lld/COFF/CallGraphSort.cpp

   1 //===- CallGraphSort.cpp --------------------------------------------------===//
   2 //
   3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
   4 // See https://llvm.org/LICENSE.txt for license information.
   5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
   6 //
   7 //===----------------------------------------------------------------------===//
   8 ///
   9 /// This is based on the ELF port, see ELF/CallGraphSort.cpp for the details
  10 /// about the algorithm.
  11 ///
  12 //===----------------------------------------------------------------------===//
  13
  14 #include "CallGraphSort.h"
  15 #include "COFFLinkerContext.h"
  16 #include "InputFiles.h"
  17 #include "SymbolTable.h"
  18 #include "Symbols.h"
  19 #include "lld/Common/ErrorHandler.h"
  20
  21 #include <numeric>
  22
  23 using namespace llvm;
  24 using namespace lld;
  25 using namespace lld::coff;
  26
  27 namespace {
  28 struct Edge {
  29   int from;
  30   uint64_t weight;
  31 };
  32
  33 struct Cluster {
  34   Cluster(int sec, size_t s) : next(sec), prev(sec), size(s) {}
  35
  36   double getDensity() const {
  37     if (size == 0)
  38       return 0;
  39     return double(weight) / double(size);
  40   }
  41
  42   int next;
  43   int prev;
  44   uint64_t size;
  45   uint64_t weight = 0;
  46   uint64_t initialWeight = 0;
  47   Edge bestPred = {-1, 0};
  48 };
  49
  50 class CallGraphSort {
  51 public:
  52   CallGraphSort(const COFFLinkerContext &ctx);
  53
  54   DenseMap<const SectionChunk *, int> run();
  55
  56 private:
  57   std::vector<Cluster> clusters;
  58   std::vector<const SectionChunk *> sections;
  59 };
  60
  61 // Maximum amount the combined cluster density can be worse than the original
  62 // cluster to consider merging.
  63 constexpr int MAX_DENSITY_DEGRADATION = 8;
  64
  65 // Maximum cluster size in bytes.
  66 constexpr uint64_t MAX_CLUSTER_SIZE = 1024 * 1024;
  67 } // end anonymous namespace
  68
  69 using SectionPair = std::pair<const SectionChunk *, const SectionChunk *>;
  70
  71 // Take the edge list in Config->CallGraphProfile, resolve symbol names to
  72 // Symbols, and generate a graph between InputSections with the provided
  73 // weights.
  74 CallGraphSort::CallGraphSort(const COFFLinkerContext &ctx) {
  75   MapVector<SectionPair, uint64_t> &profile = config->callGraphProfile;
  76   DenseMap<const SectionChunk *, int> secToCluster;
  77
  78   auto getOrCreateNode = [&](const SectionChunk *isec) -> int {
  79     auto res = secToCluster.try_emplace(isec, clusters.size());
  80     if (res.second) {
  81       sections.push_back(isec);
  82       clusters.emplace_back(clusters.size(), isec->getSize());
  83     }
  84     return res.first->second;
  85   };
  86
  87   // Create the graph.
  88   for (std::pair<SectionPair, uint64_t> &c : profile) {
  89     const auto *fromSec = cast<SectionChunk>(c.first.first->repl);
  90     const auto *toSec = cast<SectionChunk>(c.first.second->repl);
  91     uint64_t weight = c.second;
  92
  93     // Ignore edges between input sections belonging to different output
  94     // sections.  This is done because otherwise we would end up with clusters
  95     // containing input sections that can't actually be placed adjacently in the
  96     // output.  This messes with the cluster size and density calculations.  We
  97     // would also end up moving input sections in other output sections without
  98     // moving them closer to what calls them.
  99     if (ctx.getOutputSection(fromSec) != ctx.getOutputSection(toSec))
 100       continue;
 101
 102     int from = getOrCreateNode(fromSec);
 103     int to = getOrCreateNode(toSec);
 104
 105     clusters[to].weight += weight;
 106
 107     if (from == to)
 108       continue;
 109
 110     // Remember the best edge.
 111     Cluster &toC = clusters[to];
 112     if (toC.bestPred.from == -1 || toC.bestPred.weight < weight) {
 113       toC.bestPred.from = from;
 114       toC.bestPred.weight = weight;
 115     }
 116   }
 117   for (Cluster &c : clusters)
 118     c.initialWeight = c.weight;
 119 }
 120
 121 // It's bad to merge clusters which would degrade the density too much.
 122 static bool isNewDensityBad(Cluster &a, Cluster &b) {
 123   double newDensity = double(a.weight + b.weight) / double(a.size + b.size);
 124   return newDensity < a.getDensity() / MAX_DENSITY_DEGRADATION;
 125 }
 126
 127 // Find the leader of V's belonged cluster (represented as an equivalence
 128 // class). We apply union-find path-halving technique (simple to implement) in
 129 // the meantime as it decreases depths and the time complexity.
