1 //===- ProfileSummaryInfo.cpp - Global profile summary information --------===//
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 contains a pass that provides access to the global profile summary
10 // information.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "llvm/Analysis/ProfileSummaryInfo.h"
15 #include "llvm/Analysis/BlockFrequencyInfo.h"
16 #include "llvm/IR/BasicBlock.h"
17 #include "llvm/IR/Instructions.h"
18 #include "llvm/IR/Metadata.h"
19 #include "llvm/IR/Module.h"
20 #include "llvm/IR/ProfileSummary.h"
21 #include "llvm/InitializePasses.h"
22 #include "llvm/ProfileData/ProfileCommon.h"
23 #include "llvm/Support/CommandLine.h"
24 using namespace llvm;
25 
26 // Knobs for profile summary based thresholds.
27 extern cl::opt<int> ProfileSummaryCutoffHot;
28 extern cl::opt<int> ProfileSummaryCutoffCold;
29 extern cl::opt<unsigned> ProfileSummaryHugeWorkingSetSizeThreshold;
30 extern cl::opt<unsigned> ProfileSummaryLargeWorkingSetSizeThreshold;
31 extern cl::opt<int> ProfileSummaryHotCount;
32 extern cl::opt<int> ProfileSummaryColdCount;
33 
34 static cl::opt<bool> PartialProfile(
35     "partial-profile", cl::Hidden, cl::init(false),
36     cl::desc("Specify the current profile is used as a partial profile."));
37 
38 cl::opt<bool> ScalePartialSampleProfileWorkingSetSize(
39     "scale-partial-sample-profile-working-set-size", cl::Hidden, cl::init(true),
40     cl::desc(
41         "If true, scale the working set size of the partial sample profile "
42         "by the partial profile ratio to reflect the size of the program "
43         "being compiled."));
44 
45 static cl::opt<double> PartialSampleProfileWorkingSetSizeScaleFactor(
46     "partial-sample-profile-working-set-size-scale-factor", cl::Hidden,
47     cl::init(0.008),
48     cl::desc("The scale factor used to scale the working set size of the "
49              "partial sample profile along with the partial profile ratio. "
50              "This includes the factor of the profile counter per block "
51              "and the factor to scale the working set size to use the same "
52              "shared thresholds as PGO."));
53 
54 // The profile summary metadata may be attached either by the frontend or by
55 // any backend passes (IR level instrumentation, for example). This method
56 // checks if the Summary is null and if so checks if the summary metadata is now
57 // available in the module and parses it to get the Summary object.
58 void ProfileSummaryInfo::refresh() {
59   if (hasProfileSummary())
60     return;
61   // First try to get context sensitive ProfileSummary.
62   auto *SummaryMD = M->getProfileSummary(/* IsCS */ true);
63   if (SummaryMD)
64     Summary.reset(ProfileSummary::getFromMD(SummaryMD));
65 
66   if (!hasProfileSummary()) {
67     // This will actually return PSK_Instr or PSK_Sample summary.
68     SummaryMD = M->getProfileSummary(/* IsCS */ false);
69     if (SummaryMD)
70       Summary.reset(ProfileSummary::getFromMD(SummaryMD));
71   }
72   if (!hasProfileSummary())
73     return;
74   computeThresholds();
75 }
76 
77 Optional<uint64_t> ProfileSummaryInfo::getProfileCount(
78     const CallBase &Call, BlockFrequencyInfo *BFI, bool AllowSynthetic) const {
79   assert((isa<CallInst>(Call) || isa<InvokeInst>(Call)) &&
80          "We can only get profile count for call/invoke instruction.");
81   if (hasSampleProfile()) {
82     // In sample PGO mode, check if there is a profile metadata on the
83     // instruction. If it is present, determine hotness solely based on that,
84     // since the sampled entry count may not be accurate. If there is no
85     // annotated on the instruction, return None.
