1//===-- Support/FoldingSet.cpp - Uniquing Hash Set --------------*- C++ -*-===//
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 a hash set that can be used to remove duplication of
11// nodes in a graph.
12//
13//===----------------------------------------------------------------------===//
14
15#include "llvm/ADT/FoldingSet.h"
16#include "llvm/ADT/Hashing.h"
17#include "llvm/Support/Allocator.h"
18#include "llvm/Support/ErrorHandling.h"
19#include "llvm/Support/Host.h"
20#include "llvm/Support/MathExtras.h"
21#include <cassert>
22#include <cstring>
23using namespace llvm;
24
25//===----------------------------------------------------------------------===//
26// FoldingSetNodeIDRef Implementation
27
28/// ComputeHash - Compute a strong hash value for this FoldingSetNodeIDRef,
29/// used to lookup the node in the FoldingSetImpl.
30unsigned FoldingSetNodeIDRef::ComputeHash() const {
31  return static_cast<unsigned>(hash_combine_range(Data, Data+Size));
32}
33
34bool FoldingSetNodeIDRef::operator==(FoldingSetNodeIDRef RHS) const {
35  if (Size != RHS.Size) return false;
36  return memcmp(Data, RHS.Data, Size*sizeof(*Data)) == 0;
37}
38
39/// Used to compare the "ordering" of two nodes as defined by the
40/// profiled bits and their ordering defined by memcmp().
41bool FoldingSetNodeIDRef::operator<(FoldingSetNodeIDRef RHS) const {
42  if (Size != RHS.Size)
43    return Size < RHS.Size;
44  return memcmp(Data, RHS.Data, Size*sizeof(*Data)) < 0;
45}
46
47//===----------------------------------------------------------------------===//
48// FoldingSetNodeID Implementation
49
50/// Add* - Add various data types to Bit data.
51///
52void FoldingSetNodeID::AddPointer(const void *Ptr) {
53  // Note: this adds pointers to the hash using sizes and endianness that
54  // depend on the host. It doesn't matter, however, because hashing on
55  // pointer values is inherently unstable. Nothing should depend on the
56  // ordering of nodes in the folding set.
57  Bits.append(reinterpret_cast<unsigned *>(&Ptr),
58              reinterpret_cast<unsigned *>(&Ptr+1));
59}
60void FoldingSetNodeID::AddInteger(signed I) {
61  Bits.push_back(I);
62}
63void FoldingSetNodeID::AddInteger(unsigned I) {
64  Bits.push_back(I);
65}
66void FoldingSetNodeID::AddInteger(long I) {
67  AddInteger((unsigned long)I);
68}
69void FoldingSetNodeID::AddInteger(unsigned long I) {
70  if (sizeof(long) == sizeof(int))
71    AddInteger(unsigned(I));
72  else if (sizeof(long) == sizeof(long long)) {
73    AddInteger((unsigned long long)I);
74  } else {
75    llvm_unreachable("unexpected sizeof(long)");
76  }
77}
78void FoldingSetNodeID::AddInteger(long long I) {
79  AddInteger((unsigned long long)I);
80}
81void FoldingSetNodeID::AddInteger(unsigned long long I) {
82  AddInteger(unsigned(I));
83  if ((uint64_t)(unsigned)I != I)
84    Bits.push_back(unsigned(I >> 32));
85}
86
87void FoldingSetNodeID::AddString(StringRef String) {
88  unsigned Size =  String.size();
89  Bits.push_back(Size);
90  if (!Size) return;
91
92  unsigned Units = Size / 4;
93  unsigned Pos = 0;
94  const unsigned *Base = (const unsigned*) String.data();
95
96  // If the string is aligned do a bulk transfer.
97  if (!((intptr_t)Base & 3)) {
98    Bits.append(Base, Base + Units);
99    Pos = (Units + 1) * 4;
100  } else {
101    // Otherwise do it the hard way.
102    // To be compatible with above bulk transfer, we need to take endianness
103    // into account.
