MachineBlockPlacement.cpp revision dce4a407a24b04eebc6a376f8e62b41aaa7b071f
1//===-- MachineBlockPlacement.cpp - Basic Block Code Layout optimization --===//
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 basic block placement transformations using the CFG
11// structure and branch probability estimates.
12//
13// The pass strives to preserve the structure of the CFG (that is, retain
14// a topological ordering of basic blocks) in the absence of a *strong* signal
15// to the contrary from probabilities. However, within the CFG structure, it
16// attempts to choose an ordering which favors placing more likely sequences of
17// blocks adjacent to each other.
18//
19// The algorithm works from the inner-most loop within a function outward, and
20// at each stage walks through the basic blocks, trying to coalesce them into
21// sequential chains where allowed by the CFG (or demanded by heavy
22// probabilities). Finally, it walks the blocks in topological order, and the
23// first time it reaches a chain of basic blocks, it schedules them in the
24// function in-order.
25//
26//===----------------------------------------------------------------------===//
27
28#include "llvm/CodeGen/Passes.h"
29#include "llvm/ADT/DenseMap.h"
30#include "llvm/ADT/SmallPtrSet.h"
31#include "llvm/ADT/SmallVector.h"
32#include "llvm/ADT/Statistic.h"
33#include "llvm/CodeGen/MachineBasicBlock.h"
34#include "llvm/CodeGen/MachineBlockFrequencyInfo.h"
35#include "llvm/CodeGen/MachineBranchProbabilityInfo.h"
36#include "llvm/CodeGen/MachineFunction.h"
37#include "llvm/CodeGen/MachineFunctionPass.h"
38#include "llvm/CodeGen/MachineLoopInfo.h"
39#include "llvm/CodeGen/MachineModuleInfo.h"
40#include "llvm/Support/Allocator.h"
41#include "llvm/Support/CommandLine.h"
42#include "llvm/Support/Debug.h"
43#include "llvm/Target/TargetInstrInfo.h"
44#include "llvm/Target/TargetLowering.h"
45#include <algorithm>
46using namespace llvm;
47
48#define DEBUG_TYPE "block-placement2"
49
50STATISTIC(NumCondBranches, "Number of conditional branches");
51STATISTIC(NumUncondBranches, "Number of uncondittional branches");
52STATISTIC(CondBranchTakenFreq,
53          "Potential frequency of taking conditional branches");
54STATISTIC(UncondBranchTakenFreq,
55          "Potential frequency of taking unconditional branches");
56
57static cl::opt<unsigned> AlignAllBlock("align-all-blocks",
58                                       cl::desc("Force the alignment of all "
59                                                "blocks in the function."),
60                                       cl::init(0), cl::Hidden);
61
62// FIXME: Find a good default for this flag and remove the flag.
63static cl::opt<unsigned>
64ExitBlockBias("block-placement-exit-block-bias",
65              cl::desc("Block frequency percentage a loop exit block needs "
66                       "over the original exit to be considered the new exit."),
67              cl::init(0), cl::Hidden);
68
69namespace {
70class BlockChain;
71/// \brief Type for our function-wide basic block -> block chain mapping.
72typedef DenseMap<MachineBasicBlock *, BlockChain *> BlockToChainMapType;
73}
74
75namespace {
76/// \brief A chain of blocks which will be laid out contiguously.
77///
78/// This is the datastructure representing a chain of consecutive blocks that
79/// are profitable to layout together in order to maximize fallthrough
80/// probabilities and code locality. We also can use a block chain to represent
81/// a sequence of basic blocks which have some external (correctness)
82/// requirement for sequential layout.
83///
84/// Chains can be built around a single basic block and can be merged to grow
85/// them. They participate in a block-to-chain mapping, which is updated
86/// automatically as chains are merged together.
87class BlockChain {
88  /// \brief The sequence of blocks belonging to this chain.
89  ///
90  /// This is the sequence of blocks for a particular chain. These will be laid
91  /// out in-order within the function.
92  SmallVector<MachineBasicBlock *, 4> Blocks;
93
94  /// \brief A handle to the function-wide basic block to block chain mapping.
95  ///
96  /// This is retained in each block chain to simplify the computation of child
97  /// block chains for SCC-formation and iteration. We store the edges to child
98  /// basic blocks, and map them back to their associated chains using this
99  /// structure.
100  BlockToChainMapType &BlockToChain;
101
102public:
103  /// \brief Construct a new BlockChain.
104  ///
105  /// This builds a new block chain representing a single basic block in the
106  /// function. It also registers itself as the chain that block participates
107  /// in with the BlockToChain mapping.
108  BlockChain(BlockToChainMapType &BlockToChain, MachineBasicBlock *BB)
109    : Blocks(1, BB), BlockToChain(BlockToChain), LoopPredecessors(0) {
110    assert(BB && "Cannot create a chain with a null basic block");
111    BlockToChain[BB] = this;
112  }
113
114  /// \brief Iterator over blocks within the chain.
115  typedef SmallVectorImpl<MachineBasicBlock *>::iterator iterator;
116
117  /// \brief Beginning of blocks within the chain.
118  iterator begin() { return Blocks.begin(); }
119
120  /// \brief End of blocks within the chain.
121  iterator end() { return Blocks.end(); }
122
123  /// \brief Merge a block chain into this one.
124  ///
125  /// This routine merges a block chain into this one. It takes care of forming
126  /// a contiguous sequence of basic blocks, updating the edge list, and
127  /// updating the block -> chain mapping. It does not free or tear down the
128  /// old chain, but the old chain's block list is no longer valid.
129  void merge(MachineBasicBlock *BB, BlockChain *Chain) {
130    assert(BB);
131    assert(!Blocks.empty());
132
133    // Fast path in case we don't have a chain already.
134    if (!Chain) {
135      assert(!BlockToChain[BB]);
136      Blocks.push_back(BB);
137      BlockToChain[BB] = this;
138      return;
139    }
140
141    assert(BB == *Chain->begin());
142    assert(Chain->begin() != Chain->end());
143
144    // Update the incoming blocks to point to this chain, and add them to the
145    // chain structure.
146    for (BlockChain::iterator BI = Chain->begin(), BE = Chain->end();
147         BI != BE; ++BI) {
148      Blocks.push_back(*BI);
149      assert(BlockToChain[*BI] == Chain && "Incoming blocks not in chain");
150      BlockToChain[*BI] = this;
151    }
152  }
153
154#ifndef NDEBUG
155  /// \brief Dump the blocks in this chain.
156  LLVM_DUMP_METHOD void dump() {
157    for (iterator I = begin(), E = end(); I != E; ++I)
158      (*I)->dump();
159  }
160#endif // NDEBUG
161
162  /// \brief Count of predecessors within the loop currently being processed.
163  ///
164  /// This count is updated at each loop we process to represent the number of
165  /// in-loop predecessors of this chain.
166  unsigned LoopPredecessors;
167};
168}
169
170namespace {
171class MachineBlockPlacement : public MachineFunctionPass {
172  /// \brief A typedef for a block filter set.
