| 1 | //===--- SyntheticCountsUtils.cpp - synthetic counts propagation utils ---===// |
| 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 file defines utilities for propagating synthetic counts. |
| 10 | // |
| 11 | //===----------------------------------------------------------------------===// |
| 12 | |
| 13 | #include "llvm/Analysis/SyntheticCountsUtils.h" |
| 14 | #include "llvm/ADT/DenseSet.h" |
| 15 | #include "llvm/ADT/SCCIterator.h" |
| 16 | #include "llvm/Analysis/CallGraph.h" |
| 17 | #include "llvm/IR/ModuleSummaryIndex.h" |
| 18 | |
| 19 | using namespace llvm; |
| 20 | |
| 21 | // Given an SCC, propagate entry counts along the edge of the SCC nodes. |
| 22 | template <typename CallGraphType> |
| 23 | void SyntheticCountsUtils<CallGraphType>::propagateFromSCC( |
| 24 | const SccTy &SCC, GetProfCountTy GetProfCount, AddCountTy AddCount) { |
| 25 | |
| 26 | DenseSet<NodeRef> SCCNodes(llvm::from_range, SCC); |
| 27 | SmallVector<std::pair<NodeRef, EdgeRef>, 8> SCCEdges, NonSCCEdges; |
| 28 | |
| 29 | // Partition the edges coming out of the SCC into those whose destination is |
| 30 | // in the SCC and the rest. |
| 31 | for (const auto &Node : SCCNodes) { |
| 32 | for (auto &E : children_edges<CallGraphType>(Node)) { |
| 33 | if (SCCNodes.count(CGT::edge_dest(E))) |
| 34 | SCCEdges.emplace_back(Node, E); |
| 35 | else |
| 36 | NonSCCEdges.emplace_back(Node, E); |
| 37 | } |
| 38 | } |
| 39 | |
| 40 | // For nodes in the same SCC, update the counts in two steps: |
| 41 | // 1. Compute the additional count for each node by propagating the counts |
| 42 | // along all incoming edges to the node that originate from within the same |
| 43 | // SCC and summing them up. |
| 44 | // 2. Add the additional counts to the nodes in the SCC. |
| 45 | // This ensures that the order of |
| 46 | // traversal of nodes within the SCC doesn't affect the final result. |
| 47 | |
| 48 | DenseMap<NodeRef, Scaled64> AdditionalCounts; |
| 49 | for (auto &E : SCCEdges) { |
| 50 | auto OptProfCount = GetProfCount(E.first, E.second); |
| 51 | if (!OptProfCount) |
| 52 | continue; |
| 53 | auto Callee = CGT::edge_dest(E.second); |
| 54 | AdditionalCounts[Callee] += *OptProfCount; |
| 55 | } |
| 56 | |
| 57 | // Update the counts for the nodes in the SCC. |
| 58 | for (auto &Entry : AdditionalCounts) |
| 59 | AddCount(Entry.first, Entry.second); |
| 60 | |
| 61 | // Now update the counts for nodes outside the SCC. |
| 62 | for (auto &E : NonSCCEdges) { |
| 63 | auto OptProfCount = GetProfCount(E.first, E.second); |
| 64 | if (!OptProfCount) |
| 65 | continue; |
| 66 | auto Callee = CGT::edge_dest(E.second); |
| 67 | AddCount(Callee, *OptProfCount); |
| 68 | } |
| 69 | } |
| 70 | |
| 71 | /// Propgate synthetic entry counts on a callgraph \p CG. |
| 72 | /// |
| 73 | /// This performs a reverse post-order traversal of the callgraph SCC. For each |
| 74 | /// SCC, it first propagates the entry counts to the nodes within the SCC |
| 75 | /// through call edges and updates them in one shot. Then the entry counts are |
| 76 | /// propagated to nodes outside the SCC. This requires \p GraphTraits |
| 77 | /// to have a specialization for \p CallGraphType. |
| 78 | |
| 79 | template <typename CallGraphType> |
| 80 | void SyntheticCountsUtils<CallGraphType>::propagate(const CallGraphType &CG, |
| 81 | GetProfCountTy GetProfCount, |
| 82 | AddCountTy AddCount) { |
| 83 | std::vector<SccTy> SCCs; |
| 84 | |
| 85 | // Collect all the SCCs. |
| 86 | for (auto I = scc_begin(CG); !I.isAtEnd(); ++I) |
| 87 | SCCs.push_back(*I); |
| 88 | |
| 89 | // The callgraph-scc needs to be visited in top-down order for propagation. |
| 90 | // The scc iterator returns the scc in bottom-up order, so reverse the SCCs |
| 91 | // and call propagateFromSCC. |
| 92 | for (auto &SCC : reverse(SCCs)) |
| 93 | propagateFromSCC(SCC, GetProfCount, AddCount); |
| 94 | } |
| 95 | |
| 96 | template class llvm::SyntheticCountsUtils<const CallGraph *>; |
| 97 | template class llvm::SyntheticCountsUtils<ModuleSummaryIndex *>; |
| 98 | |