1 //===-- Support/FoldingSet.cpp - Uniquing Hash Set --------------*- C++ -*-===//
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 a hash set that can be used to remove duplication of
10 // nodes in a graph.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "llvm/ADT/FoldingSet.h"
15 #include "llvm/ADT/Hashing.h"
16 #include "llvm/ADT/StringRef.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.
ComputeHash() const30 unsigned FoldingSetNodeIDRef::ComputeHash() const {
31   return static_cast<unsigned>(hash_combine_range(Data, Data+Size));
32 }
33 
operator ==(FoldingSetNodeIDRef RHS) const34 bool 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().
operator <(FoldingSetNodeIDRef RHS) const41 bool 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 ///
AddPointer(const void * Ptr)52 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   AddInteger(reinterpret_cast<uintptr_t>(Ptr));
60 }
AddInteger(signed I)61 void FoldingSetNodeID::AddInteger(signed I) {
62   Bits.push_back(I);
63 }
AddInteger(unsigned I)64 void FoldingSetNodeID::AddInteger(unsigned I) {
65   Bits.push_back(I);
66 }
AddInteger(long I)67 void FoldingSetNodeID::AddInteger(long I) {
68   AddInteger((unsigned long)I);
69 }
AddInteger(unsigned long I)70 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     AddInteger((unsigned long long)I);
75   } else {
76     llvm_unreachable("unexpected sizeof(long)");
77   }
78 }
AddInteger(long long I)79 void FoldingSetNodeID::AddInteger(long long I) {
80   AddInteger((unsigned long long)I);
81 }
AddInteger(unsigned long long I)82 void FoldingSetNodeID::AddInteger(unsigned long long I) {
83   AddInteger(unsigned(I));
84   AddInteger(unsigned(I >> 32));
85 }
86 
AddString(StringRef String)87 void FoldingSetNodeID::AddString(StringRef String) {
88   unsigned Size =  String.size();
89 
90   unsigned NumInserts = 1 + divideCeil(Size, 4);
91   Bits.reserve(Bits.size() + NumInserts);
92 
93   Bits.push_back(Size);
94   if (!Size) return;
95 
96   unsigned Units = Size / 4;
97   unsigned Pos = 0;
98   const unsigned *Base = (const unsigned*) String.data();
99 
100   // If the string is aligned do a bulk transfer.
101   if (!((intptr_t)Base & 3)) {
102     Bits.append(Base, Base + Units);
103     Pos = (Units + 1) * 4;
104   } else {
105     // Otherwise do it the hard way.
106     // To be compatible with above bulk transfer, we need to take endianness
107     // into account.
108     static_assert(sys::IsBigEndianHost || sys::IsLittleEndianHost,
109                   "Unexpected host endianness");
110     if (sys::IsBigEndianHost) {
111       for (Pos += 4; Pos <= Size; Pos += 4) {
112         unsigned V = ((unsigned char)String[Pos - 4] << 24) |
113                      ((unsigned char)String[Pos - 3] << 16) |
114                      ((unsigned char)String[Pos - 2] << 8) |
115                       (unsigned char)String[Pos - 1];
116         Bits.push_back(V);
117       }
118     } else {  // Little-endian host
119       for (Pos += 4; Pos <= Size; Pos += 4) {
120         unsigned V = ((unsigned char)String[Pos - 1] << 24) |
121                      ((unsigned char)String[Pos - 2] << 16) |
122                      ((unsigned char)String[Pos - 3] << 8) |
123                       (unsigned char)String[Pos - 4];
124         Bits.push_back(V);
125       }
126     }
127   }
128 
129   // With the leftover bits.
130   unsigned V = 0;
131   // Pos will have overshot size by 4 - #bytes left over.
132   // No need to take endianness into account here - this is always executed.
