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 FoldingSetImpl.
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   Bits.append(reinterpret_cast<unsigned *>(&Ptr),
58               reinterpret_cast<unsigned *>(&Ptr+1));
59 }
AddInteger(signed I)60 void FoldingSetNodeID::AddInteger(signed I) {
61   Bits.push_back(I);
62 }
AddInteger(unsigned I)63 void FoldingSetNodeID::AddInteger(unsigned I) {
64   Bits.push_back(I);
65 }
AddInteger(long I)66 void FoldingSetNodeID::AddInteger(long I) {
67   AddInteger((unsigned long)I);
68 }
AddInteger(unsigned long I)69 void 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 }
AddInteger(long long I)78 void FoldingSetNodeID::AddInteger(long long I) {
79   AddInteger((unsigned long long)I);
80 }
AddInteger(unsigned long long I)81 void 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 
AddString(StringRef String)87 void 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.
AddNodeID(const FoldingSetNodeID & ID)140 void 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.
ComputeHash() const146 unsigned FoldingSetNodeID::ComputeHash() const {
147   return FoldingSetNodeIDRef(Bits.data(), Bits.size()).ComputeHash();
148 }
149 
150 /// operator== - Used to compare two nodes to each other.
151 ///
operator ==(const FoldingSetNodeID & RHS) const152 bool 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 ///
operator ==(FoldingSetNodeIDRef RHS) const158 bool 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().
operator <(const FoldingSetNodeID & RHS) const164 bool FoldingSetNodeID::operator<(const FoldingSetNodeID &RHS) const {
165   return *this < FoldingSetNodeIDRef(RHS.Bits.data(), RHS.Bits.size());
166 }
167 
operator <(FoldingSetNodeIDRef RHS) const168 bool 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.
175 FoldingSetNodeIDRef
Intern(BumpPtrAllocator & Allocator) const176 FoldingSetNodeID::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.
GetNextPtr(void * NextInBucketPtr)191 static 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.
GetBucketPtr(void * NextInBucketPtr)201 static 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.
GetBucketFor(unsigned Hash,void ** Buckets,unsigned NumBuckets)209 static 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.
AllocateBuckets(unsigned NumBuckets)216 static 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 
anchor()226 void FoldingSetImpl::anchor() {}
227 
FoldingSetImpl(unsigned Log2InitSize)228 FoldingSetImpl::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 }
~FoldingSetImpl()235 FoldingSetImpl::~FoldingSetImpl() {
236   free(Buckets);
237 }
clear()238 void FoldingSetImpl::clear() {
239   // Set all but the last bucket to null pointers.
240   memset(Buckets, 0, NumBuckets*sizeof(void*));
241 
242   // Set the very last bucket to be a non-null "pointer".
243   Buckets[NumBuckets] = reinterpret_cast<void*>(-1);
244 
245   // Reset the node count to zero.
246   NumNodes = 0;
247 }
248 
249 /// GrowHashTable - Double the size of the hash table and rehash everything.
250 ///
GrowHashTable()251 void FoldingSetImpl::GrowHashTable() {
252   void **OldBuckets = Buckets;
253   unsigned OldNumBuckets = NumBuckets;
254   NumBuckets <<= 1;
255 
256   // Clear out new buckets.
257   Buckets = AllocateBuckets(NumBuckets);
258   NumNodes = 0;
259 
260   // Walk the old buckets, rehashing nodes into their new place.
261   FoldingSetNodeID TempID;
262   for (unsigned i = 0; i != OldNumBuckets; ++i) {
263     void *Probe = OldBuckets[i];
264     if (!Probe) continue;
265     while (Node *NodeInBucket = GetNextPtr(Probe)) {
266       // Figure out the next link, remove NodeInBucket from the old link.
267       Probe = NodeInBucket->getNextInBucket();
268       NodeInBucket->SetNextInBucket(nullptr);
269 
270       // Insert the node into the new bucket, after recomputing the hash.
271       InsertNode(NodeInBucket,
272                  GetBucketFor(ComputeNodeHash(NodeInBucket, TempID),
273                               Buckets, NumBuckets));
274       TempID.clear();
275     }
276   }
277 
278   free(OldBuckets);
279 }
280 
281 /// FindNodeOrInsertPos - Look up the node specified by ID.  If it exists,
282 /// return it.  If not, return the insertion token that will make insertion
283 /// faster.
284 FoldingSetImpl::Node
FindNodeOrInsertPos(const FoldingSetNodeID & ID,void * & InsertPos)285 *FoldingSetImpl::FindNodeOrInsertPos(const FoldingSetNodeID &ID,
286                                      void *&InsertPos) {
287   unsigned IDHash = ID.ComputeHash();
288   void **Bucket = GetBucketFor(IDHash, Buckets, NumBuckets);
289   void *Probe = *Bucket;
290 
291   InsertPos = nullptr;
292 
293   FoldingSetNodeID TempID;
294   while (Node *NodeInBucket = GetNextPtr(Probe)) {
295     if (NodeEquals(NodeInBucket, ID, IDHash, TempID))
296       return NodeInBucket;
297     TempID.clear();
298 
299     Probe = NodeInBucket->getNextInBucket();
300   }
301 
302   // Didn't find the node, return null with the bucket as the InsertPos.
