Details | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 14 | pmbaty | 1 | //==------ llvm/CodeGen/LoopTraversal.h - Loop Traversal -*- 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 | /// \file Loop Traversal logic. |
||
| 10 | /// |
||
| 11 | /// This class provides the basic blocks traversal order used by passes like |
||
| 12 | /// ReachingDefAnalysis and ExecutionDomainFix. |
||
| 13 | /// It identifies basic blocks that are part of loops and should to be visited |
||
| 14 | /// twice and returns efficient traversal order for all the blocks. |
||
| 15 | // |
||
| 16 | //===----------------------------------------------------------------------===// |
||
| 17 | |||
| 18 | #ifndef LLVM_CODEGEN_LOOPTRAVERSAL_H |
||
| 19 | #define LLVM_CODEGEN_LOOPTRAVERSAL_H |
||
| 20 | |||
| 21 | #include "llvm/ADT/SmallVector.h" |
||
| 22 | |||
| 23 | namespace llvm { |
||
| 24 | |||
| 25 | class MachineBasicBlock; |
||
| 26 | class MachineFunction; |
||
| 27 | |||
| 28 | /// This class provides the basic blocks traversal order used by passes like |
||
| 29 | /// ReachingDefAnalysis and ExecutionDomainFix. |
||
| 30 | /// It identifies basic blocks that are part of loops and should to be visited |
||
| 31 | /// twice and returns efficient traversal order for all the blocks. |
||
| 32 | /// |
||
| 33 | /// We want to visit every instruction in every basic block in order to update |
||
| 34 | /// it's execution domain or collect clearance information. However, for the |
||
| 35 | /// clearance calculation, we need to know clearances from all predecessors |
||
| 36 | /// (including any backedges), therfore we need to visit some blocks twice. |
||
| 37 | /// As an example, consider the following loop. |
||
| 38 | /// |
||
| 39 | /// |
||
| 40 | /// PH -> A -> B (xmm<Undef> -> xmm<Def>) -> C -> D -> EXIT |
||
| 41 | /// ^ | |
||
| 42 | /// +----------------------------------+ |
||
| 43 | /// |
||
| 44 | /// The iteration order this pass will return is as follows: |
||
| 45 | /// Optimized: PH A B C A' B' C' D |
||
| 46 | /// |
||
| 47 | /// The basic block order is constructed as follows: |
||
| 48 | /// Once we finish processing some block, we update the counters in MBBInfos |
||
| 49 | /// and re-process any successors that are now 'done'. |
||
| 50 | /// We call a block that is ready for its final round of processing `done` |
||
| 51 | /// (isBlockDone), e.g. when all predecessor information is known. |
||
| 52 | /// |
||
| 53 | /// Note that a naive traversal order would be to do two complete passes over |
||
| 54 | /// all basic blocks/instructions, the first for recording clearances, the |
||
| 55 | /// second for updating clearance based on backedges. |
||
| 56 | /// However, for functions without backedges, or functions with a lot of |
||
| 57 | /// straight-line code, and a small loop, that would be a lot of unnecessary |
||
| 58 | /// work (since only the BBs that are part of the loop require two passes). |
||
| 59 | /// |
||
| 60 | /// E.g., the naive iteration order for the above exmple is as follows: |
||
| 61 | /// Naive: PH A B C D A' B' C' D' |
||
| 62 | /// |
||
| 63 | /// In the optimized approach we avoid processing D twice, because we |
||
| 64 | /// can entirely process the predecessors before getting to D. |
||
| 65 | class LoopTraversal { |
||
| 66 | private: |
||
| 67 | struct MBBInfo { |
||
| 68 | /// Whether we have gotten to this block in primary processing yet. |
||
| 69 | bool PrimaryCompleted = false; |
||
| 70 | |||
| 71 | /// The number of predecessors for which primary processing has completed |
||
| 72 | unsigned IncomingProcessed = 0; |
||
| 73 | |||
| 74 | /// The value of `IncomingProcessed` at the start of primary processing |
||
| 75 | unsigned PrimaryIncoming = 0; |
||
| 76 | |||
| 77 | /// The number of predecessors for which all processing steps are done. |
||
| 78 | unsigned IncomingCompleted = 0; |
||
| 79 | |||
| 80 | MBBInfo() = default; |
||
| 81 | }; |
||
| 82 | using MBBInfoMap = SmallVector<MBBInfo, 4>; |
||
| 83 | /// Helps keep track if we proccessed this block and all its predecessors. |
||
| 84 | MBBInfoMap MBBInfos; |
||
| 85 | |||
| 86 | public: |
||
| 87 | struct TraversedMBBInfo { |
||
| 88 | /// The basic block. |
||
| 89 | MachineBasicBlock *MBB = nullptr; |
||
| 90 | |||
| 91 | /// True if this is the first time we process the basic block. |
||
| 92 | bool PrimaryPass = true; |
||
| 93 | |||
| 94 | /// True if the block that is ready for its final round of processing. |
||
| 95 | bool IsDone = true; |
||
| 96 | |||
| 97 | TraversedMBBInfo(MachineBasicBlock *BB = nullptr, bool Primary = true, |
||
| 98 | bool Done = true) |
||
| 99 | : MBB(BB), PrimaryPass(Primary), IsDone(Done) {} |
||
| 100 | }; |
||
| 101 | LoopTraversal() = default; |
||
| 102 | |||
| 103 | /// Identifies basic blocks that are part of loops and should to be |
||
| 104 | /// visited twice and returns efficient traversal order for all the blocks. |
||
| 105 | typedef SmallVector<TraversedMBBInfo, 4> TraversalOrder; |
||
| 106 | TraversalOrder traverse(MachineFunction &MF); |
||
| 107 | |||
| 108 | private: |
||
| 109 | /// Returens true if the block is ready for its final round of processing. |
||
| 110 | bool isBlockDone(MachineBasicBlock *MBB); |
||
| 111 | }; |
||
| 112 | |||
| 113 | } // namespace llvm |
||
| 114 | |||
| 115 | #endif // LLVM_CODEGEN_LOOPTRAVERSAL_H |