133   switch (Pos - Size) {
134   case 1: V = (V << 8) | (unsigned char)String[Size - 3]; LLVM_FALLTHROUGH;
135   case 2: V = (V << 8) | (unsigned char)String[Size - 2]; LLVM_FALLTHROUGH;
136   case 3: V = (V << 8) | (unsigned char)String[Size - 1]; break;
137   default: return; // Nothing left.
138   }
139 
140   Bits.push_back(V);
141 }
142 
143 // AddNodeID - Adds the Bit data of another ID to *this.
AddNodeID(const FoldingSetNodeID & ID)144 void FoldingSetNodeID::AddNodeID(const FoldingSetNodeID &ID) {
145   Bits.append(ID.Bits.begin(), ID.Bits.end());
146 }
147 
148 /// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used to
149 /// lookup the node in the FoldingSetBase.
ComputeHash() const150 unsigned FoldingSetNodeID::ComputeHash() const {
151   return FoldingSetNodeIDRef(Bits.data(), Bits.size()).ComputeHash();
152 }
153 
154 /// operator== - Used to compare two nodes to each other.
155 ///
operator ==(const FoldingSetNodeID & RHS) const156 bool FoldingSetNodeID::operator==(const FoldingSetNodeID &RHS) const {
157   return *this == FoldingSetNodeIDRef(RHS.Bits.data(), RHS.Bits.size());
158 }
159 
160 /// operator== - Used to compare two nodes to each other.
161 ///
operator ==(FoldingSetNodeIDRef RHS) const162 bool FoldingSetNodeID::operator==(FoldingSetNodeIDRef RHS) const {
163   return FoldingSetNodeIDRef(Bits.data(), Bits.size()) == RHS;
164 }
165 
166 /// Used to compare the "ordering" of two nodes as defined by the
167 /// profiled bits and their ordering defined by memcmp().
operator <(const FoldingSetNodeID & RHS) const168 bool FoldingSetNodeID::operator<(const FoldingSetNodeID &RHS) const {
169   return *this < FoldingSetNodeIDRef(RHS.Bits.data(), RHS.Bits.size());
170 }
171 
operator <(FoldingSetNodeIDRef RHS) const172 bool FoldingSetNodeID::operator<(FoldingSetNodeIDRef RHS) const {
173   return FoldingSetNodeIDRef(Bits.data(), Bits.size()) < RHS;
174 }
175 
176 /// Intern - Copy this node's data to a memory region allocated from the
177 /// given allocator and return a FoldingSetNodeIDRef describing the
178 /// interned data.
179 FoldingSetNodeIDRef
Intern(BumpPtrAllocator & Allocator) const180 FoldingSetNodeID::Intern(BumpPtrAllocator &Allocator) const {
181   unsigned *New = Allocator.Allocate<unsigned>(Bits.size());
182   std::uninitialized_copy(Bits.begin(), Bits.end(), New);
183   return FoldingSetNodeIDRef(New, Bits.size());
184 }
185 
186 //===----------------------------------------------------------------------===//
187 /// Helper functions for FoldingSetBase.
188 
189 /// GetNextPtr - In order to save space, each bucket is a
190 /// singly-linked-list. In order to make deletion more efficient, we make
191 /// the list circular, so we can delete a node without computing its hash.
192 /// The problem with this is that the start of the hash buckets are not
193 /// Nodes.  If NextInBucketPtr is a bucket pointer, this method returns null:
194 /// use GetBucketPtr when this happens.
GetNextPtr(void * NextInBucketPtr)195 static FoldingSetBase::Node *GetNextPtr(void *NextInBucketPtr) {
196   // The low bit is set if this is the pointer back to the bucket.
197   if (reinterpret_cast<intptr_t>(NextInBucketPtr) & 1)
198     return nullptr;
199 
200   return static_cast<FoldingSetBase::Node*>(NextInBucketPtr);
201 }
202 
203 
204 /// testing.
