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/DJB.h"
16#include "llvm/Support/MathExtras.h"
17
18using namespace llvm;
19
20/// Returns the number of buckets to allocate to ensure that the DenseMap can
21/// accommodate \p NumEntries without need to grow().
22static unsigned getMinBucketToReserveForEntries(unsigned NumEntries) {
23  // Ensure that "NumEntries * 4 < NumBuckets * 3"
24  if (NumEntries == 0)
25    return 0;
26  // +1 is required because of the strict equality.
27  // For example if NumEntries is 48, we need to return 401.
28  return NextPowerOf2(NumEntries * 4 / 3 + 1);
29}
30
31StringMapImpl::StringMapImpl(unsigned InitSize, unsigned itemSize) {
32  ItemSize = itemSize;
33
34  // If a size is specified, initialize the table with that many buckets.
35  if (InitSize) {
36    // The table will grow when the number of entries reach 3/4 of the number of
37    // buckets. To guarantee that "InitSize" number of entries can be inserted
38    // in the table without growing, we allocate just what is needed here.
39    init(getMinBucketToReserveForEntries(InitSize));
40    return;
41  }
42
43  // Otherwise, initialize it with zero buckets to avoid the allocation.
44  TheTable = nullptr;
45  NumBuckets = 0;
46  NumItems = 0;
47  NumTombstones = 0;
48}
49
50void StringMapImpl::init(unsigned InitSize) {
51  assert((InitSize & (InitSize - 1)) == 0 &&
52         "Init Size must be a power of 2 or zero!");
53
54  unsigned NewNumBuckets = InitSize ? InitSize : 16;
55  NumItems = 0;
56  NumTombstones = 0;
57
58  TheTable = static_cast<StringMapEntryBase **>(safe_calloc(
59      NewNumBuckets + 1, sizeof(StringMapEntryBase **) + sizeof(unsigned)));
60
61  // Set the member only if TheTable was successfully allocated
62  NumBuckets = NewNumBuckets;
63
64  // Allocate one extra bucket, set it to look filled so the iterators stop at
65  // end.
66  TheTable[NumBuckets] = (StringMapEntryBase *)2;
67}
68
69/// LookupBucketFor - Look up the bucket that the specified string should end
70/// up in.  If it already exists as a key in the map, the Item pointer for the
71/// specified bucket will be non-null.  Otherwise, it will be null.  In either
72/// case, the FullHashValue field of the bucket will be set to the hash value
73/// of the string.
74unsigned StringMapImpl::LookupBucketFor(StringRef Name) {
75  unsigned HTSize = NumBuckets;
76  if (HTSize == 0) { // Hash table unallocated so far?
77    init(16);
78    HTSize = NumBuckets;
79  }
80  unsigned FullHashValue = djbHash(Name, 0);
81  unsigned BucketNo = FullHashValue & (HTSize - 1);
82  unsigned *HashTable = (unsigned *)(TheTable + NumBuckets + 1);
83
84  unsigned ProbeAmt = 1;
85  int FirstTombstone = -1;
86  while (true) {
87    StringMapEntryBase *BucketItem = TheTable[BucketNo];
88    // If we found an empty bucket, this key isn't in the table yet, return it.
89    if (LLVM_LIKELY(!BucketItem)) {
90      // If we found a tombstone, we want to reuse the tombstone instead of an
91      // empty bucket.  This reduces probing.
92      if (FirstTombstone != -1) {
93        HashTable[FirstTombstone] = FullHashValue;
94        return FirstTombstone;
95      }
96
97      HashTable[BucketNo] = FullHashValue;
98      return BucketNo;
99    }
100
101    if (BucketItem == getTombstoneVal()) {
102      // Skip over tombstones.  However, remember the first one we see.
103      if (FirstTombstone == -1)
104        FirstTombstone = BucketNo;
105    } else if (LLVM_LIKELY(HashTable[BucketNo] == FullHashValue)) {
106      // If the full hash value matches, check deeply for a match.  The common
107      // case here is that we are only looking at the buckets (for item info
108      // being non-null and for the full hash value) not at the items.  This
109      // is important for cache locality.
110
111      // Do the comparison like this because Name isn't necessarily
112      // null-terminated!
113      char *ItemStr = (char *)BucketItem + ItemSize;
114      if (Name == StringRef(ItemStr, BucketItem->getKeyLength())) {
115        // We found a match!
116        return BucketNo;
117      }
118    }
119
120    // Okay, we didn't find the item.  Probe to the next bucket.
121    BucketNo = (BucketNo + ProbeAmt) & (HTSize - 1);
122
123    // Use quadratic probing, it has fewer clumping artifacts than linear
124    // probing and has good cache behavior in the common case.
125    ++ProbeAmt;
126  }
127}
128
129/// FindKey - Look up the bucket that contains the specified key. If it exists
130/// in the map, return the bucket number of the key.  Otherwise return -1.
131/// This does not modify the map.
132int StringMapImpl::FindKey(StringRef Key) const {
133  unsigned HTSize = NumBuckets;
134  if (HTSize == 0)
135    return -1; // Really empty table?
136  unsigned FullHashValue = djbHash(Key, 0);
137  unsigned BucketNo = FullHashValue & (HTSize - 1);
138  unsigned *HashTable = (unsigned *)(TheTable + NumBuckets + 1);
139
140  unsigned ProbeAmt = 1;
141  while (true) {
142    StringMapEntryBase *BucketItem = TheTable[BucketNo];
143    // If we found an empty bucket, this key isn't in the table yet, return.
144    if (LLVM_LIKELY(!BucketItem))
145      return -1;
146
147    if (BucketItem == getTombstoneVal()) {
148      // Ignore tombstones.
149    } else if (LLVM_LIKELY(HashTable[BucketNo] == FullHashValue)) {
150      // If the full hash value matches, check deeply for a match.  The common
151      // case here is that we are only looking at the buckets (for item info
152      // being non-null and for the full hash value) not at the items.  This
153      // is important for cache locality.
154
155      // Do the comparison like this because NameStart isn't necessarily
156      // null-terminated!
157      char *ItemStr = (char *)BucketItem + ItemSize;
158      if (Key == StringRef(ItemStr, BucketItem->getKeyLength())) {
159        // We found a match!
160        return BucketNo;
161      }
162    }
163
164    // Okay, we didn't find the item.  Probe to the next bucket.
165    BucketNo = (BucketNo + ProbeAmt) & (HTSize - 1);
166
167    // Use quadratic probing, it has fewer clumping artifacts than linear
168    // probing and has good cache behavior in the common case.
169    ++ProbeAmt;
170  }
171}
172
173/// RemoveKey - Remove the specified StringMapEntry from the table, but do not
174/// delete it.  This aborts if the value isn't in the table.
175void StringMapImpl::RemoveKey(StringMapEntryBase *V) {
176  const char *VStr = (char *)V + ItemSize;
177  StringMapEntryBase *V2 = RemoveKey(StringRef(VStr, V->getKeyLength()));
178  (void)V2;
179  assert(V == V2 && "Didn't find key?");
180}
181
182/// RemoveKey - Remove the StringMapEntry for the specified key from the
183/// table, returning it.  If the key is not in the table, this returns null.
184StringMapEntryBase *StringMapImpl::RemoveKey(StringRef Key) {
185  int Bucket = FindKey(Key);
186  if (Bucket == -1)
187    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.
200unsigned 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 **>(safe_calloc(
220      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}
262