86     uint64_t TotalCount;
87     if (Call.extractProfTotalWeight(TotalCount))
88       return TotalCount;
89     return None;
90   }
91   if (BFI)
92     return BFI->getBlockProfileCount(Call.getParent(), AllowSynthetic);
93   return None;
94 }
95 
96 /// Returns true if the function's entry is hot. If it returns false, it
97 /// either means it is not hot or it is unknown whether it is hot or not (for
98 /// example, no profile data is available).
99 bool ProfileSummaryInfo::isFunctionEntryHot(const Function *F) const {
100   if (!F || !hasProfileSummary())
101     return false;
102   auto FunctionCount = F->getEntryCount();
103   // FIXME: The heuristic used below for determining hotness is based on
104   // preliminary SPEC tuning for inliner. This will eventually be a
105   // convenience method that calls isHotCount.
106   return FunctionCount && isHotCount(FunctionCount.getCount());
107 }
108 
109 /// Returns true if the function contains hot code. This can include a hot
110 /// function entry count, hot basic block, or (in the case of Sample PGO)
111 /// hot total call edge count.
112 /// If it returns false, it either means it is not hot or it is unknown
113 /// (for example, no profile data is available).
114 bool ProfileSummaryInfo::isFunctionHotInCallGraph(
115     const Function *F, BlockFrequencyInfo &BFI) const {
116   if (!F || !hasProfileSummary())
117     return false;
118   if (auto FunctionCount = F->getEntryCount())
119     if (isHotCount(FunctionCount.getCount()))
120       return true;
121 
122   if (hasSampleProfile()) {
123     uint64_t TotalCallCount = 0;
124     for (const auto &BB : *F)
125       for (const auto &I : BB)
126         if (isa<CallInst>(I) || isa<InvokeInst>(I))
127           if (auto CallCount = getProfileCount(cast<CallBase>(I), nullptr))
128             TotalCallCount += CallCount.getValue();
129     if (isHotCount(TotalCallCount))
130       return true;
131   }
132   for (const auto &BB : *F)
133     if (isHotBlock(&BB, &BFI))
134       return true;
135   return false;
136 }
137 
138 /// Returns true if the function only contains cold code. This means that
139 /// the function entry and blocks are all cold, and (in the case of Sample PGO)
140 /// the total call edge count is cold.
141 /// If it returns false, it either means it is not cold or it is unknown
142 /// (for example, no profile data is available).
143 bool ProfileSummaryInfo::isFunctionColdInCallGraph(
144     const Function *F, BlockFrequencyInfo &BFI) const {
145   if (!F || !hasProfileSummary())
146     return false;
147   if (auto FunctionCount = F->getEntryCount())
148     if (!isColdCount(FunctionCount.getCount()))
149       return false;
150 
151   if (hasSampleProfile()) {
152     uint64_t TotalCallCount = 0;
153     for (const auto &BB : *F)
154       for (const auto &I : BB)
155         if (isa<CallInst>(I) || isa<InvokeInst>(I))
156           if (auto CallCount = getProfileCount(cast<CallBase>(I), nullptr))
157             TotalCallCount += CallCount.getValue();
158     if (!isColdCount(TotalCallCount))
159       return false;
160   }
161   for (const auto &BB : *F)
162     if (!isColdBlock(&BB, &BFI))
163       return false;
164   return true;
165 }
166 
167 bool ProfileSummaryInfo::isFunctionHotnessUnknown(const Function &F) const {
168   assert(hasPartialSampleProfile() && "Expect partial sample profile");
169   return !F.getEntryCount().hasValue();
170 }
171 
172 template <bool isHot>
173 bool ProfileSummaryInfo::isFunctionHotOrColdInCallGraphNthPercentile(
174     int PercentileCutoff, const Function *F, BlockFrequencyInfo &BFI) const {
