Line data Source code
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>
23 : using 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 FoldingSetBase.
30 1334048764 : unsigned FoldingSetNodeIDRef::ComputeHash() const {
31 1334048764 : return static_cast<unsigned>(hash_combine_range(Data, Data+Size));
32 : }
33 :
34 1287946753 : bool FoldingSetNodeIDRef::operator==(FoldingSetNodeIDRef RHS) const {
35 1287946753 : if (Size != RHS.Size) return false;
36 1101014582 : 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().
41 276 : bool FoldingSetNodeIDRef::operator<(FoldingSetNodeIDRef RHS) const {
42 276 : if (Size != RHS.Size)
43 151 : return Size < RHS.Size;
44 125 : 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 : ///
52 4946548633 : void 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 : static_assert(sizeof(uintptr_t) <= sizeof(unsigned long long),
58 : "unexpected pointer size");
59 4946548633 : AddInteger(reinterpret_cast<uintptr_t>(Ptr));
60 4946548717 : }
61 1060375629 : void FoldingSetNodeID::AddInteger(signed I) {
62 1060375629 : Bits.push_back(I);
63 1060375629 : }
64 12989978484 : void FoldingSetNodeID::AddInteger(unsigned I) {
65 12989978484 : Bits.push_back(I);
66 12989979010 : }
67 66629681 : void FoldingSetNodeID::AddInteger(long I) {
68 66629681 : AddInteger((unsigned long)I);
69 66629682 : }
70 5240782148 : void FoldingSetNodeID::AddInteger(unsigned long I) {
71 : if (sizeof(long) == sizeof(int))
72 : AddInteger(unsigned(I));
73 : else if (sizeof(long) == sizeof(long long)) {
74 5240782148 : AddInteger((unsigned long long)I);
75 : } else {
76 : llvm_unreachable("unexpected sizeof(long)");
77 : }
78 5240782559 : }
79 0 : void FoldingSetNodeID::AddInteger(long long I) {
80 0 : AddInteger((unsigned long long)I);
81 0 : }
82 5240787713 : void FoldingSetNodeID::AddInteger(unsigned long long I) {
83 5240787713 : AddInteger(unsigned(I));
84 5240787715 : AddInteger(unsigned(I >> 32));
85 5240787743 : }
86 :
87 295750186 : void FoldingSetNodeID::AddString(StringRef String) {
88 295750186 : unsigned Size = String.size();
89 295750186 : Bits.push_back(Size);
90 316130025 : if (!Size) return;
91 :
92 295737447 : 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 295737447 : if (!((intptr_t)Base & 3)) {
98 294896102 : Bits.append(Base, Base + Units);
99 294896102 : 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 3378200 : for (Pos += 4; Pos <= Size; Pos += 4) {
116 2536855 : unsigned V = ((unsigned char)String[Pos - 1] << 24) |
117 2536855 : ((unsigned char)String[Pos - 2] << 16) |
118 2536855 : ((unsigned char)String[Pos - 3] << 8) |
119 2536855 : (unsigned char)String[Pos - 4];
120 2536855 : Bits.push_back(V);
121 : }
122 : }
123 : }
124 :
125 : // With the leftover bits.
126 295737447 : 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 295737447 : switch (Pos - Size) {
130 104539814 : case 1: V = (V << 8) | (unsigned char)String[Size - 3]; LLVM_FALLTHROUGH;
131 270539874 : case 2: V = (V << 8) | (unsigned char)String[Size - 2]; LLVM_FALLTHROUGH;
132 275357608 : case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
133 : default: return; // Nothing left.
134 : }
135 :
136 275357608 : Bits.push_back(V);
137 : }
138 :
139 : // AddNodeID - Adds the Bit data of another ID to *this.
140 11649675 : void FoldingSetNodeID::AddNodeID(const FoldingSetNodeID &ID) {
141 23299350 : Bits.append(ID.Bits.begin(), ID.Bits.end());
142 11649675 : }
143 :
144 : /// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to
145 : /// lookup the node in the FoldingSetBase.
146 1333432226 : unsigned FoldingSetNodeID::ComputeHash() const {
147 4000296678 : return FoldingSetNodeIDRef(Bits.data(), Bits.size()).ComputeHash();
148 : }
149 :
150 : /// operator== - Used to compare two nodes to each other.
151 : ///
152 1254730450 : bool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS) const {
153 3764191350 : return *this == FoldingSetNodeIDRef(RHS.Bits.data(), RHS.Bits.size());
154 : }
155 :
156 : /// operator== - Used to compare two nodes to each other.
