1 //=- SyntheticCountsPropagation.cpp - Propagate function counts --*- C++ -*-=//
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 //
10 // This file implements a transformation that synthesizes entry counts for
11 // functions and attaches !prof metadata to functions with the synthesized
12 // counts. The presence of !prof metadata with counter name set to
13 // 'synthesized_function_entry_count' indicate that the value of the counter is
14 // an estimation of the likely execution count of the function. This transform
15 // is applied only in non PGO mode as functions get 'real' profile-based
16 // function entry counts in the PGO mode.
17 //
18 // The transformation works by first assigning some initial values to the entry
19 // counts of all functions and then doing a top-down traversal of the
20 // callgraph-scc to propagate the counts. For each function the set of callsites
21 // and their relative block frequency is gathered. The relative block frequency
22 // multiplied by the entry count of the caller and added to the callee's entry
23 // count. For non-trivial SCCs, the new counts are computed from the previous
24 // counts and updated in one shot.
25 //
26 //===----------------------------------------------------------------------===//
27 
28 #include "llvm/Transforms/IPO/SyntheticCountsPropagation.h"
29 #include "llvm/ADT/DenseSet.h"
30 #include "llvm/ADT/STLExtras.h"
31 #include "llvm/Analysis/BlockFrequencyInfo.h"
32 #include "llvm/Analysis/CallGraph.h"
33 #include "llvm/Analysis/ProfileSummaryInfo.h"
34 #include "llvm/Analysis/SyntheticCountsUtils.h"
35 #include "llvm/IR/CallSite.h"
36 #include "llvm/IR/Function.h"
37 #include "llvm/IR/Instructions.h"
38 #include "llvm/IR/Module.h"
39 #include "llvm/Support/CommandLine.h"
40 #include "llvm/Support/Debug.h"
41 #include "llvm/Support/raw_ostream.h"
42 
43 using namespace llvm;
44 using Scaled64 = ScaledNumber<uint64_t>;
45 using ProfileCount = Function::ProfileCount;
46 
47 #define DEBUG_TYPE "synthetic-counts-propagation"
48 
49 /// Initial synthetic count assigned to functions.
50 cl::opt<int>
51     InitialSyntheticCount("initial-synthetic-count", cl::Hidden, cl::init(10),
52                           cl::ZeroOrMore,
53                           cl::desc("Initial value of synthetic entry count."));
54 
55 /// Initial synthetic count assigned to inline functions.
56 static cl::opt<int> InlineSyntheticCount(
57     "inline-synthetic-count", cl::Hidden, cl::init(15), cl::ZeroOrMore,
58     cl::desc("Initial synthetic entry count for inline functions."));
59 
60 /// Initial synthetic count assigned to cold functions.
61 static cl::opt<int> ColdSyntheticCount(
62     "cold-synthetic-count", cl::Hidden, cl::init(5), cl::ZeroOrMore,
63     cl::desc("Initial synthetic entry count for cold functions."));
64 
65 // Assign initial synthetic entry counts to functions.
66 static void
67 initializeCounts(Module &M, function_ref<void(Function *, uint64_t)> SetCount) {
68   auto MayHaveIndirectCalls = [](Function &F) {
69     for (auto *U : F.users()) {
70       if (!isa<CallInst>(U) && !isa<InvokeInst>(U))
71         return true;
72     }
73     return false;
74   };
75 
76   for (Function &F : M) {
77     uint64_t InitialCount = InitialSyntheticCount;
78     if (F.isDeclaration())
79       continue;
80     if (F.hasFnAttribute(Attribute::AlwaysInline) ||
81         F.hasFnAttribute(Attribute::InlineHint)) {
82       // Use a higher value for inline functions to account for the fact that
83       // these are usually beneficial to inline.
84       InitialCount = InlineSyntheticCount;
85     } else if (F.hasLocalLinkage() && !MayHaveIndirectCalls(F)) {
86       // Local functions without inline hints get counts only through
87       // propagation.
88       InitialCount = 0;
89     } else if (F.hasFnAttribute(Attribute::Cold) ||
90                F.hasFnAttribute(Attribute::NoInline)) {
91       // Use a lower value for noinline and cold functions.
92       InitialCount = ColdSyntheticCount;
93     }
94     SetCount(&F, InitialCount);
95   }
96 }
97 
98 PreservedAnalyses SyntheticCountsPropagation::run(Module &M,
99                                                   ModuleAnalysisManager &MAM) {
100   FunctionAnalysisManager &FAM =
101       MAM.getResult<FunctionAnalysisManagerModuleProxy>(M).getManager();
102   DenseMap<Function *, Scaled64> Counts;
103   // Set initial entry counts.
104   initializeCounts(
105       M, [&](Function *F, uint64_t Count) { Counts[F] = Scaled64(Count, 0); });
106 
107   // Edge includes information about the source. Hence ignore the first
108   // parameter.
109   auto GetCallSiteProfCount = [&](const CallGraphNode *,
110                                   const CallGraphNode::CallRecord &Edge) {
111     Optional<Scaled64> Res = None;
112     if (!Edge.first)
113       return Res;
114     assert(isa<Instruction>(Edge.first));
115     CallSite CS(cast<Instruction>(Edge.first));
116     Function *Caller = CS.getCaller();
117     auto &BFI = FAM.getResult<BlockFrequencyAnalysis>(*Caller);
118 
119     // Now compute the callsite count from relative frequency and
120     // entry count:
121     BasicBlock *CSBB = CS.getInstruction()->getParent();
122     Scaled64 EntryFreq(BFI.getEntryFreq(), 0);
123     Scaled64 BBCount(BFI.getBlockFreq(CSBB).getFrequency(), 0);
124     BBCount /= EntryFreq;
125     BBCount *= Counts[Caller];
126     return Optional<Scaled64>(BBCount);
127   };
128 
129   CallGraph CG(M);
130   // Propgate the entry counts on the callgraph.
131   SyntheticCountsUtils<const CallGraph *>::propagate(
132       &CG, GetCallSiteProfCount, [&](const CallGraphNode *N, Scaled64 New) {
133         auto F = N->getFunction();
134         if (!F || F->isDeclaration())
135           return;
136 
137         Counts[F] += New;
138       });
139 
140   // Set the counts as metadata.
141   for (auto Entry : Counts) {
142     Entry.first->setEntryCount(ProfileCount(
143         Entry.second.template toInt<uint64_t>(), Function::PCT_Synthetic));
144   }
145 
146   return PreservedAnalyses::all();
147 }
148