173  typedef SmallPtrSet<MachineBasicBlock *, 16> BlockFilterSet;
174
175  /// \brief A handle to the branch probability pass.
176  const MachineBranchProbabilityInfo *MBPI;
177
178  /// \brief A handle to the function-wide block frequency pass.
179  const MachineBlockFrequencyInfo *MBFI;
180
181  /// \brief A handle to the loop info.
182  const MachineLoopInfo *MLI;
183
184  /// \brief A handle to the target's instruction info.
185  const TargetInstrInfo *TII;
186
187  /// \brief A handle to the target's lowering info.
188  const TargetLoweringBase *TLI;
189
190  /// \brief Allocator and owner of BlockChain structures.
191  ///
192  /// We build BlockChains lazily while processing the loop structure of
193  /// a function. To reduce malloc traffic, we allocate them using this
194  /// slab-like allocator, and destroy them after the pass completes. An
195  /// important guarantee is that this allocator produces stable pointers to
196  /// the chains.
197  SpecificBumpPtrAllocator<BlockChain> ChainAllocator;
198
199  /// \brief Function wide BasicBlock to BlockChain mapping.
200  ///
201  /// This mapping allows efficiently moving from any given basic block to the
202  /// BlockChain it participates in, if any. We use it to, among other things,
203  /// allow implicitly defining edges between chains as the existing edges
204  /// between basic blocks.
205  DenseMap<MachineBasicBlock *, BlockChain *> BlockToChain;
206
207  void markChainSuccessors(BlockChain &Chain,
208                           MachineBasicBlock *LoopHeaderBB,
209                           SmallVectorImpl<MachineBasicBlock *> &BlockWorkList,
210                           const BlockFilterSet *BlockFilter = nullptr);
211  MachineBasicBlock *selectBestSuccessor(MachineBasicBlock *BB,
212                                         BlockChain &Chain,
213                                         const BlockFilterSet *BlockFilter);
214  MachineBasicBlock *selectBestCandidateBlock(
215      BlockChain &Chain, SmallVectorImpl<MachineBasicBlock *> &WorkList,
216      const BlockFilterSet *BlockFilter);
217  MachineBasicBlock *getFirstUnplacedBlock(
218      MachineFunction &F,
219      const BlockChain &PlacedChain,
220      MachineFunction::iterator &PrevUnplacedBlockIt,
221      const BlockFilterSet *BlockFilter);
222  void buildChain(MachineBasicBlock *BB, BlockChain &Chain,
223                  SmallVectorImpl<MachineBasicBlock *> &BlockWorkList,
224                  const BlockFilterSet *BlockFilter = nullptr);
225  MachineBasicBlock *findBestLoopTop(MachineLoop &L,
226                                     const BlockFilterSet &LoopBlockSet);
227  MachineBasicBlock *findBestLoopExit(MachineFunction &F,
228                                      MachineLoop &L,
229                                      const BlockFilterSet &LoopBlockSet);
230  void buildLoopChains(MachineFunction &F, MachineLoop &L);
231  void rotateLoop(BlockChain &LoopChain, MachineBasicBlock *ExitingBB,
232                  const BlockFilterSet &LoopBlockSet);
233  void buildCFGChains(MachineFunction &F);
234
235public:
236  static char ID; // Pass identification, replacement for typeid
237  MachineBlockPlacement() : MachineFunctionPass(ID) {
238    initializeMachineBlockPlacementPass(*PassRegistry::getPassRegistry());
239  }
240
241  bool runOnMachineFunction(MachineFunction &F) override;
242
243  void getAnalysisUsage(AnalysisUsage &AU) const override {
244    AU.addRequired<MachineBranchProbabilityInfo>();
245    AU.addRequired<MachineBlockFrequencyInfo>();
246    AU.addRequired<MachineLoopInfo>();
247    MachineFunctionPass::getAnalysisUsage(AU);
248  }
249};
250}
251
252char MachineBlockPlacement::ID = 0;
253char &llvm::MachineBlockPlacementID = MachineBlockPlacement::ID;
254INITIALIZE_PASS_BEGIN(MachineBlockPlacement, "block-placement2",
255                      "Branch Probability Basic Block Placement", false, false)
256INITIALIZE_PASS_DEPENDENCY(MachineBranchProbabilityInfo)
257INITIALIZE_PASS_DEPENDENCY(MachineBlockFrequencyInfo)
258INITIALIZE_PASS_DEPENDENCY(MachineLoopInfo)
259INITIALIZE_PASS_END(MachineBlockPlacement, "block-placement2",
260                    "Branch Probability Basic Block Placement", false, false)
261
262#ifndef NDEBUG
263/// \brief Helper to print the name of a MBB.
264///
265/// Only used by debug logging.
266static std::string getBlockName(MachineBasicBlock *BB) {
267  std::string Result;
268  raw_string_ostream OS(Result);
269  OS << "BB#" << BB->getNumber()
270     << " (derived from LLVM BB '" << BB->getName() << "')";
271  OS.flush();
272  return Result;
273}
274
275/// \brief Helper to print the number of a MBB.
276///
277/// Only used by debug logging.
278static std::string getBlockNum(MachineBasicBlock *BB) {
279  std::string Result;
280  raw_string_ostream OS(Result);
281  OS << "BB#" << BB->getNumber();
282  OS.flush();
283  return Result;
284}
285#endif
286
287/// \brief Mark a chain's successors as having one fewer preds.
288///
289/// When a chain is being merged into the "placed" chain, this routine will
290/// quickly walk the successors of each block in the chain and mark them as
291/// having one fewer active predecessor. It also adds any successors of this
292/// chain which reach the zero-predecessor state to the worklist passed in.
293void MachineBlockPlacement::markChainSuccessors(
294    BlockChain &Chain,
295    MachineBasicBlock *LoopHeaderBB,
296    SmallVectorImpl<MachineBasicBlock *> &BlockWorkList,
297    const BlockFilterSet *BlockFilter) {
298  // Walk all the blocks in this chain, marking their successors as having
299  // a predecessor placed.
300  for (BlockChain::iterator CBI = Chain.begin(), CBE = Chain.end();
301       CBI != CBE; ++CBI) {
302    // Add any successors for which this is the only un-placed in-loop
303    // predecessor to the worklist as a viable candidate for CFG-neutral
304    // placement. No subsequent placement of this block will violate the CFG
305    // shape, so we get to use heuristics to choose a favorable placement.
306    for (MachineBasicBlock::succ_iterator SI = (*CBI)->succ_begin(),
307                                          SE = (*CBI)->succ_end();
308         SI != SE; ++SI) {
309      if (BlockFilter && !BlockFilter->count(*SI))
310        continue;
311      BlockChain &SuccChain = *BlockToChain[*SI];
312      // Disregard edges within a fixed chain, or edges to the loop header.
313      if (&Chain == &SuccChain || *SI == LoopHeaderBB)
314        continue;
315
316      // This is a cross-chain edge that is within the loop, so decrement the
317      // loop predecessor count of the destination chain.
