// structure and branch probability estimates.
//
// The pass strives to preserve the structure of the CFG (that is, retain
-// a topological ordering of basic blocks) in the absense of a *strong* signal
+// a topological ordering of basic blocks) in the absence of a *strong* signal
// to the contrary from probabilities. However, within the CFG structure, it
// attempts to choose an ordering which favors placing more likely sequences of
// blocks adjacent to each other.
//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "block-placement2"
+#include "llvm/CodeGen/Passes.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/Statistic.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
#include "llvm/CodeGen/MachineBlockFrequencyInfo.h"
#include "llvm/CodeGen/MachineBranchProbabilityInfo.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineLoopInfo.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
-#include "llvm/CodeGen/Passes.h"
#include "llvm/Support/Allocator.h"
+#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
-#include "llvm/Support/ErrorHandling.h"
-#include "llvm/ADT/DenseMap.h"
-#include "llvm/ADT/PostOrderIterator.h"
-#include "llvm/ADT/SCCIterator.h"
-#include "llvm/ADT/SmallPtrSet.h"
-#include "llvm/ADT/SmallVector.h"
-#include "llvm/ADT/Statistic.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetLowering.h"
+#include "llvm/Target/TargetSubtargetInfo.h"
#include <algorithm>
using namespace llvm;
+#define DEBUG_TYPE "block-placement2"
+
STATISTIC(NumCondBranches, "Number of conditional branches");
STATISTIC(NumUncondBranches, "Number of uncondittional branches");
STATISTIC(CondBranchTakenFreq,
STATISTIC(UncondBranchTakenFreq,
"Potential frequency of taking unconditional branches");
-namespace {
-/// \brief A structure for storing a weighted edge.
-///
-/// This stores an edge and its weight, computed as the product of the
-/// frequency that the starting block is entered with the probability of
-/// a particular exit block.
-struct WeightedEdge {
- BlockFrequency EdgeFrequency;
- MachineBasicBlock *From, *To;
-
- bool operator<(const WeightedEdge &RHS) const {
- return EdgeFrequency < RHS.EdgeFrequency;
- }
-};
-}
+static cl::opt<unsigned> AlignAllBlock("align-all-blocks",
+ cl::desc("Force the alignment of all "
+ "blocks in the function."),
+ cl::init(0), cl::Hidden);
+
+// FIXME: Find a good default for this flag and remove the flag.
+static cl::opt<unsigned>
+ExitBlockBias("block-placement-exit-block-bias",
+ cl::desc("Block frequency percentage a loop exit block needs "
+ "over the original exit to be considered the new exit."),
+ cl::init(0), cl::Hidden);
namespace {
class BlockChain;
///
/// This is the datastructure representing a chain of consecutive blocks that
/// are profitable to layout together in order to maximize fallthrough
-/// probabilities. We also can use a block chain to represent a sequence of
-/// basic blocks which have some external (correctness) requirement for
-/// sequential layout.
-///
-/// Eventually, the block chains will form a directed graph over the function.
-/// We provide an SCC-supporting-iterator in order to quicky build and walk the
-/// SCCs of block chains within a function.
+/// probabilities and code locality. We also can use a block chain to represent
+/// a sequence of basic blocks which have some external (correctness)
+/// requirement for sequential layout.
///
-/// The block chains also have support for calculating and caching probability
-/// information related to the chain itself versus other chains. This is used
-/// for ranking during the final layout of block chains.
+/// Chains can be built around a single basic block and can be merged to grow
+/// them. They participate in a block-to-chain mapping, which is updated
+/// automatically as chains are merged together.
class BlockChain {
/// \brief The sequence of blocks belonging to this chain.
///
}
/// \brief Iterator over blocks within the chain.
- typedef SmallVectorImpl<MachineBasicBlock *>::const_iterator iterator;
+ typedef SmallVectorImpl<MachineBasicBlock *>::iterator iterator;
/// \brief Beginning of blocks within the chain.
- iterator begin() const { return Blocks.begin(); }
+ iterator begin() { return Blocks.begin(); }
/// \brief End of blocks within the chain.
- iterator end() const { return Blocks.end(); }
+ iterator end() { return Blocks.end(); }
/// \brief Merge a block chain into this one.
///
}
}
+#ifndef NDEBUG
+ /// \brief Dump the blocks in this chain.
+ LLVM_DUMP_METHOD void dump() {
+ for (iterator I = begin(), E = end(); I != E; ++I)
+ (*I)->dump();
+ }
+#endif // NDEBUG
+
/// \brief Count of predecessors within the loop currently being processed.
///
/// This count is updated at each loop we process to represent the number of
const TargetInstrInfo *TII;
/// \brief A handle to the target's lowering info.
- const TargetLowering *TLI;
+ const TargetLoweringBase *TLI;
/// \brief Allocator and owner of BlockChain structures.
