include/llvm/Transforms/Scalar/SROA.h

   1 //===- SROA.h - Scalar Replacement Of Aggregates ----------------*- 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 /// \file
   9 /// This file provides the interface for LLVM's Scalar Replacement of
  10 /// Aggregates pass. This pass provides both aggregate splitting and the
  11 /// primary SSA formation used in the compiler.
  12 ///
  13 //===----------------------------------------------------------------------===//
  14
  15 #ifndef LLVM_TRANSFORMS_SCALAR_SROA_H
  16 #define LLVM_TRANSFORMS_SCALAR_SROA_H
  17
  18 #include "llvm/ADT/SetVector.h"
  19 #include "llvm/ADT/SmallVector.h"
  20 #include "llvm/IR/PassManager.h"
  21 #include <vector>
  22
  23 namespace llvm {
  24
  25 class AllocaInst;
  26 class AssumptionCache;
  27 class DominatorTree;
  28 class Function;
  29 class Instruction;
  30 class LLVMContext;
  31 class PHINode;
  32 class SelectInst;
  33 class Use;
  34
  35 /// A private "module" namespace for types and utilities used by SROA. These
  36 /// are implementation details and should not be used by clients.
  37 namespace sroa LLVM_LIBRARY_VISIBILITY {
  38
  39 class AllocaSliceRewriter;
  40 class AllocaSlices;
  41 class Partition;
  42 class SROALegacyPass;
  43
  44 } // end namespace sroa
  45
  46 /// An optimization pass providing Scalar Replacement of Aggregates.
  47 ///
  48 /// This pass takes allocations which can be completely analyzed (that is, they
  49 /// don't escape) and tries to turn them into scalar SSA values. There are
  50 /// a few steps to this process.
  51 ///
  52 /// 1) It takes allocations of aggregates and analyzes the ways in which they
  53 ///    are used to try to split them into smaller allocations, ideally of
  54 ///    a single scalar data type. It will split up memcpy and memset accesses
  55 ///    as necessary and try to isolate individual scalar accesses.
  56 /// 2) It will transform accesses into forms which are suitable for SSA value
  57 ///    promotion. This can be replacing a memset with a scalar store of an
  58 ///    integer value, or it can involve speculating operations on a PHI or
  59 ///    select to be a PHI or select of the results.
  60 /// 3) Finally, this will try to detect a pattern of accesses which map cleanly
  61 ///    onto insert and extract operations on a vector value, and convert them to
  62 ///    this form. By doing so, it will enable promotion of vector aggregates to
  63 ///    SSA vector values.
  64 class SROA : public PassInfoMixin<SROA> {
  65   LLVMContext *C = nullptr;
  66   DominatorTree *DT = nullptr;
  67   AssumptionCache *AC = nullptr;
  68
  69   /// Worklist of alloca instructions to simplify.
  70   ///
  71   /// Each alloca in the function is added to this. Each new alloca formed gets
  72   /// added to it as well to recursively simplify unless that alloca can be
  73   /// directly promoted. Finally, each time we rewrite a use of an alloca other
  74   /// the one being actively rewritten, we add it back onto the list if not
  75   /// already present to ensure it is re-visited.
  76   SetVector<AllocaInst *, SmallVector<AllocaInst *, 16>> Worklist;
  77
  78   /// A collection of instructions to delete.
  79   /// We try to batch deletions to simplify code and make things a bit more
  80   /// efficient.
  81   SetVector<Instruction *, SmallVector<Instruction *, 8>> DeadInsts;
  82
  83   /// Post-promotion worklist.
  84   ///
  85   /// Sometimes we discover an alloca which has a high probability of becoming
  86   /// viable for SROA after a round of promotion takes place. In those cases,
  87   /// the alloca is enqueued here for re-processing.
  88   ///
  89   /// Note that we have to be very careful to clear allocas out of this list in
  90   /// the event they are deleted.
  91   SetVector<AllocaInst *, SmallVector<AllocaInst *, 16>> PostPromotionWorklist;
  92
  93   /// A collection of alloca instructions we can directly promote.
  94   std::vector<AllocaInst *> PromotableAllocas;
  95
  96   /// A worklist of PHIs to speculate prior to promoting allocas.
  97   ///
  98   /// All of these PHIs have been checked for the safety of speculation and by
  99   /// being speculated will allow promoting allocas currently in the promotable
 100   /// queue.
 101   SetVector<PHINode *, SmallVector<PHINode *, 2>> SpeculatablePHIs;
 102
 103   /// A worklist of select instructions to speculate prior to promoting
 104   /// allocas.
 105   ///
 106   /// All of these select instructions have been checked for the safety of
 107   /// speculation and by being speculated will allow promoting allocas
 108   /// currently in the promotable queue.
 109   SetVector<SelectInst *, SmallVector<SelectInst *, 2>> SpeculatableSelects;
 110
 111 public:
 112   SROA() = default;
 113
 114   /// Run the pass over the function.
 115   PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM);
 116
 117 private:
 118   friend class sroa::AllocaSliceRewriter;
 119   friend class sroa::SROALegacyPass;
 120
 121   /// Helper used by both the public run method and by the legacy pass.
 122   PreservedAnalyses runImpl(Function &F, DominatorTree &RunDT,
 123                             AssumptionCache &RunAC);
 124
 125   bool presplitLoadsAndStores(AllocaInst &AI, sroa::AllocaSlices &AS);
 126   AllocaInst *rewritePartition(AllocaInst &AI, sroa::AllocaSlices &AS,
 127                                sroa::Partition &P);
 128   bool splitAlloca(AllocaInst &AI, sroa::AllocaSlices &AS);
 129   bool runOnAlloca(AllocaInst &AI);
 130   void clobberUse(Use &U);
 131   bool deleteDeadInstructions(SmallPtrSetImpl<AllocaInst *> &DeletedAllocas);
 132   bool promoteAllocas(Function &F);
 133 };
 134
 135 } // end namespace llvm
 136
 137 #endif // LLVM_TRANSFORMS_SCALAR_SROA_H