318      if (SuccChain.LoopPredecessors > 0 && --SuccChain.LoopPredecessors == 0)
319        BlockWorkList.push_back(*SuccChain.begin());
320    }
321  }
322}
323
324/// \brief Select the best successor for a block.
325///
326/// This looks across all successors of a particular block and attempts to
327/// select the "best" one to be the layout successor. It only considers direct
328/// successors which also pass the block filter. It will attempt to avoid
329/// breaking CFG structure, but cave and break such structures in the case of
330/// very hot successor edges.
331///
332/// \returns The best successor block found, or null if none are viable.
333MachineBasicBlock *MachineBlockPlacement::selectBestSuccessor(
334    MachineBasicBlock *BB, BlockChain &Chain,
335    const BlockFilterSet *BlockFilter) {
336  const BranchProbability HotProb(4, 5); // 80%
337
338  MachineBasicBlock *BestSucc = nullptr;
339  // FIXME: Due to the performance of the probability and weight routines in
340  // the MBPI analysis, we manually compute probabilities using the edge
341  // weights. This is suboptimal as it means that the somewhat subtle
342  // definition of edge weight semantics is encoded here as well. We should
343  // improve the MBPI interface to efficiently support query patterns such as
344  // this.
345  uint32_t BestWeight = 0;
346  uint32_t WeightScale = 0;
347  uint32_t SumWeight = MBPI->getSumForBlock(BB, WeightScale);
348  DEBUG(dbgs() << "Attempting merge from: " << getBlockName(BB) << "\n");
349  for (MachineBasicBlock::succ_iterator SI = BB->succ_begin(),
350                                        SE = BB->succ_end();
351       SI != SE; ++SI) {
352    if (BlockFilter && !BlockFilter->count(*SI))
353      continue;
354    BlockChain &SuccChain = *BlockToChain[*SI];
355    if (&SuccChain == &Chain) {
356      DEBUG(dbgs() << "    " << getBlockName(*SI) << " -> Already merged!\n");
357      continue;
358    }
359    if (*SI != *SuccChain.begin()) {
360      DEBUG(dbgs() << "    " << getBlockName(*SI) << " -> Mid chain!\n");
361      continue;
362    }
363
364    uint32_t SuccWeight = MBPI->getEdgeWeight(BB, *SI);
365    BranchProbability SuccProb(SuccWeight / WeightScale, SumWeight);
366
367    // Only consider successors which are either "hot", or wouldn't violate
368    // any CFG constraints.
369    if (SuccChain.LoopPredecessors != 0) {
370      if (SuccProb < HotProb) {
371        DEBUG(dbgs() << "    " << getBlockName(*SI) << " -> " << SuccProb
372                     << " (prob) (CFG conflict)\n");
373        continue;
374      }
375
376      // Make sure that a hot successor doesn't have a globally more important
377      // predecessor.
378      BlockFrequency CandidateEdgeFreq
379        = MBFI->getBlockFreq(BB) * SuccProb * HotProb.getCompl();
380      bool BadCFGConflict = false;
381      for (MachineBasicBlock::pred_iterator PI = (*SI)->pred_begin(),
382                                            PE = (*SI)->pred_end();
383           PI != PE; ++PI) {
384        if (*PI == *SI || (BlockFilter && !BlockFilter->count(*PI)) ||
385            BlockToChain[*PI] == &Chain)
386          continue;
387        BlockFrequency PredEdgeFreq
388          = MBFI->getBlockFreq(*PI) * MBPI->getEdgeProbability(*PI, *SI);
389        if (PredEdgeFreq >= CandidateEdgeFreq) {
390          BadCFGConflict = true;
391          break;
392        }
393      }
394      if (BadCFGConflict) {
395        DEBUG(dbgs() << "    " << getBlockName(*SI) << " -> " << SuccProb
396                     << " (prob) (non-cold CFG conflict)\n");
397        continue;
398      }
399    }
400
401    DEBUG(dbgs() << "    " << getBlockName(*SI) << " -> " << SuccProb
402                 << " (prob)"
403                 << (SuccChain.LoopPredecessors != 0 ? " (CFG break)" : "")
404                 << "\n");
405    if (BestSucc && BestWeight >= SuccWeight)
406      continue;
407    BestSucc = *SI;
408    BestWeight = SuccWeight;
409  }
410  return BestSucc;
411}
412
413/// \brief Select the best block from a worklist.
414///
415/// This looks through the provided worklist as a list of candidate basic
416/// blocks and select the most profitable one to place. The definition of
417/// profitable only really makes sense in the context of a loop. This returns
418/// the most frequently visited block in the worklist, which in the case of
419/// a loop, is the one most desirable to be physically close to the rest of the
420/// loop body in order to improve icache behavior.
421///
422/// \returns The best block found, or null if none are viable.
423MachineBasicBlock *MachineBlockPlacement::selectBestCandidateBlock(
424    BlockChain &Chain, SmallVectorImpl<MachineBasicBlock *> &WorkList,
425    const BlockFilterSet *BlockFilter) {
426  // Once we need to walk the worklist looking for a candidate, cleanup the
427  // worklist of already placed entries.
428  // FIXME: If this shows up on profiles, it could be folded (at the cost of
429  // some code complexity) into the loop below.
430  WorkList.erase(std::remove_if(WorkList.begin(), WorkList.end(),
431                                [&](MachineBasicBlock *BB) {
432                   return BlockToChain.lookup(BB) == &Chain;
433                 }),
434                 WorkList.end());
435
436  MachineBasicBlock *BestBlock = nullptr;
437  BlockFrequency BestFreq;
438  for (SmallVectorImpl<MachineBasicBlock *>::iterator WBI = WorkList.begin(),
439                                                      WBE = WorkList.end();
440       WBI != WBE; ++WBI) {
441    BlockChain &SuccChain = *BlockToChain[*WBI];
442    if (&SuccChain == &Chain) {
443      DEBUG(dbgs() << "    " << getBlockName(*WBI)
444                   << " -> Already merged!\n");
445      continue;
446    }
447    assert(SuccChain.LoopPredecessors == 0 && "Found CFG-violating block");
448
449    BlockFrequency CandidateFreq = MBFI->getBlockFreq(*WBI);
450    DEBUG(dbgs() << "    " << getBlockName(*WBI) << " -> ";
451                 MBFI->printBlockFreq(dbgs(), CandidateFreq) << " (freq)\n");
452    if (BestBlock && BestFreq >= CandidateFreq)
453      continue;
454    BestBlock = *WBI;
455    BestFreq = CandidateFreq;
456  }
457  return BestBlock;
458}
459
460/// \brief Retrieve the first unplaced basic block.
461///
462/// This routine is called when we are unable to use the CFG to walk through
463/// all of the basic blocks and form a chain due to unnatural loops in the CFG.
464/// We walk through the function's blocks in order, starting from the
465/// LastUnplacedBlockIt. We update this iterator on each call to avoid
466/// re-scanning the entire sequence on repeated calls to this routine.