///
- /// We build BlockChains lazily by merging together high probability BB
- /// sequences acording to the "Algo2" in the paper mentioned at the top of
- /// the file. To reduce malloc traffic, we allocate them using this slab-like
- /// allocator, and destroy them after the pass completes.
+ /// We build BlockChains lazily while processing the loop structure of
+ /// a function. To reduce malloc traffic, we allocate them using this
+ /// slab-like allocator, and destroy them after the pass completes. An
+ /// important guarantee is that this allocator produces stable pointers to
+ /// the chains.
SpecificBumpPtrAllocator<BlockChain> ChainAllocator;
/// \brief Function wide BasicBlock to BlockChain mapping.
void markChainSuccessors(BlockChain &Chain,
MachineBasicBlock *LoopHeaderBB,
SmallVectorImpl<MachineBasicBlock *> &BlockWorkList,
- const BlockFilterSet *BlockFilter = 0);
+ const BlockFilterSet *BlockFilter = nullptr);
MachineBasicBlock *selectBestSuccessor(MachineBasicBlock *BB,
BlockChain &Chain,
const BlockFilterSet *BlockFilter);
const BlockFilterSet *BlockFilter);
void buildChain(MachineBasicBlock *BB, BlockChain &Chain,
SmallVectorImpl<MachineBasicBlock *> &BlockWorkList,
- const BlockFilterSet *BlockFilter = 0);
+ const BlockFilterSet *BlockFilter = nullptr);
+ MachineBasicBlock *findBestLoopTop(MachineLoop &L,
+ const BlockFilterSet &LoopBlockSet);
+ MachineBasicBlock *findBestLoopExit(MachineFunction &F,
+ MachineLoop &L,
+ const BlockFilterSet &LoopBlockSet);
void buildLoopChains(MachineFunction &F, MachineLoop &L);
+ void rotateLoop(BlockChain &LoopChain, MachineBasicBlock *ExitingBB,
+ const BlockFilterSet &LoopBlockSet);
void buildCFGChains(MachineFunction &F);
- void AlignLoops(MachineFunction &F);
public:
static char ID; // Pass identification, replacement for typeid
initializeMachineBlockPlacementPass(*PassRegistry::getPassRegistry());
}
- bool runOnMachineFunction(MachineFunction &F);
+ bool runOnMachineFunction(MachineFunction &F) override;
- void getAnalysisUsage(AnalysisUsage &AU) const {
+ void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.addRequired<MachineBranchProbabilityInfo>();
AU.addRequired<MachineBlockFrequencyInfo>();
AU.addRequired<MachineLoopInfo>();
MachineFunctionPass::getAnalysisUsage(AU);
}
-
- const char *getPassName() const { return "Block Placement"; }
};
}
char MachineBlockPlacement::ID = 0;
+char &llvm::MachineBlockPlacementID = MachineBlockPlacement::ID;
INITIALIZE_PASS_BEGIN(MachineBlockPlacement, "block-placement2",
"Branch Probability Basic Block Placement", false, false)
INITIALIZE_PASS_DEPENDENCY(MachineBranchProbabilityInfo)
INITIALIZE_PASS_END(MachineBlockPlacement, "block-placement2",
"Branch Probability Basic Block Placement", false, false)
-FunctionPass *llvm::createMachineBlockPlacementPass() {
- return new MachineBlockPlacement();
-}
-
#ifndef NDEBUG
/// \brief Helper to print the name of a MBB.
///
const BlockFilterSet *BlockFilter) {
const BranchProbability HotProb(4, 5); // 80%
- MachineBasicBlock *BestSucc = 0;
+ MachineBasicBlock *BestSucc = nullptr;
// FIXME: Due to the performance of the probability and weight routines in
// the MBPI analysis, we manually compute probabilities using the edge
// weights. This is suboptimal as it means that the somewhat subtle
// definition of edge weight semantics is encoded here as well. We should
- // improve the MBPI interface to effeciently support query patterns such as
+ // improve the MBPI interface to efficiently support query patterns such as
// this.
uint32_t BestWeight = 0;
uint32_t WeightScale = 0;
// any CFG constraints.
if (SuccChain.LoopPredecessors != 0) {
if (SuccProb < HotProb) {
- DEBUG(dbgs() << " " << getBlockName(*SI) << " -> CFG conflict\n");
+ DEBUG(dbgs() << " " << getBlockName(*SI) << " -> " << SuccProb
+ << " (prob) (CFG conflict)\n");
continue;
}
}
}
if (BadCFGConflict) {
- DEBUG(dbgs() << " " << getBlockName(*SI)
- << " -> non-cold CFG conflict\n");
+ DEBUG(dbgs() << " " << getBlockName(*SI) << " -> " << SuccProb
+ << " (prob) (non-cold CFG conflict)\n");
continue;
}
}
return BestSucc;
}
-namespace {
-/// \brief Predicate struct to detect blocks already placed.
