1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-25 20:23:11 +01:00
llvm-mirror/lib/ProfileData/ProfileSummaryBuilder.cpp
Hongtao Yu 45a66978f9 [CSSPGO] Report zero-count probe in profile instead of dangling probes.
Previously dangling samples were represented by INT64_MAX in sample profile while probes never executed were not reported. This was based on an observation that dangling probes were only at a smaller portion than zero-count probes. However, with compiler optimizations, dangling probes end up becoming at large portion of all probes in general and reporting them does not make sense from profile size point of view. This change flips sample reporting by reporting zero-count probes instead. This enabled dangling probe to be represented by none (missing entry in profile). This has a couple benefits:

1. Reducing sample profile size in optimize mode, even when the number of non-executed probes outperform the number of dangling probes, since INT64_MAX takes more space over 0 to encode.

2. Binary size savings. No need to encode dangling probe anymore, since missing probes are treated as dangling in the profile reader.

3. Reducing compiler work to track dangling probes. However, for probes that are real dead and removed, we still need the compiler to identify them so that they can be reported as zero-count, instead of mistreated as dangling probes.

4. Improving counts quality by respecting the counts already collected on the non-dangling copy of a probe. A probe, when duplicated, gets two copies at runtime. If one of them is dangling while the other is not, merging the two probes at profile generation time will cause the real samples collected on the non-dangling one to be discarded. Not reporting the dangling counterpart will keep the real samples.

5. Better readability.

6. Be consistent with non-CS dwarf line number based profile. Zero counts are trusted by the compiler counts inferencer while missing counts will be inferred by the compiler.

Note that the current patch does include any work for #3. There will be follow-up changes.

For #1, I've seen for a large Facebook service, the text profile is reduced by 7%. For extbinary profile, the size of  LBRProfileSection is reduced by 35%.

For #4, I have seen general counts quality for SPEC2017 is improved by 10%.

Reviewed By: wenlei, wlei, wmi

Differential Revision: https://reviews.llvm.org/D104129
2021-06-16 11:45:29 -07:00

