Home | History | Annotate | Download | only in Analysis
      1 //===- llvm/Analysis/MemoryDependenceAnalysis.h - Memory Deps ---*- 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 defines the MemoryDependenceAnalysis analysis pass.
     11 //
     12 //===----------------------------------------------------------------------===//
     13 
     14 #ifndef LLVM_ANALYSIS_MEMORYDEPENDENCEANALYSIS_H
     15 #define LLVM_ANALYSIS_MEMORYDEPENDENCEANALYSIS_H
     16 
     17 #include "llvm/ADT/DenseMap.h"
     18 #include "llvm/ADT/Optional.h"
     19 #include "llvm/ADT/PointerEmbeddedInt.h"
     20 #include "llvm/ADT/PointerIntPair.h"
     21 #include "llvm/ADT/PointerSumType.h"
     22 #include "llvm/ADT/SmallPtrSet.h"
     23 #include "llvm/Analysis/AliasAnalysis.h"
     24 #include "llvm/Analysis/MemoryLocation.h"
     25 #include "llvm/IR/BasicBlock.h"
     26 #include "llvm/IR/Metadata.h"
     27 #include "llvm/IR/PassManager.h"
     28 #include "llvm/IR/PredIteratorCache.h"
     29 #include "llvm/Pass.h"
     30 #include "llvm/Support/ErrorHandling.h"
     31 #include <cassert>
     32 #include <cstdint>
     33 #include <utility>
     34 #include <vector>
     35 
     36 namespace llvm {
     37 
     38 class AssumptionCache;
     39 class CallSite;
     40 class DominatorTree;
     41 class Function;
     42 class Instruction;
     43 class LoadInst;
     44 class PHITransAddr;
     45 class TargetLibraryInfo;
     46 class Value;
     47 
     48 /// A memory dependence query can return one of three different answers.
     49 class MemDepResult {
     50   enum DepType {
     51     /// Clients of MemDep never see this.
     52     ///
     53     /// Entries with this marker occur in a LocalDeps map or NonLocalDeps map
     54     /// when the instruction they previously referenced was removed from
     55     /// MemDep.  In either case, the entry may include an instruction pointer.
     56     /// If so, the pointer is an instruction in the block where scanning can
     57     /// start from, saving some work.
     58     ///
     59     /// In a default-constructed MemDepResult object, the type will be Invalid
     60     /// and the instruction pointer will be null.
     61     Invalid = 0,
     62 
     63     /// This is a dependence on the specified instruction which clobbers the
     64     /// desired value.  The pointer member of the MemDepResult pair holds the
     65     /// instruction that clobbers the memory.  For example, this occurs when we
     66     /// see a may-aliased store to the memory location we care about.
     67     ///
     68     /// There are several cases that may be interesting here:
     69     ///   1. Loads are clobbered by may-alias stores.
     70     ///   2. Loads are considered clobbered by partially-aliased loads.  The
     71     ///      client may choose to analyze deeper into these cases.
     72     Clobber,
     73 
     74     /// This is a dependence on the specified instruction which defines or
     75     /// produces the desired memory location.  The pointer member of the
     76     /// MemDepResult pair holds the instruction that defines the memory.
     77     ///
     78     /// Cases of interest:
     79     ///   1. This could be a load or store for dependence queries on
     80     ///      load/store.  The value loaded or stored is the produced value.
     81     ///      Note that the pointer operand may be different than that of the
     82     ///      queried pointer due to must aliases and phi translation. Note
     83     ///      that the def may not be the same type as the query, the pointers
     84     ///      may just be must aliases.
     85     ///   2. For loads and stores, this could be an allocation instruction. In
     86     ///      this case, the load is loading an undef value or a store is the
     87     ///      first store to (that part of) the allocation.
     88     ///   3. Dependence queries on calls return Def only when they are readonly
     89     ///      calls or memory use intrinsics with identical callees and no
     90     ///      intervening clobbers.  No validation is done that the operands to
     91     ///      the calls are the same.
