void Analysis::writeSnippet(raw_ostream &OS, ArrayRef<uint8_t> Bytes,
const char *Separator) const {
SmallVector<std::string, 3> Lines;
- const auto &SI = State_.getSubtargetInfo();
// Parse the asm snippet and print it.
while (!Bytes.empty()) {
MCInst MI;
}
SmallString<128> InstPrinterStr; // FIXME: magic number.
raw_svector_ostream OSS(InstPrinterStr);
- InstPrinter_->printInst(&MI, 0, "", SI, OSS);
+ InstPrinter_->printInst(&MI, 0, "", *SubtargetInfo_, OSS);
Bytes = Bytes.drop_front(MISize);
Lines.emplace_back(InstPrinterStr.str().trim());
}
const MCInst &MCI = Point.keyInstruction();
unsigned SchedClassId;
std::tie(SchedClassId, std::ignore) = ResolvedSchedClass::resolveSchedClassId(
- State_.getSubtargetInfo(), State_.getInstrInfo(), MCI);
+ *SubtargetInfo_, *InstrInfo_, MCI);
#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
const MCSchedClassDesc *const SCDesc =
- State_.getSubtargetInfo().getSchedModel().getSchedClassDesc(SchedClassId);
+ SubtargetInfo_->getSchedModel().getSchedClassDesc(SchedClassId);
writeEscaped<kEscapeCsv>(OS, SCDesc->Name);
#else
OS << SchedClassId;
OS << "\n";
}
-Analysis::Analysis(const LLVMState &State,
+Analysis::Analysis(const Target &Target,
+ std::unique_ptr<MCSubtargetInfo> SubtargetInfo,
+ std::unique_ptr<MCInstrInfo> InstrInfo,
const InstructionBenchmarkClustering &Clustering,
double AnalysisInconsistencyEpsilon,
- bool AnalysisDisplayUnstableOpcodes)
- : Clustering_(Clustering), State_(State),
+ bool AnalysisDisplayUnstableOpcodes,
+ const std::string &ForceCpuName)
+ : Clustering_(Clustering), SubtargetInfo_(std::move(SubtargetInfo)),
+ InstrInfo_(std::move(InstrInfo)),
AnalysisInconsistencyEpsilonSquared_(AnalysisInconsistencyEpsilon *
AnalysisInconsistencyEpsilon),
AnalysisDisplayUnstableOpcodes_(AnalysisDisplayUnstableOpcodes) {
if (Clustering.getPoints().empty())
return;
+ const InstructionBenchmark &FirstPoint = Clustering.getPoints().front();
+ const std::string CpuName =
+ ForceCpuName.empty() ? FirstPoint.CpuName : ForceCpuName;
+ RegInfo_.reset(Target.createMCRegInfo(FirstPoint.LLVMTriple));
MCTargetOptions MCOptions;
- const auto &TM = State.getTargetMachine();
- const auto &Triple = TM.getTargetTriple();
- AsmInfo_.reset(TM.getTarget().createMCAsmInfo(State_.getRegInfo(),
- Triple.str(), MCOptions));
- InstPrinter_.reset(TM.getTarget().createMCInstPrinter(
- Triple, 0 /*default variant*/, *AsmInfo_, State_.getInstrInfo(),
- State_.getRegInfo()));
-
- Context_ = std::make_unique<MCContext>(
- Triple, AsmInfo_.get(), &State_.getRegInfo(), &State_.getSubtargetInfo());
- Disasm_.reset(TM.getTarget().createMCDisassembler(State_.getSubtargetInfo(),
- *Context_));
+ AsmInfo_.reset(
+ Target.createMCAsmInfo(*RegInfo_, FirstPoint.LLVMTriple, MCOptions));
+ SubtargetInfo_.reset(
+ Target.createMCSubtargetInfo(FirstPoint.LLVMTriple, CpuName, ""));
+ InstPrinter_.reset(Target.createMCInstPrinter(
+ Triple(FirstPoint.LLVMTriple), 0 /*default variant*/, *AsmInfo_,
+ *InstrInfo_, *RegInfo_));
+
+ Context_ =
+ std::make_unique<MCContext>(Triple(FirstPoint.LLVMTriple), AsmInfo_.get(),
+ RegInfo_.get(), SubtargetInfo_.get());
+ Disasm_.reset(Target.createMCDisassembler(*SubtargetInfo_, *Context_));
assert(Disasm_ && "cannot create MCDisassembler. missing call to "
"InitializeXXXTargetDisassembler ?");
}
unsigned SchedClassId;
bool WasVariant;
std::tie(SchedClassId, WasVariant) =
- ResolvedSchedClass::resolveSchedClassId(State_.getSubtargetInfo(),
- State_.getInstrInfo(), MCI);
+ ResolvedSchedClass::resolveSchedClassId(*SubtargetInfo_, *InstrInfo_,
+ MCI);
const auto IndexIt = SchedClassIdToIndex.find(SchedClassId);
if (IndexIt == SchedClassIdToIndex.end()) {
// Create a new entry.