467MachineBasicBlock *MachineBlockPlacement::getFirstUnplacedBlock(
468    MachineFunction &F, const BlockChain &PlacedChain,
469    MachineFunction::iterator &PrevUnplacedBlockIt,
470    const BlockFilterSet *BlockFilter) {
471  for (MachineFunction::iterator I = PrevUnplacedBlockIt, E = F.end(); I != E;
472       ++I) {
473    if (BlockFilter && !BlockFilter->count(I))
474      continue;
475    if (BlockToChain[I] != &PlacedChain) {
476      PrevUnplacedBlockIt = I;
477      // Now select the head of the chain to which the unplaced block belongs
478      // as the block to place. This will force the entire chain to be placed,
479      // and satisfies the requirements of merging chains.
480      return *BlockToChain[I]->begin();
481    }
482  }
483  return nullptr;
484}
485
486void MachineBlockPlacement::buildChain(
487    MachineBasicBlock *BB,
488    BlockChain &Chain,
489    SmallVectorImpl<MachineBasicBlock *> &BlockWorkList,
490    const BlockFilterSet *BlockFilter) {
491  assert(BB);
492  assert(BlockToChain[BB] == &Chain);
493  MachineFunction &F = *BB->getParent();
494  MachineFunction::iterator PrevUnplacedBlockIt = F.begin();
495
496  MachineBasicBlock *LoopHeaderBB = BB;
497  markChainSuccessors(Chain, LoopHeaderBB, BlockWorkList, BlockFilter);
498  BB = *std::prev(Chain.end());
499  for (;;) {
500    assert(BB);
501    assert(BlockToChain[BB] == &Chain);
502    assert(*std::prev(Chain.end()) == BB);
503
504    // Look for the best viable successor if there is one to place immediately
505    // after this block.
506    MachineBasicBlock *BestSucc = selectBestSuccessor(BB, Chain, BlockFilter);
507
508    // If an immediate successor isn't available, look for the best viable
509    // block among those we've identified as not violating the loop's CFG at
510    // this point. This won't be a fallthrough, but it will increase locality.
511    if (!BestSucc)
512      BestSucc = selectBestCandidateBlock(Chain, BlockWorkList, BlockFilter);
513
514    if (!BestSucc) {
515      BestSucc = getFirstUnplacedBlock(F, Chain, PrevUnplacedBlockIt,
516                                       BlockFilter);
517      if (!BestSucc)
518        break;
519
520      DEBUG(dbgs() << "Unnatural loop CFG detected, forcibly merging the "
521                      "layout successor until the CFG reduces\n");
522    }
523
524    // Place this block, updating the datastructures to reflect its placement.
525    BlockChain &SuccChain = *BlockToChain[BestSucc];
526    // Zero out LoopPredecessors for the successor we're about to merge in case
527    // we selected a successor that didn't fit naturally into the CFG.
528    SuccChain.LoopPredecessors = 0;
529    DEBUG(dbgs() << "Merging from " << getBlockNum(BB)
530                 << " to " << getBlockNum(BestSucc) << "\n");
531    markChainSuccessors(SuccChain, LoopHeaderBB, BlockWorkList, BlockFilter);
532    Chain.merge(BestSucc, &SuccChain);
533    BB = *std::prev(Chain.end());
534  }
535
536  DEBUG(dbgs() << "Finished forming chain for header block "
537               << getBlockNum(*Chain.begin()) << "\n");
538}
539
540/// \brief Find the best loop top block for layout.
541///
542/// Look for a block which is strictly better than the loop header for laying
543/// out at the top of the loop. This looks for one and only one pattern:
544/// a latch block with no conditional exit. This block will cause a conditional
545/// jump around it or will be the bottom of the loop if we lay it out in place,
546/// but if it it doesn't end up at the bottom of the loop for any reason,
547/// rotation alone won't fix it. Because such a block will always result in an
548/// unconditional jump (for the backedge) rotating it in front of the loop
549/// header is always profitable.
550MachineBasicBlock *
551MachineBlockPlacement::findBestLoopTop(MachineLoop &L,
552                                       const BlockFilterSet &LoopBlockSet) {
553  // Check that the header hasn't been fused with a preheader block due to
554  // crazy branches. If it has, we need to start with the header at the top to
555  // prevent pulling the preheader into the loop body.
556  BlockChain &HeaderChain = *BlockToChain[L.getHeader()];
557  if (!LoopBlockSet.count(*HeaderChain.begin()))
558    return L.getHeader();
559
560  DEBUG(dbgs() << "Finding best loop top for: "
561               << getBlockName(L.getHeader()) << "\n");
562
563  BlockFrequency BestPredFreq;
564  MachineBasicBlock *BestPred = nullptr;
565  for (MachineBasicBlock::pred_iterator PI = L.getHeader()->pred_begin(),
566                                        PE = L.getHeader()->pred_end();
567       PI != PE; ++PI) {
568    MachineBasicBlock *Pred = *PI;
569    if (!LoopBlockSet.count(Pred))
570      continue;
571    DEBUG(dbgs() << "    header pred: " << getBlockName(Pred) << ", "
572                 << Pred->succ_size() << " successors, ";
573                 MBFI->printBlockFreq(dbgs(), Pred) << " freq\n");
574    if (Pred->succ_size() > 1)
575      continue;
576
577    BlockFrequency PredFreq = MBFI->getBlockFreq(Pred);
578    if (!BestPred || PredFreq > BestPredFreq ||
579        (!(PredFreq < BestPredFreq) &&
580         Pred->isLayoutSuccessor(L.getHeader()))) {
581      BestPred = Pred;
582      BestPredFreq = PredFreq;
583    }
584  }
585
586  // If no direct predecessor is fine, just use the loop header.
587  if (!BestPred)
588    return L.getHeader();
589
590  // Walk backwards through any straight line of predecessors.
591  while (BestPred->pred_size() == 1 &&
592         (*BestPred->pred_begin())->succ_size() == 1 &&
593         *BestPred->pred_begin() != L.getHeader())
594    BestPred = *BestPred->pred_begin();
595
596  DEBUG(dbgs() << "    final top: " << getBlockName(BestPred) << "\n");
597  return BestPred;
598}
599
600
601/// \brief Find the best loop exiting block for layout.
602///
603/// This routine implements the logic to analyze the loop looking for the best
604/// block to layout at the top of the loop. Typically this is done to maximize
605/// fallthrough opportunities.
606MachineBasicBlock *
607MachineBlockPlacement::findBestLoopExit(MachineFunction &F,
608                                        MachineLoop &L,
609                                        const BlockFilterSet &LoopBlockSet) {
610  // We don't want to layout the loop linearly in all cases. If the loop header
611  // is just a normal basic block in the loop, we want to look for what block
612  // within the loop is the best one to layout at the top. However, if the loop
613  // header has be pre-merged into a chain due to predecessors not having
614  // analyzable branches, *and* the predecessor it is merged with is *not* part
615  // of the loop, rotating the header into the middle of the loop will create
616  // a non-contiguous range of blocks which is Very Bad. So start with the
617  // header and only rotate if safe.