     92     Def,
     93 
     94     /// This marker indicates that the query has no known dependency in the
     95     /// specified block.
     96     ///
     97     /// More detailed state info is encoded in the upper part of the pair (i.e.
     98     /// the Instruction*)
     99     Other
    100   };
    101 
    102   /// If DepType is "Other", the upper part of the sum type is an encoding of
    103   /// the following more detailed type information.
    104   enum OtherType {
    105     /// This marker indicates that the query has no dependency in the specified
    106     /// block.
    107     ///
    108     /// To find out more, the client should query other predecessor blocks.
    109     NonLocal = 1,
    110     /// This marker indicates that the query has no dependency in the specified
    111     /// function.
    112     NonFuncLocal,
    113     /// This marker indicates that the query dependency is unknown.
    114     Unknown
    115   };
    116 
    117   using ValueTy = PointerSumType<
    118       DepType, PointerSumTypeMember<Invalid, Instruction *>,
    119       PointerSumTypeMember<Clobber, Instruction *>,
    120       PointerSumTypeMember<Def, Instruction *>,
    121       PointerSumTypeMember<Other, PointerEmbeddedInt<OtherType, 3>>>;
    122   ValueTy Value;
    123 
    124   explicit MemDepResult(ValueTy V) : Value(V) {}
    125 
    126 public:
    127   MemDepResult() = default;
    128 
    129   /// get methods: These are static ctor methods for creating various
    130   /// MemDepResult kinds.
    131   static MemDepResult getDef(Instruction *Inst) {
    132     assert(Inst && "Def requires inst");
    133     return MemDepResult(ValueTy::create<Def>(Inst));
    134   }
    135   static MemDepResult getClobber(Instruction *Inst) {
    136     assert(Inst && "Clobber requires inst");
    137     return MemDepResult(ValueTy::create<Clobber>(Inst));
    138   }
    139   static MemDepResult getNonLocal() {
    140     return MemDepResult(ValueTy::create<Other>(NonLocal));
    141   }
    142   static MemDepResult getNonFuncLocal() {
    143     return MemDepResult(ValueTy::create<Other>(NonFuncLocal));
    144   }
    145   static MemDepResult getUnknown() {
    146     return MemDepResult(ValueTy::create<Other>(Unknown));
    147   }
    148 
    149   /// Tests if this MemDepResult represents a query that is an instruction
    150   /// clobber dependency.
    151   bool isClobber() const { return Value.is<Clobber>(); }
    152 
    153   /// Tests if this MemDepResult represents a query that is an instruction
    154   /// definition dependency.
    155   bool isDef() const { return Value.is<Def>(); }
    156 
    157   /// Tests if this MemDepResult represents a query that is transparent to the
    158   /// start of the block, but where a non-local hasn't been done.
    159   bool isNonLocal() const {
    160     return Value.is<Other>() && Value.cast<Other>() == NonLocal;
    161   }
    162 
    163   /// Tests if this MemDepResult represents a query that is transparent to the
    164   /// start of the function.
    165   bool isNonFuncLocal() const {
    166     return Value.is<Other>() && Value.cast<Other>() == NonFuncLocal;
    167   }
    168 
    169   /// Tests if this MemDepResult represents a query which cannot and/or will
    170   /// not be computed.
    171   bool isUnknown() const {
    172     return Value.is<Other>() && Value.cast<Other>() == Unknown;
    173   }
    174 
    175   /// If this is a normal dependency, returns the instruction that is depended
    176   /// on.  Otherwise, returns null.
    177   Instruction *getInst() const {
    178     switch (Value.getTag()) {
    179     case Invalid:
    180       return Value.cast<Invalid>();
    181     case Clobber:
    182       return Value.cast<Clobber>();
    183     case Def:
    184       return Value.cast<Def>();
    185     case Other:
    186       return nullptr;
    187     }
    188     llvm_unreachable("Unknown discriminant!");
    189   }
    190 
    191   bool operator==(const MemDepResult &M) const { return Value == M.Value; }
    192   bool operator!=(const MemDepResult &M) const { return Value != M.Value; }
    193   bool operator<(const MemDepResult &M) const { return Value < M.Value; }
    194   bool operator>(const MemDepResult &M) const { return Value > M.Value; }
    195 
    196 private:
    197   friend class MemoryDependenceResults;
    198 
    199   /// Tests if this is a MemDepResult in its dirty/invalid. state.
