1 //===--------------------- SchedulerStatistics.cpp --------------*- 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 /// \file
10 ///
11 /// This file implements the SchedulerStatistics interface.
12 ///
13 //===----------------------------------------------------------------------===//
14 
15 #include "Views/SchedulerStatistics.h"
16 #include "llvm/Support/Format.h"
17 #include "llvm/Support/FormattedStream.h"
18 
19 namespace llvm {
20 namespace mca {
21 
22 SchedulerStatistics::SchedulerStatistics(const llvm::MCSubtargetInfo &STI)
23     : SM(STI.getSchedModel()), LQResourceID(0), SQResourceID(0), NumIssued(0),
24       NumCycles(0), MostRecentLoadDispatched(~0U),
25       MostRecentStoreDispatched(~0U),
26       IssuedPerCycle(STI.getSchedModel().NumProcResourceKinds, 0),
27       Usage(STI.getSchedModel().NumProcResourceKinds, {0, 0, 0}) {
28   if (SM.hasExtraProcessorInfo()) {
29     const MCExtraProcessorInfo &EPI = SM.getExtraProcessorInfo();
30     LQResourceID = EPI.LoadQueueID;
31     SQResourceID = EPI.StoreQueueID;
32   }
33 }
34 
35 // FIXME: This implementation works under the assumption that load/store queue
36 // entries are reserved at 'instruction dispatched' stage, and released at
37 // 'instruction executed' stage. This currently matches the behavior of LSUnit.
38 //
39 // The current design minimizes the number of events generated by the
40 // Dispatch/Execute stages, at the cost of doing extra bookkeeping in method
41 // `onEvent`. However, it introduces a subtle dependency between this view and
42 // how the LSUnit works.
43 //
44 // In future we should add a new "memory queue" event type, so that we stop
45 // making assumptions on how LSUnit internally works (See PR39828).
46 void SchedulerStatistics::onEvent(const HWInstructionEvent &Event) {
47   if (Event.Type == HWInstructionEvent::Issued)
48     ++NumIssued;
49   else if (Event.Type == HWInstructionEvent::Dispatched) {
50     const Instruction &Inst = *Event.IR.getInstruction();
51     const unsigned Index = Event.IR.getSourceIndex();
52     if (LQResourceID && Inst.getDesc().MayLoad &&
53         MostRecentLoadDispatched != Index) {
54       Usage[LQResourceID].SlotsInUse++;
55       MostRecentLoadDispatched = Index;
56     }
57     if (SQResourceID && Inst.getDesc().MayStore &&
58         MostRecentStoreDispatched != Index) {
59       Usage[SQResourceID].SlotsInUse++;
60       MostRecentStoreDispatched = Index;
61     }
62   } else if (Event.Type == HWInstructionEvent::Executed) {
63     const Instruction &Inst = *Event.IR.getInstruction();
64     if (LQResourceID && Inst.getDesc().MayLoad) {
65       assert(Usage[LQResourceID].SlotsInUse);
66       Usage[LQResourceID].SlotsInUse--;
67     }
68     if (SQResourceID && Inst.getDesc().MayStore) {
69       assert(Usage[SQResourceID].SlotsInUse);
70       Usage[SQResourceID].SlotsInUse--;
71     }
72   }
73 }
74 
75 void SchedulerStatistics::onReservedBuffers(const InstRef & /* unused */,
76                                             ArrayRef<unsigned> Buffers) {
77   for (const unsigned Buffer : Buffers) {
78     if (Buffer == LQResourceID || Buffer == SQResourceID)
79       continue;
80     Usage[Buffer].SlotsInUse++;
81   }
82 }
83 
84 void SchedulerStatistics::onReleasedBuffers(const InstRef & /* unused */,
85                                             ArrayRef<unsigned> Buffers) {
86   for (const unsigned Buffer : Buffers) {
