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/Function.h"
18 #include "llvm/IR/InstIterator.h"
19 #include "llvm/IR/Instructions.h"
20 #include "llvm/IR/ModuleSummaryIndex.h"
21
22 using namespace llvm;
23
24 // Given an SCC, propagate entry counts along the edge of the SCC nodes.
25 template <typename CallGraphType>
propagateFromSCC(const SccTy & SCC,GetProfCountTy GetProfCount,AddCountTy AddCount)26 void SyntheticCountsUtils<CallGraphType>::propagateFromSCC(
27 const SccTy &SCC, GetProfCountTy GetProfCount, AddCountTy AddCount) {
28
29 DenseSet<NodeRef> SCCNodes;
30 SmallVector<std::pair<NodeRef, EdgeRef>, 8> SCCEdges, NonSCCEdges;
31
32 for (auto &Node : SCC)
33 SCCNodes.insert(Node);
34
35 // Partition the edges coming out of the SCC into those whose destination is
36 // in the SCC and the rest.
37 for (const auto &Node : SCCNodes) {
38 for (auto &E : children_edges<CallGraphType>(Node)) {
39 if (SCCNodes.count(CGT::edge_dest(E)))
40 SCCEdges.emplace_back(Node, E);
41 else
42 NonSCCEdges.emplace_back(Node, E);
43 }
44 }
45
46 // For nodes in the same SCC, update the counts in two steps:
47 // 1. Compute the additional count for each node by propagating the counts
48 // along all incoming edges to the node that originate from within the same
49 // SCC and summing them up.
50 // 2. Add the additional counts to the nodes in the SCC.
51 // This ensures that the order of
52 // traversal of nodes within the SCC doesn't affect the final result.
53
54 DenseMap<NodeRef, Scaled64> AdditionalCounts;
55 for (auto &E : SCCEdges) {
56 auto OptProfCount = GetProfCount(E.first, E.second);
57 if (!OptProfCount)
58 continue;
59 auto Callee = CGT::edge_dest(E.second);
60 AdditionalCounts[Callee] += OptProfCount.getValue();
61 }
62
63 // Update the counts for the nodes in the SCC.
64 for (auto &Entry : AdditionalCounts)
65 AddCount(Entry.first, Entry.second);
66
67 // Now update the counts for nodes outside the SCC.
68 for (auto &E : NonSCCEdges) {
69 auto OptProfCount = GetProfCount(E.first, E.second);
70 if (!OptProfCount)
71 continue;
72 auto Callee = CGT::edge_dest(E.second);
73 AddCount(Callee, OptProfCount.getValue());
74 }
75 }
76
77 /// Propgate synthetic entry counts on a callgraph \p CG.
78 ///
79 /// This performs a reverse post-order traversal of the callgraph SCC. For each
80 /// SCC, it first propagates the entry counts to the nodes within the SCC
81 /// through call edges and updates them in one shot. Then the entry counts are
82 /// propagated to nodes outside the SCC. This requires \p GraphTraits
83 /// to have a specialization for \p CallGraphType.
84
85 template <typename CallGraphType>
propagate(const CallGraphType & CG,GetProfCountTy GetProfCount,AddCountTy AddCount)86 void SyntheticCountsUtils<CallGraphType>::propagate(const CallGraphType &CG,
87 GetProfCountTy GetProfCount,
88 AddCountTy AddCount) {
89 std::vector<SccTy> SCCs;
90
91 // Collect all the SCCs.
92 for (auto I = scc_begin(CG); !I.isAtEnd(); ++I)
93 SCCs.push_back(*I);
94
95 // The callgraph-scc needs to be visited in top-down order for propagation.
96 // The scc iterator returns the scc in bottom-up order, so reverse the SCCs
97 // and call propagateFromSCC.
98 for (auto &SCC : reverse(SCCs))
99 propagateFromSCC(SCC, GetProfCount, AddCount);
100 }
101
102 template class llvm::SyntheticCountsUtils<const CallGraph *>;
103 template class llvm::SyntheticCountsUtils<ModuleSummaryIndex *>;
104