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/Support/MathExtras.h" |
15 | #include "llvm/Support/ReverseIteration.h" |
16 | #include "llvm/Support/xxhash.h" |
17 | |
18 | using 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(). |
22 | static inline 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(A: NumEntries * 4 / 3 + 1); |
29 | } |
30 | |
31 | static inline StringMapEntryBase **createTable(unsigned NewNumBuckets) { |
32 | auto **Table = static_cast<StringMapEntryBase **>(safe_calloc( |
33 | Count: NewNumBuckets + 1, Sz: sizeof(StringMapEntryBase **) + sizeof(unsigned))); |
34 | |
35 | // Allocate one extra bucket, set it to look filled so the iterators stop at |
36 | // end. |
37 | Table[NewNumBuckets] = (StringMapEntryBase *)2; |
38 | return Table; |
39 | } |
40 | |
41 | static inline unsigned *getHashTable(StringMapEntryBase **TheTable, |
42 | unsigned NumBuckets) { |
43 | return reinterpret_cast<unsigned *>(TheTable + NumBuckets + 1); |
44 | } |
45 | |
46 | uint32_t StringMapImpl::hash(StringRef Key) { return xxh3_64bits(data: Key); } |
47 | |
48 | StringMapImpl::StringMapImpl(unsigned InitSize, unsigned itemSize) { |
49 | ItemSize = itemSize; |
50 | |
51 | // If a size is specified, initialize the table with that many buckets. |
52 | if (InitSize) { |
53 | // The table will grow when the number of entries reach 3/4 of the number of |
54 | // buckets. To guarantee that "InitSize" number of entries can be inserted |
55 | // in the table without growing, we allocate just what is needed here. |
56 | init(Size: getMinBucketToReserveForEntries(NumEntries: InitSize)); |
57 | return; |
58 | } |
59 | |
60 | // Otherwise, initialize it with zero buckets to avoid the allocation. |
61 | TheTable = nullptr; |
62 | NumBuckets = 0; |
63 | NumItems = 0; |
64 | NumTombstones = 0; |
65 | } |
66 | |
67 | void StringMapImpl::init(unsigned InitSize) { |
68 | assert((InitSize & (InitSize - 1)) == 0 && |
69 | "Init Size must be a power of 2 or zero!" ); |
70 | |
71 | unsigned NewNumBuckets = InitSize ? InitSize : 16; |
72 | NumItems = 0; |
73 | NumTombstones = 0; |
74 | |
75 | TheTable = createTable(NewNumBuckets); |
76 | |
77 | // Set the member only if TheTable was successfully allocated |
78 | NumBuckets = NewNumBuckets; |
79 | } |
80 | |
81 | /// LookupBucketFor - Look up the bucket that the specified string should end |
82 | /// up in. If it already exists as a key in the map, the Item pointer for the |
83 | /// specified bucket will be non-null. Otherwise, it will be null. In either |
84 | /// case, the FullHashValue field of the bucket will be set to the hash value |
85 | /// of the string. |
86 | unsigned StringMapImpl::LookupBucketFor(StringRef Name, |
87 | uint32_t FullHashValue) { |
88 | #ifdef EXPENSIVE_CHECKS |
89 | assert(FullHashValue == hash(Name)); |
90 | #endif |
91 | // Hash table unallocated so far? |
92 | if (NumBuckets == 0) |
93 | init(InitSize: 16); |
94 | if (shouldReverseIterate()) |
95 | FullHashValue = ~FullHashValue; |
96 | unsigned BucketNo = FullHashValue & (NumBuckets - 1); |
97 | unsigned *HashTable = getHashTable(TheTable, NumBuckets); |
98 | |
99 | unsigned ProbeAmt = 1; |
100 | int FirstTombstone = -1; |
101 | while (true) { |
102 | StringMapEntryBase *BucketItem = TheTable[BucketNo]; |
103 | // If we found an empty bucket, this key isn't in the table yet, return it. |
