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