-class IsBlockPlaced {
- const BlockChain &PlacedChain;
- const BlockToChainMapType &BlockToChain;
-
-public:
- IsBlockPlaced(const BlockChain &PlacedChain,
- const BlockToChainMapType &BlockToChain)
- : PlacedChain(PlacedChain), BlockToChain(BlockToChain) {}
-
- bool operator()(MachineBasicBlock *BB) const {
- return BlockToChain.lookup(BB) == &PlacedChain;
- }
-};
-}
-
/// \brief Select the best block from a worklist.
///
/// This looks through the provided worklist as a list of candidate basic
// FIXME: If this shows up on profiles, it could be folded (at the cost of
// some code complexity) into the loop below.
WorkList.erase(std::remove_if(WorkList.begin(), WorkList.end(),
- IsBlockPlaced(Chain, BlockToChain)),
+ [&](MachineBasicBlock *BB) {
+ return BlockToChain.lookup(BB) == &Chain;
+ }),
WorkList.end());
- MachineBasicBlock *BestBlock = 0;
+ MachineBasicBlock *BestBlock = nullptr;
BlockFrequency BestFreq;
for (SmallVectorImpl<MachineBasicBlock *>::iterator WBI = WorkList.begin(),
WBE = WorkList.end();
WBI != WBE; ++WBI) {
- assert(!BlockFilter || BlockFilter->count(*WBI));
BlockChain &SuccChain = *BlockToChain[*WBI];
if (&SuccChain == &Chain) {
DEBUG(dbgs() << " " << getBlockName(*WBI)
assert(SuccChain.LoopPredecessors == 0 && "Found CFG-violating block");
BlockFrequency CandidateFreq = MBFI->getBlockFreq(*WBI);
- DEBUG(dbgs() << " " << getBlockName(*WBI) << " -> " << CandidateFreq
- << " (freq)\n");
+ DEBUG(dbgs() << " " << getBlockName(*WBI) << " -> ";
+ MBFI->printBlockFreq(dbgs(), CandidateFreq) << " (freq)\n");
if (BestBlock && BestFreq >= CandidateFreq)
continue;
BestBlock = *WBI;
return *BlockToChain[I]->begin();
}
}
- return 0;
+ return nullptr;
}
void MachineBlockPlacement::buildChain(
MachineBasicBlock *LoopHeaderBB = BB;
markChainSuccessors(Chain, LoopHeaderBB, BlockWorkList, BlockFilter);
- BB = *llvm::prior(Chain.end());
+ BB = *std::prev(Chain.end());
for (;;) {
assert(BB);
assert(BlockToChain[BB] == &Chain);
- assert(*llvm::prior(Chain.end()) == BB);
- MachineBasicBlock *BestSucc = 0;
+ assert(*std::prev(Chain.end()) == BB);
// Look for the best viable successor if there is one to place immediately
// after this block.
- BestSucc = selectBestSuccessor(BB, Chain, BlockFilter);
+ MachineBasicBlock *BestSucc = selectBestSuccessor(BB, Chain, BlockFilter);
// If an immediate successor isn't available, look for the best viable
// block among those we've identified as not violating the loop's CFG at
<< " to " << getBlockNum(BestSucc) << "\n");
markChainSuccessors(SuccChain, LoopHeaderBB, BlockWorkList, BlockFilter);
Chain.merge(BestSucc, &SuccChain);
- BB = *llvm::prior(Chain.end());
- };
+ BB = *std::prev(Chain.end());
+ }
DEBUG(dbgs() << "Finished forming chain for header block "
<< getBlockNum(*Chain.begin()) << "\n");
}
+/// \brief Find the best loop top block for layout.
+///
+/// Look for a block which is strictly better than the loop header for laying
+/// out at the top of the loop. This looks for one and only one pattern:
+/// a latch block with no conditional exit. This block will cause a conditional
+/// jump around it or will be the bottom of the loop if we lay it out in place,
+/// but if it it doesn't end up at the bottom of the loop for any reason,
+/// rotation alone won't fix it. Because such a block will always result in an
+/// unconditional jump (for the backedge) rotating it in front of the loop
+/// header is always profitable.
+MachineBasicBlock *
+MachineBlockPlacement::findBestLoopTop(MachineLoop &L,
+ const BlockFilterSet &LoopBlockSet) {
+ // Check that the header hasn't been fused with a preheader block due to
+ // crazy branches. If it has, we need to start with the header at the top to
+ // prevent pulling the preheader into the loop body.
