Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 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 | // |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 10 | // This file implements basic block placement transformations using the CFG |
| 11 | // structure and branch probability estimates. |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 12 | // |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 13 | // The pass strives to preserve the structure of the CFG (that is, retain |
| 14 | // a topological ordering of basic blocks) in the absense 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. |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 25 | // |
| 26 | //===----------------------------------------------------------------------===// |
| 27 | |
| 28 | #define DEBUG_TYPE "block-placement2" |
Chandler Carruth | 4a85cc9 | 2011-10-21 08:57:37 +0000 | [diff] [blame] | 29 | #include "llvm/CodeGen/MachineBasicBlock.h" |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 30 | #include "llvm/CodeGen/MachineBlockFrequencyInfo.h" |
| 31 | #include "llvm/CodeGen/MachineBranchProbabilityInfo.h" |
| 32 | #include "llvm/CodeGen/MachineFunction.h" |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 33 | #include "llvm/CodeGen/MachineFunctionPass.h" |
Chandler Carruth | 4a85cc9 | 2011-10-21 08:57:37 +0000 | [diff] [blame] | 34 | #include "llvm/CodeGen/MachineLoopInfo.h" |
| 35 | #include "llvm/CodeGen/MachineModuleInfo.h" |
| 36 | #include "llvm/CodeGen/Passes.h" |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 37 | #include "llvm/Support/Allocator.h" |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 38 | #include "llvm/Support/Debug.h" |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 39 | #include "llvm/Support/ErrorHandling.h" |
| 40 | #include "llvm/ADT/DenseMap.h" |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 41 | #include "llvm/ADT/PostOrderIterator.h" |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 42 | #include "llvm/ADT/SCCIterator.h" |
| 43 | #include "llvm/ADT/SmallPtrSet.h" |
| 44 | #include "llvm/ADT/SmallVector.h" |
| 45 | #include "llvm/ADT/Statistic.h" |
| 46 | #include "llvm/Target/TargetInstrInfo.h" |
Chandler Carruth | 4a85cc9 | 2011-10-21 08:57:37 +0000 | [diff] [blame] | 47 | #include "llvm/Target/TargetLowering.h" |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 48 | #include <algorithm> |
| 49 | using namespace llvm; |
| 50 | |
Chandler Carruth | 37efc9f | 2011-11-02 07:17:12 +0000 | [diff] [blame^] | 51 | STATISTIC(NumCondBranches, "Number of conditional branches"); |
| 52 | STATISTIC(NumUncondBranches, "Number of uncondittional branches"); |
| 53 | STATISTIC(CondBranchTakenFreq, |
| 54 | "Potential frequency of taking conditional branches"); |
| 55 | STATISTIC(UncondBranchTakenFreq, |
| 56 | "Potential frequency of taking unconditional branches"); |
| 57 | |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 58 | namespace { |
| 59 | /// \brief A structure for storing a weighted edge. |
| 60 | /// |
| 61 | /// This stores an edge and its weight, computed as the product of the |
| 62 | /// frequency that the starting block is entered with the probability of |
| 63 | /// a particular exit block. |
| 64 | struct WeightedEdge { |
| 65 | BlockFrequency EdgeFrequency; |
| 66 | MachineBasicBlock *From, *To; |
| 67 | |
| 68 | bool operator<(const WeightedEdge &RHS) const { |
| 69 | return EdgeFrequency < RHS.EdgeFrequency; |
| 70 | } |
| 71 | }; |
| 72 | } |
| 73 | |
| 74 | namespace { |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 75 | class BlockChain; |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 76 | /// \brief Type for our function-wide basic block -> block chain mapping. |
| 77 | typedef DenseMap<MachineBasicBlock *, BlockChain *> BlockToChainMapType; |