 130 static int getLeader(std::vector<int> &leaders, int v) {
 131   while (leaders[v] != v) {
 132     leaders[v] = leaders[leaders[v]];
 133     v = leaders[v];
 134   }
 135   return v;
 136 }
 137
 138 static void mergeClusters(std::vector<Cluster> &cs, Cluster &into, int intoIdx,
 139                           Cluster &from, int fromIdx) {
 140   int tail1 = into.prev, tail2 = from.prev;
 141   into.prev = tail2;
 142   cs[tail2].next = intoIdx;
 143   from.prev = tail1;
 144   cs[tail1].next = fromIdx;
 145   into.size += from.size;
 146   into.weight += from.weight;
 147   from.size = 0;
 148   from.weight = 0;
 149 }
 150
 151 // Group InputSections into clusters using the Call-Chain Clustering heuristic
 152 // then sort the clusters by density.
 153 DenseMap<const SectionChunk *, int> CallGraphSort::run() {
 154   std::vector<int> sorted(clusters.size());
 155   std::vector<int> leaders(clusters.size());
 156
 157   std::iota(leaders.begin(), leaders.end(), 0);
 158   std::iota(sorted.begin(), sorted.end(), 0);
 159   llvm::stable_sort(sorted, [&](int a, int b) {
 160     return clusters[a].getDensity() > clusters[b].getDensity();
 161   });
 162
 163   for (int l : sorted) {
 164     // The cluster index is the same as the index of its leader here because
 165     // clusters[L] has not been merged into another cluster yet.
 166     Cluster &c = clusters[l];
 167
 168     // Don't consider merging if the edge is unlikely.
 169     if (c.bestPred.from == -1 || c.bestPred.weight * 10 <= c.initialWeight)
 170       continue;
 171
 172     int predL = getLeader(leaders, c.bestPred.from);
 173     if (l == predL)
 174       continue;
 175
 176     Cluster *predC = &clusters[predL];
 177     if (c.size + predC->size > MAX_CLUSTER_SIZE)
 178       continue;
 179
 180     if (isNewDensityBad(*predC, c))
 181       continue;
 182
 183     leaders[l] = predL;
 184     mergeClusters(clusters, *predC, predL, c, l);
 185   }
 186
 187   // Sort remaining non-empty clusters by density.
 188   sorted.clear();
 189   for (int i = 0, e = (int)clusters.size(); i != e; ++i)
 190     if (clusters[i].size > 0)
 191       sorted.push_back(i);
 192   llvm::stable_sort(sorted, [&](int a, int b) {
 193     return clusters[a].getDensity() > clusters[b].getDensity();
 194   });
 195
 196   DenseMap<const SectionChunk *, int> orderMap;
 197   // Sections will be sorted by increasing order. Absent sections will have
 198   // priority 0 and be placed at the end of sections.
 199   int curOrder = INT_MIN;
 200   for (int leader : sorted) {
 201     for (int i = leader;;) {
 202       orderMap[sections[i]] = curOrder++;
 203       i = clusters[i].next;
 204       if (i == leader)
 205         break;
 206     }
 207   }
 208   if (!config->printSymbolOrder.empty()) {
 209     std::error_code ec;
 210     raw_fd_ostream os(config->printSymbolOrder, ec, sys::fs::OF_None);
 211     if (ec) {
 212       error("cannot open " + config->printSymbolOrder + ": " + ec.message());
 213       return orderMap;
 214     }
 215     // Print the symbols ordered by C3, in the order of increasing curOrder
 216     // Instead of sorting all the orderMap, just repeat the loops above.
 217     for (int leader : sorted)
 218       for (int i = leader;;) {
 219         const SectionChunk *sc = sections[i];
 220
 221         // Search all the symbols in the file of the section
 222         // and find out a DefinedCOFF symbol with name that is within the
 223         // section.
 224         for (Symbol *sym : sc->file->getSymbols())
 225           if (auto *d = dyn_cast_or_null<DefinedCOFF>(sym))
 226             // Filter out non-COMDAT symbols and section symbols.
 227             if (d->isCOMDAT && !d->getCOFFSymbol().isSection() &&
 228                 sc == d->getChunk())
 229               os << sym->getName() << "\n";
 230         i = clusters[i].next;
 231         if (i == leader)
 232           break;
 233       }
 234   }
 235
 236   return orderMap;
 237 }
 238
 239 // Sort sections by the profile data provided by  /call-graph-ordering-file
 240 //
 241 // This first builds a call graph based on the profile data then merges sections
 242 // according to the C³ heuristic. All clusters are then sorted by a density
 243 // metric to further improve locality.
 244 DenseMap<const SectionChunk *, int>
 245 coff::computeCallGraphProfileOrder(const COFFLinkerContext &ctx) {
 246   return CallGraphSort(ctx).run();
 247 }