157 : ///
158 1287946753 : bool FoldingSetNodeID::operator==(FoldingSetNodeIDRef RHS) const {
159 3863840259 : 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().
164 276 : bool FoldingSetNodeID::operator<(const FoldingSetNodeID &RHS) const {
165 828 : return *this < FoldingSetNodeIDRef(RHS.Bits.data(), RHS.Bits.size());
166 : }
167 :
168 276 : bool FoldingSetNodeID::operator<(FoldingSetNodeIDRef RHS) const {
169 828 : 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.
175 : FoldingSetNodeIDRef
176 1507833 : FoldingSetNodeID::Intern(BumpPtrAllocator &Allocator) const {
177 1507833 : unsigned *New = Allocator.Allocate<unsigned>(Bits.size());
178 : std::uninitialized_copy(Bits.begin(), Bits.end(), New);
179 1507833 : return FoldingSetNodeIDRef(New, Bits.size());
180 : }
181 :
182 : //===----------------------------------------------------------------------===//
183 : /// Helper functions for FoldingSetBase.
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.
191 : static FoldingSetBase::Node *GetNextPtr(void *NextInBucketPtr) {
192 : // The low bit is set if this is the pointer back to the bucket.
193 2163731078 : if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1)
194 : return nullptr;
195 :
196 : return static_cast<FoldingSetBase::Node*>(NextInBucketPtr);
197 : }
198 :
199 :
200 : /// testing.
201 : static void **GetBucketPtr(void *NextInBucketPtr) {
202 : intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr);
203 : assert((Ptr & 1) && "Not a bucket pointer");
204 1089351 : 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.
209 : static void **GetBucketFor(unsigned Hash, void **Buckets, unsigned NumBuckets) {
210 : // NumBuckets is always a power of 2.
211 1312908575 : unsigned BucketNum = Hash & (NumBuckets-1);
212 335499 : return Buckets + BucketNum;
213 : }
214 :
215 : /// AllocateBuckets - Allocated initialized bucket memory.
216 : static void **AllocateBuckets(unsigned NumBuckets) {
217 11472062 : void **Buckets = static_cast<void**>(safe_calloc(NumBuckets + 1,
218 : sizeof(void*)));
219 : // Set the very last bucket to be a non-null "pointer".
220 11472064 : Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
221 : return Buckets;
222 : }
223 :
224 : //===----------------------------------------------------------------------===//
225 : // FoldingSetBase Implementation
226 :
227 0 : void FoldingSetBase::anchor() {}
228 :
229 11126810 : FoldingSetBase::FoldingSetBase(unsigned Log2InitSize) {
230 : assert(5 < Log2InitSize && Log2InitSize < 32 &&
231 : "Initial hash table size out of range");
232 11126810 : NumBuckets = 1 << Log2InitSize;
233 11126812 : Buckets = AllocateBuckets(NumBuckets);
234 11126812 : NumNodes = 0;
235 11126812 : }
236 :
237 3652 : FoldingSetBase::FoldingSetBase(FoldingSetBase &&Arg)
238 3652 : : Buckets(Arg.Buckets), NumBuckets(Arg.NumBuckets), NumNodes(Arg.NumNodes) {
239 3652 : Arg.Buckets = nullptr;
240 3652 : Arg.NumBuckets = 0;
241 3652 : Arg.NumNodes = 0;
242 3652 : }
243 :
244 0 : FoldingSetBase &FoldingSetBase::operator=(FoldingSetBase &&RHS) {
245 0 : free(Buckets); // This may be null if the set is in a moved-from state.
246 0 : Buckets = RHS.Buckets;
247 0 : NumBuckets = RHS.NumBuckets;
248 0 : NumNodes = RHS.NumNodes;
249 0 : RHS.Buckets = nullptr;
250 0 : RHS.NumBuckets = 0;
251 0 : RHS.NumNodes = 0;
252 0 : return *this;
253 : }
254 :
255 6722894 : FoldingSetBase::~FoldingSetBase() {
256 3361447 : free(Buckets);
257 3361447 : }
258 :
259 1270699 : void FoldingSetBase::clear() {
260 : // Set all but the last bucket to null pointers.
261 1270699 : memset(Buckets, 0, NumBuckets*sizeof(void*));
262 :
263 : // Set the very last bucket to be a non-null "pointer".
264 1270699 : Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
265 :
266 : // Reset the node count to zero.