618  BlockChain &HeaderChain = *BlockToChain[L.getHeader()];
619  if (!LoopBlockSet.count(*HeaderChain.begin()))
620    return nullptr;
621
622  BlockFrequency BestExitEdgeFreq;
623  unsigned BestExitLoopDepth = 0;
624  MachineBasicBlock *ExitingBB = nullptr;
625  // If there are exits to outer loops, loop rotation can severely limit
626  // fallthrough opportunites unless it selects such an exit. Keep a set of
627  // blocks where rotating to exit with that block will reach an outer loop.
628  SmallPtrSet<MachineBasicBlock *, 4> BlocksExitingToOuterLoop;
629
630  DEBUG(dbgs() << "Finding best loop exit for: "
631               << getBlockName(L.getHeader()) << "\n");
632  for (MachineLoop::block_iterator I = L.block_begin(),
633                                   E = L.block_end();
634       I != E; ++I) {
635    BlockChain &Chain = *BlockToChain[*I];
636    // Ensure that this block is at the end of a chain; otherwise it could be
637    // mid-way through an inner loop or a successor of an analyzable branch.
638    if (*I != *std::prev(Chain.end()))
639      continue;
640
641    // Now walk the successors. We need to establish whether this has a viable
642    // exiting successor and whether it has a viable non-exiting successor.
643    // We store the old exiting state and restore it if a viable looping
644    // successor isn't found.
645    MachineBasicBlock *OldExitingBB = ExitingBB;
646    BlockFrequency OldBestExitEdgeFreq = BestExitEdgeFreq;
647    bool HasLoopingSucc = false;
648    // FIXME: Due to the performance of the probability and weight routines in
649    // the MBPI analysis, we use the internal weights and manually compute the
650    // probabilities to avoid quadratic behavior.
651    uint32_t WeightScale = 0;
652    uint32_t SumWeight = MBPI->getSumForBlock(*I, WeightScale);
653    for (MachineBasicBlock::succ_iterator SI = (*I)->succ_begin(),
654                                          SE = (*I)->succ_end();
655         SI != SE; ++SI) {
656      if ((*SI)->isLandingPad())
657        continue;
658      if (*SI == *I)
659        continue;
660      BlockChain &SuccChain = *BlockToChain[*SI];
661      // Don't split chains, either this chain or the successor's chain.
662      if (&Chain == &SuccChain) {
663        DEBUG(dbgs() << "    exiting: " << getBlockName(*I) << " -> "
664                     << getBlockName(*SI) << " (chain conflict)\n");
665        continue;
666      }
667
668      uint32_t SuccWeight = MBPI->getEdgeWeight(*I, *SI);
669      if (LoopBlockSet.count(*SI)) {
670        DEBUG(dbgs() << "    looping: " << getBlockName(*I) << " -> "
671                     << getBlockName(*SI) << " (" << SuccWeight << ")\n");
672        HasLoopingSucc = true;
673        continue;
674      }
675
676      unsigned SuccLoopDepth = 0;
677      if (MachineLoop *ExitLoop = MLI->getLoopFor(*SI)) {
678        SuccLoopDepth = ExitLoop->getLoopDepth();
679        if (ExitLoop->contains(&L))
680          BlocksExitingToOuterLoop.insert(*I);
681      }
682
683      BranchProbability SuccProb(SuccWeight / WeightScale, SumWeight);
684      BlockFrequency ExitEdgeFreq = MBFI->getBlockFreq(*I) * SuccProb;
685      DEBUG(dbgs() << "    exiting: " << getBlockName(*I) << " -> "
686                   << getBlockName(*SI) << " [L:" << SuccLoopDepth
687                   << "] (";
688                   MBFI->printBlockFreq(dbgs(), ExitEdgeFreq) << ")\n");
689      // Note that we bias this toward an existing layout successor to retain
690      // incoming order in the absence of better information. The exit must have
691      // a frequency higher than the current exit before we consider breaking
692      // the layout.
693      BranchProbability Bias(100 - ExitBlockBias, 100);
694      if (!ExitingBB || BestExitLoopDepth < SuccLoopDepth ||
695          ExitEdgeFreq > BestExitEdgeFreq ||
696          ((*I)->isLayoutSuccessor(*SI) &&
697           !(ExitEdgeFreq < BestExitEdgeFreq * Bias))) {
698        BestExitEdgeFreq = ExitEdgeFreq;
699        ExitingBB = *I;
700      }
701    }
702
703    // Restore the old exiting state, no viable looping successor was found.
704    if (!HasLoopingSucc) {
705      ExitingBB = OldExitingBB;
706      BestExitEdgeFreq = OldBestExitEdgeFreq;
707      continue;
708    }
709  }
710  // Without a candidate exiting block or with only a single block in the
711  // loop, just use the loop header to layout the loop.
712  if (!ExitingBB || L.getNumBlocks() == 1)
713    return nullptr;
714
715  // Also, if we have exit blocks which lead to outer loops but didn't select
716  // one of them as the exiting block we are rotating toward, disable loop
717  // rotation altogether.
718  if (!BlocksExitingToOuterLoop.empty() &&
719      !BlocksExitingToOuterLoop.count(ExitingBB))
720    return nullptr;
721
722  DEBUG(dbgs() << "  Best exiting block: " << getBlockName(ExitingBB) << "\n");
723  return ExitingBB;
724}
725
726/// \brief Attempt to rotate an exiting block to the bottom of the loop.
727///
728/// Once we have built a chain, try to rotate it to line up the hot exit block
729/// with fallthrough out of the loop if doing so doesn't introduce unnecessary
730/// branches. For example, if the loop has fallthrough into its header and out
731/// of its bottom already, don't rotate it.
732void MachineBlockPlacement::rotateLoop(BlockChain &LoopChain,
733                                       MachineBasicBlock *ExitingBB,
734                                       const BlockFilterSet &LoopBlockSet) {
735  if (!ExitingBB)
736    return;
737
738  MachineBasicBlock *Top = *LoopChain.begin();
739  bool ViableTopFallthrough = false;
740  for (MachineBasicBlock::pred_iterator PI = Top->pred_begin(),
741                                        PE = Top->pred_end();
742       PI != PE; ++PI) {
743    BlockChain *PredChain = BlockToChain[*PI];
744    if (!LoopBlockSet.count(*PI) &&
745        (!PredChain || *PI == *std::prev(PredChain->end()))) {
746      ViableTopFallthrough = true;
747      break;
748    }
749  }
750
751  // If the header has viable fallthrough, check whether the current loop
752  // bottom is a viable exiting block. If so, bail out as rotating will
753  // introduce an unnecessary branch.
754  if (ViableTopFallthrough) {
755    MachineBasicBlock *Bottom = *std::prev(LoopChain.end());
756    for (MachineBasicBlock::succ_iterator SI = Bottom->succ_begin(),
757                                          SE = Bottom->succ_end();
758         SI != SE; ++SI) {
759      BlockChain *SuccChain = BlockToChain[*SI];
760      if (!LoopBlockSet.count(*SI) &&
761          (!SuccChain || *SI == *SuccChain->begin()))
762        return;
763    }
764  }
765
766  BlockChain::iterator ExitIt = std::find(LoopChain.begin(), LoopChain.end(),
767                                          ExitingBB);
768  if (ExitIt == LoopChain.end())
769    return;
770
771  std::rotate(LoopChain.begin(), std::next(ExitIt), LoopChain.end());
772}
773
774/// \brief Forms basic block chains from the natural loop structures.