| 78 | } |
| 79 | |
| 80 | namespace { |
| 81 | /// \brief A chain of blocks which will be laid out contiguously. |
| 82 | /// |
| 83 | /// This is the datastructure representing a chain of consecutive blocks that |
| 84 | /// are profitable to layout together in order to maximize fallthrough |
| 85 | /// probabilities. We also can use a block chain to represent a sequence of |
| 86 | /// basic blocks which have some external (correctness) requirement for |
| 87 | /// sequential layout. |
| 88 | /// |
| 89 | /// Eventually, the block chains will form a directed graph over the function. |
| 90 | /// We provide an SCC-supporting-iterator in order to quicky build and walk the |
| 91 | /// SCCs of block chains within a function. |
| 92 | /// |
| 93 | /// The block chains also have support for calculating and caching probability |
| 94 | /// information related to the chain itself versus other chains. This is used |
| 95 | /// for ranking during the final layout of block chains. |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 96 | class BlockChain { |
| 97 | /// \brief The sequence of blocks belonging to this chain. |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 98 | /// |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 99 | /// This is the sequence of blocks for a particular chain. These will be laid |
| 100 | /// out in-order within the function. |
| 101 | SmallVector<MachineBasicBlock *, 4> Blocks; |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 102 | |
| 103 | /// \brief A handle to the function-wide basic block to block chain mapping. |
| 104 | /// |
| 105 | /// This is retained in each block chain to simplify the computation of child |
| 106 | /// block chains for SCC-formation and iteration. We store the edges to child |
| 107 | /// basic blocks, and map them back to their associated chains using this |
| 108 | /// structure. |
| 109 | BlockToChainMapType &BlockToChain; |
| 110 | |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 111 | public: |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 112 | /// \brief Construct a new BlockChain. |
| 113 | /// |
| 114 | /// This builds a new block chain representing a single basic block in the |
| 115 | /// function. It also registers itself as the chain that block participates |
| 116 | /// in with the BlockToChain mapping. |
| 117 | BlockChain(BlockToChainMapType &BlockToChain, MachineBasicBlock *BB) |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 118 | : Blocks(1, BB), BlockToChain(BlockToChain) { |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 119 | assert(BB && "Cannot create a chain with a null basic block"); |
| 120 | BlockToChain[BB] = this; |
| 121 | } |
| 122 | |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 123 | /// \brief Iterator over blocks within the chain. |
| 124 | typedef SmallVectorImpl<MachineBasicBlock *>::const_iterator iterator; |
| 125 | |
| 126 | /// \brief Beginning of blocks within the chain. |
| 127 | iterator begin() const { return Blocks.begin(); } |
| 128 | |
| 129 | /// \brief End of blocks within the chain. |
| 130 | iterator end() const { return Blocks.end(); } |
| 131 | |
| 132 | /// \brief Merge a block chain into this one. |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 133 | /// |
| 134 | /// This routine merges a block chain into this one. It takes care of forming |
| 135 | /// a contiguous sequence of basic blocks, updating the edge list, and |
| 136 | /// updating the block -> chain mapping. It does not free or tear down the |
| 137 | /// old chain, but the old chain's block list is no longer valid. |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 138 | void merge(MachineBasicBlock *BB, BlockChain *Chain) { |
| 139 | assert(BB); |
| 140 | assert(!Blocks.empty()); |