104    static_assert(sys::IsBigEndianHost || sys::IsLittleEndianHost,
105                  "Unexpected host endianness");
106    if (sys::IsBigEndianHost) {
107      for (Pos += 4; Pos <= Size; Pos += 4) {
108        unsigned V = ((unsigned char)String[Pos - 4] << 24) |
109                     ((unsigned char)String[Pos - 3] << 16) |
110                     ((unsigned char)String[Pos - 2] << 8) |
111                      (unsigned char)String[Pos - 1];
112        Bits.push_back(V);
113      }
114    } else {  // Little-endian host
115      for (Pos += 4; Pos <= Size; Pos += 4) {
116        unsigned V = ((unsigned char)String[Pos - 1] << 24) |
117                     ((unsigned char)String[Pos - 2] << 16) |
118                     ((unsigned char)String[Pos - 3] << 8) |
119                      (unsigned char)String[Pos - 4];
120        Bits.push_back(V);
121      }
122    }
123  }
124
125  // With the leftover bits.
126  unsigned V = 0;
127  // Pos will have overshot size by 4 - #bytes left over.
128  // No need to take endianness into account here - this is always executed.
129  switch (Pos - Size) {
130  case 1: V = (V << 8) | (unsigned char)String[Size - 3]; // Fall thru.
131  case 2: V = (V << 8) | (unsigned char)String[Size - 2]; // Fall thru.
132  case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
133  default: return; // Nothing left.
134  }
135
136  Bits.push_back(V);
137}
138
139// AddNodeID - Adds the Bit data of another ID to *this.
140void FoldingSetNodeID::AddNodeID(const FoldingSetNodeID &ID) {
141  Bits.append(ID.Bits.begin(), ID.Bits.end());
142}
143
144/// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to
145/// lookup the node in the FoldingSetImpl.
146unsigned FoldingSetNodeID::ComputeHash() const {
147  return FoldingSetNodeIDRef(Bits.data(), Bits.size()).ComputeHash();
148}
149
150/// operator== - Used to compare two nodes to each other.
151///
152bool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS) const {
153  return *this == FoldingSetNodeIDRef(RHS.Bits.data(), RHS.Bits.size());
154}
155
156/// operator== - Used to compare two nodes to each other.
157///
158bool FoldingSetNodeID::operator==(FoldingSetNodeIDRef RHS) const {
159  return FoldingSetNodeIDRef(Bits.data(), Bits.size()) == RHS;
160}
161
162/// Used to compare the "ordering" of two nodes as defined by the
163/// profiled bits and their ordering defined by memcmp().
164bool FoldingSetNodeID::operator<(const FoldingSetNodeID &RHS) const {
165  return *this < FoldingSetNodeIDRef(RHS.Bits.data(), RHS.Bits.size());
166}
167
168bool FoldingSetNodeID::operator<(FoldingSetNodeIDRef RHS) const {
169  return FoldingSetNodeIDRef(Bits.data(), Bits.size()) < RHS;
170}
171
172/// Intern - Copy this node's data to a memory region allocated from the
173/// given allocator and return a FoldingSetNodeIDRef describing the
174/// interned data.
175FoldingSetNodeIDRef
176FoldingSetNodeID::Intern(BumpPtrAllocator &Allocator) const {
177  unsigned *New = Allocator.Allocate<unsigned>(Bits.size());
178  std::uninitialized_copy(Bits.begin(), Bits.end(), New);
179  return FoldingSetNodeIDRef(New, Bits.size());
180}
181
182//===----------------------------------------------------------------------===//
183/// Helper functions for FoldingSetImpl.
184
185/// GetNextPtr - In order to save space, each bucket is a
186/// singly-linked-list. In order to make deletion more efficient, we make
187/// the list circular, so we can delete a node without computing its hash.
188/// The problem with this is that the start of the hash buckets are not
189/// Nodes.  If NextInBucketPtr is a bucket pointer, this method returns null:
190/// use GetBucketPtr when this happens.
191static FoldingSetImpl::Node *GetNextPtr(void *NextInBucketPtr) {
192  // The low bit is set if this is the pointer back to the bucket.
193  if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1)
194    return nullptr;
195
196  return static_cast<FoldingSetImpl::Node*>(NextInBucketPtr);
197}
198
199
200/// testing.
201static void **GetBucketPtr(void *NextInBucketPtr) {
202  intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr);
203  assert((Ptr & 1) && "Not a bucket pointer");
204  return reinterpret_cast<void**>(Ptr & ~intptr_t(1));
205}
206
207/// GetBucketFor - Hash the specified node ID and return the hash bucket for
208/// the specified ID.
209static void **GetBucketFor(unsigned Hash, void **Buckets, unsigned NumBuckets) {
210  // NumBuckets is always a power of 2.