104 | if (LLVM_LIKELY(!BucketItem)) { |
105 | // If we found a tombstone, we want to reuse the tombstone instead of an |
106 | // empty bucket. This reduces probing. |
107 | if (FirstTombstone != -1) { |
108 | HashTable[FirstTombstone] = FullHashValue; |
109 | return FirstTombstone; |
110 | } |
111 | |
112 | HashTable[BucketNo] = FullHashValue; |
113 | return BucketNo; |
114 | } |
115 | |
116 | if (BucketItem == getTombstoneVal()) { |
117 | // Skip over tombstones. However, remember the first one we see. |
118 | if (FirstTombstone == -1) |
119 | FirstTombstone = BucketNo; |
120 | } else if (LLVM_LIKELY(HashTable[BucketNo] == FullHashValue)) { |
121 | // If the full hash value matches, check deeply for a match. The common |
122 | // case here is that we are only looking at the buckets (for item info |
123 | // being non-null and for the full hash value) not at the items. This |
124 | // is important for cache locality. |
125 | |
126 | // Do the comparison like this because Name isn't necessarily |
127 | // null-terminated! |
128 | char *ItemStr = (char *)BucketItem + ItemSize; |
129 | if (Name == StringRef(ItemStr, BucketItem->getKeyLength())) { |
130 | // We found a match! |
131 | return BucketNo; |
132 | } |
133 | } |
134 | |
135 | // Okay, we didn't find the item. Probe to the next bucket. |
136 | BucketNo = (BucketNo + ProbeAmt) & (NumBuckets - 1); |
137 | |
138 | // Use quadratic probing, it has fewer clumping artifacts than linear |
139 | // probing and has good cache behavior in the common case. |
140 | ++ProbeAmt; |
141 | } |
142 | } |
143 | |
144 | /// FindKey - Look up the bucket that contains the specified key. If it exists |
145 | /// in the map, return the bucket number of the key. Otherwise return -1. |
146 | /// This does not modify the map. |
147 | int StringMapImpl::FindKey(StringRef Key, uint32_t FullHashValue) const { |
148 | if (NumBuckets == 0) |
149 | return -1; // Really empty table? |
150 | #ifdef EXPENSIVE_CHECKS |
151 | assert(FullHashValue == hash(Key)); |
152 | #endif |
153 | if (shouldReverseIterate()) |
154 | FullHashValue = ~FullHashValue; |
155 | unsigned BucketNo = FullHashValue & (NumBuckets - 1); |
156 | unsigned *HashTable = getHashTable(TheTable, NumBuckets); |
157 | |
158 | unsigned ProbeAmt = 1; |
159 | while (true) { |
160 | StringMapEntryBase *BucketItem = TheTable[BucketNo]; |
161 | // If we found an empty bucket, this key isn't in the table yet, return. |
162 | if (LLVM_LIKELY(!BucketItem)) |
163 | return -1; |
164 | |
165 | if (BucketItem == getTombstoneVal()) { |
166 | // Ignore tombstones. |
167 | } else if (LLVM_LIKELY(HashTable[BucketNo] == FullHashValue)) { |
168 | // If the full hash value matches, check deeply for a match. The common |
169 | // case here is that we are only looking at the buckets (for item info |
170 | // being non-null and for the full hash value) not at the items. This |
171 | // is important for cache locality. |
172 | |
173 | // Do the comparison like this because NameStart isn't necessarily |
174 | // null-terminated! |
175 | char *ItemStr = (char *)BucketItem + ItemSize; |
176 | if (Key == StringRef(ItemStr, BucketItem->getKeyLength())) { |
177 | // We found a match! |
178 | return BucketNo; |
179 | } |
180 | } |
181 | |
182 | // Okay, we didn't find the item. Probe to the next bucket. |