175   if (!F || !hasProfileSummary())
176     return false;
177   if (auto FunctionCount = F->getEntryCount()) {
178     if (isHot &&
179         isHotCountNthPercentile(PercentileCutoff, FunctionCount.getCount()))
180       return true;
181     if (!isHot &&
182         !isColdCountNthPercentile(PercentileCutoff, FunctionCount.getCount()))
183       return false;
184   }
185   if (hasSampleProfile()) {
186     uint64_t TotalCallCount = 0;
187     for (const auto &BB : *F)
188       for (const auto &I : BB)
189         if (isa<CallInst>(I) || isa<InvokeInst>(I))
190           if (auto CallCount = getProfileCount(cast<CallBase>(I), nullptr))
191             TotalCallCount += CallCount.getValue();
192     if (isHot && isHotCountNthPercentile(PercentileCutoff, TotalCallCount))
193       return true;
194     if (!isHot && !isColdCountNthPercentile(PercentileCutoff, TotalCallCount))
195       return false;
196   }
197   for (const auto &BB : *F) {
198     if (isHot && isHotBlockNthPercentile(PercentileCutoff, &BB, &BFI))
199       return true;
200     if (!isHot && !isColdBlockNthPercentile(PercentileCutoff, &BB, &BFI))
201       return false;
202   }
203   return !isHot;
204 }
205 
206 // Like isFunctionHotInCallGraph but for a given cutoff.
207 bool ProfileSummaryInfo::isFunctionHotInCallGraphNthPercentile(
208     int PercentileCutoff, const Function *F, BlockFrequencyInfo &BFI) const {
209   return isFunctionHotOrColdInCallGraphNthPercentile<true>(
210       PercentileCutoff, F, BFI);
211 }
212 
213 bool ProfileSummaryInfo::isFunctionColdInCallGraphNthPercentile(
214     int PercentileCutoff, const Function *F, BlockFrequencyInfo &BFI) const {
215   return isFunctionHotOrColdInCallGraphNthPercentile<false>(
216       PercentileCutoff, F, BFI);
217 }
218 
219 /// Returns true if the function's entry is a cold. If it returns false, it
220 /// either means it is not cold or it is unknown whether it is cold or not (for
221 /// example, no profile data is available).
222 bool ProfileSummaryInfo::isFunctionEntryCold(const Function *F) const {
223   if (!F)
224     return false;
225   if (F->hasFnAttribute(Attribute::Cold))
226     return true;
227   if (!hasProfileSummary())
228     return false;
229   auto FunctionCount = F->getEntryCount();
230   // FIXME: The heuristic used below for determining coldness is based on
231   // preliminary SPEC tuning for inliner. This will eventually be a
232   // convenience method that calls isHotCount.
233   return FunctionCount && isColdCount(FunctionCount.getCount());
234 }
235 
236 /// Compute the hot and cold thresholds.
237 void ProfileSummaryInfo::computeThresholds() {
238   auto &DetailedSummary = Summary->getDetailedSummary();
239   auto &HotEntry = ProfileSummaryBuilder::getEntryForPercentile(
240       DetailedSummary, ProfileSummaryCutoffHot);
241   HotCountThreshold =
242       ProfileSummaryBuilder::getHotCountThreshold(DetailedSummary);
243   ColdCountThreshold =
244       ProfileSummaryBuilder::getColdCountThreshold(DetailedSummary);
245   assert(ColdCountThreshold <= HotCountThreshold &&
246          "Cold count threshold cannot exceed hot count threshold!");
247   if (!hasPartialSampleProfile() || !ScalePartialSampleProfileWorkingSetSize) {
248     HasHugeWorkingSetSize =
249         HotEntry.NumCounts > ProfileSummaryHugeWorkingSetSizeThreshold;
250     HasLargeWorkingSetSize =
251         HotEntry.NumCounts > ProfileSummaryLargeWorkingSetSizeThreshold;
252   } else {
253     // Scale the working set size of the partial sample profile to reflect the
254     // size of the program being compiled.
