xref: /freebsd/contrib/llvm-project/llvm/lib/Support/StringMap.cpp (revision 3e8eb5c7f4909209c042403ddee340b2ee7003a5)
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/DJB.h"
15 #include "llvm/Support/MathExtras.h"
16 
17 using namespace llvm;
18 
19 /// Returns the number of buckets to allocate to ensure that the DenseMap can
20 /// accommodate \p NumEntries without need to grow().
21 static unsigned getMinBucketToReserveForEntries(unsigned NumEntries) {
22   // Ensure that "NumEntries * 4 < NumBuckets * 3"
23   if (NumEntries == 0)
24     return 0;
25   // +1 is required because of the strict equality.
26   // For example if NumEntries is 48, we need to return 401.
27   return NextPowerOf2(NumEntries * 4 / 3 + 1);
28 }
29 
30 StringMapImpl::StringMapImpl(unsigned InitSize, unsigned itemSize) {
31   ItemSize = itemSize;
32 
33   // If a size is specified, initialize the table with that many buckets.
34   if (InitSize) {
35     // The table will grow when the number of entries reach 3/4 of the number of
36     // buckets. To guarantee that "InitSize" number of entries can be inserted
37     // in the table without growing, we allocate just what is needed here.
38     init(getMinBucketToReserveForEntries(InitSize));
39     return;
40   }
41 
42   // Otherwise, initialize it with zero buckets to avoid the allocation.
43   TheTable = nullptr;
44   NumBuckets = 0;
45   NumItems = 0;
46   NumTombstones = 0;
47 }
48 
49 void StringMapImpl::init(unsigned InitSize) {
50   assert((InitSize & (InitSize - 1)) == 0 &&
51          "Init Size must be a power of 2 or zero!");
52 
53   unsigned NewNumBuckets = InitSize ? InitSize : 16;
54   NumItems = 0;
55   NumTombstones = 0;
56 
57   TheTable = static_cast<StringMapEntryBase **>(safe_calloc(
58       NewNumBuckets + 1, sizeof(StringMapEntryBase **) + sizeof(unsigned)));
59 
60   // Set the member only if TheTable was successfully allocated
61   NumBuckets = NewNumBuckets;
62 
63   // Allocate one extra bucket, set it to look filled so the iterators stop at
64   // end.
65   TheTable[NumBuckets] = (StringMapEntryBase *)2;
66 }
67 
68 /// LookupBucketFor - Look up the bucket that the specified string should end
69 /// up in.  If it already exists as a key in the map, the Item pointer for the
70 /// specified bucket will be non-null.  Otherwise, it will be null.  In either
71 /// case, the FullHashValue field of the bucket will be set to the hash value
72 /// of the string.
73 unsigned StringMapImpl::LookupBucketFor(StringRef Name) {
74   unsigned HTSize = NumBuckets;
75   if (HTSize == 0) { // Hash table unallocated so far?
76     init(16);
77     HTSize = NumBuckets;
78   }
79   unsigned FullHashValue = djbHash(Name, 0);
80   unsigned BucketNo = FullHashValue & (HTSize - 1);
81   unsigned *HashTable = (unsigned *)(TheTable + NumBuckets + 1);
82 
83   unsigned ProbeAmt = 1;
84   int FirstTombstone = -1;
85   while (true) {
86     StringMapEntryBase *BucketItem = TheTable[BucketNo];
87     // If we found an empty bucket, this key isn't in the table yet, return it.
88     if (LLVM_LIKELY(!BucketItem)) {
89       // If we found a tombstone, we want to reuse the tombstone instead of an
90       // empty bucket.  This reduces probing.
91       if (FirstTombstone != -1) {
92         HashTable[FirstTombstone] = FullHashValue;
93         return FirstTombstone;
94       }
95 
96       HashTable[BucketNo] = FullHashValue;
97       return BucketNo;
98     }
99 
100     if (BucketItem == getTombstoneVal()) {
101       // Skip over tombstones.  However, remember the first one we see.
102       if (FirstTombstone == -1)
103         FirstTombstone = BucketNo;
104     } else if (LLVM_LIKELY(HashTable[BucketNo] == FullHashValue)) {
105       // If the full hash value matches, check deeply for a match.  The common
106       // case here is that we are only looking at the buckets (for item info
107       // being non-null and for the full hash value) not at the items.  This
108       // is important for cache locality.
109 
110       // Do the comparison like this because Name isn't necessarily
111       // null-terminated!
112       char *ItemStr = (char *)BucketItem + ItemSize;
113       if (Name == StringRef(ItemStr, BucketItem->getKeyLength())) {
114         // We found a match!
115         return BucketNo;
116       }
117     }
118 
119     // Okay, we didn't find the item.  Probe to the next bucket.
120     BucketNo = (BucketNo + ProbeAmt) & (HTSize - 1);
121 
122     // Use quadratic probing, it has fewer clumping artifacts than linear
123     // probing and has good cache behavior in the common case.
124     ++ProbeAmt;
125   }
126 }
127 
128 /// FindKey - Look up the bucket that contains the specified key. If it exists
129 /// in the map, return the bucket number of the key.  Otherwise return -1.
130 /// This does not modify the map.
131 int StringMapImpl::FindKey(StringRef Key) const {
132   unsigned HTSize = NumBuckets;
133   if (HTSize == 0)
134     return -1; // Really empty table?