211  unsigned BucketNum = Hash & (NumBuckets-1);
212  return Buckets + BucketNum;
213}
214
215/// AllocateBuckets - Allocated initialized bucket memory.
216static void **AllocateBuckets(unsigned NumBuckets) {
217  void **Buckets = static_cast<void**>(calloc(NumBuckets+1, sizeof(void*)));
218  // Set the very last bucket to be a non-null "pointer".
219  Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
220  return Buckets;
221}
222
223//===----------------------------------------------------------------------===//
224// FoldingSetImpl Implementation
225
226void FoldingSetImpl::anchor() {}
227
228FoldingSetImpl::FoldingSetImpl(unsigned Log2InitSize) {
229  assert(5 < Log2InitSize && Log2InitSize < 32 &&
230         "Initial hash table size out of range");
231  NumBuckets = 1 << Log2InitSize;
232  Buckets = AllocateBuckets(NumBuckets);
233  NumNodes = 0;
234}
235
236FoldingSetImpl::FoldingSetImpl(FoldingSetImpl &&Arg)
237    : Buckets(Arg.Buckets), NumBuckets(Arg.NumBuckets), NumNodes(Arg.NumNodes) {
238  Arg.Buckets = nullptr;
239  Arg.NumBuckets = 0;
240  Arg.NumNodes = 0;
241}
242
243FoldingSetImpl &FoldingSetImpl::operator=(FoldingSetImpl &&RHS) {
244  free(Buckets); // This may be null if the set is in a moved-from state.
245  Buckets = RHS.Buckets;
246  NumBuckets = RHS.NumBuckets;
247  NumNodes = RHS.NumNodes;
248  RHS.Buckets = nullptr;
249  RHS.NumBuckets = 0;
250  RHS.NumNodes = 0;
251  return *this;
252}
253
254FoldingSetImpl::~FoldingSetImpl() {
255  free(Buckets);
256}
257
258void FoldingSetImpl::clear() {
259  // Set all but the last bucket to null pointers.
260  memset(Buckets, 0, NumBuckets*sizeof(void*));
261
262  // Set the very last bucket to be a non-null "pointer".
263  Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
264
265  // Reset the node count to zero.
266  NumNodes = 0;
267}
268
269/// GrowHashTable - Double the size of the hash table and rehash everything.
270///
271void FoldingSetImpl::GrowHashTable() {
272  void **OldBuckets = Buckets;
273  unsigned OldNumBuckets = NumBuckets;
274  NumBuckets <<= 1;
275
276  // Clear out new buckets.
277  Buckets = AllocateBuckets(NumBuckets);
278  NumNodes = 0;
279
280  // Walk the old buckets, rehashing nodes into their new place.
281  FoldingSetNodeID TempID;
282  for (unsigned i = 0; i != OldNumBuckets; ++i) {
283    void *Probe = OldBuckets[i];
284    if (!Probe) continue;
285    while (Node *NodeInBucket = GetNextPtr(Probe)) {
286      // Figure out the next link, remove NodeInBucket from the old link.
287      Probe = NodeInBucket->getNextInBucket();
288      NodeInBucket->SetNextInBucket(nullptr);
289
290      // Insert the node into the new bucket, after recomputing the hash.
291      InsertNode(NodeInBucket,
292                 GetBucketFor(ComputeNodeHash(NodeInBucket, TempID),
293                              Buckets, NumBuckets));
294      TempID.clear();
295    }
296  }
297
298  free(OldBuckets);
299}
300
301/// FindNodeOrInsertPos - Look up the node specified by ID.  If it exists,
302/// return it.  If not, return the insertion token that will make insertion
303/// faster.
304FoldingSetImpl::Node
305*FoldingSetImpl::FindNodeOrInsertPos(const FoldingSetNodeID &ID,
306                                     void *&InsertPos) {
307  unsigned IDHash = ID.ComputeHash();
308  void **Bucket = GetBucketFor(IDHash, Buckets, NumBuckets);
309  void *Probe = *Bucket;
310
311  InsertPos = nullptr;
312
313  FoldingSetNodeID TempID;
314  while (Node *NodeInBucket = GetNextPtr(Probe)) {
315    if (NodeEquals(NodeInBucket, ID, IDHash, TempID))
316      return NodeInBucket;
317    TempID.clear();
318
319    Probe = NodeInBucket->getNextInBucket();
320  }
321
322  // Didn't find the node, return null with the bucket as the InsertPos.