GetBucketPtr(void * NextInBucketPtr)205 static void **GetBucketPtr(void *NextInBucketPtr) {
206   intptr_t Ptr = reinterpret_cast<intptr_t>(NextInBucketPtr);
207   assert((Ptr & 1) && "Not a bucket pointer");
208   return reinterpret_cast<void**>(Ptr & ~intptr_t(1));
209 }
210 
211 /// GetBucketFor - Hash the specified node ID and return the hash bucket for
212 /// the specified ID.
GetBucketFor(unsigned Hash,void ** Buckets,unsigned NumBuckets)213 static void **GetBucketFor(unsigned Hash, void **Buckets, unsigned NumBuckets) {
214   // NumBuckets is always a power of 2.
215   unsigned BucketNum = Hash & (NumBuckets-1);
216   return Buckets + BucketNum;
217 }
218 
219 /// AllocateBuckets - Allocated initialized bucket memory.
AllocateBuckets(unsigned NumBuckets)220 static void **AllocateBuckets(unsigned NumBuckets) {
221   void **Buckets = static_cast<void**>(safe_calloc(NumBuckets + 1,
222                                                    sizeof(void*)));
223   // Set the very last bucket to be a non-null "pointer".
224   Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
225   return Buckets;
226 }
227 
228 //===----------------------------------------------------------------------===//
229 // FoldingSetBase Implementation
230 
FoldingSetBase(unsigned Log2InitSize)231 FoldingSetBase::FoldingSetBase(unsigned Log2InitSize) {
232   assert(5 < Log2InitSize && Log2InitSize < 32 &&
233          "Initial hash table size out of range");
234   NumBuckets = 1 << Log2InitSize;
235   Buckets = AllocateBuckets(NumBuckets);
236   NumNodes = 0;
237 }
238 
FoldingSetBase(FoldingSetBase && Arg)239 FoldingSetBase::FoldingSetBase(FoldingSetBase &&Arg)
240     : Buckets(Arg.Buckets), NumBuckets(Arg.NumBuckets), NumNodes(Arg.NumNodes) {
241   Arg.Buckets = nullptr;
242   Arg.NumBuckets = 0;
243   Arg.NumNodes = 0;
244 }
245 
operator =(FoldingSetBase && RHS)246 FoldingSetBase &FoldingSetBase::operator=(FoldingSetBase &&RHS) {
247   free(Buckets); // This may be null if the set is in a moved-from state.
248   Buckets = RHS.Buckets;
249   NumBuckets = RHS.NumBuckets;
250   NumNodes = RHS.NumNodes;
251   RHS.Buckets = nullptr;
252   RHS.NumBuckets = 0;
253   RHS.NumNodes = 0;
254   return *this;
255 }
256 
~FoldingSetBase()257 FoldingSetBase::~FoldingSetBase() {
258   free(Buckets);
259 }
260 
clear()261 void FoldingSetBase::clear() {
262   // Set all but the last bucket to null pointers.
263   memset(Buckets, 0, NumBuckets*sizeof(void*));
264 
265   // Set the very last bucket to be a non-null "pointer".
266   Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
267 
268   // Reset the node count to zero.
269   NumNodes = 0;
270 }
271 
GrowBucketCount(unsigned NewBucketCount,const FoldingSetInfo & Info)272 void FoldingSetBase::GrowBucketCount(unsigned NewBucketCount,
273                                      const FoldingSetInfo &Info) {
274   assert((NewBucketCount > NumBuckets) &&
275          "Can't shrink a folding set with GrowBucketCount");
276   assert(isPowerOf2_32(NewBucketCount) && "Bad bucket count!");
277   void **OldBuckets = Buckets;
278   unsigned OldNumBuckets = NumBuckets;
279 
280   // Clear out new buckets.
281   Buckets = AllocateBuckets(NewBucketCount);
282   // Set NumBuckets only if allocation of new buckets was successful.
283   NumBuckets = NewBucketCount;
284   NumNodes = 0;
285 
286   // Walk the old buckets, rehashing nodes into their new place.