SchedClassIdToIndex.emplace(SchedClassId, Entries.size());
- ResolvedSchedClassAndPoints Entry(ResolvedSchedClass(
- State_.getSubtargetInfo(), SchedClassId, WasVariant));
+ ResolvedSchedClassAndPoints Entry(
+ ResolvedSchedClass(*SubtargetInfo_, SchedClassId, WasVariant));
Entry.PointIds.push_back(PointId);
Entries.push_back(std::move(Entry));
} else {
OS << "\">";
switch (Point.Mode) {
case InstructionBenchmark::Latency:
- writeLatencySnippetHtml(OS, Point.Key.Instructions, State_.getInstrInfo());
+ writeLatencySnippetHtml(OS, Point.Key.Instructions, *InstrInfo_);
break;
case InstructionBenchmark::Uops:
case InstructionBenchmark::InverseThroughput:
- writeParallelSnippetHtml(OS, Point.Key.Instructions, State_.getInstrInfo());
+ writeParallelSnippetHtml(OS, Point.Key.Instructions, *InstrInfo_);
break;
default:
llvm_unreachable("invalid mode");
OS << "</tr>";
for (const SchedClassCluster &Cluster : Clusters) {
OS << "<tr class=\""
- << (Cluster.measurementsMatch(State_.getSubtargetInfo(), RSC,
- Clustering_,
+ << (Cluster.measurementsMatch(*SubtargetInfo_, RSC, Clustering_,
AnalysisInconsistencyEpsilonSquared_)
? "good-cluster"
: "bad-cluster")
"idealized unit resource (port) pressure assuming ideal "
"distribution\">Idealized Resource Pressure</th></tr>";
if (RSC.SCDesc->isValid()) {
- const auto &SI = State_.getSubtargetInfo();
- const auto &SM = SI.getSchedModel();
+ const auto &SM = SubtargetInfo_->getSchedModel();
OS << "<tr><td>✔</td>";
OS << "<td>" << (RSC.WasVariant ? "✔" : "✕") << "</td>";
OS << "<td>" << RSC.SCDesc->NumMicroOps << "</td>";
// Latencies.
OS << "<td><ul>";
for (int I = 0, E = RSC.SCDesc->NumWriteLatencyEntries; I < E; ++I) {
- const auto *const Entry = SI.getWriteLatencyEntry(RSC.SCDesc, I);
+ const auto *const Entry =
+ SubtargetInfo_->getWriteLatencyEntry(RSC.SCDesc, I);
OS << "<li>" << Entry->Cycles;
if (RSC.SCDesc->NumWriteLatencyEntries > 1) {
// Dismabiguate if more than 1 latency.
// inverse throughput.
OS << "<td>";
writeMeasurementValue<kEscapeHtml>(
- OS, MCSchedModel::getReciprocalThroughput(SI, *RSC.SCDesc));
+ OS,
+ MCSchedModel::getReciprocalThroughput(*SubtargetInfo_, *RSC.SCDesc));
OS << "</td>";
// WriteProcRes.