255     double PartialProfileRatio = Summary->getPartialProfileRatio();
256     uint64_t ScaledHotEntryNumCounts =
257         static_cast<uint64_t>(HotEntry.NumCounts * PartialProfileRatio *
258                               PartialSampleProfileWorkingSetSizeScaleFactor);
259     HasHugeWorkingSetSize =
260         ScaledHotEntryNumCounts > ProfileSummaryHugeWorkingSetSizeThreshold;
261     HasLargeWorkingSetSize =
262         ScaledHotEntryNumCounts > ProfileSummaryLargeWorkingSetSizeThreshold;
263   }
264 }
265 
266 Optional<uint64_t>
267 ProfileSummaryInfo::computeThreshold(int PercentileCutoff) const {
268   if (!hasProfileSummary())
269     return None;
270   auto iter = ThresholdCache.find(PercentileCutoff);
271   if (iter != ThresholdCache.end()) {
272     return iter->second;
273   }
274   auto &DetailedSummary = Summary->getDetailedSummary();
275   auto &Entry = ProfileSummaryBuilder::getEntryForPercentile(DetailedSummary,
276                                                              PercentileCutoff);
277   uint64_t CountThreshold = Entry.MinCount;
278   ThresholdCache[PercentileCutoff] = CountThreshold;
279   return CountThreshold;
280 }
281 
282 bool ProfileSummaryInfo::hasHugeWorkingSetSize() const {
283   return HasHugeWorkingSetSize && HasHugeWorkingSetSize.getValue();
284 }
285 
286 bool ProfileSummaryInfo::hasLargeWorkingSetSize() const {
287   return HasLargeWorkingSetSize && HasLargeWorkingSetSize.getValue();
288 }
289 
290 bool ProfileSummaryInfo::isHotCount(uint64_t C) const {
291   return HotCountThreshold && C >= HotCountThreshold.getValue();
292 }
293 
294 bool ProfileSummaryInfo::isColdCount(uint64_t C) const {
295   return ColdCountThreshold && C <= ColdCountThreshold.getValue();
296 }
297 
298 template <bool isHot>
299 bool ProfileSummaryInfo::isHotOrColdCountNthPercentile(int PercentileCutoff,
300                                                        uint64_t C) const {
301   auto CountThreshold = computeThreshold(PercentileCutoff);
302   if (isHot)
303     return CountThreshold && C >= CountThreshold.getValue();
304   else
305     return CountThreshold && C <= CountThreshold.getValue();
306 }
307 
308 bool ProfileSummaryInfo::isHotCountNthPercentile(int PercentileCutoff,
309                                                  uint64_t C) const {
310   return isHotOrColdCountNthPercentile<true>(PercentileCutoff, C);
311 }
312 
313 bool ProfileSummaryInfo::isColdCountNthPercentile(int PercentileCutoff,
314                                                   uint64_t C) const {
315   return isHotOrColdCountNthPercentile<false>(PercentileCutoff, C);
316 }
317 
318 uint64_t ProfileSummaryInfo::getOrCompHotCountThreshold() const {
319   return HotCountThreshold ? HotCountThreshold.getValue() : UINT64_MAX;
320 }
321 
322 uint64_t ProfileSummaryInfo::getOrCompColdCountThreshold() const {
323   return ColdCountThreshold ? ColdCountThreshold.getValue() : 0;
324 }
325 
326 bool ProfileSummaryInfo::isHotBlock(const BasicBlock *BB,
327                                     BlockFrequencyInfo *BFI) const {
328   auto Count = BFI->getBlockProfileCount(BB);
329   return Count && isHotCount(*Count);
330 }
331 
332 bool ProfileSummaryInfo::isColdBlock(const BasicBlock *BB,
333                                      BlockFrequencyInfo *BFI) const {