267 1270699 : NumNodes = 0;
268 1270699 : }
269 :
270 345252 : void FoldingSetBase::GrowBucketCount(unsigned NewBucketCount) {
271 : assert((NewBucketCount > NumBuckets) && "Can't shrink a folding set with GrowBucketCount");
272 : assert(isPowerOf2_32(NewBucketCount) && "Bad bucket count!");
273 345252 : void **OldBuckets = Buckets;
274 345252 : unsigned OldNumBuckets = NumBuckets;
275 :
276 : // Clear out new buckets.
277 345252 : Buckets = AllocateBuckets(NewBucketCount);
278 : // Set NumBuckets only if allocation of new buckets was successful.
279 345252 : NumBuckets = NewBucketCount;
280 345252 : NumNodes = 0;
281 :
282 : // Walk the old buckets, rehashing nodes into their new place.
283 : FoldingSetNodeID TempID;
284 150409188 : for (unsigned i = 0; i != OldNumBuckets; ++i) {
285 150063936 : void *Probe = OldBuckets[i];
286 150063936 : if (!Probe) continue;
287 298889342 : while (Node *NodeInBucket = GetNextPtr(Probe)) {
288 : // Figure out the next link, remove NodeInBucket from the old link.
289 298889342 : Probe = NodeInBucket->getNextInBucket();
290 : NodeInBucket->SetNextInBucket(nullptr);
291 :
292 : // Insert the node into the new bucket, after recomputing the hash.
293 298889342 : InsertNode(NodeInBucket,
294 298889342 : GetBucketFor(ComputeNodeHash(NodeInBucket, TempID),
295 : Buckets, NumBuckets));
296 : TempID.clear();
297 : }
298 : }
299 :
300 345252 : free(OldBuckets);
301 345252 : }
302 :
303 : /// GrowHashTable - Double the size of the hash table and rehash everything.
304 : ///
305 335499 : void FoldingSetBase::GrowHashTable() {
306 335499 : GrowBucketCount(NumBuckets * 2);
307 335499 : }
308 :
309 9761 : void FoldingSetBase::reserve(unsigned EltCount) {
310 : // This will give us somewhere between EltCount / 2 and
311 : // EltCount buckets. This puts us in the load factor
312 : // range of 1.0 - 2.0.
313 19522 : if(EltCount < capacity())
314 : return;
315 19506 : GrowBucketCount(PowerOf2Floor(EltCount));
316 : }
317 :
318 : /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists,
319 : /// return it. If not, return the insertion token that will make insertion
320 : /// faster.
321 : FoldingSetBase::Node *
322 1013683738 : FoldingSetBase::FindNodeOrInsertPos(const FoldingSetNodeID &ID,
323 : void *&InsertPos) {
324 1013683738 : unsigned IDHash = ID.ComputeHash();
325 1013683734 : void **Bucket = GetBucketFor(IDHash, Buckets, NumBuckets);
326 1013683734 : void *Probe = *Bucket;
327 :
328 1013683734 : InsertPos = nullptr;
329 :
330 : FoldingSetNodeID TempID;
331 1427647413 : while (Node *NodeInBucket = GetNextPtr(Probe)) {
332 1288747572 : if (NodeEquals(NodeInBucket, ID, IDHash, TempID))
333 626590313 : return NodeInBucket;
334 : TempID.clear();
335 :
336 662157258 : Probe = NodeInBucket->getNextInBucket();
337 662157258 : }
338 :
339 : // Didn't find the node, return null with the bucket as the InsertPos.
340 387093420 : InsertPos = Bucket;
341 387093420 : return nullptr;
342 : }
343 :
344 : /// InsertNode - Insert the specified node into the folding set, knowing that it
345 : /// is not already in the map. InsertPos must be obtained from
346 : /// FindNodeOrInsertPos.
347 605411302 : void FoldingSetBase::InsertNode(Node *N, void *InsertPos) {
348 : assert(!N->getNextInBucket());
349 : // Do we need to grow the hashtable?
350 1210822604 : if (NumNodes+1 > capacity()) {
351 335499 : GrowHashTable();
352 : FoldingSetNodeID TempID;
353 335499 : InsertPos = GetBucketFor(ComputeNodeHash(N, TempID), Buckets, NumBuckets);
354 : }
355 :
356 605411302 : ++NumNodes;
357 :
358 : /// The insert position is actually a bucket pointer.
