMapImpl::RemoveKey(StringRef Key) {
 186   int Bucket = FindKey(Key);
 187   if (Bucket == -1) return nullptr;
 188
 189   StringMapEntryBase *Result = TheTable[Bucket];
 190   TheTable[Bucket] = getTombstoneVal();
 191   --NumItems;
 192   ++NumTombstones;
 193   assert(NumItems + NumTombstones <= NumBuckets);
 194
 195   return Result;
 196 }
 197
 198 /// RehashTable - Grow the table, redistributing values into the buckets with
 199 /// the appropriate mod-of-hashtable-size.
 200 unsigned StringMapImpl::RehashTable(unsigned BucketNo) {
 201   unsigned NewSize;
 202   unsigned *HashTable = (unsigned *)(TheTable + NumBuckets + 1);
 203
 204   // If the hash table is now more than 3/4 full, or if fewer than 1/8 of
 205   // the buckets are empty (meaning that many are filled with tombstones),
 206   // grow/rehash the table.
 207   if (LLVM_UNLIKELY(NumItems * 4 > NumBuckets * 3)) {
 208     NewSize = NumBuckets*2;
 209   } else if (LLVM_UNLIKELY(NumBuckets - (NumItems + NumTombstones) <=
 210                            NumBuckets / 8)) {
 211     NewSize = NumBuckets;
 212   } else {
 213     return BucketNo;
 214   }
 215
 216   unsigned NewBucketNo = BucketNo;
 217   // Allocate one extra bucket which will always be non-empty.  This allows the
 218   // iterators to stop at end.
 219   auto NewTableArray = static_cast<StringMapEntryBase **>(
 220       safe_calloc(NewSize+1, sizeof(StringMapEntryBase *) + sizeof(unsigned)));
 221
 222   unsigned *NewHashArray = (unsigned *)(NewTableArray + NewSize + 1);
 223   NewTableArray[NewSize] = (StringMapEntryBase*)2;
 224
 225   // Rehash all the items into their new buckets.  Luckily :) we already have
 226   // the hash values available, so we don't have to rehash any strings.
 227   for (unsigned I = 0, E = NumBuckets; I != E; ++I) {
 228     StringMapEntryBase *Bucket = TheTable[I];
 229     if (Bucket && Bucket != getTombstoneVal()) {
 230       // Fast case, bucket available.
 231       unsigned FullHash = HashTable[I];
 232       unsigned NewBucket = FullHash & (NewSize-1);
 233       if (!NewTableArray[NewBucket]) {
 234         NewTableArray[FullHash & (NewSize-1)] = Bucket;
 235         NewHashArray[FullHash & (NewSize-1)] = FullHash;
 236         if (I == BucketNo)
 237           NewBucketNo = NewBucket;
 238         continue;
 239       }
 240
 241       // Otherwise probe for a spot.
 242       unsigned ProbeSize = 1;
 243       do {
 244         NewBucket = (NewBucket + ProbeSize++) & (NewSize-1);
 245       } while (NewTableArray[NewBucket]);
 246
 247       // Finally found a slot.  Fill it in.
 248       NewTableArray[NewBucket] = Bucket;
 249       NewHashArray[NewBucket] = FullHash;
 250       if (I == BucketNo)
 251         NewBucketNo = NewBucket;
 252     }
 253   }
 254
 255   free(TheTable);
 256
 257   TheTable = NewTableArray;
 258   NumBuckets = NewSize;
 259   NumTombstones = 0;
 260   return NewBucketNo;
 261 }