775///
776/// These chains are designed to preserve the existing *structure* of the code
777/// as much as possible. We can then stitch the chains together in a way which
778/// both preserves the topological structure and minimizes taken conditional
779/// branches.
780void MachineBlockPlacement::buildLoopChains(MachineFunction &F,
781                                            MachineLoop &L) {
782  // First recurse through any nested loops, building chains for those inner
783  // loops.
784  for (MachineLoop::iterator LI = L.begin(), LE = L.end(); LI != LE; ++LI)
785    buildLoopChains(F, **LI);
786
787  SmallVector<MachineBasicBlock *, 16> BlockWorkList;
788  BlockFilterSet LoopBlockSet(L.block_begin(), L.block_end());
789
790  // First check to see if there is an obviously preferable top block for the
791  // loop. This will default to the header, but may end up as one of the
792  // predecessors to the header if there is one which will result in strictly
793  // fewer branches in the loop body.
794  MachineBasicBlock *LoopTop = findBestLoopTop(L, LoopBlockSet);
795
796  // If we selected just the header for the loop top, look for a potentially
797  // profitable exit block in the event that rotating the loop can eliminate
798  // branches by placing an exit edge at the bottom.
799  MachineBasicBlock *ExitingBB = nullptr;
800  if (LoopTop == L.getHeader())
801    ExitingBB = findBestLoopExit(F, L, LoopBlockSet);
802
803  BlockChain &LoopChain = *BlockToChain[LoopTop];
804
805  // FIXME: This is a really lame way of walking the chains in the loop: we
806  // walk the blocks, and use a set to prevent visiting a particular chain
807  // twice.
808  SmallPtrSet<BlockChain *, 4> UpdatedPreds;
809  assert(LoopChain.LoopPredecessors == 0);
810  UpdatedPreds.insert(&LoopChain);
811  for (MachineLoop::block_iterator BI = L.block_begin(),
812                                   BE = L.block_end();
813       BI != BE; ++BI) {
814    BlockChain &Chain = *BlockToChain[*BI];
815    if (!UpdatedPreds.insert(&Chain))
816      continue;
817
818    assert(Chain.LoopPredecessors == 0);
819    for (BlockChain::iterator BCI = Chain.begin(), BCE = Chain.end();
820         BCI != BCE; ++BCI) {
821      assert(BlockToChain[*BCI] == &Chain);
822      for (MachineBasicBlock::pred_iterator PI = (*BCI)->pred_begin(),
823                                            PE = (*BCI)->pred_end();
824           PI != PE; ++PI) {
825        if (BlockToChain[*PI] == &Chain || !LoopBlockSet.count(*PI))
826          continue;
827        ++Chain.LoopPredecessors;
828      }
829    }
830
831    if (Chain.LoopPredecessors == 0)
832      BlockWorkList.push_back(*Chain.begin());
833  }
834
835  buildChain(LoopTop, LoopChain, BlockWorkList, &LoopBlockSet);
836  rotateLoop(LoopChain, ExitingBB, LoopBlockSet);
837
838  DEBUG({
839    // Crash at the end so we get all of the debugging output first.
840    bool BadLoop = false;
841    if (LoopChain.LoopPredecessors) {
842      BadLoop = true;
843      dbgs() << "Loop chain contains a block without its preds placed!\n"
844             << "  Loop header:  " << getBlockName(*L.block_begin()) << "\n"
845             << "  Chain header: " << getBlockName(*LoopChain.begin()) << "\n";
846    }
847    for (BlockChain::iterator BCI = LoopChain.begin(), BCE = LoopChain.end();
848         BCI != BCE; ++BCI) {
849      dbgs() << "          ... " << getBlockName(*BCI) << "\n";
850      if (!LoopBlockSet.erase(*BCI)) {
851        // We don't mark the loop as bad here because there are real situations
852        // where this can occur. For example, with an unanalyzable fallthrough
853        // from a loop block to a non-loop block or vice versa.
854        dbgs() << "Loop chain contains a block not contained by the loop!\n"
855               << "  Loop header:  " << getBlockName(*L.block_begin()) << "\n"
856               << "  Chain header: " << getBlockName(*LoopChain.begin()) << "\n"
857               << "  Bad block:    " << getBlockName(*BCI) << "\n";
858      }
859    }
860
861    if (!LoopBlockSet.empty()) {
862      BadLoop = true;
863      for (BlockFilterSet::iterator LBI = LoopBlockSet.begin(),
864                                    LBE = LoopBlockSet.end();
865           LBI != LBE; ++LBI)
866        dbgs() << "Loop contains blocks never placed into a chain!\n"
867               << "  Loop header:  " << getBlockName(*L.block_begin()) << "\n"
868               << "  Chain header: " << getBlockName(*LoopChain.begin()) << "\n"
869               << "  Bad block:    " << getBlockName(*LBI) << "\n";
870    }
871    assert(!BadLoop && "Detected problems with the placement of this loop.");
872  });
873}
874
875void MachineBlockPlacement::buildCFGChains(MachineFunction &F) {
876  // Ensure that every BB in the function has an associated chain to simplify
877  // the assumptions of the remaining algorithm.
878  SmallVector<MachineOperand, 4> Cond; // For AnalyzeBranch.
879  for (MachineFunction::iterator FI = F.begin(), FE = F.end(); FI != FE; ++FI) {
880    MachineBasicBlock *BB = FI;
881    BlockChain *Chain
882      = new (ChainAllocator.Allocate()) BlockChain(BlockToChain, BB);
883    // Also, merge any blocks which we cannot reason about and must preserve
884    // the exact fallthrough behavior for.
885    for (;;) {
886      Cond.clear();
887      MachineBasicBlock *TBB = nullptr, *FBB = nullptr; // For AnalyzeBranch.
888      if (!TII->AnalyzeBranch(*BB, TBB, FBB, Cond) || !FI->canFallThrough())
889        break;
890
891      MachineFunction::iterator NextFI(std::next(FI));
892      MachineBasicBlock *NextBB = NextFI;
893      // Ensure that the layout successor is a viable block, as we know that
894      // fallthrough is a possibility.
895      assert(NextFI != FE && "Can't fallthrough past the last block.");
896      DEBUG(dbgs() << "Pre-merging due to unanalyzable fallthrough: "
897                   << getBlockName(BB) << " -> " << getBlockName(NextBB)
898                   << "\n");
899      Chain->merge(NextBB, nullptr);
900      FI = NextFI;
901      BB = NextBB;
902    }
903  }
904
905  // Build any loop-based chains.