334   auto Count = BFI->getBlockProfileCount(BB);
335   return Count && isColdCount(*Count);
336 }
337 
338 template <bool isHot>
339 bool ProfileSummaryInfo::isHotOrColdBlockNthPercentile(
340     int PercentileCutoff, const BasicBlock *BB, BlockFrequencyInfo *BFI) const {
341   auto Count = BFI->getBlockProfileCount(BB);
342   if (isHot)
343     return Count && isHotCountNthPercentile(PercentileCutoff, *Count);
344   else
345     return Count && isColdCountNthPercentile(PercentileCutoff, *Count);
346 }
347 
348 bool ProfileSummaryInfo::isHotBlockNthPercentile(
349     int PercentileCutoff, const BasicBlock *BB, BlockFrequencyInfo *BFI) const {
350   return isHotOrColdBlockNthPercentile<true>(PercentileCutoff, BB, BFI);
351 }
352 
353 bool ProfileSummaryInfo::isColdBlockNthPercentile(
354     int PercentileCutoff, const BasicBlock *BB, BlockFrequencyInfo *BFI) const {
355   return isHotOrColdBlockNthPercentile<false>(PercentileCutoff, BB, BFI);
356 }
357 
358 bool ProfileSummaryInfo::isHotCallSite(const CallBase &CB,
359                                        BlockFrequencyInfo *BFI) const {
360   auto C = getProfileCount(CB, BFI);
361   return C && isHotCount(*C);
362 }
363 
364 bool ProfileSummaryInfo::isColdCallSite(const CallBase &CB,
365                                         BlockFrequencyInfo *BFI) const {
366   auto C = getProfileCount(CB, BFI);
367   if (C)
368     return isColdCount(*C);
369 
370   // In SamplePGO, if the caller has been sampled, and there is no profile
371   // annotated on the callsite, we consider the callsite as cold.
372   return hasSampleProfile() && CB.getCaller()->hasProfileData();
373 }
374 
375 bool ProfileSummaryInfo::hasPartialSampleProfile() const {
376   return hasProfileSummary() &&
377          Summary->getKind() == ProfileSummary::PSK_Sample &&
378          (PartialProfile || Summary->isPartialProfile());
379 }
380 
381 INITIALIZE_PASS(ProfileSummaryInfoWrapperPass, "profile-summary-info",
382                 "Profile summary info", false, true)
383 
384 ProfileSummaryInfoWrapperPass::ProfileSummaryInfoWrapperPass()
385     : ImmutablePass(ID) {
386   initializeProfileSummaryInfoWrapperPassPass(*PassRegistry::getPassRegistry());
387 }
388 
389 bool ProfileSummaryInfoWrapperPass::doInitialization(Module &M) {
390   PSI.reset(new ProfileSummaryInfo(M));
391   return false;
392 }
393 
394 bool ProfileSummaryInfoWrapperPass::doFinalization(Module &M) {
395   PSI.reset();
396   return false;
397 }
398 
399 AnalysisKey ProfileSummaryAnalysis::Key;
400 ProfileSummaryInfo ProfileSummaryAnalysis::run(Module &M,
401                                                ModuleAnalysisManager &) {
402   return ProfileSummaryInfo(M);
403 }
404 
405 PreservedAnalyses ProfileSummaryPrinterPass::run(Module &M,
406                                                  ModuleAnalysisManager &AM) {
407   ProfileSummaryInfo &PSI = AM.getResult<ProfileSummaryAnalysis>(M);
408 
409   OS << "Functions in " << M.getName() << " with hot/cold annotations: \n";
410   for (auto &F : M) {
411     OS << F.getName();
412     if (PSI.isFunctionEntryHot(&F))
413       OS << " :hot entry ";
414     else if (PSI.isFunctionEntryCold(&F))
415       OS << " :cold entry ";
416     OS << "\n";
417   }
418   return PreservedAnalyses::all();
419 }
420 
421 char ProfileSummaryInfoWrapperPass::ID = 0;
422