| 141 | assert(Blocks.back()->isSuccessor(BB)); |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 142 | |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 143 | // Fast path in case we don't have a chain already. |
| 144 | if (!Chain) { |
| 145 | assert(!BlockToChain[BB]); |
| 146 | Blocks.push_back(BB); |
| 147 | BlockToChain[BB] = this; |
| 148 | return; |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 149 | } |
| 150 | |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 151 | assert(BB == *Chain->begin()); |
| 152 | assert(Chain->begin() != Chain->end()); |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 153 | |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 154 | // Update the incoming blocks to point to this chain, and add them to the |
| 155 | // chain structure. |
| 156 | for (BlockChain::iterator BI = Chain->begin(), BE = Chain->end(); |
| 157 | BI != BE; ++BI) { |
| 158 | Blocks.push_back(*BI); |
| 159 | assert(BlockToChain[*BI] == Chain && "Incoming blocks not in chain"); |
| 160 | BlockToChain[*BI] = this; |
| 161 | } |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 162 | } |
| 163 | }; |
| 164 | } |
| 165 | |
| 166 | namespace { |
| 167 | class MachineBlockPlacement : public MachineFunctionPass { |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 168 | /// \brief A typedef for a block filter set. |
| 169 | typedef SmallPtrSet<MachineBasicBlock *, 16> BlockFilterSet; |
| 170 | |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 171 | /// \brief A handle to the branch probability pass. |
| 172 | const MachineBranchProbabilityInfo *MBPI; |
| 173 | |
| 174 | /// \brief A handle to the function-wide block frequency pass. |
| 175 | const MachineBlockFrequencyInfo *MBFI; |
| 176 | |
Chandler Carruth | 4a85cc9 | 2011-10-21 08:57:37 +0000 | [diff] [blame] | 177 | /// \brief A handle to the loop info. |
| 178 | const MachineLoopInfo *MLI; |
| 179 | |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 180 | /// \brief A handle to the target's instruction info. |
| 181 | const TargetInstrInfo *TII; |
| 182 | |
Chandler Carruth | 4a85cc9 | 2011-10-21 08:57:37 +0000 | [diff] [blame] | 183 | /// \brief A handle to the target's lowering info. |
| 184 | const TargetLowering *TLI; |
| 185 | |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 186 | /// \brief Allocator and owner of BlockChain structures. |
| 187 | /// |
| 188 | /// We build BlockChains lazily by merging together high probability BB |
| 189 | /// sequences acording to the "Algo2" in the paper mentioned at the top of |
| 190 | /// the file. To reduce malloc traffic, we allocate them using this slab-like |
| 191 | /// allocator, and destroy them after the pass completes. |
| 192 | SpecificBumpPtrAllocator<BlockChain> ChainAllocator; |
| 193 | |
| 194 | /// \brief Function wide BasicBlock to BlockChain mapping. |
| 195 | /// |
| 196 | /// This mapping allows efficiently moving from any given basic block to the |
| 197 | /// BlockChain it participates in, if any. We use it to, among other things, |
| 198 | /// allow implicitly defining edges between chains as the existing edges |
| 199 | /// between basic blocks. |
| 200 | DenseMap<MachineBasicBlock *, BlockChain *> BlockToChain; |
| 201 | |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 202 | BlockChain *CreateChain(MachineBasicBlock *BB); |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 203 | void mergeSuccessor(MachineBasicBlock *BB, BlockChain *Chain, |
| 204 | BlockFilterSet *Filter = 0); |
| 205 | void buildLoopChains(MachineFunction &F, MachineLoop &L); |
| 206 | void buildCFGChains(MachineFunction &F); |
| 207 | void placeChainsTopologically(MachineFunction &F); |
Chandler Carruth | 4a85cc9 | 2011-10-21 08:57:37 +0000 | [diff] [blame] | 208 | void AlignLoops(MachineFunction &F); |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 209 | |