+ BlockChain &HeaderChain = *BlockToChain[L.getHeader()];
+ if (!LoopBlockSet.count(*HeaderChain.begin()))
+ return L.getHeader();
+
+ DEBUG(dbgs() << "Finding best loop top for: "
+ << getBlockName(L.getHeader()) << "\n");
+
+ BlockFrequency BestPredFreq;
+ MachineBasicBlock *BestPred = nullptr;
+ for (MachineBasicBlock::pred_iterator PI = L.getHeader()->pred_begin(),
+ PE = L.getHeader()->pred_end();
+ PI != PE; ++PI) {
+ MachineBasicBlock *Pred = *PI;
+ if (!LoopBlockSet.count(Pred))
+ continue;
+ DEBUG(dbgs() << " header pred: " << getBlockName(Pred) << ", "
+ << Pred->succ_size() << " successors, ";
+ MBFI->printBlockFreq(dbgs(), Pred) << " freq\n");
+ if (Pred->succ_size() > 1)
+ continue;
+
+ BlockFrequency PredFreq = MBFI->getBlockFreq(Pred);
+ if (!BestPred || PredFreq > BestPredFreq ||
+ (!(PredFreq < BestPredFreq) &&
+ Pred->isLayoutSuccessor(L.getHeader()))) {
+ BestPred = Pred;
+ BestPredFreq = PredFreq;
+ }
+ }
+
+ // If no direct predecessor is fine, just use the loop header.
+ if (!BestPred)
+ return L.getHeader();
+
+ // Walk backwards through any straight line of predecessors.
+ while (BestPred->pred_size() == 1 &&
+ (*BestPred->pred_begin())->succ_size() == 1 &&
+ *BestPred->pred_begin() != L.getHeader())
+ BestPred = *BestPred->pred_begin();
+
+ DEBUG(dbgs() << " final top: " << getBlockName(BestPred) << "\n");
+ return BestPred;
+}
+
+
+/// \brief Find the best loop exiting block for layout.
+///
+/// This routine implements the logic to analyze the loop looking for the best
+/// block to layout at the top of the loop. Typically this is done to maximize
+/// fallthrough opportunities.
+MachineBasicBlock *
+MachineBlockPlacement::findBestLoopExit(MachineFunction &F,
+ MachineLoop &L,
+ const BlockFilterSet &LoopBlockSet) {
+ // We don't want to layout the loop linearly in all cases. If the loop header
+ // is just a normal basic block in the loop, we want to look for what block
+ // within the loop is the best one to layout at the top. However, if the loop
+ // header has be pre-merged into a chain due to predecessors not having
+ // analyzable branches, *and* the predecessor it is merged with is *not* part
+ // of the loop, rotating the header into the middle of the loop will create
+ // a non-contiguous range of blocks which is Very Bad. So start with the
+ // header and only rotate if safe.
+ BlockChain &HeaderChain = *BlockToChain[L.getHeader()];
+ if (!LoopBlockSet.count(*HeaderChain.begin()))
+ return nullptr;
+
+ BlockFrequency BestExitEdgeFreq;
+ unsigned BestExitLoopDepth = 0;
+ MachineBasicBlock *ExitingBB = nullptr;
+ // If there are exits to outer loops, loop rotation can severely limit
+ // fallthrough opportunites unless it selects such an exit. Keep a set of
+ // blocks where rotating to exit with that block will reach an outer loop.
+ SmallPtrSet<MachineBasicBlock *, 4> BlocksExitingToOuterLoop;
+
+ DEBUG(dbgs() << "Finding best loop exit for: "
+ << getBlockName(L.getHeader()) << "\n");
+ for (MachineLoop::block_iterator I = L.block_begin(),
+ E = L.block_end();
+ I != E; ++I) {
+ BlockChain &Chain = *BlockToChain[*I];
+ // Ensure that this block is at the end of a chain; otherwise it could be
+ // mid-way through an inner loop or a successor of an analyzable branch.
+ if (*I != *std::prev(Chain.end()))
+ continue;
+
+ // Now walk the successors. We need to establish whether this has a viable
+ // exiting successor and whether it has a viable non-exiting successor.
+ // We store the old exiting state and restore it if a viable looping
+ // successor isn't found.
+ MachineBasicBlock *OldExitingBB = ExitingBB;
+ BlockFrequency OldBestExitEdgeFreq = BestExitEdgeFreq;
+ bool HasLoopingSucc = false;
+ // FIXME: Due to the performance of the probability and weight routines in
+ // the MBPI analysis, we use the internal weights and manually compute the
+ // probabilities to avoid quadratic behavior.
+ uint32_t WeightScale = 0;
+ uint32_t SumWeight = MBPI->getSumForBlock(*I, WeightScale);
+ for (MachineBasicBlock::succ_iterator SI = (*I)->succ_begin(),
+ SE = (*I)->succ_end();
+ SI != SE; ++SI) {
+ if ((*SI)->isLandingPad())
+ continue;
+ if (*SI == *I)
+ continue;
+ BlockChain &SuccChain = *BlockToChain[*SI];
+ // Don't split chains, either this chain or the successor's chain.