359 : void **Bucket = static_cast<void**>(InsertPos);
360 :
361 605411302 : void *Next = *Bucket;
362 :
363 : // If this is the first insertion into this bucket, its next pointer will be
364 : // null. Pretend as if it pointed to itself, setting the low bit to indicate
365 : // that it is a pointer to the bucket.
366 605411302 : if (!Next)
367 305131132 : Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1);
368 :
369 : // Set the node's next pointer, and make the bucket point to the node.
370 : N->SetNextInBucket(Next);
371 605411302 : *Bucket = N;
372 605411302 : }
373 :
374 : /// RemoveNode - Remove a node from the folding set, returning true if one was
375 : /// removed or false if the node was not in the folding set.
376 44306201 : bool FoldingSetBase::RemoveNode(Node *N) {
377 : // Because each bucket is a circular list, we don't need to compute N's hash
378 : // to remove it.
379 44306201 : void *Ptr = N->getNextInBucket();
380 44306201 : if (!Ptr) return false; // Not in folding set.
381 :
382 38756108 : --NumNodes;
383 : N->SetNextInBucket(nullptr);
384 :
385 : // Remember what N originally pointed to, either a bucket or another node.
386 : void *NodeNextPtr = Ptr;
387 :
388 : // Chase around the list until we find the node (or bucket) which points to N.
389 : while (true) {
390 19873097 : if (Node *NodeInBucket = GetNextPtr(Ptr)) {
391 : // Advance pointer.
392 19873097 : Ptr = NodeInBucket->getNextInBucket();
393 :
394 : // We found a node that points to N, change it to point to N's next node,
395 : // removing N from the list.
396 19873097 : if (Ptr == N) {
397 : NodeInBucket->SetNextInBucket(NodeNextPtr);
398 7197156 : return true;
399 : }
400 : } else {
401 : void **Bucket = GetBucketPtr(Ptr);
402 38756108 : Ptr = *Bucket;
403 :
404 : // If we found that the bucket points to N, update the bucket to point to
405 : // whatever is next.
406 38756108 : if (Ptr == N) {
407 31558952 : *Bucket = NodeNextPtr;
408 31558952 : return true;
409 : }
410 : }
411 : }
412 : }
413 :
414 : /// GetOrInsertNode - If there is an existing simple Node exactly
415 : /// equal to the specified node, return it. Otherwise, insert 'N' and it
416 : /// instead.
417 10937652 : FoldingSetBase::Node *FoldingSetBase::GetOrInsertNode(FoldingSetBase::Node *N) {
418 : FoldingSetNodeID ID;
419 10937652 : GetNodeProfile(N, ID);
420 : void *IP;
421 10937652 : if (Node *E = FindNodeOrInsertPos(ID, IP))
422 : return E;
423 10861012 : InsertNode(N, IP);
424 10861012 : return N;
425 : }
426 :
427 : //===----------------------------------------------------------------------===//
428 : // FoldingSetIteratorImpl Implementation
429 :
430 556399 : FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) {
431 : // Skip to the first non-null non-self-cycle bucket.
432 13526706 : while (*Bucket != reinterpret_cast<void*>(-1) &&
433 : (!*Bucket || !GetNextPtr(*Bucket)))
434 12970307 : ++Bucket;
435 :
436 556399 : NodePtr = static_cast<FoldingSetNode*>(*Bucket);
437 556399 : }
438 :
439 1434331 : void FoldingSetIteratorImpl::advance() {
440 : // If there is another link within this bucket, go to it.
441 1434331 : void *Probe = NodePtr->getNextInBucket();
442 :
443 344981 : if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe))
444 344980 : NodePtr = NextNodeInBucket;
445 : else {
446 : // Otherwise, this is the last link in this bucket.
447 : void **Bucket = GetBucketPtr(Probe);
448 :
449 : // Skip to the next non-null non-self-cycle bucket.
450 : do {
451 48195295 : ++Bucket;
452 48195295 : } while (*Bucket != reinterpret_cast<void*>(-1) &&
453 : (!*Bucket || !GetNextPtr(*Bucket)));
454 :
455 1089351 : NodePtr = static_cast<FoldingSetNode*>(*Bucket);
456 : }
457 1434331 : }
458 :
459 : //===----------------------------------------------------------------------===//
460 : // FoldingSetBucketIteratorImpl Implementation
461 :
462 0 : FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) {
463 0 : Ptr = (!*Bucket || !GetNextPtr(*Bucket)) ? (void*) Bucket : *Bucket;
464 0 : }
|