| 210 | public: |
| 211 | static char ID; // Pass identification, replacement for typeid |
| 212 | MachineBlockPlacement() : MachineFunctionPass(ID) { |
| 213 | initializeMachineBlockPlacementPass(*PassRegistry::getPassRegistry()); |
| 214 | } |
| 215 | |
| 216 | bool runOnMachineFunction(MachineFunction &F); |
| 217 | |
| 218 | void getAnalysisUsage(AnalysisUsage &AU) const { |
| 219 | AU.addRequired<MachineBranchProbabilityInfo>(); |
| 220 | AU.addRequired<MachineBlockFrequencyInfo>(); |
Chandler Carruth | 4a85cc9 | 2011-10-21 08:57:37 +0000 | [diff] [blame] | 221 | AU.addRequired<MachineLoopInfo>(); |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 222 | MachineFunctionPass::getAnalysisUsage(AU); |
| 223 | } |
| 224 | |
| 225 | const char *getPassName() const { return "Block Placement"; } |
| 226 | }; |
| 227 | } |
| 228 | |
| 229 | char MachineBlockPlacement::ID = 0; |
| 230 | INITIALIZE_PASS_BEGIN(MachineBlockPlacement, "block-placement2", |
| 231 | "Branch Probability Basic Block Placement", false, false) |
| 232 | INITIALIZE_PASS_DEPENDENCY(MachineBranchProbabilityInfo) |
| 233 | INITIALIZE_PASS_DEPENDENCY(MachineBlockFrequencyInfo) |
Chandler Carruth | 4a85cc9 | 2011-10-21 08:57:37 +0000 | [diff] [blame] | 234 | INITIALIZE_PASS_DEPENDENCY(MachineLoopInfo) |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 235 | INITIALIZE_PASS_END(MachineBlockPlacement, "block-placement2", |
| 236 | "Branch Probability Basic Block Placement", false, false) |
| 237 | |
| 238 | FunctionPass *llvm::createMachineBlockPlacementPass() { |
| 239 | return new MachineBlockPlacement(); |
| 240 | } |
| 241 | |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 242 | #ifndef NDEBUG |
| 243 | /// \brief Helper to print the name of a MBB. |
| 244 | /// |
| 245 | /// Only used by debug logging. |
| 246 | static std::string getBlockName(MachineBasicBlock *BB) { |
| 247 | std::string Result; |
| 248 | raw_string_ostream OS(Result); |
| 249 | OS << "BB#" << BB->getNumber() |
| 250 | << " (derived from LLVM BB '" << BB->getName() << "')"; |
| 251 | OS.flush(); |
| 252 | return Result; |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 253 | } |
| 254 | |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 255 | /// \brief Helper to print the number of a MBB. |
| 256 | /// |
| 257 | /// Only used by debug logging. |
| 258 | static std::string getBlockNum(MachineBasicBlock *BB) { |
| 259 | std::string Result; |
| 260 | raw_string_ostream OS(Result); |
| 261 | OS << "BB#" << BB->getNumber(); |
| 262 | OS.flush(); |
| 263 | return Result; |
| 264 | } |
| 265 | #endif |
| 266 | |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 267 | /// \brief Helper to create a new chain for a single BB. |
| 268 | /// |
| 269 | /// Takes care of growing the Chains, setting up the BlockChain object, and any |
| 270 | /// debug checking logic. |
| 271 | /// \returns A pointer to the new BlockChain. |
| 272 | BlockChain *MachineBlockPlacement::CreateChain(MachineBasicBlock *BB) { |
| 273 | BlockChain *Chain = |
| 274 | new (ChainAllocator.Allocate()) BlockChain(BlockToChain, BB); |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 275 | return Chain; |
| 276 | } |
| 277 | |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 278 | /// \brief Merge a chain with any viable successor. |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 279 | /// |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 280 | /// This routine walks the predecessors of the current block, looking for |
| 281 | /// viable merge candidates. It has strict rules it uses to determine when |
| 282 | /// a predecessor can be merged with the current block, which center around |
| 283 | /// preserving the CFG structure. It performs the merge if any viable candidate |