287   FoldingSetNodeID TempID;
288   for (unsigned i = 0; i != OldNumBuckets; ++i) {
289     void *Probe = OldBuckets[i];
290     if (!Probe) continue;
291     while (Node *NodeInBucket = GetNextPtr(Probe)) {
292       // Figure out the next link, remove NodeInBucket from the old link.
293       Probe = NodeInBucket->getNextInBucket();
294       NodeInBucket->SetNextInBucket(nullptr);
295 
296       // Insert the node into the new bucket, after recomputing the hash.
297       InsertNode(NodeInBucket,
298                  GetBucketFor(Info.ComputeNodeHash(this, NodeInBucket, TempID),
299                               Buckets, NumBuckets),
300                  Info);
301       TempID.clear();
302     }
303   }
304 
305   free(OldBuckets);
306 }
307 
308 /// GrowHashTable - Double the size of the hash table and rehash everything.
309 ///
GrowHashTable(const FoldingSetInfo & Info)310 void FoldingSetBase::GrowHashTable(const FoldingSetInfo &Info) {
311   GrowBucketCount(NumBuckets * 2, Info);
312 }
313 
reserve(unsigned EltCount,const FoldingSetInfo & Info)314 void FoldingSetBase::reserve(unsigned EltCount, const FoldingSetInfo &Info) {
315   // This will give us somewhere between EltCount / 2 and
316   // EltCount buckets.  This puts us in the load factor
317   // range of 1.0 - 2.0.
318   if(EltCount < capacity())
319     return;
320   GrowBucketCount(PowerOf2Floor(EltCount), Info);
321 }
322 
323 /// FindNodeOrInsertPos - Look up the node specified by ID.  If it exists,
324 /// return it.  If not, return the insertion token that will make insertion
325 /// faster.
FindNodeOrInsertPos(const FoldingSetNodeID & ID,void * & InsertPos,const FoldingSetInfo & Info)326 FoldingSetBase::Node *FoldingSetBase::FindNodeOrInsertPos(
327     const FoldingSetNodeID &ID, void *&InsertPos, const FoldingSetInfo &Info) {
328   unsigned IDHash = ID.ComputeHash();
329   void **Bucket = GetBucketFor(IDHash, Buckets, NumBuckets);
330   void *Probe = *Bucket;
331 
332   InsertPos = nullptr;
333 
334   FoldingSetNodeID TempID;
335   while (Node *NodeInBucket = GetNextPtr(Probe)) {
336     if (Info.NodeEquals(this, NodeInBucket, ID, IDHash, TempID))
337       return NodeInBucket;
338     TempID.clear();
339 
340     Probe = NodeInBucket->getNextInBucket();
341   }
342 
343   // Didn't find the node, return null with the bucket as the InsertPos.
344   InsertPos = Bucket;
345   return nullptr;
346 }
347 
348 /// InsertNode - Insert the specified node into the folding set, knowing that it
349 /// is not already in the map.  InsertPos must be obtained from
350 /// FindNodeOrInsertPos.
InsertNode(Node * N,void * InsertPos,const FoldingSetInfo & Info)351 void FoldingSetBase::InsertNode(Node *N, void *InsertPos,
352                                 const FoldingSetInfo &Info) {
353   assert(!N->getNextInBucket());
354   // Do we need to grow the hashtable?
355   if (NumNodes+1 > capacity()) {
356     GrowHashTable(Info);
357     FoldingSetNodeID TempID;
358     InsertPos = GetBucketFor(Info.ComputeNodeHash(this, N, TempID), Buckets,
359                              NumBuckets);
360   }
361 
362   ++NumNodes;
363 
364   /// The insert position is actually a bucket pointer.
365   void **Bucket = static_cast<void**>(InsertPos);
366 
367   void *Next = *Bucket;
368 
369   // If this is the first insertion into this bucket, its next pointer will be
370   // null.  Pretend as if it pointed to itself, setting the low bit to indicate
371   // that it is a pointer to the bucket.