87     if (Buffer == LQResourceID || Buffer == SQResourceID)
88       continue;
89     Usage[Buffer].SlotsInUse--;
90   }
91 }
92 
93 void SchedulerStatistics::updateHistograms() {
94   for (BufferUsage &BU : Usage) {
95     BU.CumulativeNumUsedSlots += BU.SlotsInUse;
96     BU.MaxUsedSlots = std::max(BU.MaxUsedSlots, BU.SlotsInUse);
97   }
98 
99   IssuedPerCycle[NumIssued]++;
100   NumIssued = 0;
101 }
102 
103 void SchedulerStatistics::printSchedulerStats(raw_ostream &OS) const {
104   OS << "\n\nSchedulers - "
105      << "number of cycles where we saw N instructions issued:\n";
106   OS << "[# issued], [# cycles]\n";
107 
108   const auto It =
109       std::max_element(IssuedPerCycle.begin(), IssuedPerCycle.end());
110   unsigned Index = std::distance(IssuedPerCycle.begin(), It);
111 
112   bool HasColors = OS.has_colors();
113   for (unsigned I = 0, E = IssuedPerCycle.size(); I < E; ++I) {
114     unsigned IPC = IssuedPerCycle[I];
115     if (!IPC)
116       continue;
117 
118     if (I == Index && HasColors)
119       OS.changeColor(raw_ostream::SAVEDCOLOR, true, false);
120 
121     OS << " " << I << ",          " << IPC << "  ("
122        << format("%.1f", ((double)IPC / NumCycles) * 100) << "%)\n";
123     if (HasColors)
124       OS.resetColor();
125   }
126 }
127 
128 void SchedulerStatistics::printSchedulerUsage(raw_ostream &OS) const {
129   assert(NumCycles && "Unexpected number of cycles!");
130 
131   OS << "\nScheduler's queue usage:\n";
132   if (all_of(Usage, [](const BufferUsage &BU) { return !BU.MaxUsedSlots; })) {
133     OS << "No scheduler resources used.\n";
134     return;
135   }
136 
137   OS << "[1] Resource name.\n"
138      << "[2] Average number of used buffer entries.\n"
139      << "[3] Maximum number of used buffer entries.\n"
140      << "[4] Total number of buffer entries.\n\n"
141      << " [1]            [2]        [3]        [4]\n";
142 
143   formatted_raw_ostream FOS(OS);
144   bool HasColors = FOS.has_colors();
145   for (unsigned I = 0, E = SM.getNumProcResourceKinds(); I < E; ++I) {
146     const MCProcResourceDesc &ProcResource = *SM.getProcResource(I);
147     if (ProcResource.BufferSize <= 0)
148       continue;
149 
150     const BufferUsage &BU = Usage[I];
151     double AvgUsage = (double)BU.CumulativeNumUsedSlots / NumCycles;
152     double AlmostFullThreshold = (double)(ProcResource.BufferSize * 4) / 5;
153     unsigned NormalizedAvg = floor((AvgUsage * 10) + 0.5) / 10;
154     unsigned NormalizedThreshold = floor((AlmostFullThreshold * 10) + 0.5) / 10;
155 
156     FOS << ProcResource.Name;
157     FOS.PadToColumn(17);
158     if (HasColors && NormalizedAvg >= NormalizedThreshold)
159       FOS.changeColor(raw_ostream::YELLOW, true, false);
160     FOS << NormalizedAvg;
161     if (HasColors)
162       FOS.resetColor();
163     FOS.PadToColumn(28);
164     if (HasColors &&
165         BU.MaxUsedSlots == static_cast<unsigned>(ProcResource.BufferSize))
166       FOS.changeColor(raw_ostream::RED, true, false);
167     FOS << BU.MaxUsedSlots;
168     if (HasColors)
169       FOS.resetColor();
170     FOS.PadToColumn(39);
171     FOS << ProcResource.BufferSize << '\n';
172   }
173 
174   FOS.flush();
175 }
176 
177 void SchedulerStatistics::printView(raw_ostream &OS) const {
178   printSchedulerStats(OS);
179   printSchedulerUsage(OS);
180 }
181 
182 } // namespace mca
183 } // namespace llvm
184