| 284 | /// is found. |
| 285 | void MachineBlockPlacement::mergeSuccessor(MachineBasicBlock *BB, |
| 286 | BlockChain *Chain, |
| 287 | BlockFilterSet *Filter) { |
| 288 | assert(BB); |
| 289 | assert(Chain); |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 290 | |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 291 | // If this block is not at the end of its chain, it cannot merge with any |
| 292 | // other chain. |
| 293 | if (Chain && *llvm::prior(Chain->end()) != BB) |
| 294 | return; |
| 295 | |
| 296 | // Walk through the successors looking for the highest probability edge. |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 297 | MachineBasicBlock *Successor = 0; |
Chandler Carruth | 66d847c | 2011-10-23 20:10:34 +0000 | [diff] [blame] | 298 | BranchProbability BestProb = BranchProbability::getZero(); |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 299 | DEBUG(dbgs() << "Attempting merge from: " << getBlockName(BB) << "\n"); |
| 300 | for (MachineBasicBlock::succ_iterator SI = BB->succ_begin(), |
| 301 | SE = BB->succ_end(); |
| 302 | SI != SE; ++SI) { |
| 303 | if (BB == *SI || (Filter && !Filter->count(*SI))) |
| 304 | continue; |
| 305 | |
Chandler Carruth | 66d847c | 2011-10-23 20:10:34 +0000 | [diff] [blame] | 306 | BranchProbability SuccProb = MBPI->getEdgeProbability(BB, *SI); |
| 307 | DEBUG(dbgs() << " " << getBlockName(*SI) << " -> " << SuccProb << "\n"); |
| 308 | if (!Successor || SuccProb > BestProb || (!(SuccProb < BestProb) && |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 309 | BB->isLayoutSuccessor(*SI))) { |
| 310 | Successor = *SI; |
Chandler Carruth | 66d847c | 2011-10-23 20:10:34 +0000 | [diff] [blame] | 311 | BestProb = SuccProb; |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 312 | } |
| 313 | } |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 314 | if (!Successor) |
| 315 | return; |
| 316 | |
| 317 | // Grab a chain if it exists already for this successor and make sure the |
| 318 | // successor is at the start of the chain as we can't merge mid-chain. Also, |
| 319 | // if the successor chain is the same as our chain, we're already merged. |
| 320 | BlockChain *SuccChain = BlockToChain[Successor]; |
| 321 | if (SuccChain && (SuccChain == Chain || Successor != *SuccChain->begin())) |
| 322 | return; |
| 323 | |
| 324 | // We only merge chains across a CFG merge when the desired merge path is |
| 325 | // significantly hotter than the incoming edge. We define a hot edge more |
| 326 | // strictly than the BranchProbabilityInfo does, as the two predecessor |
| 327 | // blocks may have dramatically different incoming probabilities we need to |
| 328 | // account for. Therefor we use the "global" edge weight which is the |
| 329 | // branch's probability times the block frequency of the predecessor. |
| 330 | BlockFrequency MergeWeight = MBFI->getBlockFreq(BB); |
| 331 | MergeWeight *= MBPI->getEdgeProbability(BB, Successor); |
| 332 | // We only want to consider breaking the CFG when the merge weight is much |
| 333 | // higher (80% vs. 20%), so multiply it by 1/4. This will require the merged |
| 334 | // edge to be 4x more likely before we disrupt the CFG. This number matches |
| 335 | // the definition of "hot" in BranchProbabilityAnalysis (80% vs. 20%). |
| 336 | MergeWeight *= BranchProbability(1, 4); |
| 337 | for (MachineBasicBlock::pred_iterator PI = Successor->pred_begin(), |
| 338 | PE = Successor->pred_end(); |
| 339 | PI != PE; ++PI) { |
| 340 | if (BB == *PI || Successor == *PI) continue; |
| 341 | BlockFrequency PredWeight = MBFI->getBlockFreq(*PI); |
| 342 | PredWeight *= MBPI->getEdgeProbability(*PI, Successor); |
| 343 | |
| 344 | // Return on the first predecessor we find which outstrips our merge weight. |
| 345 | if (MergeWeight < PredWeight) |
| 346 | return; |