135   unsigned FullHashValue = djbHash(Key, 0);
136   unsigned BucketNo = FullHashValue & (HTSize - 1);
137   unsigned *HashTable = (unsigned *)(TheTable + NumBuckets + 1);
138 
139   unsigned ProbeAmt = 1;
140   while (true) {
141     StringMapEntryBase *BucketItem = TheTable[BucketNo];
142     // If we found an empty bucket, this key isn't in the table yet, return.
143     if (LLVM_LIKELY(!BucketItem))
144       return -1;
145 
146     if (BucketItem == getTombstoneVal()) {
147       // Ignore tombstones.
148     } else if (LLVM_LIKELY(HashTable[BucketNo] == FullHashValue)) {
149       // If the full hash value matches, check deeply for a match.  The common
150       // case here is that we are only looking at the buckets (for item info
151       // being non-null and for the full hash value) not at the items.  This
152       // is important for cache locality.
153 
154       // Do the comparison like this because NameStart isn't necessarily
155       // null-terminated!
156       char *ItemStr = (char *)BucketItem + ItemSize;
157       if (Key == StringRef(ItemStr, BucketItem->getKeyLength())) {
158         // We found a match!
159         return BucketNo;
160       }
161     }
162 
163     // Okay, we didn't find the item.  Probe to the next bucket.
164     BucketNo = (BucketNo + ProbeAmt) & (HTSize - 1);
165 
166     // Use quadratic probing, it has fewer clumping artifacts than linear
167     // probing and has good cache behavior in the common case.
168     ++ProbeAmt;
169   }
170 }
171 
172 /// RemoveKey - Remove the specified StringMapEntry from the table, but do not
173 /// delete it.  This aborts if the value isn't in the table.
174 void StringMapImpl::RemoveKey(StringMapEntryBase *V) {
175   const char *VStr = (char *)V + ItemSize;
176   StringMapEntryBase *V2 = RemoveKey(StringRef(VStr, V->getKeyLength()));
177   (void)V2;
178   assert(V == V2 && "Didn't find key?");
179 }
180 
181 /// RemoveKey - Remove the StringMapEntry for the specified key from the
182 /// table, returning it.  If the key is not in the table, this returns null.
183 StringMapEntryBase *StringMapImpl::RemoveKey(StringRef Key) {
184   int Bucket = FindKey(Key);
185   if (Bucket == -1)
186     return nullptr;
187 
188   StringMapEntryBase *Result = TheTable[Bucket];
189   TheTable[Bucket] = getTombstoneVal();
190   --NumItems;
191   ++NumTombstones;
192   assert(NumItems + NumTombstones <= NumBuckets);
193 
194   return Result;
195 }
196 
197 /// RehashTable - Grow the table, redistributing values into the buckets with
198 /// the appropriate mod-of-hashtable-size.
199 unsigned StringMapImpl::RehashTable(unsigned BucketNo) {
200   unsigned NewSize;
201   unsigned *HashTable = (unsigned *)(TheTable + NumBuckets + 1);
202 
203   // If the hash table is now more than 3/4 full, or if fewer than 1/8 of
204   // the buckets are empty (meaning that many are filled with tombstones),
205   // grow/rehash the table.
206   if (LLVM_UNLIKELY(NumItems * 4 > NumBuckets * 3)) {
207     NewSize = NumBuckets * 2;
208   } else if (LLVM_UNLIKELY(NumBuckets - (NumItems + NumTombstones) <=
209                            NumBuckets / 8)) {
210     NewSize = NumBuckets;
211   } else {
212     return BucketNo;
213   }
214 
215   unsigned NewBucketNo = BucketNo;
216   // Allocate one extra bucket which will always be non-empty.  This allows the
217   // iterators to stop at end.
218   auto NewTableArray = static_cast<StringMapEntryBase **>(safe_calloc(
219       NewSize + 1, sizeof(StringMapEntryBase *) + sizeof(unsigned)));
220 
221   unsigned *NewHashArray = (unsigned *)(NewTableArray + NewSize + 1);
222   NewTableArray[NewSize] = (StringMapEntryBase *)2;
223 
224   // Rehash all the items into their new buckets.  Luckily :) we already have
225   // the hash values available, so we don't have to rehash any strings.
226   for (unsigned I = 0, E = NumBuckets; I != E; ++I) {
227     StringMapEntryBase *Bucket = TheTable[I];
228     if (Bucket && Bucket != getTombstoneVal()) {
229       // Fast case, bucket available.
230       unsigned FullHash = HashTable[I];
231       unsigned NewBucket = FullHash & (NewSize - 1);
232       if (!NewTableArray[NewBucket]) {
233         NewTableArray[FullHash & (NewSize - 1)] = Bucket;
234         NewHashArray[FullHash & (NewSize - 1)] = FullHash;
235         if (I == BucketNo)
236           NewBucketNo = NewBucket;
237         continue;
238       }
239 
240       // Otherwise probe for a spot.
241       unsigned ProbeSize = 1;
242       do {
243         NewBucket = (NewBucket + ProbeSize++) & (NewSize - 1);
244       } while (NewTableArray[NewBucket]);
245 
246       // Finally found a slot.  Fill it in.
247       NewTableArray[NewBucket] = Bucket;
248       NewHashArray[NewBucket] = FullHash;
249       if (I == BucketNo)
250         NewBucketNo = NewBucket;
251     }
252   }
253 
254   free(TheTable);
255 
256   TheTable = NewTableArray;
257   NumBuckets = NewSize;
258   NumTombstones = 0;
259   return NewBucketNo;
260 }
261