239 lines
9.2 KiB
C++

//=-- ProfilesummaryBuilder.cpp - Profile summary computation ---------------=//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file contains support for computing profile summary data.
//
//===----------------------------------------------------------------------===//
#include "llvm/IR/Attributes.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/Metadata.h"
#include "llvm/IR/Type.h"
#include "llvm/ProfileData/InstrProf.h"
#include "llvm/ProfileData/ProfileCommon.h"
#include "llvm/ProfileData/SampleProf.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/CommandLine.h"
using namespace llvm;
cl::opt<bool> UseContextLessSummary(
"profile-summary-contextless", cl::Hidden, cl::init(false), cl::ZeroOrMore,
cl::desc("Merge context profiles before calculating thresholds."));
// The following two parameters determine the threshold for a count to be
// considered hot/cold. These two parameters are percentile values (multiplied
// by 10000). If the counts are sorted in descending order, the minimum count to
// reach ProfileSummaryCutoffHot gives the threshold to determine a hot count.
// Similarly, the minimum count to reach ProfileSummaryCutoffCold gives the
// threshold for determining cold count (everything <= this threshold is
// considered cold).
cl::opt<int> ProfileSummaryCutoffHot(
"profile-summary-cutoff-hot", cl::Hidden, cl::init(990000), cl::ZeroOrMore,
cl::desc("A count is hot if it exceeds the minimum count to"
" reach this percentile of total counts."));
cl::opt<int> ProfileSummaryCutoffCold(
"profile-summary-cutoff-cold", cl::Hidden, cl::init(999999), cl::ZeroOrMore,
cl::desc("A count is cold if it is below the minimum count"
" to reach this percentile of total counts."));
cl::opt<unsigned> ProfileSummaryHugeWorkingSetSizeThreshold(
"profile-summary-huge-working-set-size-threshold", cl::Hidden,
cl::init(15000), cl::ZeroOrMore,
cl::desc("The code working set size is considered huge if the number of"
" blocks required to reach the -profile-summary-cutoff-hot"
" percentile exceeds this count."));
cl::opt<unsigned> ProfileSummaryLargeWorkingSetSizeThreshold(
"profile-summary-large-working-set-size-threshold", cl::Hidden,
cl::init(12500), cl::ZeroOrMore,
cl::desc("The code working set size is considered large if the number of"
" blocks required to reach the -profile-summary-cutoff-hot"
" percentile exceeds this count."));
// The next two options override the counts derived from summary computation and
// are useful for debugging purposes.
cl::opt<int> ProfileSummaryHotCount(
"profile-summary-hot-count", cl::ReallyHidden, cl::ZeroOrMore,
cl::desc("A fixed hot count that overrides the count derived from"
" profile-summary-cutoff-hot"));
cl::opt<int> ProfileSummaryColdCount(
"profile-summary-cold-count", cl::ReallyHidden, cl::ZeroOrMore,
cl::desc("A fixed cold count that overrides the count derived from"
" profile-summary-cutoff-cold"));
// A set of cutoff values. Each value, when divided by ProfileSummary::Scale
// (which is 1000000) is a desired percentile of total counts.
static const uint32_t DefaultCutoffsData[] = {
10000, /* 1% */
100000, /* 10% */
200000, 300000, 400000, 500000, 600000, 700000, 800000,
900000, 950000, 990000, 999000, 999900, 999990, 999999};
const ArrayRef<uint32_t> ProfileSummaryBuilder::DefaultCutoffs =
DefaultCutoffsData;
const ProfileSummaryEntry &
ProfileSummaryBuilder::getEntryForPercentile(SummaryEntryVector &DS,
uint64_t Percentile) {
auto It = partition_point(DS, [=](const ProfileSummaryEntry &Entry) {
return Entry.Cutoff < Percentile;
});
// The required percentile has to be <= one of the percentiles in the
// detailed summary.
if (It == DS.end())
report_fatal_error("Desired percentile exceeds the maximum cutoff");
return *It;
}
void InstrProfSummaryBuilder::addRecord(const InstrProfRecord &R) {
// The first counter is not necessarily an entry count for IR
// instrumentation profiles.
// Eventually MaxFunctionCount will become obsolete and this can be
// removed.
addEntryCount(R.Counts[0]);
for (size_t I = 1, E = R.Counts.size(); I < E; ++I)
addInternalCount(R.Counts[I]);
}
// To compute the detailed summary, we consider each line containing samples as
// equivalent to a block with a count in the instrumented profile.