| 347 | DEBUG(dbgs() << "Breaking CFG edge!\n" |
| 348 | << " Edge from " << getBlockNum(BB) << " to " |
| 349 | << getBlockNum(Successor) << ": " << MergeWeight << "\n" |
| 350 | << " vs. " << getBlockNum(BB) << " to " |
| 351 | << getBlockNum(*PI) << ": " << PredWeight << "\n"); |
| 352 | } |
| 353 | |
| 354 | DEBUG(dbgs() << "Merging from " << getBlockNum(BB) << " to " |
| 355 | << getBlockNum(Successor) << "\n"); |
| 356 | Chain->merge(Successor, SuccChain); |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 357 | } |
| 358 | |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 359 | /// \brief Forms basic block chains from the natural loop structures. |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 360 | /// |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 361 | /// These chains are designed to preserve the existing *structure* of the code |
| 362 | /// as much as possible. We can then stitch the chains together in a way which |
| 363 | /// both preserves the topological structure and minimizes taken conditional |
| 364 | /// branches. |
| 365 | void MachineBlockPlacement::buildLoopChains(MachineFunction &F, MachineLoop &L) { |
| 366 | // First recurse through any nested loops, building chains for those inner |
| 367 | // loops. |
| 368 | for (MachineLoop::iterator LI = L.begin(), LE = L.end(); LI != LE; ++LI) |
| 369 | buildLoopChains(F, **LI); |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 370 | |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 371 | SmallPtrSet<MachineBasicBlock *, 16> LoopBlockSet(L.block_begin(), |
| 372 | L.block_end()); |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 373 | |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 374 | // Begin building up a set of chains of blocks within this loop which should |
| 375 | // remain contiguous. Some of the blocks already belong to a chain which |
| 376 | // represents an inner loop. |
| 377 | for (MachineLoop::block_iterator BI = L.block_begin(), BE = L.block_end(); |
| 378 | BI != BE; ++BI) { |
| 379 | MachineBasicBlock *BB = *BI; |
| 380 | BlockChain *Chain = BlockToChain[BB]; |
| 381 | if (!Chain) Chain = CreateChain(BB); |
| 382 | mergeSuccessor(BB, Chain, &LoopBlockSet); |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 383 | } |
| 384 | } |
| 385 | |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 386 | void MachineBlockPlacement::buildCFGChains(MachineFunction &F) { |
| 387 | // First build any loop-based chains. |
| 388 | for (MachineLoopInfo::iterator LI = MLI->begin(), LE = MLI->end(); LI != LE; |
| 389 | ++LI) |
| 390 | buildLoopChains(F, **LI); |
| 391 | |
| 392 | // Now walk the blocks of the function forming chains where they don't |
| 393 | // violate any CFG structure. |
| 394 | for (MachineFunction::iterator BI = F.begin(), BE = F.end(); |
| 395 | BI != BE; ++BI) { |
| 396 | MachineBasicBlock *BB = BI; |
| 397 | BlockChain *Chain = BlockToChain[BB]; |
| 398 | if (!Chain) Chain = CreateChain(BB); |
| 399 | mergeSuccessor(BB, Chain); |
| 400 | } |
| 401 | } |
| 402 | |
| 403 | void MachineBlockPlacement::placeChainsTopologically(MachineFunction &F) { |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 404 | MachineBasicBlock *EntryB = &F.front(); |
Chandler Carruth | e4617c0 | 2011-10-24 16:51:55 +0000 | [diff] [blame] | 405 | assert(BlockToChain[EntryB] && "Missing chain for entry block"); |
| 406 | assert(*BlockToChain[EntryB]->begin() == EntryB && |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 407 | "Entry block is not the head of the entry block chain"); |
| 408 | |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 409 | // Walk the blocks in RPO, and insert each block for a chain in order the |