    200   bool isDirty() const { return Value.is<Invalid>(); }
    201 
    202   static MemDepResult getDirty(Instruction *Inst) {
    203     return MemDepResult(ValueTy::create<Invalid>(Inst));
    204   }
    205 };
    206 
    207 /// This is an entry in the NonLocalDepInfo cache.
    208 ///
    209 /// For each BasicBlock (the BB entry) it keeps a MemDepResult.
    210 class NonLocalDepEntry {
    211   BasicBlock *BB;
    212   MemDepResult Result;
    213 
    214 public:
    215   NonLocalDepEntry(BasicBlock *bb, MemDepResult result)
    216       : BB(bb), Result(result) {}
    217 
    218   // This is used for searches.
    219   NonLocalDepEntry(BasicBlock *bb) : BB(bb) {}
    220 
    221   // BB is the sort key, it can't be changed.
    222   BasicBlock *getBB() const { return BB; }
    223 
    224   void setResult(const MemDepResult &R) { Result = R; }
    225 
    226   const MemDepResult &getResult() const { return Result; }
    227 
    228   bool operator<(const NonLocalDepEntry &RHS) const { return BB < RHS.BB; }
    229 };
    230 
    231 /// This is a result from a NonLocal dependence query.
    232 ///
    233 /// For each BasicBlock (the BB entry) it keeps a MemDepResult and the
    234 /// (potentially phi translated) address that was live in the block.
    235 class NonLocalDepResult {
    236   NonLocalDepEntry Entry;
    237   Value *Address;
    238 
    239 public:
    240   NonLocalDepResult(BasicBlock *bb, MemDepResult result, Value *address)
    241       : Entry(bb, result), Address(address) {}
    242 
    243   // BB is the sort key, it can't be changed.
    244   BasicBlock *getBB() const { return Entry.getBB(); }
    245 
    246   void setResult(const MemDepResult &R, Value *Addr) {
    247     Entry.setResult(R);
    248     Address = Addr;
    249   }
    250 
    251   const MemDepResult &getResult() const { return Entry.getResult(); }
    252 
    253   /// Returns the address of this pointer in this block.
    254   ///
    255   /// This can be different than the address queried for the non-local result
    256   /// because of phi translation.  This returns null if the address was not
    257   /// available in a block (i.e. because phi translation failed) or if this is
    258   /// a cached result and that address was deleted.
    259   ///
    260   /// The address is always null for a non-local 'call' dependence.
    261   Value *getAddress() const { return Address; }
    262 };
    263 
    264 /// Provides a lazy, caching interface for making common memory aliasing
    265 /// information queries, backed by LLVM's alias analysis passes.
    266 ///
    267 /// The dependency information returned is somewhat unusual, but is pragmatic.
    268 /// If queried about a store or call that might modify memory, the analysis
    269 /// will return the instruction[s] that may either load from that memory or
    270 /// store to it.  If queried with a load or call that can never modify memory,
    271 /// the analysis will return calls and stores that might modify the pointer,
    272 /// but generally does not return loads unless a) they are volatile, or
    273 /// b) they load from *must-aliased* pointers.  Returning a dependence on
    274 /// must-alias'd pointers instead of all pointers interacts well with the
    275 /// internal caching mechanism.
    276 class MemoryDependenceResults {
    277   // A map from instructions to their dependency.
    278   using LocalDepMapType = DenseMap<Instruction *, MemDepResult>;
    279   LocalDepMapType LocalDeps;
    280 
    281 public:
    282   using NonLocalDepInfo = std::vector<NonLocalDepEntry>;
    283 
    284 private:
    285   /// A pair<Value*, bool> where the bool is true if the dependence is a read
    286   /// only dependence, false if read/write.