+ if (&Chain == &SuccChain) {
+ DEBUG(dbgs() << " exiting: " << getBlockName(*I) << " -> "
+ << getBlockName(*SI) << " (chain conflict)\n");
+ continue;
+ }
+
+ uint32_t SuccWeight = MBPI->getEdgeWeight(*I, *SI);
+ if (LoopBlockSet.count(*SI)) {
+ DEBUG(dbgs() << " looping: " << getBlockName(*I) << " -> "
+ << getBlockName(*SI) << " (" << SuccWeight << ")\n");
+ HasLoopingSucc = true;
+ continue;
+ }
+
+ unsigned SuccLoopDepth = 0;
+ if (MachineLoop *ExitLoop = MLI->getLoopFor(*SI)) {
+ SuccLoopDepth = ExitLoop->getLoopDepth();
+ if (ExitLoop->contains(&L))
+ BlocksExitingToOuterLoop.insert(*I);
+ }
+
+ BranchProbability SuccProb(SuccWeight / WeightScale, SumWeight);
+ BlockFrequency ExitEdgeFreq = MBFI->getBlockFreq(*I) * SuccProb;
+ DEBUG(dbgs() << " exiting: " << getBlockName(*I) << " -> "
+ << getBlockName(*SI) << " [L:" << SuccLoopDepth
+ << "] (";
+ MBFI->printBlockFreq(dbgs(), ExitEdgeFreq) << ")\n");
+ // Note that we bias this toward an existing layout successor to retain
+ // incoming order in the absence of better information. The exit must have
+ // a frequency higher than the current exit before we consider breaking
+ // the layout.
+ BranchProbability Bias(100 - ExitBlockBias, 100);
+ if (!ExitingBB || BestExitLoopDepth < SuccLoopDepth ||
+ ExitEdgeFreq > BestExitEdgeFreq ||
+ ((*I)->isLayoutSuccessor(*SI) &&
+ !(ExitEdgeFreq < BestExitEdgeFreq * Bias))) {
+ BestExitEdgeFreq = ExitEdgeFreq;
+ ExitingBB = *I;
+ }
+ }
+
+ // Restore the old exiting state, no viable looping successor was found.
+ if (!HasLoopingSucc) {
+ ExitingBB = OldExitingBB;
+ BestExitEdgeFreq = OldBestExitEdgeFreq;
+ continue;
+ }
+ }
+ // Without a candidate exiting block or with only a single block in the
+ // loop, just use the loop header to layout the loop.
+ if (!ExitingBB || L.getNumBlocks() == 1)
+ return nullptr;
+
+ // Also, if we have exit blocks which lead to outer loops but didn't select
+ // one of them as the exiting block we are rotating toward, disable loop
+ // rotation altogether.
+ if (!BlocksExitingToOuterLoop.empty() &&
+ !BlocksExitingToOuterLoop.count(ExitingBB))
+ return nullptr;
+
+ DEBUG(dbgs() << " Best exiting block: " << getBlockName(ExitingBB) << "\n");
+ return ExitingBB;
+}
+
+/// \brief Attempt to rotate an exiting block to the bottom of the loop.
+///
+/// Once we have built a chain, try to rotate it to line up the hot exit block
+/// with fallthrough out of the loop if doing so doesn't introduce unnecessary
+/// branches. For example, if the loop has fallthrough into its header and out
+/// of its bottom already, don't rotate it.
+void MachineBlockPlacement::rotateLoop(BlockChain &LoopChain,
+ MachineBasicBlock *ExitingBB,
+ const BlockFilterSet &LoopBlockSet) {
+ if (!ExitingBB)
+ return;
+
+ MachineBasicBlock *Top = *LoopChain.begin();
+ bool ViableTopFallthrough = false;
+ for (MachineBasicBlock::pred_iterator PI = Top->pred_begin(),
+ PE = Top->pred_end();
+ PI != PE; ++PI) {
+ BlockChain *PredChain = BlockToChain[*PI];
+ if (!LoopBlockSet.count(*PI) &&
+ (!PredChain || *PI == *std::prev(PredChain->end()))) {
+ ViableTopFallthrough = true;
+ break;
+ }
+ }
+
+ // If the header has viable fallthrough, check whether the current loop
+ // bottom is a viable exiting block. If so, bail out as rotating will
+ // introduce an unnecessary branch.
+ if (ViableTopFallthrough) {
+ MachineBasicBlock *Bottom = *std::prev(LoopChain.end());
+ for (MachineBasicBlock::succ_iterator SI = Bottom->succ_begin(),
+ SE = Bottom->succ_end();
+ SI != SE; ++SI) {
+ BlockChain *SuccChain = BlockToChain[*SI];
+ if (!LoopBlockSet.count(*SI) &&
+ (!SuccChain || *SI == *SuccChain->begin()))
+ return;
+ }
+ }
+
+ BlockChain::iterator ExitIt = std::find(LoopChain.begin(), LoopChain.end(),
+ ExitingBB);
+ if (ExitIt == LoopChain.end())
+ return;
+
+ std::rotate(LoopChain.begin(), std::next(ExitIt), LoopChain.end());
+}
+
/// \brief Forms basic block chains from the natural loop structures.