906  for (MachineLoopInfo::iterator LI = MLI->begin(), LE = MLI->end(); LI != LE;
907       ++LI)
908    buildLoopChains(F, **LI);
909
910  SmallVector<MachineBasicBlock *, 16> BlockWorkList;
911
912  SmallPtrSet<BlockChain *, 4> UpdatedPreds;
913  for (MachineFunction::iterator FI = F.begin(), FE = F.end(); FI != FE; ++FI) {
914    MachineBasicBlock *BB = &*FI;
915    BlockChain &Chain = *BlockToChain[BB];
916    if (!UpdatedPreds.insert(&Chain))
917      continue;
918
919    assert(Chain.LoopPredecessors == 0);
920    for (BlockChain::iterator BCI = Chain.begin(), BCE = Chain.end();
921         BCI != BCE; ++BCI) {
922      assert(BlockToChain[*BCI] == &Chain);
923      for (MachineBasicBlock::pred_iterator PI = (*BCI)->pred_begin(),
924                                            PE = (*BCI)->pred_end();
925           PI != PE; ++PI) {
926        if (BlockToChain[*PI] == &Chain)
927          continue;
928        ++Chain.LoopPredecessors;
929      }
930    }
931
932    if (Chain.LoopPredecessors == 0)
933      BlockWorkList.push_back(*Chain.begin());
934  }
935
936  BlockChain &FunctionChain = *BlockToChain[&F.front()];
937  buildChain(&F.front(), FunctionChain, BlockWorkList);
938
939#ifndef NDEBUG
940  typedef SmallPtrSet<MachineBasicBlock *, 16> FunctionBlockSetType;
941#endif
942  DEBUG({
943    // Crash at the end so we get all of the debugging output first.
944    bool BadFunc = false;
945    FunctionBlockSetType FunctionBlockSet;
946    for (MachineFunction::iterator FI = F.begin(), FE = F.end(); FI != FE; ++FI)
947      FunctionBlockSet.insert(FI);
948
949    for (BlockChain::iterator BCI = FunctionChain.begin(),
950                              BCE = FunctionChain.end();
951         BCI != BCE; ++BCI)
952      if (!FunctionBlockSet.erase(*BCI)) {
953        BadFunc = true;
954        dbgs() << "Function chain contains a block not in the function!\n"
955               << "  Bad block:    " << getBlockName(*BCI) << "\n";
956      }
957
958    if (!FunctionBlockSet.empty()) {
959      BadFunc = true;
960      for (FunctionBlockSetType::iterator FBI = FunctionBlockSet.begin(),
961                                          FBE = FunctionBlockSet.end();
962           FBI != FBE; ++FBI)
963        dbgs() << "Function contains blocks never placed into a chain!\n"
964               << "  Bad block:    " << getBlockName(*FBI) << "\n";
965    }
966    assert(!BadFunc && "Detected problems with the block placement.");
967  });
968
969  // Splice the blocks into place.
970  MachineFunction::iterator InsertPos = F.begin();
971  for (BlockChain::iterator BI = FunctionChain.begin(),
972                            BE = FunctionChain.end();
973       BI != BE; ++BI) {
974    DEBUG(dbgs() << (BI == FunctionChain.begin() ? "Placing chain "
975                                                  : "          ... ")
976          << getBlockName(*BI) << "\n");
977    if (InsertPos != MachineFunction::iterator(*BI))
978      F.splice(InsertPos, *BI);
979    else
980      ++InsertPos;
981
982    // Update the terminator of the previous block.
983    if (BI == FunctionChain.begin())
984      continue;
985    MachineBasicBlock *PrevBB = std::prev(MachineFunction::iterator(*BI));
986
987    // FIXME: It would be awesome of updateTerminator would just return rather
988    // than assert when the branch cannot be analyzed in order to remove this
989    // boiler plate.
990    Cond.clear();
991    MachineBasicBlock *TBB = nullptr, *FBB = nullptr; // For AnalyzeBranch.
992    if (!TII->AnalyzeBranch(*PrevBB, TBB, FBB, Cond)) {
993      // The "PrevBB" is not yet updated to reflect current code layout, so,
994      //   o. it may fall-through to a block without explict "goto" instruction
995      //      before layout, and no longer fall-through it after layout; or
996      //   o. just opposite.
997      //
998      // AnalyzeBranch() may return erroneous value for FBB when these two
999      // situations take place. For the first scenario FBB is mistakenly set
1000      // NULL; for the 2nd scenario, the FBB, which is expected to be NULL,
1001      // is mistakenly pointing to "*BI".
1002      //
1003      bool needUpdateBr = true;
1004      if (!Cond.empty() && (!FBB || FBB == *BI)) {
1005        PrevBB->updateTerminator();
1006        needUpdateBr = false;
1007        Cond.clear();
1008        TBB = FBB = nullptr;
1009        if (TII->AnalyzeBranch(*PrevBB, TBB, FBB, Cond)) {
1010          // FIXME: This should never take place.
1011          TBB = FBB = nullptr;
1012        }
1013      }
1014
1015      // If PrevBB has a two-way branch, try to re-order the branches
1016      // such that we branch to the successor with higher weight first.
1017      if (TBB && !Cond.empty() && FBB &&
1018          MBPI->getEdgeWeight(PrevBB, FBB) > MBPI->getEdgeWeight(PrevBB, TBB) &&
1019          !TII->ReverseBranchCondition(Cond)) {
1020        DEBUG(dbgs() << "Reverse order of the two branches: "
1021                     << getBlockName(PrevBB) << "\n");
1022        DEBUG(dbgs() << "    Edge weight: " << MBPI->getEdgeWeight(PrevBB, FBB)
1023                     << " vs " << MBPI->getEdgeWeight(PrevBB, TBB) << "\n");
1024        DebugLoc dl;  // FIXME: this is nowhere
1025        TII->RemoveBranch(*PrevBB);
1026        TII->InsertBranch(*PrevBB, FBB, TBB, Cond, dl);
1027        needUpdateBr = true;
1028      }
1029      if (needUpdateBr)
1030        PrevBB->updateTerminator();
1031    }
1032  }
1033
1034  // Fixup the last block.
1035  Cond.clear();
1036  MachineBasicBlock *TBB = nullptr, *FBB = nullptr; // For AnalyzeBranch.
1037  if (!TII->AnalyzeBranch(F.back(), TBB, FBB, Cond))
1038    F.back().updateTerminator();
1039
1040  // Walk through the backedges of the function now that we have fully laid out
1041  // the basic blocks and align the destination of each backedge. We don't rely
1042  // exclusively on the loop info here so that we can align backedges in
1043  // unnatural CFGs and backedges that were introduced purely because of the
1044  // loop rotations done during this layout pass.
1045  if (F.getFunction()->getAttributes().
1046        hasAttribute(AttributeSet::FunctionIndex, Attribute::OptimizeForSize))
1047    return;
1048  unsigned Align = TLI->getPrefLoopAlignment();
1049  if (!Align)
1050    return;  // Don't care about loop alignment.
1051  if (FunctionChain.begin() == FunctionChain.end())
1052    return;  // Empty chain.