    287   using ValueIsLoadPair = PointerIntPair<const Value *, 1, bool>;
    288 
    289   /// This pair is used when caching information for a block.
    290   ///
    291   /// If the pointer is null, the cache value is not a full query that starts
    292   /// at the specified block.  If non-null, the bool indicates whether or not
    293   /// the contents of the block was skipped.
    294   using BBSkipFirstBlockPair = PointerIntPair<BasicBlock *, 1, bool>;
    295 
    296   /// This record is the information kept for each (value, is load) pair.
    297   struct NonLocalPointerInfo {
    298     /// The pair of the block and the skip-first-block flag.
    299     BBSkipFirstBlockPair Pair;
    300     /// The results of the query for each relevant block.
    301     NonLocalDepInfo NonLocalDeps;
    302     /// The maximum size of the dereferences of the pointer.
    303     ///
    304     /// May be UnknownSize if the sizes are unknown.
    305     uint64_t Size = MemoryLocation::UnknownSize;
    306     /// The AA tags associated with dereferences of the pointer.
    307     ///
    308     /// The members may be null if there are no tags or conflicting tags.
    309     AAMDNodes AATags;
    310 
    311     NonLocalPointerInfo() = default;
    312   };
    313 
    314   /// Cache storing single nonlocal def for the instruction.
    315   /// It is set when nonlocal def would be found in function returning only
    316   /// local dependencies.
    317   DenseMap<Instruction *, NonLocalDepResult> NonLocalDefsCache;
    318 
    319   /// This map stores the cached results of doing a pointer lookup at the
    320   /// bottom of a block.
    321   ///
    322   /// The key of this map is the pointer+isload bit, the value is a list of
    323   /// <bb->result> mappings.
    324   using CachedNonLocalPointerInfo =
    325       DenseMap<ValueIsLoadPair, NonLocalPointerInfo>;
    326   CachedNonLocalPointerInfo NonLocalPointerDeps;
    327 
    328   // A map from instructions to their non-local pointer dependencies.
    329   using ReverseNonLocalPtrDepTy =
    330       DenseMap<Instruction *, SmallPtrSet<ValueIsLoadPair, 4>>;
    331   ReverseNonLocalPtrDepTy ReverseNonLocalPtrDeps;
    332 
    333   /// This is the instruction we keep for each cached access that we have for
    334   /// an instruction.
    335   ///
    336   /// The pointer is an owning pointer and the bool indicates whether we have
    337   /// any dirty bits in the set.
    338   using PerInstNLInfo = std::pair<NonLocalDepInfo, bool>;
    339 
    340   // A map from instructions to their non-local dependencies.
    341   using NonLocalDepMapType = DenseMap<Instruction *, PerInstNLInfo>;
    342 
    343   NonLocalDepMapType NonLocalDeps;
    344 
    345   // A reverse mapping from dependencies to the dependees.  This is
    346   // used when removing instructions to keep the cache coherent.
    347   using ReverseDepMapType =
    348       DenseMap<Instruction *, SmallPtrSet<Instruction *, 4>>;
    349   ReverseDepMapType ReverseLocalDeps;
    350 
    351   // A reverse mapping from dependencies to the non-local dependees.
    352   ReverseDepMapType ReverseNonLocalDeps;
    353 
    354   /// Current AA implementation, just a cache.
    355   AliasAnalysis &AA;
    356   AssumptionCache &AC;
    357   const TargetLibraryInfo &TLI;
    358   DominatorTree &DT;
    359   PredIteratorCache PredCache;
    360 
    361 public:
    362   MemoryDependenceResults(AliasAnalysis &AA, AssumptionCache &AC,
    363                           const TargetLibraryInfo &TLI,
    364                           DominatorTree &DT)
    365       : AA(AA), AC(AC), TLI(TLI), DT(DT) {}
    366 
    367   /// Handle invalidation in the new PM.
    368   bool invalidate(Function &F, const PreservedAnalyses &PA,
    369                   FunctionAnalysisManager::Invalidator &Inv);
    370 
    371   /// Some methods limit the number of instructions they will examine.