323  InsertPos = Bucket;
324  return nullptr;
325}
326
327/// InsertNode - Insert the specified node into the folding set, knowing that it
328/// is not already in the map.  InsertPos must be obtained from
329/// FindNodeOrInsertPos.
330void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) {
331  assert(!N->getNextInBucket());
332  // Do we need to grow the hashtable?
333  if (NumNodes+1 > NumBuckets*2) {
334    GrowHashTable();
335    FoldingSetNodeID TempID;
336    InsertPos = GetBucketFor(ComputeNodeHash(N, TempID), Buckets, NumBuckets);
337  }
338
339  ++NumNodes;
340
341  /// The insert position is actually a bucket pointer.
342  void **Bucket = static_cast<void**>(InsertPos);
343
344  void *Next = *Bucket;
345
346  // If this is the first insertion into this bucket, its next pointer will be
347  // null.  Pretend as if it pointed to itself, setting the low bit to indicate
348  // that it is a pointer to the bucket.
349  if (!Next)
350    Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1);
351
352  // Set the node's next pointer, and make the bucket point to the node.
353  N->SetNextInBucket(Next);
354  *Bucket = N;
355}
356
357/// RemoveNode - Remove a node from the folding set, returning true if one was
358/// removed or false if the node was not in the folding set.
359bool FoldingSetImpl::RemoveNode(Node *N) {
360  // Because each bucket is a circular list, we don't need to compute N's hash
361  // to remove it.
362  void *Ptr = N->getNextInBucket();
363  if (!Ptr) return false;  // Not in folding set.
364
365  --NumNodes;
366  N->SetNextInBucket(nullptr);
367
368  // Remember what N originally pointed to, either a bucket or another node.
369  void *NodeNextPtr = Ptr;
370
371  // Chase around the list until we find the node (or bucket) which points to N.
372  while (true) {
373    if (Node *NodeInBucket = GetNextPtr(Ptr)) {
374      // Advance pointer.
375      Ptr = NodeInBucket->getNextInBucket();
376
377      // We found a node that points to N, change it to point to N's next node,
378      // removing N from the list.
379      if (Ptr == N) {
380        NodeInBucket->SetNextInBucket(NodeNextPtr);
381        return true;
382      }
383    } else {
384      void **Bucket = GetBucketPtr(Ptr);
385      Ptr = *Bucket;
386
387      // If we found that the bucket points to N, update the bucket to point to
388      // whatever is next.
389      if (Ptr == N) {
390        *Bucket = NodeNextPtr;
391        return true;
392      }
393    }
394  }
395}
396
397/// GetOrInsertNode - If there is an existing simple Node exactly
398/// equal to the specified node, return it.  Otherwise, insert 'N' and it
399/// instead.
400FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) {
401  FoldingSetNodeID ID;
402  GetNodeProfile(N, ID);
403  void *IP;
404  if (Node *E = FindNodeOrInsertPos(ID, IP))
405    return E;
406  InsertNode(N, IP);
407  return N;
408}
409
410//===----------------------------------------------------------------------===//
411// FoldingSetIteratorImpl Implementation
412
413FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) {
414  // Skip to the first non-null non-self-cycle bucket.
415  while (*Bucket != reinterpret_cast<void*>(-1) &&
416         (!*Bucket || !GetNextPtr(*Bucket)))
417    ++Bucket;
418
419  NodePtr = static_cast<FoldingSetNode*>(*Bucket);
420}
421
422void FoldingSetIteratorImpl::advance() {
423  // If there is another link within this bucket, go to it.
424  void *Probe = NodePtr->getNextInBucket();
425
426  if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe))
427    NodePtr = NextNodeInBucket;
428  else {
429    // Otherwise, this is the last link in this bucket.
430    void **Bucket = GetBucketPtr(Probe);
431
432    // Skip to the next non-null non-self-cycle bucket.
433    do {
434      ++Bucket;
435    } while (*Bucket != reinterpret_cast<void*>(-1) &&
436             (!*Bucket || !GetNextPtr(*Bucket)));
437
438    NodePtr = static_cast<FoldingSetNode*>(*Bucket);
439  }
440}
441
442//===----------------------------------------------------------------------===//
443// FoldingSetBucketIteratorImpl Implementation
444
445FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) {
446  Ptr = (!*Bucket || !GetNextPtr(*Bucket)) ? (void*) Bucket : *Bucket;
447}
448