372   if (!Next)
373     Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1);
374 
375   // Set the node's next pointer, and make the bucket point to the node.
376   N->SetNextInBucket(Next);
377   *Bucket = N;
378 }
379 
380 /// RemoveNode - Remove a node from the folding set, returning true if one was
381 /// removed or false if the node was not in the folding set.
RemoveNode(Node * N)382 bool FoldingSetBase::RemoveNode(Node *N) {
383   // Because each bucket is a circular list, we don't need to compute N's hash
384   // to remove it.
385   void *Ptr = N->getNextInBucket();
386   if (!Ptr) return false;  // Not in folding set.
387 
388   --NumNodes;
389   N->SetNextInBucket(nullptr);
390 
391   // Remember what N originally pointed to, either a bucket or another node.
392   void *NodeNextPtr = Ptr;
393 
394   // Chase around the list until we find the node (or bucket) which points to N.
395   while (true) {
396     if (Node *NodeInBucket = GetNextPtr(Ptr)) {
397       // Advance pointer.
398       Ptr = NodeInBucket->getNextInBucket();
399 
400       // We found a node that points to N, change it to point to N's next node,
401       // removing N from the list.
402       if (Ptr == N) {
403         NodeInBucket->SetNextInBucket(NodeNextPtr);
404         return true;
405       }
406     } else {
407       void **Bucket = GetBucketPtr(Ptr);
408       Ptr = *Bucket;
409 
410       // If we found that the bucket points to N, update the bucket to point to
411       // whatever is next.
412       if (Ptr == N) {
413         *Bucket = NodeNextPtr;
414         return true;
415       }
416     }
417   }
418 }
419 
420 /// GetOrInsertNode - If there is an existing simple Node exactly
421 /// equal to the specified node, return it.  Otherwise, insert 'N' and it
422 /// instead.
423 FoldingSetBase::Node *
GetOrInsertNode(FoldingSetBase::Node * N,const FoldingSetInfo & Info)424 FoldingSetBase::GetOrInsertNode(FoldingSetBase::Node *N,
425                                 const FoldingSetInfo &Info) {
426   FoldingSetNodeID ID;
427   Info.GetNodeProfile(this, N, ID);
428   void *IP;
429   if (Node *E = FindNodeOrInsertPos(ID, IP, Info))
430     return E;
431   InsertNode(N, IP, Info);
432   return N;
433 }
434 
435 //===----------------------------------------------------------------------===//
436 // FoldingSetIteratorImpl Implementation
437 
FoldingSetIteratorImpl(void ** Bucket)438 FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) {
439   // Skip to the first non-null non-self-cycle bucket.
440   while (*Bucket != reinterpret_cast<void*>(-1) &&
441          (!*Bucket || !GetNextPtr(*Bucket)))
442     ++Bucket;
443 
444   NodePtr = static_cast<FoldingSetNode*>(*Bucket);
445 }
446 
advance()447 void FoldingSetIteratorImpl::advance() {
448   // If there is another link within this bucket, go to it.
449   void *Probe = NodePtr->getNextInBucket();
450 
451   if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe))
452     NodePtr = NextNodeInBucket;
453   else {
454     // Otherwise, this is the last link in this bucket.
455     void **Bucket = GetBucketPtr(Probe);
456 
457     // Skip to the next non-null non-self-cycle bucket.
458     do {
459       ++Bucket;
460     } while (*Bucket != reinterpret_cast<void*>(-1) &&
461              (!*Bucket || !GetNextPtr(*Bucket)));
462 
463     NodePtr = static_cast<FoldingSetNode*>(*Bucket);
464   }
465 }
466 
467 //===----------------------------------------------------------------------===//
468 // FoldingSetBucketIteratorImpl Implementation
469 
FoldingSetBucketIteratorImpl(void ** Bucket)470 FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) {
471   Ptr = (!*Bucket || !GetNextPtr(*Bucket)) ? (void*) Bucket : *Bucket;
472 }
473