void SampleProfileSummaryBuilder::addRecord(
const sampleprof::FunctionSamples &FS, bool isCallsiteSample) {
if (!isCallsiteSample) {
NumFunctions++;
if (FS.getHeadSamples() > MaxFunctionCount)
MaxFunctionCount = FS.getHeadSamples();
}
for (const auto &I : FS.getBodySamples()) {
uint64_t Count = I.second.getSamples();
addCount(Count);
}
for (const auto &I : FS.getCallsiteSamples())
for (const auto &CS : I.second)
addRecord(CS.second, true);
}
// The argument to this method is a vector of cutoff percentages and the return
// value is a vector of (Cutoff, MinCount, NumCounts) triplets.
void ProfileSummaryBuilder::computeDetailedSummary() {
if (DetailedSummaryCutoffs.empty())
return;
llvm::sort(DetailedSummaryCutoffs);
auto Iter = CountFrequencies.begin();
const auto End = CountFrequencies.end();
uint32_t CountsSeen = 0;
uint64_t CurrSum = 0, Count = 0;
for (const uint32_t Cutoff : DetailedSummaryCutoffs) {
assert(Cutoff <= 999999);
APInt Temp(128, TotalCount);
APInt N(128, Cutoff);
APInt D(128, ProfileSummary::Scale);
Temp *= N;
Temp = Temp.sdiv(D);
uint64_t DesiredCount = Temp.getZExtValue();
assert(DesiredCount <= TotalCount);
while (CurrSum < DesiredCount && Iter != End) {
Count = Iter->first;
uint32_t Freq = Iter->second;
CurrSum += (Count * Freq);
CountsSeen += Freq;
Iter++;
}
assert(CurrSum >= DesiredCount);
ProfileSummaryEntry PSE = {Cutoff, Count, CountsSeen};
DetailedSummary.push_back(PSE);
}
}
uint64_t ProfileSummaryBuilder::getHotCountThreshold(SummaryEntryVector &DS) {
auto &HotEntry =
ProfileSummaryBuilder::getEntryForPercentile(DS, ProfileSummaryCutoffHot);
uint64_t HotCountThreshold = HotEntry.MinCount;
if (ProfileSummaryHotCount.getNumOccurrences() > 0)
HotCountThreshold = ProfileSummaryHotCount;
return HotCountThreshold;
}
uint64_t ProfileSummaryBuilder::getColdCountThreshold(SummaryEntryVector &DS) {
auto &ColdEntry = ProfileSummaryBuilder::getEntryForPercentile(
DS, ProfileSummaryCutoffCold);
uint64_t ColdCountThreshold = ColdEntry.MinCount;
if (ProfileSummaryColdCount.getNumOccurrences() > 0)
ColdCountThreshold = ProfileSummaryColdCount;
return ColdCountThreshold;
}
std::unique_ptr<ProfileSummary> SampleProfileSummaryBuilder::getSummary() {
computeDetailedSummary();
return std::make_unique<ProfileSummary>(
ProfileSummary::PSK_Sample, DetailedSummary, TotalCount, MaxCount, 0,
MaxFunctionCount, NumCounts, NumFunctions);
}
std::unique_ptr<ProfileSummary>
SampleProfileSummaryBuilder::computeSummaryForProfiles(
const StringMap<sampleprof::FunctionSamples> &Profiles) {
assert(NumFunctions == 0 &&
"This can only be called on an empty summary builder");
StringMap<sampleprof::FunctionSamples> ContextLessProfiles;
const StringMap<sampleprof::FunctionSamples> *ProfilesToUse = &Profiles;
// For CSSPGO, context-sensitive profile effectively split a function profile
// into many copies each representing the CFG profile of a particular calling
// context. That makes the count distribution looks more flat as we now have
// more function profiles each with lower counts, which in turn leads to lower
// hot thresholds. To compensate for that, by defauly we merge context
// profiles before coumputing profile summary.
if (UseContextLessSummary || (sampleprof::FunctionSamples::ProfileIsCS &&
!UseContextLessSummary.getNumOccurrences())) {
for (const auto &I : Profiles) {
ContextLessProfiles[I.second.getName()].merge(I.second);
}
ProfilesToUse = &ContextLessProfiles;
}
for (const auto &I : *ProfilesToUse) {
const sampleprof::FunctionSamples &Profile = I.second;
addRecord(Profile);
}
return getSummary();
}
std::unique_ptr<ProfileSummary> InstrProfSummaryBuilder::getSummary() {
computeDetailedSummary();
return std::make_unique<ProfileSummary>(
ProfileSummary::PSK_Instr, DetailedSummary, TotalCount, MaxCount,
MaxInternalBlockCount, MaxFunctionCount, NumCounts, NumFunctions);
}
void InstrProfSummaryBuilder::addEntryCount(uint64_t Count) {
NumFunctions++;
// Skip invalid count.
if (Count == (uint64_t)-1)
return;
addCount(Count);
if (Count > MaxFunctionCount)
MaxFunctionCount = Count;
}
void InstrProfSummaryBuilder::addInternalCount(uint64_t Count) {
// Skip invalid count.
if (Count == (uint64_t)-1)
return;
addCount(Count);
if (Count > MaxInternalBlockCount)
MaxInternalBlockCount = Count;
}