    372   /// The return value of this method is the default limit that will be
    373   /// used if no limit is explicitly passed in.
    374   unsigned getDefaultBlockScanLimit() const;
    375 
    376   /// Returns the instruction on which a memory operation depends.
    377   ///
    378   /// See the class comment for more details. It is illegal to call this on
    379   /// non-memory instructions.
    380   MemDepResult getDependency(Instruction *QueryInst);
    381 
    382   /// Perform a full dependency query for the specified call, returning the set
    383   /// of blocks that the value is potentially live across.
    384   ///
    385   /// The returned set of results will include a "NonLocal" result for all
    386   /// blocks where the value is live across.
    387   ///
    388   /// This method assumes the instruction returns a "NonLocal" dependency
    389   /// within its own block.
    390   ///
    391   /// This returns a reference to an internal data structure that may be
    392   /// invalidated on the next non-local query or when an instruction is
    393   /// removed.  Clients must copy this data if they want it around longer than
    394   /// that.
    395   const NonLocalDepInfo &getNonLocalCallDependency(CallSite QueryCS);
    396 
    397   /// Perform a full dependency query for an access to the QueryInst's
    398   /// specified memory location, returning the set of instructions that either
    399   /// define or clobber the value.
    400   ///
    401   /// Warning: For a volatile query instruction, the dependencies will be
    402   /// accurate, and thus usable for reordering, but it is never legal to
    403   /// remove the query instruction.
    404   ///
    405   /// This method assumes the pointer has a "NonLocal" dependency within
    406   /// QueryInst's parent basic block.
    407   void getNonLocalPointerDependency(Instruction *QueryInst,
    408                                     SmallVectorImpl<NonLocalDepResult> &Result);
    409 
    410   /// Removes an instruction from the dependence analysis, updating the
    411   /// dependence of instructions that previously depended on it.
    412   void removeInstruction(Instruction *InstToRemove);
    413 
    414   /// Invalidates cached information about the specified pointer, because it
    415   /// may be too conservative in memdep.
    416   ///
    417   /// This is an optional call that can be used when the client detects an
    418   /// equivalence between the pointer and some other value and replaces the
    419   /// other value with ptr. This can make Ptr available in more places that
    420   /// cached info does not necessarily keep.
    421   void invalidateCachedPointerInfo(Value *Ptr);
    422 
    423   /// Clears the PredIteratorCache info.
    424   ///
    425   /// This needs to be done when the CFG changes, e.g., due to splitting
    426   /// critical edges.
    427   void invalidateCachedPredecessors();
    428 
    429   /// Returns the instruction on which a memory location depends.
    430   ///
    431   /// If isLoad is true, this routine ignores may-aliases with read-only
    432   /// operations.  If isLoad is false, this routine ignores may-aliases
    433   /// with reads from read-only locations. If possible, pass the query
    434   /// instruction as well; this function may take advantage of the metadata
    435   /// annotated to the query instruction to refine the result. \p Limit
    436   /// can be used to set the maximum number of instructions that will be
    437   /// examined to find the pointer dependency. On return, it will be set to
    438   /// the number of instructions left to examine. If a null pointer is passed
    439   /// in, the limit will default to the value of -memdep-block-scan-limit.
    440   ///
    441   /// Note that this is an uncached query, and thus may be inefficient.
    442   MemDepResult getPointerDependencyFrom(const MemoryLocation &Loc, bool isLoad,
    443                                         BasicBlock::iterator ScanIt,
    444                                         BasicBlock *BB,
    445                                         Instruction *QueryInst = nullptr,
    446                                         unsigned *Limit = nullptr);
    447 
    448   MemDepResult getSimplePointerDependencyFrom(const MemoryLocation &MemLoc,
    449                                               bool isLoad,
    450                                               BasicBlock::iterator ScanIt,
    451                                               BasicBlock *BB,
    452                                               Instruction *QueryInst,
    453                                               unsigned *Limit = nullptr);
    454 
    455   /// This analysis looks for other loads and stores with invariant.group
    456   /// metadata and the same pointer operand. Returns Unknown if it does not
    457   /// find anything, and Def if it can be assumed that 2 instructions load or
    458   /// store the same value and NonLocal which indicate that non-local Def was
    459   /// found, which can be retrieved by calling getNonLocalPointerDependency
    460   /// with the same queried instruction.
