/src/keystone/llvm/lib/Support/StringMap.cpp
Line | Count | Source |
1 | | //===--- StringMap.cpp - String Hash table map implementation -------------===// |
2 | | // |
3 | | // The LLVM Compiler Infrastructure |
4 | | // |
5 | | // This file is distributed under the University of Illinois Open Source |
6 | | // License. See LICENSE.TXT for details. |
7 | | // |
8 | | //===----------------------------------------------------------------------===// |
9 | | // |
10 | | // This file implements the StringMap class. |
11 | | // |
12 | | //===----------------------------------------------------------------------===// |
13 | | |
14 | | #include "llvm/ADT/StringMap.h" |
15 | | #include "llvm/ADT/StringExtras.h" |
16 | | #include "llvm/Support/Compiler.h" |
17 | | #include <cassert> |
18 | | using namespace llvm_ks; |
19 | | |
20 | 0 | StringMapImpl::StringMapImpl(unsigned InitSize, unsigned itemSize) { |
21 | 0 | ItemSize = itemSize; |
22 | | |
23 | | // If a size is specified, initialize the table with that many buckets. |
24 | 0 | if (InitSize) { |
25 | 0 | init(InitSize); |
26 | 0 | return; |
27 | 0 | } |
28 | | |
29 | | // Otherwise, initialize it with zero buckets to avoid the allocation. |
30 | 0 | TheTable = nullptr; |
31 | 0 | NumBuckets = 0; |
32 | 0 | NumItems = 0; |
33 | 0 | NumTombstones = 0; |
34 | 0 | } |
35 | | |
36 | 657k | void StringMapImpl::init(unsigned InitSize) { |
37 | 657k | assert((InitSize & (InitSize-1)) == 0 && |
38 | 657k | "Init Size must be a power of 2 or zero!"); |
39 | 657k | NumBuckets = InitSize ? InitSize : 16; |
40 | 657k | NumItems = 0; |
41 | 657k | NumTombstones = 0; |
42 | | |
43 | 657k | TheTable = (StringMapEntryBase **)calloc(NumBuckets+1, |
44 | 657k | sizeof(StringMapEntryBase **) + |
45 | 657k | sizeof(unsigned)); |
46 | | |
47 | | // Allocate one extra bucket, set it to look filled so the iterators stop at |
48 | | // end. |
49 | 657k | TheTable[NumBuckets] = (StringMapEntryBase*)2; |
50 | 657k | } |
51 | | |
52 | | |
53 | | /// LookupBucketFor - Look up the bucket that the specified string should end |
54 | | /// up in. If it already exists as a key in the map, the Item pointer for the |
55 | | /// specified bucket will be non-null. Otherwise, it will be null. In either |
56 | | /// case, the FullHashValue field of the bucket will be set to the hash value |
57 | | /// of the string. |
58 | 32.6M | unsigned StringMapImpl::LookupBucketFor(StringRef Name) { |
59 | 32.6M | unsigned HTSize = NumBuckets; |
60 | 32.6M | if (HTSize == 0) { // Hash table unallocated so far? |
61 | 657k | init(16); |
62 | 657k | HTSize = NumBuckets; |
63 | 657k | } |
64 | 32.6M | unsigned FullHashValue = HashString(Name); |
65 | 32.6M | unsigned BucketNo = FullHashValue & (HTSize-1); |
66 | 32.6M | unsigned *HashTable = (unsigned *)(TheTable + NumBuckets + 1); |
67 | | |
68 | 32.6M | unsigned ProbeAmt = 1; |
69 | 32.6M | int FirstTombstone = -1; |
70 | 75.3M | while (1) { |
71 | 75.3M | StringMapEntryBase *BucketItem = TheTable[BucketNo]; |
72 | | // If we found an empty bucket, this key isn't in the table yet, return it. |
73 | 75.3M | if (LLVM_LIKELY(!BucketItem)) { |
74 | | // If we found a tombstone, we want to reuse the tombstone instead of an |
75 | | // empty bucket. This reduces probing. |
76 | 28.6M | if (FirstTombstone != -1) { |
77 | 0 | HashTable[FirstTombstone] = FullHashValue; |
78 | 0 | return FirstTombstone; |
79 | 0 | } |
80 | | |
81 | 28.6M | HashTable[BucketNo] = FullHashValue; |
82 | 28.6M | return BucketNo; |
83 | 28.6M | } |
84 | | |
85 | 46.6M | if (BucketItem == getTombstoneVal()) { |
86 | | // Skip over tombstones. However, remember the first one we see. |