| 410 | // first time we see that chain. |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 411 | MachineFunction::iterator InsertPos = F.begin(); |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 412 | SmallPtrSet<BlockChain *, 16> VisitedChains; |
| 413 | ReversePostOrderTraversal<MachineBasicBlock *> RPOT(EntryB); |
| 414 | typedef ReversePostOrderTraversal<MachineBasicBlock *>::rpo_iterator |
| 415 | rpo_iterator; |
| 416 | for (rpo_iterator I = RPOT.begin(), E = RPOT.end(); I != E; ++I) { |
| 417 | BlockChain *Chain = BlockToChain[*I]; |
| 418 | assert(Chain); |
| 419 | if(!VisitedChains.insert(Chain)) |
| 420 | continue; |
| 421 | for (BlockChain::iterator BI = Chain->begin(), BE = Chain->end(); BI != BE; |
| 422 | ++BI) { |
| 423 | DEBUG(dbgs() << (BI == Chain->begin() ? "Placing chain " |
| 424 | : " ... ") |
| 425 | << getBlockName(*BI) << "\n"); |
| 426 | if (InsertPos != MachineFunction::iterator(*BI)) |
| 427 | F.splice(InsertPos, *BI); |
| 428 | else |
| 429 | ++InsertPos; |
| 430 | } |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 431 | } |
| 432 | |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 433 | // Now that every block is in its final position, update all of the |
| 434 | // terminators. |
| 435 | SmallVector<MachineOperand, 4> Cond; // For AnalyzeBranch. |
| 436 | for (MachineFunction::iterator FI = F.begin(), FE = F.end(); FI != FE; ++FI) { |
| 437 | // FIXME: It would be awesome of updateTerminator would just return rather |
| 438 | // than assert when the branch cannot be analyzed in order to remove this |
| 439 | // boiler plate. |
| 440 | Cond.clear(); |
| 441 | MachineBasicBlock *TBB = 0, *FBB = 0; // For AnalyzeBranch. |
| 442 | if (!TII->AnalyzeBranch(*FI, TBB, FBB, Cond)) |
| 443 | FI->updateTerminator(); |
| 444 | } |
| 445 | } |
| 446 | |
Chandler Carruth | 4a85cc9 | 2011-10-21 08:57:37 +0000 | [diff] [blame] | 447 | /// \brief Recursive helper to align a loop and any nested loops. |
| 448 | static void AlignLoop(MachineFunction &F, MachineLoop *L, unsigned Align) { |
| 449 | // Recurse through nested loops. |
| 450 | for (MachineLoop::iterator I = L->begin(), E = L->end(); I != E; ++I) |
| 451 | AlignLoop(F, *I, Align); |
| 452 | |
| 453 | L->getTopBlock()->setAlignment(Align); |
| 454 | } |
| 455 | |
| 456 | /// \brief Align loop headers to target preferred alignments. |
| 457 | void MachineBlockPlacement::AlignLoops(MachineFunction &F) { |
| 458 | if (F.getFunction()->hasFnAttr(Attribute::OptimizeForSize)) |
| 459 | return; |
| 460 | |
| 461 | unsigned Align = TLI->getPrefLoopAlignment(); |
| 462 | if (!Align) |
| 463 | return; // Don't care about loop alignment. |
| 464 | |
| 465 | for (MachineLoopInfo::iterator I = MLI->begin(), E = MLI->end(); I != E; ++I) |
| 466 | AlignLoop(F, *I, Align); |
| 467 | } |
| 468 | |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 469 | bool MachineBlockPlacement::runOnMachineFunction(MachineFunction &F) { |
| 470 | // Check for single-block functions and skip them. |
| 471 | if (llvm::next(F.begin()) == F.end()) |
| 472 | return false; |
| 473 | |
| 474 | MBPI = &getAnalysis<MachineBranchProbabilityInfo>(); |
| 475 | MBFI = &getAnalysis<MachineBlockFrequencyInfo>(); |
Chandler Carruth | 4a85cc9 | 2011-10-21 08:57:37 +0000 | [diff] [blame] | 476 | MLI = &getAnalysis<MachineLoopInfo>(); |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 477 | TII = F.getTarget().getInstrInfo(); |
Chandler Carruth | 4a85cc9 | 2011-10-21 08:57:37 +0000 | [diff] [blame] | 478 | TLI = F.getTarget().getTargetLowering(); |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 479 | assert(BlockToChain.empty()); |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 480 | |
Chandler Carruth | 3071363 | 2011-10-23 09:18:45 +0000 | [diff] [blame] | 481 | buildCFGChains(F); |
| 482 | placeChainsTopologically(F); |