183 | BucketNo = (BucketNo + ProbeAmt) & (NumBuckets - 1); |
184 | |
185 | // Use quadratic probing, it has fewer clumping artifacts than linear |
186 | // probing and has good cache behavior in the common case. |
187 | ++ProbeAmt; |
188 | } |
189 | } |
190 | |
191 | /// RemoveKey - Remove the specified StringMapEntry from the table, but do not |
192 | /// delete it. This aborts if the value isn't in the table. |
193 | void StringMapImpl::RemoveKey(StringMapEntryBase *V) { |
194 | const char *VStr = (char *)V + ItemSize; |
195 | StringMapEntryBase *V2 = RemoveKey(Key: StringRef(VStr, V->getKeyLength())); |
196 | (void)V2; |
197 | assert(V == V2 && "Didn't find key?" ); |
198 | } |
199 | |
200 | /// RemoveKey - Remove the StringMapEntry for the specified key from the |
201 | /// table, returning it. If the key is not in the table, this returns null. |
202 | StringMapEntryBase *StringMapImpl::RemoveKey(StringRef Key) { |
203 | int Bucket = FindKey(Key); |
204 | if (Bucket == -1) |
205 | return nullptr; |
206 | |
207 | StringMapEntryBase *Result = TheTable[Bucket]; |
208 | TheTable[Bucket] = getTombstoneVal(); |
209 | --NumItems; |
210 | ++NumTombstones; |
211 | assert(NumItems + NumTombstones <= NumBuckets); |
212 | |
213 | return Result; |
214 | } |
215 | |
216 | /// RehashTable - Grow the table, redistributing values into the buckets with |
217 | /// the appropriate mod-of-hashtable-size. |
218 | unsigned StringMapImpl::RehashTable(unsigned BucketNo) { |
219 | unsigned NewSize; |
220 | // If the hash table is now more than 3/4 full, or if fewer than 1/8 of |
221 | // the buckets are empty (meaning that many are filled with tombstones), |
222 | // grow/rehash the table. |
223 | if (LLVM_UNLIKELY(NumItems * 4 > NumBuckets * 3)) { |
224 | NewSize = NumBuckets * 2; |
225 | } else if (LLVM_UNLIKELY(NumBuckets - (NumItems + NumTombstones) <= |
226 | NumBuckets / 8)) { |
227 | NewSize = NumBuckets; |
228 | } else { |
229 | return BucketNo; |
230 | } |
231 | |
232 | unsigned NewBucketNo = BucketNo; |
233 | auto **NewTableArray = createTable(NewNumBuckets: NewSize); |
234 | unsigned *NewHashArray = getHashTable(TheTable: NewTableArray, NumBuckets: NewSize); |
235 | unsigned *HashTable = getHashTable(TheTable, NumBuckets); |
236 | |
237 | // Rehash all the items into their new buckets. Luckily :) we already have |
238 | // the hash values available, so we don't have to rehash any strings. |
239 | for (unsigned I = 0, E = NumBuckets; I != E; ++I) { |
240 | StringMapEntryBase *Bucket = TheTable[I]; |
241 | if (Bucket && Bucket != getTombstoneVal()) { |
242 | // If the bucket is not available, probe for a spot. |
243 | unsigned FullHash = HashTable[I]; |
244 | unsigned NewBucket = FullHash & (NewSize - 1); |
245 | if (NewTableArray[NewBucket]) { |
246 | unsigned ProbeSize = 1; |
247 | do { |
248 | NewBucket = (NewBucket + ProbeSize++) & (NewSize - 1); |
249 | } while (NewTableArray[NewBucket]); |
250 | } |
251 | |
252 | // Finally found a slot. Fill it in. |
253 | NewTableArray[NewBucket] = Bucket; |
254 | NewHashArray[NewBucket] = FullHash; |
255 | if (I == BucketNo) |
256 | NewBucketNo = NewBucket; |
257 | } |
258 | } |
259 | |
260 | free(ptr: TheTable); |
261 | |
262 | TheTable = NewTableArray; |
263 | NumBuckets = NewSize; |
264 | NumTombstones = 0; |
265 | return NewBucketNo; |
266 | } |
267 | |