1053
1054  const BranchProbability ColdProb(1, 5); // 20%
1055  BlockFrequency EntryFreq = MBFI->getBlockFreq(F.begin());
1056  BlockFrequency WeightedEntryFreq = EntryFreq * ColdProb;
1057  for (BlockChain::iterator BI = std::next(FunctionChain.begin()),
1058                            BE = FunctionChain.end();
1059       BI != BE; ++BI) {
1060    // Don't align non-looping basic blocks. These are unlikely to execute
1061    // enough times to matter in practice. Note that we'll still handle
1062    // unnatural CFGs inside of a natural outer loop (the common case) and
1063    // rotated loops.
1064    MachineLoop *L = MLI->getLoopFor(*BI);
1065    if (!L)
1066      continue;
1067
1068    // If the block is cold relative to the function entry don't waste space
1069    // aligning it.
1070    BlockFrequency Freq = MBFI->getBlockFreq(*BI);
1071    if (Freq < WeightedEntryFreq)
1072      continue;
1073
1074    // If the block is cold relative to its loop header, don't align it
1075    // regardless of what edges into the block exist.
1076    MachineBasicBlock *LoopHeader = L->getHeader();
1077    BlockFrequency LoopHeaderFreq = MBFI->getBlockFreq(LoopHeader);
1078    if (Freq < (LoopHeaderFreq * ColdProb))
1079      continue;
1080
1081    // Check for the existence of a non-layout predecessor which would benefit
1082    // from aligning this block.
1083    MachineBasicBlock *LayoutPred = *std::prev(BI);
1084
1085    // Force alignment if all the predecessors are jumps. We already checked
1086    // that the block isn't cold above.
1087    if (!LayoutPred->isSuccessor(*BI)) {
1088      (*BI)->setAlignment(Align);
1089      continue;
1090    }
1091
1092    // Align this block if the layout predecessor's edge into this block is
1093    // cold relative to the block. When this is true, other predecessors make up
1094    // all of the hot entries into the block and thus alignment is likely to be
1095    // important.
1096    BranchProbability LayoutProb = MBPI->getEdgeProbability(LayoutPred, *BI);
1097    BlockFrequency LayoutEdgeFreq = MBFI->getBlockFreq(LayoutPred) * LayoutProb;
1098    if (LayoutEdgeFreq <= (Freq * ColdProb))
1099      (*BI)->setAlignment(Align);
1100  }
1101}
1102
1103bool MachineBlockPlacement::runOnMachineFunction(MachineFunction &F) {
1104  // Check for single-block functions and skip them.
1105  if (std::next(F.begin()) == F.end())
1106    return false;
1107
1108  if (skipOptnoneFunction(*F.getFunction()))
1109    return false;
1110
1111  MBPI = &getAnalysis<MachineBranchProbabilityInfo>();
1112  MBFI = &getAnalysis<MachineBlockFrequencyInfo>();
1113  MLI = &getAnalysis<MachineLoopInfo>();
1114  TII = F.getTarget().getInstrInfo();
1115  TLI = F.getTarget().getTargetLowering();
1116  assert(BlockToChain.empty());
1117
1118  buildCFGChains(F);
1119
1120  BlockToChain.clear();
1121  ChainAllocator.DestroyAll();
1122
1123  if (AlignAllBlock)
1124    // Align all of the blocks in the function to a specific alignment.
1125    for (MachineFunction::iterator FI = F.begin(), FE = F.end();
1126         FI != FE; ++FI)
1127      FI->setAlignment(AlignAllBlock);
1128
1129  // We always return true as we have no way to track whether the final order
1130  // differs from the original order.
1131  return true;
1132}
1133
1134namespace {
1135/// \brief A pass to compute block placement statistics.
1136///
1137/// A separate pass to compute interesting statistics for evaluating block
1138/// placement. This is separate from the actual placement pass so that they can
1139/// be computed in the absence of any placement transformations or when using
1140/// alternative placement strategies.
1141class MachineBlockPlacementStats : public MachineFunctionPass {
1142  /// \brief A handle to the branch probability pass.
1143  const MachineBranchProbabilityInfo *MBPI;
1144
1145  /// \brief A handle to the function-wide block frequency pass.
1146  const MachineBlockFrequencyInfo *MBFI;
1147
1148public:
1149  static char ID; // Pass identification, replacement for typeid
1150  MachineBlockPlacementStats() : MachineFunctionPass(ID) {
1151    initializeMachineBlockPlacementStatsPass(*PassRegistry::getPassRegistry());
1152  }
1153
1154  bool runOnMachineFunction(MachineFunction &F) override;
1155
1156  void getAnalysisUsage(AnalysisUsage &AU) const override {
1157    AU.addRequired<MachineBranchProbabilityInfo>();
1158    AU.addRequired<MachineBlockFrequencyInfo>();
1159    AU.setPreservesAll();
1160    MachineFunctionPass::getAnalysisUsage(AU);
1161  }
1162};
1163}
1164
1165char MachineBlockPlacementStats::ID = 0;
1166char &llvm::MachineBlockPlacementStatsID = MachineBlockPlacementStats::ID;
1167INITIALIZE_PASS_BEGIN(MachineBlockPlacementStats, "block-placement-stats",
1168                      "Basic Block Placement Stats", false, false)
1169INITIALIZE_PASS_DEPENDENCY(MachineBranchProbabilityInfo)
1170INITIALIZE_PASS_DEPENDENCY(MachineBlockFrequencyInfo)
1171INITIALIZE_PASS_END(MachineBlockPlacementStats, "block-placement-stats",
1172                    "Basic Block Placement Stats", false, false)
1173
1174bool MachineBlockPlacementStats::runOnMachineFunction(MachineFunction &F) {
1175  // Check for single-block functions and skip them.
1176  if (std::next(F.begin()) == F.end())
1177    return false;
1178
1179  MBPI = &getAnalysis<MachineBranchProbabilityInfo>();
1180  MBFI = &getAnalysis<MachineBlockFrequencyInfo>();
1181
1182  for (MachineFunction::iterator I = F.begin(), E = F.end(); I != E; ++I) {
1183    BlockFrequency BlockFreq = MBFI->getBlockFreq(I);
1184    Statistic &NumBranches = (I->succ_size() > 1) ? NumCondBranches
1185                                                  : NumUncondBranches;
1186    Statistic &BranchTakenFreq = (I->succ_size() > 1) ? CondBranchTakenFreq
1187                                                      : UncondBranchTakenFreq;
1188    for (MachineBasicBlock::succ_iterator SI = I->succ_begin(),
1189                                          SE = I->succ_end();
1190         SI != SE; ++SI) {
1191      // Skip if this successor is a fallthrough.
1192      if (I->isLayoutSuccessor(*SI))
1193        continue;
1194
1195      BlockFrequency EdgeFreq = BlockFreq * MBPI->getEdgeProbability(I, *SI);
1196      ++NumBranches;
1197      BranchTakenFreq += EdgeFreq.getFrequency();
1198    }
1199  }
1200
1201  return false;
1202}
1203
1204