StringMap.cpp revision bb28a81ba3c112853f0eb3d8df0190accc0379c9 
1//===--- StringMap.cpp - String Hash table map implementation -------------===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file was developed by Chris Lattner and is distributed under
6// the University of Illinois Open Source 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 <cassert>
16using namespace llvm;
17
18StringMapVisitor::~StringMapVisitor() {
19}
20
21StringMapImpl::StringMapImpl(unsigned InitSize, unsigned itemSize) {
22  assert((InitSize & (InitSize-1)) == 0 &&
23         "Init Size must be a power of 2 or zero!");
24  NumBuckets = InitSize ? InitSize : 512;
25  ItemSize = itemSize;
26  NumItems = 0;
27
28  TheTable = new ItemBucket[NumBuckets]();
29  memset(TheTable, 0, NumBuckets*sizeof(ItemBucket));
30}
31
32
33/// HashString - Compute a hash code for the specified string.
34///
35static unsigned HashString(const char *Start, const char *End) {
36  // Bernstein hash function.
37  unsigned int Result = 0;
38  // TODO: investigate whether a modified bernstein hash function performs
39  // better: http://eternallyconfuzzled.com/tuts/algorithms/jsw_tut_hashing.aspx
40  //   X*33+c -> X*33^c
41  while (Start != End)
42    Result = Result * 33 + *Start++;
43  Result = Result + (Result >> 5);
44  return Result;
45}
46
47/// LookupBucketFor - Look up the bucket that the specified string should end
48/// up in.  If it already exists as a key in the map, the Item pointer for the
49/// specified bucket will be non-null.  Otherwise, it will be null.  In either
50/// case, the FullHashValue field of the bucket will be set to the hash value
51/// of the string.
52unsigned StringMapImpl::LookupBucketFor(const char *NameStart,
53                                         const char *NameEnd) {
54  unsigned HTSize = NumBuckets;
55  unsigned FullHashValue = HashString(NameStart, NameEnd);
56  unsigned BucketNo = FullHashValue & (HTSize-1);
57
58  unsigned ProbeAmt = 1;
59  while (1) {
60    ItemBucket &Bucket = TheTable[BucketNo];
61    StringMapEntryBase *BucketItem = Bucket.Item;
62    // If we found an empty bucket, this key isn't in the table yet, return it.
63    if (BucketItem == 0) {
64      Bucket.FullHashValue = FullHashValue;
65      return BucketNo;
66    }
67
68    // If the full hash value matches, check deeply for a match.  The common
69    // case here is that we are only looking at the buckets (for item info
70    // being non-null and for the full hash value) not at the items.  This
71    // is important for cache locality.
72    if (Bucket.FullHashValue == FullHashValue) {
73      // Do the comparison like this because NameStart isn't necessarily
74      // null-terminated!
75      char *ItemStr = (char*)BucketItem+ItemSize;
76      unsigned ItemStrLen = BucketItem->getKeyLength();
77      if (unsigned(NameEnd-NameStart) == ItemStrLen &&
78          memcmp(ItemStr, NameStart, ItemStrLen) == 0) {
79        // We found a match!
80        return BucketNo;
81      }
82    }
83
84    // Okay, we didn't find the item.  Probe to the next bucket.
85    BucketNo = (BucketNo+ProbeAmt) & (HTSize-1);
86
87    // Use quadratic probing, it has fewer clumping artifacts than linear
88    // probing and has good cache behavior in the common case.
89    ++ProbeAmt;
90  }
91}
92
93/// RehashTable - Grow the table, redistributing values into the buckets with
94/// the appropriate mod-of-hashtable-size.
95void StringMapImpl::RehashTable() {
96  unsigned NewSize = NumBuckets*2;
97  ItemBucket *NewTableArray = new ItemBucket[NewSize]();
98  memset(NewTableArray, 0, NewSize*sizeof(ItemBucket));
99
100  // Rehash all the items into their new buckets.  Luckily :) we already have
101  // the hash values available, so we don't have to rehash any strings.
102  for (ItemBucket *IB = TheTable, *E = TheTable+NumBuckets; IB != E; ++IB) {
103    if (IB->Item) {
104      // Fast case, bucket available.
105      unsigned FullHash = IB->FullHashValue;
106      unsigned NewBucket = FullHash & (NewSize-1);
107      if (NewTableArray[NewBucket].Item == 0) {
108        NewTableArray[FullHash & (NewSize-1)].Item = IB->Item;
109        NewTableArray[FullHash & (NewSize-1)].FullHashValue = FullHash;
110        continue;
111      }
112
113      unsigned ProbeSize = 1;
114      do {
115        NewBucket = (NewBucket + ProbeSize++) & (NewSize-1);
116      } while (NewTableArray[NewBucket].Item);
117
118      // Finally found a slot.  Fill it in.
119      NewTableArray[NewBucket].Item = IB->Item;
120      NewTableArray[NewBucket].FullHashValue = FullHash;
121    }
122  }
123
124  delete[] TheTable;
125
126  TheTable = NewTableArray;
127  NumBuckets = NewSize;
128}
129
130
131/// VisitEntries - This method walks through all of the items,
132/// invoking Visitor.Visit for each of them.
133void StringMapImpl::VisitEntries(const StringMapVisitor &Visitor) const {
134  for (ItemBucket *IB = TheTable, *E = TheTable+NumBuckets; IB != E; ++IB) {
135    if (StringMapEntryBase *Id = IB->Item)
136      Visitor.Visit((char*)Id + ItemSize, Id);
137  }
138}
139