///
/// These chains are designed to preserve the existing *structure* of the code
SmallVector<MachineBasicBlock *, 16> BlockWorkList;
BlockFilterSet LoopBlockSet(L.block_begin(), L.block_end());
- BlockChain &LoopChain = *BlockToChain[L.getHeader()];
+
+ // First check to see if there is an obviously preferable top block for the
+ // loop. This will default to the header, but may end up as one of the
+ // predecessors to the header if there is one which will result in strictly
+ // fewer branches in the loop body.
+ MachineBasicBlock *LoopTop = findBestLoopTop(L, LoopBlockSet);
+
+ // If we selected just the header for the loop top, look for a potentially
+ // profitable exit block in the event that rotating the loop can eliminate
+ // branches by placing an exit edge at the bottom.
+ MachineBasicBlock *ExitingBB = nullptr;
+ if (LoopTop == L.getHeader())
+ ExitingBB = findBestLoopExit(F, L, LoopBlockSet);
+
+ BlockChain &LoopChain = *BlockToChain[LoopTop];
// FIXME: This is a really lame way of walking the chains in the loop: we
// walk the blocks, and use a set to prevent visiting a particular chain
// twice.
SmallPtrSet<BlockChain *, 4> UpdatedPreds;
+ assert(LoopChain.LoopPredecessors == 0);
+ UpdatedPreds.insert(&LoopChain);
for (MachineLoop::block_iterator BI = L.block_begin(),
BE = L.block_end();
BI != BE; ++BI) {
BlockChain &Chain = *BlockToChain[*BI];
- if (!UpdatedPreds.insert(&Chain) || BI == L.block_begin())
+ if (!UpdatedPreds.insert(&Chain))
continue;
assert(Chain.LoopPredecessors == 0);
BlockWorkList.push_back(*Chain.begin());
}
- buildChain(*L.block_begin(), LoopChain, BlockWorkList, &LoopBlockSet);
+ buildChain(LoopTop, LoopChain, BlockWorkList, &LoopBlockSet);
+ rotateLoop(LoopChain, ExitingBB, LoopBlockSet);
DEBUG({
// Crash at the end so we get all of the debugging output first.
<< " Chain header: " << getBlockName(*LoopChain.begin()) << "\n";
}
for (BlockChain::iterator BCI = LoopChain.begin(), BCE = LoopChain.end();
- BCI != BCE; ++BCI)
+ BCI != BCE; ++BCI) {
+ dbgs() << " ... " << getBlockName(*BCI) << "\n";
if (!LoopBlockSet.erase(*BCI)) {
// We don't mark the loop as bad here because there are real situations
// where this can occur. For example, with an unanalyzable fallthrough
<< " Chain header: " << getBlockName(*LoopChain.begin()) << "\n"
<< " Bad block: " << getBlockName(*BCI) << "\n";
}
+ }
if (!LoopBlockSet.empty()) {
BadLoop = true;
SmallVector<MachineOperand, 4> Cond; // For AnalyzeBranch.
for (MachineFunction::iterator FI = F.begin(), FE = F.end(); FI != FE; ++FI) {
MachineBasicBlock *BB = FI;
- BlockChain *&Chain = BlockToChain[BB];
- Chain = new (ChainAllocator.Allocate()) BlockChain(BlockToChain, BB);
+ BlockChain *Chain
+ = new (ChainAllocator.Allocate()) BlockChain(BlockToChain, BB);
// Also, merge any blocks which we cannot reason about and must preserve
// the exact fallthrough behavior for.
for (;;) {
Cond.clear();
- MachineBasicBlock *TBB = 0, *FBB = 0; // For AnalyzeBranch.
+ MachineBasicBlock *TBB = nullptr, *FBB = nullptr; // For AnalyzeBranch.
if (!TII->AnalyzeBranch(*BB, TBB, FBB, Cond) || !FI->canFallThrough())
break;
- MachineFunction::iterator NextFI(llvm::next(FI));
+ MachineFunction::iterator NextFI(std::next(FI));
MachineBasicBlock *NextBB = NextFI;
// Ensure that the layout successor is a viable block, as we know that
// fallthrough is a possibility.
DEBUG(dbgs() << "Pre-merging due to unanalyzable fallthrough: "
<< getBlockName(BB) << " -> " << getBlockName(NextBB)
<< "\n");
- Chain->merge(NextBB, 0);
+ Chain->merge(NextBB, nullptr);
FI = NextFI;
BB = NextBB;
}
BlockChain &FunctionChain = *BlockToChain[&F.front()];
buildChain(&F.front(), FunctionChain, BlockWorkList);
+#ifndef NDEBUG
typedef SmallPtrSet<MachineBasicBlock *, 16> FunctionBlockSetType;
+#endif
DEBUG({
// Crash at the end so we get all of the debugging output first.
bool BadFunc = false;
// Update the terminator of the previous block.
if (BI == FunctionChain.begin())
continue;
- MachineBasicBlock *PrevBB = llvm::prior(MachineFunction::iterator(*BI));
+ MachineBasicBlock *PrevBB = std::prev(MachineFunction::iterator(*BI));
// FIXME: It would be awesome of updateTerminator would just return rather
// than assert when the branch cannot be analyzed in order to remove this
// boiler plate.