Chandler Carruth | 4a85cc9 | 2011-10-21 08:57:37 +0000 | [diff] [blame] | 483 | AlignLoops(F); |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 484 | |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 485 | BlockToChain.clear(); |
Chandler Carruth | db35087 | 2011-10-21 06:46:38 +0000 | [diff] [blame] | 486 | |
| 487 | // We always return true as we have no way to track whether the final order |
| 488 | // differs from the original order. |
| 489 | return true; |
| 490 | } |
Chandler Carruth | 37efc9f | 2011-11-02 07:17:12 +0000 | [diff] [blame^] | 491 | |
| 492 | namespace { |
| 493 | /// \brief A pass to compute block placement statistics. |
| 494 | /// |
| 495 | /// A separate pass to compute interesting statistics for evaluating block |
| 496 | /// placement. This is separate from the actual placement pass so that they can |
| 497 | /// be computed in the absense of any placement transformations or when using |
| 498 | /// alternative placement strategies. |
| 499 | class MachineBlockPlacementStats : public MachineFunctionPass { |
| 500 | /// \brief A handle to the branch probability pass. |
| 501 | const MachineBranchProbabilityInfo *MBPI; |
| 502 | |
| 503 | /// \brief A handle to the function-wide block frequency pass. |
| 504 | const MachineBlockFrequencyInfo *MBFI; |
| 505 | |
| 506 | public: |
| 507 | static char ID; // Pass identification, replacement for typeid |
| 508 | MachineBlockPlacementStats() : MachineFunctionPass(ID) { |
| 509 | initializeMachineBlockPlacementStatsPass(*PassRegistry::getPassRegistry()); |
| 510 | } |
| 511 | |
| 512 | bool runOnMachineFunction(MachineFunction &F); |
| 513 | |
| 514 | void getAnalysisUsage(AnalysisUsage &AU) const { |
| 515 | AU.addRequired<MachineBranchProbabilityInfo>(); |
| 516 | AU.addRequired<MachineBlockFrequencyInfo>(); |
| 517 | AU.setPreservesAll(); |
| 518 | MachineFunctionPass::getAnalysisUsage(AU); |
| 519 | } |
| 520 | |
| 521 | const char *getPassName() const { return "Block Placement Stats"; } |
| 522 | }; |
| 523 | } |
| 524 | |
| 525 | char MachineBlockPlacementStats::ID = 0; |
| 526 | INITIALIZE_PASS_BEGIN(MachineBlockPlacementStats, "block-placement-stats", |
| 527 | "Basic Block Placement Stats", false, false) |
| 528 | INITIALIZE_PASS_DEPENDENCY(MachineBranchProbabilityInfo) |
| 529 | INITIALIZE_PASS_DEPENDENCY(MachineBlockFrequencyInfo) |
| 530 | INITIALIZE_PASS_END(MachineBlockPlacementStats, "block-placement-stats", |
| 531 | "Basic Block Placement Stats", false, false) |
| 532 | |
| 533 | FunctionPass *llvm::createMachineBlockPlacementStatsPass() { |
| 534 | return new MachineBlockPlacementStats(); |
| 535 | } |
| 536 | |
| 537 | bool MachineBlockPlacementStats::runOnMachineFunction(MachineFunction &F) { |
| 538 | // Check for single-block functions and skip them. |
| 539 | if (llvm::next(F.begin()) == F.end()) |
| 540 | return false; |
| 541 | |
| 542 | MBPI = &getAnalysis<MachineBranchProbabilityInfo>(); |
| 543 | MBFI = &getAnalysis<MachineBlockFrequencyInfo>(); |
| 544 | |
| 545 | for (MachineFunction::iterator I = F.begin(), E = F.end(); I != E; ++I) { |
| 546 | BlockFrequency BlockFreq = MBFI->getBlockFreq(I); |
| 547 | Statistic &NumBranches = (I->succ_size() > 1) ? NumCondBranches |
| 548 | : NumUncondBranches; |
| 549 | Statistic &BranchTakenFreq = (I->succ_size() > 1) ? CondBranchTakenFreq |
| 550 | : UncondBranchTakenFreq; |
| 551 | for (MachineBasicBlock::succ_iterator SI = I->succ_begin(), |
| 552 | SE = I->succ_end(); |
| 553 | SI != SE; ++SI) { |
| 554 | // Skip if this successor is a fallthrough. |
| 555 | if (I->isLayoutSuccessor(*SI)) |
| 556 | continue; |
| 557 | |
| 558 | BlockFrequency EdgeFreq = BlockFreq * MBPI->getEdgeProbability(I, *SI); |
| 559 | ++NumBranches; |
| 560 | BranchTakenFreq += EdgeFreq.getFrequency(); |
| 561 | } |
| 562 | } |
| 563 | |
| 564 | return false; |
| 565 | } |
| 566 | |