87 | 0 | if (FirstTombstone == -1) FirstTombstone = BucketNo; |
88 | 46.6M | } else if (LLVM_LIKELY(HashTable[BucketNo] == FullHashValue)) { |
89 | | // If the full hash value matches, check deeply for a match. The common |
90 | | // case here is that we are only looking at the buckets (for item info |
91 | | // being non-null and for the full hash value) not at the items. This |
92 | | // is important for cache locality. |
93 | | |
94 | | // Do the comparison like this because Name isn't necessarily |
95 | | // null-terminated! |
96 | 3.95M | char *ItemStr = (char*)BucketItem+ItemSize; |
97 | 3.95M | if (Name == StringRef(ItemStr, BucketItem->getKeyLength())) { |
98 | | // We found a match! |
99 | 3.95M | return BucketNo; |
100 | 3.95M | } |
101 | 3.95M | } |
102 | | |
103 | | // Okay, we didn't find the item. Probe to the next bucket. |
104 | 42.7M | BucketNo = (BucketNo+ProbeAmt) & (HTSize-1); |
105 | | |
106 | | // Use quadratic probing, it has fewer clumping artifacts than linear |
107 | | // probing and has good cache behavior in the common case. |
108 | 42.7M | ++ProbeAmt; |
109 | 42.7M | } |
110 | 32.6M | } |
111 | | |
112 | | |
113 | | /// FindKey - Look up the bucket that contains the specified key. If it exists |
114 | | /// in the map, return the bucket number of the key. Otherwise return -1. |
115 | | /// This does not modify the map. |
116 | 28.7M | int StringMapImpl::FindKey(StringRef Key) const { |
117 | 28.7M | unsigned HTSize = NumBuckets; |
118 | 28.7M | if (HTSize == 0) return -1; // Really empty table? |
119 | 21.0M | unsigned FullHashValue = HashString(Key); |
120 | 21.0M | unsigned BucketNo = FullHashValue & (HTSize-1); |
121 | 21.0M | unsigned *HashTable = (unsigned *)(TheTable + NumBuckets + 1); |
122 | | |
123 | 21.0M | unsigned ProbeAmt = 1; |
124 | 42.5M | while (1) { |
125 | 42.5M | StringMapEntryBase *BucketItem = TheTable[BucketNo]; |
126 | | // If we found an empty bucket, this key isn't in the table yet, return. |
127 | 42.5M | if (LLVM_LIKELY(!BucketItem)) |
128 | 16.5M | return -1; |
129 | | |
130 | 25.9M | if (BucketItem == getTombstoneVal()) { |
131 | | // Ignore tombstones. |
132 | 25.9M | } else if (LLVM_LIKELY(HashTable[BucketNo] == FullHashValue)) { |
133 | | // If the full hash value matches, check deeply for a match. The common |
134 | | // case here is that we are only looking at the buckets (for item info |
135 | | // being non-null and for the full hash value) not at the items. This |
136 | | // is important for cache locality. |
137 | | |
138 | | // Do the comparison like this because NameStart isn't necessarily |
139 | | // null-terminated! |
140 | 4.51M | char *ItemStr = (char*)BucketItem+ItemSize; |
141 | 4.51M | if (Key == StringRef(ItemStr, BucketItem->getKeyLength())) { |
142 | | // We found a match! |
143 | 4.50M | return BucketNo; |
144 | 4.50M | } |
145 | 4.51M | } |
146 | | |
147 | | // Okay, we didn't find the item. Probe to the next bucket. |
148 | 21.4M | BucketNo = (BucketNo+ProbeAmt) & (HTSize-1); |
149 | | |
150 | | // Use quadratic probing, it has fewer clumping artifacts than linear |
151 | | // probing and has good cache behavior in the common case. |
152 | 21.4M | ++ProbeAmt; |
153 | 21.4M | } |
154 | 21.0M | } |
155 | | |
156 | | /// RemoveKey - Remove the specified StringMapEntry from the table, but do not |
157 | | /// delete it. This aborts if the value isn't in the table. |
158 | 0 | void StringMapImpl::RemoveKey(StringMapEntryBase *V) { |
159 | 0 | const char *VStr = (char*)V + ItemSize; |
160 | 0 | StringMapEntryBase *V2 = RemoveKey(StringRef(VStr, V->getKeyLength())); |