Cond.clear();
- MachineBasicBlock *TBB = 0, *FBB = 0; // For AnalyzeBranch.
- if (!TII->AnalyzeBranch(*PrevBB, TBB, FBB, Cond))
- PrevBB->updateTerminator();
+ MachineBasicBlock *TBB = nullptr, *FBB = nullptr; // For AnalyzeBranch.
+ if (!TII->AnalyzeBranch(*PrevBB, TBB, FBB, Cond)) {
+ // The "PrevBB" is not yet updated to reflect current code layout, so,
+ // o. it may fall-through to a block without explict "goto" instruction
+ // before layout, and no longer fall-through it after layout; or
+ // o. just opposite.
+ //
+ // AnalyzeBranch() may return erroneous value for FBB when these two
+ // situations take place. For the first scenario FBB is mistakenly set
+ // NULL; for the 2nd scenario, the FBB, which is expected to be NULL,
+ // is mistakenly pointing to "*BI".
+ //
+ bool needUpdateBr = true;
+ if (!Cond.empty() && (!FBB || FBB == *BI)) {
+ PrevBB->updateTerminator();
+ needUpdateBr = false;
+ Cond.clear();
+ TBB = FBB = nullptr;
+ if (TII->AnalyzeBranch(*PrevBB, TBB, FBB, Cond)) {
+ // FIXME: This should never take place.
+ TBB = FBB = nullptr;
+ }
+ }
+
+ // If PrevBB has a two-way branch, try to re-order the branches
+ // such that we branch to the successor with higher weight first.
+ if (TBB && !Cond.empty() && FBB &&
+ MBPI->getEdgeWeight(PrevBB, FBB) > MBPI->getEdgeWeight(PrevBB, TBB) &&
+ !TII->ReverseBranchCondition(Cond)) {
+ DEBUG(dbgs() << "Reverse order of the two branches: "
+ << getBlockName(PrevBB) << "\n");
+ DEBUG(dbgs() << " Edge weight: " << MBPI->getEdgeWeight(PrevBB, FBB)
+ << " vs " << MBPI->getEdgeWeight(PrevBB, TBB) << "\n");
+ DebugLoc dl; // FIXME: this is nowhere
+ TII->RemoveBranch(*PrevBB);
+ TII->InsertBranch(*PrevBB, FBB, TBB, Cond, dl);
+ needUpdateBr = true;
+ }
+ if (needUpdateBr)
+ PrevBB->updateTerminator();
+ }
}
// Fixup the last block.
Cond.clear();
- MachineBasicBlock *TBB = 0, *FBB = 0; // For AnalyzeBranch.
+ MachineBasicBlock *TBB = nullptr, *FBB = nullptr; // For AnalyzeBranch.
if (!TII->AnalyzeBranch(F.back(), TBB, FBB, Cond))
F.back().updateTerminator();
-}
-
-/// \brief Recursive helper to align a loop and any nested loops.
-static void AlignLoop(MachineFunction &F, MachineLoop *L, unsigned Align) {
- // Recurse through nested loops.
- for (MachineLoop::iterator I = L->begin(), E = L->end(); I != E; ++I)
- AlignLoop(F, *I, Align);
- L->getTopBlock()->setAlignment(Align);
-}
-
-/// \brief Align loop headers to target preferred alignments.
-void MachineBlockPlacement::AlignLoops(MachineFunction &F) {
- if (F.getFunction()->hasFnAttr(Attribute::OptimizeForSize))
+ // Walk through the backedges of the function now that we have fully laid out
+ // the basic blocks and align the destination of each backedge. We don't rely
+ // exclusively on the loop info here so that we can align backedges in
+ // unnatural CFGs and backedges that were introduced purely because of the
+ // loop rotations done during this layout pass.
+ if (F.getFunction()->getAttributes().
+ hasAttribute(AttributeSet::FunctionIndex, Attribute::OptimizeForSize))
return;
-
unsigned Align = TLI->getPrefLoopAlignment();
if (!Align)
return; // Don't care about loop alignment.
+ if (FunctionChain.begin() == FunctionChain.end())
+ return; // Empty chain.
+
+ const BranchProbability ColdProb(1, 5); // 20%
+ BlockFrequency EntryFreq = MBFI->getBlockFreq(F.begin());
+ BlockFrequency WeightedEntryFreq = EntryFreq * ColdProb;
+ for (BlockChain::iterator BI = std::next(FunctionChain.begin()),
+ BE = FunctionChain.end();
+ BI != BE; ++BI) {
+ // Don't align non-looping basic blocks. These are unlikely to execute
+ // enough times to matter in practice. Note that we'll still handle
+ // unnatural CFGs inside of a natural outer loop (the common case) and
+ // rotated loops.