    461   MemDepResult getInvariantGroupPointerDependency(LoadInst *LI, BasicBlock *BB);
    462 
    463   /// Looks at a memory location for a load (specified by MemLocBase, Offs, and
    464   /// Size) and compares it against a load.
    465   ///
    466   /// If the specified load could be safely widened to a larger integer load
    467   /// that is 1) still efficient, 2) safe for the target, and 3) would provide
    468   /// the specified memory location value, then this function returns the size
    469   /// in bytes of the load width to use.  If not, this returns zero.
    470   static unsigned getLoadLoadClobberFullWidthSize(const Value *MemLocBase,
    471                                                   int64_t MemLocOffs,
    472                                                   unsigned MemLocSize,
    473                                                   const LoadInst *LI);
    474 
    475   /// Release memory in caches.
    476   void releaseMemory();
    477 
    478 private:
    479   MemDepResult getCallSiteDependencyFrom(CallSite C, bool isReadOnlyCall,
    480                                          BasicBlock::iterator ScanIt,
    481                                          BasicBlock *BB);
    482   bool getNonLocalPointerDepFromBB(Instruction *QueryInst,
    483                                    const PHITransAddr &Pointer,
    484                                    const MemoryLocation &Loc, bool isLoad,
    485                                    BasicBlock *BB,
    486                                    SmallVectorImpl<NonLocalDepResult> &Result,
    487                                    DenseMap<BasicBlock *, Value *> &Visited,
    488                                    bool SkipFirstBlock = false);
    489   MemDepResult GetNonLocalInfoForBlock(Instruction *QueryInst,
    490                                        const MemoryLocation &Loc, bool isLoad,
    491                                        BasicBlock *BB, NonLocalDepInfo *Cache,
    492                                        unsigned NumSortedEntries);
    493 
    494   void RemoveCachedNonLocalPointerDependencies(ValueIsLoadPair P);
    495 
    496   void verifyRemoved(Instruction *Inst) const;
    497 };
    498 
    499 /// An analysis that produces \c MemoryDependenceResults for a function.
    500 ///
    501 /// This is essentially a no-op because the results are computed entirely
    502 /// lazily.
    503 class MemoryDependenceAnalysis
    504     : public AnalysisInfoMixin<MemoryDependenceAnalysis> {
    505   friend AnalysisInfoMixin<MemoryDependenceAnalysis>;
    506 
    507   static AnalysisKey Key;
    508 
    509 public:
    510   using Result = MemoryDependenceResults;
    511 
    512   MemoryDependenceResults run(Function &F, FunctionAnalysisManager &AM);
    513 };
    514 
    515 /// A wrapper analysis pass for the legacy pass manager that exposes a \c
    516 /// MemoryDepnedenceResults instance.
    517 class MemoryDependenceWrapperPass : public FunctionPass {
    518   Optional<MemoryDependenceResults> MemDep;
    519 
    520 public:
    521   static char ID;
    522 
    523   MemoryDependenceWrapperPass();
    524   ~MemoryDependenceWrapperPass() override;
    525 
    526   /// Pass Implementation stuff.  This doesn't do any analysis eagerly.
    527   bool runOnFunction(Function &) override;
    528 
    529   /// Clean up memory in between runs
    530   void releaseMemory() override;
    531 
    532   /// Does not modify anything.  It uses Value Numbering and Alias Analysis.
    533   void getAnalysisUsage(AnalysisUsage &AU) const override;
    534 
    535   MemoryDependenceResults &getMemDep() { return *MemDep; }
    536 };
    537 
    538 } // end namespace llvm
    539 
    540 #endif // LLVM_ANALYSIS_MEMORYDEPENDENCEANALYSIS_H
    541