303   InsertPos = Bucket;
304   return nullptr;
305 }
306 
307 /// InsertNode - Insert the specified node into the folding set, knowing that it
308 /// is not already in the map.  InsertPos must be obtained from
309 /// FindNodeOrInsertPos.
InsertNode(Node * N,void * InsertPos)310 void FoldingSetImpl::InsertNode(Node *N, void *InsertPos) {
311   assert(!N->getNextInBucket());
312   // Do we need to grow the hashtable?
313   if (NumNodes+1 > NumBuckets*2) {
314     GrowHashTable();
315     FoldingSetNodeID TempID;
316     InsertPos = GetBucketFor(ComputeNodeHash(N, TempID), Buckets, NumBuckets);
317   }
318 
319   ++NumNodes;
320 
321   /// The insert position is actually a bucket pointer.
322   void **Bucket = static_cast<void**>(InsertPos);
323 
324   void *Next = *Bucket;
325 
326   // If this is the first insertion into this bucket, its next pointer will be
327   // null.  Pretend as if it pointed to itself, setting the low bit to indicate
328   // that it is a pointer to the bucket.
329   if (!Next)
330     Next = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(Bucket)|1);
331 
332   // Set the node's next pointer, and make the bucket point to the node.
333   N->SetNextInBucket(Next);
334   *Bucket = N;
335 }
336 
337 /// RemoveNode - Remove a node from the folding set, returning true if one was
338 /// removed or false if the node was not in the folding set.
RemoveNode(Node * N)339 bool FoldingSetImpl::RemoveNode(Node *N) {
340   // Because each bucket is a circular list, we don't need to compute N's hash
341   // to remove it.
342   void *Ptr = N->getNextInBucket();
343   if (!Ptr) return false;  // Not in folding set.
344 
345   --NumNodes;
346   N->SetNextInBucket(nullptr);
347 
348   // Remember what N originally pointed to, either a bucket or another node.
349   void *NodeNextPtr = Ptr;
350 
351   // Chase around the list until we find the node (or bucket) which points to N.
352   while (true) {
353     if (Node *NodeInBucket = GetNextPtr(Ptr)) {
354       // Advance pointer.
355       Ptr = NodeInBucket->getNextInBucket();
356 
357       // We found a node that points to N, change it to point to N's next node,
358       // removing N from the list.
359       if (Ptr == N) {
360         NodeInBucket->SetNextInBucket(NodeNextPtr);
361         return true;
362       }
363     } else {
364       void **Bucket = GetBucketPtr(Ptr);
365       Ptr = *Bucket;
366 
367       // If we found that the bucket points to N, update the bucket to point to
368       // whatever is next.
369       if (Ptr == N) {
370         *Bucket = NodeNextPtr;
371         return true;
372       }
373     }
374   }
375 }
376 
377 /// GetOrInsertNode - If there is an existing simple Node exactly
378 /// equal to the specified node, return it.  Otherwise, insert 'N' and it
379 /// instead.
GetOrInsertNode(FoldingSetImpl::Node * N)380 FoldingSetImpl::Node *FoldingSetImpl::GetOrInsertNode(FoldingSetImpl::Node *N) {
381   FoldingSetNodeID ID;
382   GetNodeProfile(N, ID);
383   void *IP;
384   if (Node *E = FindNodeOrInsertPos(ID, IP))
385     return E;
386   InsertNode(N, IP);
387   return N;
388 }
389 
390 //===----------------------------------------------------------------------===//
391 // FoldingSetIteratorImpl Implementation
392 
FoldingSetIteratorImpl(void ** Bucket)393 FoldingSetIteratorImpl::FoldingSetIteratorImpl(void **Bucket) {
394   // Skip to the first non-null non-self-cycle bucket.
395   while (*Bucket != reinterpret_cast<void*>(-1) &&
396          (!*Bucket || !GetNextPtr(*Bucket)))
397     ++Bucket;
398 
399   NodePtr = static_cast<FoldingSetNode*>(*Bucket);
400 }
401 
advance()402 void FoldingSetIteratorImpl::advance() {
403   // If there is another link within this bucket, go to it.
404   void *Probe = NodePtr->getNextInBucket();
405 
406   if (FoldingSetNode *NextNodeInBucket = GetNextPtr(Probe))
407     NodePtr = NextNodeInBucket;
408   else {
409     // Otherwise, this is the last link in this bucket.
410     void **Bucket = GetBucketPtr(Probe);
411 
412     // Skip to the next non-null non-self-cycle bucket.
413     do {
414       ++Bucket;
415     } while (*Bucket != reinterpret_cast<void*>(-1) &&
416              (!*Bucket || !GetNextPtr(*Bucket)));
417 
418     NodePtr = static_cast<FoldingSetNode*>(*Bucket);
419   }
420 }
421 
422 //===----------------------------------------------------------------------===//
423 // FoldingSetBucketIteratorImpl Implementation
424 
FoldingSetBucketIteratorImpl(void ** Bucket)425 FoldingSetBucketIteratorImpl::FoldingSetBucketIteratorImpl(void **Bucket) {
426   Ptr = (!*Bucket || !GetNextPtr(*Bucket)) ? (void*) Bucket : *Bucket;
427 }
428