+ MachineLoop *L = MLI->getLoopFor(*BI);
+ if (!L)
+ continue;
- for (MachineLoopInfo::iterator I = MLI->begin(), E = MLI->end(); I != E; ++I)
- AlignLoop(F, *I, Align);
+ // If the block is cold relative to the function entry don't waste space
+ // aligning it.
+ BlockFrequency Freq = MBFI->getBlockFreq(*BI);
+ if (Freq < WeightedEntryFreq)
+ continue;
+
+ // If the block is cold relative to its loop header, don't align it
+ // regardless of what edges into the block exist.
+ MachineBasicBlock *LoopHeader = L->getHeader();
+ BlockFrequency LoopHeaderFreq = MBFI->getBlockFreq(LoopHeader);
+ if (Freq < (LoopHeaderFreq * ColdProb))
+ continue;
+
+ // Check for the existence of a non-layout predecessor which would benefit
+ // from aligning this block.
+ MachineBasicBlock *LayoutPred = *std::prev(BI);
+
+ // Force alignment if all the predecessors are jumps. We already checked
+ // that the block isn't cold above.
+ if (!LayoutPred->isSuccessor(*BI)) {
+ (*BI)->setAlignment(Align);
+ continue;
+ }
+
+ // Align this block if the layout predecessor's edge into this block is
+ // cold relative to the block. When this is true, other predecessors make up
+ // all of the hot entries into the block and thus alignment is likely to be
+ // important.
+ BranchProbability LayoutProb = MBPI->getEdgeProbability(LayoutPred, *BI);
+ BlockFrequency LayoutEdgeFreq = MBFI->getBlockFreq(LayoutPred) * LayoutProb;
+ if (LayoutEdgeFreq <= (Freq * ColdProb))
+ (*BI)->setAlignment(Align);
+ }
}
bool MachineBlockPlacement::runOnMachineFunction(MachineFunction &F) {
// Check for single-block functions and skip them.
- if (llvm::next(F.begin()) == F.end())
+ if (std::next(F.begin()) == F.end())
+ return false;
+
+ if (skipOptnoneFunction(*F.getFunction()))
return false;
MBPI = &getAnalysis<MachineBranchProbabilityInfo>();
MBFI = &getAnalysis<MachineBlockFrequencyInfo>();
MLI = &getAnalysis<MachineLoopInfo>();
- TII = F.getTarget().getInstrInfo();
- TLI = F.getTarget().getTargetLowering();
+ TII = F.getSubtarget().getInstrInfo();
+ TLI = F.getSubtarget().getTargetLowering();
assert(BlockToChain.empty());
buildCFGChains(F);
- AlignLoops(F);
BlockToChain.clear();
ChainAllocator.DestroyAll();
+ if (AlignAllBlock)
+ // Align all of the blocks in the function to a specific alignment.
+ for (MachineFunction::iterator FI = F.begin(), FE = F.end();
+ FI != FE; ++FI)
+ FI->setAlignment(AlignAllBlock);
+
// We always return true as we have no way to track whether the final order
// differs from the original order.
return true;
///
/// A separate pass to compute interesting statistics for evaluating block
/// placement. This is separate from the actual placement pass so that they can
-/// be computed in the absense of any placement transformations or when using
+/// be computed in the absence of any placement transformations or when using
/// alternative placement strategies.
class MachineBlockPlacementStats : public MachineFunctionPass {
/// \brief A handle to the branch probability pass.
initializeMachineBlockPlacementStatsPass(*PassRegistry::getPassRegistry());
}
- bool runOnMachineFunction(MachineFunction &F);
+ bool runOnMachineFunction(MachineFunction &F) override;
- void getAnalysisUsage(AnalysisUsage &AU) const {
+ void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.addRequired<MachineBranchProbabilityInfo>();
AU.addRequired<MachineBlockFrequencyInfo>();
AU.setPreservesAll();
MachineFunctionPass::getAnalysisUsage(AU);
}
-
- const char *getPassName() const { return "Block Placement Stats"; }
};
}
char MachineBlockPlacementStats::ID = 0;
+char &llvm::MachineBlockPlacementStatsID = MachineBlockPlacementStats::ID;
INITIALIZE_PASS_BEGIN(MachineBlockPlacementStats, "block-placement-stats",
"Basic Block Placement Stats", false, false)
INITIALIZE_PASS_DEPENDENCY(MachineBranchProbabilityInfo)
INITIALIZE_PASS_END(MachineBlockPlacementStats, "block-placement-stats",
"Basic Block Placement Stats", false, false)
-FunctionPass *llvm::createMachineBlockPlacementStatsPass() {
- return new MachineBlockPlacementStats();
-}
-
bool MachineBlockPlacementStats::runOnMachineFunction(MachineFunction &F) {
// Check for single-block functions and skip them.
- if (llvm::next(F.begin()) == F.end())
+ if (std::next(F.begin()) == F.end())
return false;
MBPI = &getAnalysis<MachineBranchProbabilityInfo>();