161 | 0 | (void)V2; |
162 | 0 | assert(V == V2 && "Didn't find key?"); |
163 | 0 | } |
164 | | |
165 | | /// RemoveKey - Remove the StringMapEntry for the specified key from the |
166 | | /// table, returning it. If the key is not in the table, this returns null. |
167 | 0 | StringMapEntryBase *StringMapImpl::RemoveKey(StringRef Key) { |
168 | 0 | int Bucket = FindKey(Key); |
169 | 0 | if (Bucket == -1) return nullptr; |
170 | | |
171 | 0 | StringMapEntryBase *Result = TheTable[Bucket]; |
172 | 0 | TheTable[Bucket] = getTombstoneVal(); |
173 | 0 | --NumItems; |
174 | 0 | ++NumTombstones; |
175 | 0 | assert(NumItems + NumTombstones <= NumBuckets); |
176 | | |
177 | 0 | return Result; |
178 | 0 | } |
179 | | |
180 | | |
181 | | |
182 | | /// RehashTable - Grow the table, redistributing values into the buckets with |
183 | | /// the appropriate mod-of-hashtable-size. |
184 | 28.6M | unsigned StringMapImpl::RehashTable(unsigned BucketNo) { |
185 | 28.6M | unsigned NewSize; |
186 | 28.6M | unsigned *HashTable = (unsigned *)(TheTable + NumBuckets + 1); |
187 | | |
188 | | // If the hash table is now more than 3/4 full, or if fewer than 1/8 of |
189 | | // the buckets are empty (meaning that many are filled with tombstones), |
190 | | // grow/rehash the table. |
191 | 28.6M | if (LLVM_UNLIKELY(NumItems * 4 > NumBuckets * 3)) { |
192 | 965k | NewSize = NumBuckets*2; |
193 | 27.6M | } else if (LLVM_UNLIKELY(NumBuckets - (NumItems + NumTombstones) <= |
194 | 27.6M | NumBuckets / 8)) { |
195 | 0 | NewSize = NumBuckets; |
196 | 27.6M | } else { |
197 | 27.6M | return BucketNo; |
198 | 27.6M | } |
199 | | |
200 | 965k | unsigned NewBucketNo = BucketNo; |
201 | | // Allocate one extra bucket which will always be non-empty. This allows the |
202 | | // iterators to stop at end. |
203 | 965k | StringMapEntryBase **NewTableArray = |
204 | 965k | (StringMapEntryBase **)calloc(NewSize+1, sizeof(StringMapEntryBase *) + |
205 | 965k | sizeof(unsigned)); |
206 | 965k | unsigned *NewHashArray = (unsigned *)(NewTableArray + NewSize + 1); |
207 | 965k | NewTableArray[NewSize] = (StringMapEntryBase*)2; |
208 | | |
209 | | // Rehash all the items into their new buckets. Luckily :) we already have |
210 | | // the hash values available, so we don't have to rehash any strings. |
211 | 49.1M | for (unsigned I = 0, E = NumBuckets; I != E; ++I) { |
212 | 48.2M | StringMapEntryBase *Bucket = TheTable[I]; |
213 | 48.2M | if (Bucket && Bucket != getTombstoneVal()) { |
214 | | // Fast case, bucket available. |
215 | 37.1M | unsigned FullHash = HashTable[I]; |
216 | 37.1M | unsigned NewBucket = FullHash & (NewSize-1); |
217 | 37.1M | if (!NewTableArray[NewBucket]) { |
218 | 29.0M | NewTableArray[FullHash & (NewSize-1)] = Bucket; |
219 | 29.0M | NewHashArray[FullHash & (NewSize-1)] = FullHash; |
220 | 29.0M | if (I == BucketNo) |
221 | 824k | NewBucketNo = NewBucket; |
222 | 29.0M | continue; |
223 | 29.0M | } |
224 | | |
225 | | // Otherwise probe for a spot. |
226 | 8.08M | unsigned ProbeSize = 1; |
227 | 11.7M | do { |
228 | 11.7M | NewBucket = (NewBucket + ProbeSize++) & (NewSize-1); |
229 | 11.7M | } while (NewTableArray[NewBucket]); |
230 | | |
231 | | // Finally found a slot. Fill it in. |
232 | 8.08M | NewTableArray[NewBucket] = Bucket; |
233 | 8.08M | NewHashArray[NewBucket] = FullHash; |
234 | 8.08M | if (I == BucketNo) |
235 | 141k | NewBucketNo = NewBucket; |
236 | 8.08M | } |
237 | 48.2M | } |
238 | | |
239 | 965k | free(TheTable); |
240 | | |
241 | 965k | TheTable = NewTableArray; |
242 | 965k | NumBuckets = NewSize; |
243 | 965k | NumTombstones = 0; |
244 | 965k | return NewBucketNo; |
245 | 28.6M | } |