OS << "<td><ul>";
OS << "<td><ul>";
for (const auto &Pressure : RSC.IdealizedProcResPressure) {
OS << "<li><span class=\"mono\">";
- writeEscaped<kEscapeHtml>(
- OS, SI.getSchedModel().getProcResource(Pressure.first)->Name);
+ writeEscaped<kEscapeHtml>(OS, SubtargetInfo_->getSchedModel()
+ .getProcResource(Pressure.first)
+ ->Name);
OS << "</span>: ";
writeMeasurementValue<kEscapeHtml>(OS, Pressure.second);
OS << "</li>";
writeEscaped<kEscapeHtml>(OS, FirstPoint.CpuName);
OS << "</span></h3>";
- const auto &SI = State_.getSubtargetInfo();
for (const auto &RSCAndPoints : makePointsPerSchedClass()) {
if (!RSCAndPoints.RSC.SCDesc)
continue;
// Print any scheduling class that has at least one cluster that does not
// match the checked-in data.
- if (all_of(SchedClassClusters, [this, &RSCAndPoints,
- &SI](const SchedClassCluster &C) {
- return C.measurementsMatch(SI, RSCAndPoints.RSC, Clustering_,
+ if (all_of(SchedClassClusters, [this,
+ &RSCAndPoints](const SchedClassCluster &C) {
+ return C.measurementsMatch(*SubtargetInfo_, RSCAndPoints.RSC,
+ Clustering_,
AnalysisInconsistencyEpsilonSquared_);
}))
continue; // Nothing weird.
// A helper class to analyze benchmark results for a target.
class Analysis {
public:
- Analysis(const LLVMState &State,
+ Analysis(const Target &Target, std::unique_ptr<MCSubtargetInfo> SubtargetInfo,
+ std::unique_ptr<MCInstrInfo> InstrInfo,
const InstructionBenchmarkClustering &Clustering,
double AnalysisInconsistencyEpsilon,
- bool AnalysisDisplayUnstableOpcodes);
+ bool AnalysisDisplayUnstableOpcodes,
+ const std::string &ForceCpuName = "");
// Prints a csv of instructions for each cluster.
struct PrintClusters {};
const char *Separator) const;
const InstructionBenchmarkClustering &Clustering_;
- const LLVMState &State_;
std::unique_ptr<MCContext> Context_;
+ std::unique_ptr<MCSubtargetInfo> SubtargetInfo_;
+ std::unique_ptr<MCInstrInfo> InstrInfo_;
+ std::unique_ptr<MCRegisterInfo> RegInfo_;
std::unique_ptr<MCAsmInfo> AsmInfo_;
std::unique_ptr<MCInstPrinter> InstPrinter_;
std::unique_ptr<MCDisassembler> Disasm_;
}
};
-template <> struct MappingTraits<exegesis::InstructionBenchmark::TripleAndCpu> {
- static void mapping(IO &Io,
- exegesis::InstructionBenchmark::TripleAndCpu &Obj) {
- assert(!Io.outputting() && "can only read TripleAndCpu");
- // Read triple.
- Io.mapRequired("llvm_triple", Obj.LLVMTriple);
- Io.mapRequired("cpu_name", Obj.CpuName);
- // Drop everything else.
- }
-};
-
} // namespace yaml
namespace exegesis {
-Expected<std::set<InstructionBenchmark::TripleAndCpu>>
-InstructionBenchmark::readTriplesAndCpusFromYamls(MemoryBufferRef Buffer) {
- // We're only mapping a field, drop other fields and silence the corresponding
- // warnings.
- yaml::Input Yin(
- Buffer, nullptr, +[](const SMDiagnostic &, void *Context) {});
- Yin.setAllowUnknownKeys(true);
- std::set<TripleAndCpu> Result;
- yaml::EmptyContext Context;
- while (Yin.setCurrentDocument()) {
- TripleAndCpu TC;
- yamlize(Yin, TC, /*unused*/ true, Context);
- if (Yin.error())
- return errorCodeToError(Yin.error());
- Result.insert(TC);
- Yin.nextDocument();
- }
- return Result;
-}
-
Expected<InstructionBenchmark>
-InstructionBenchmark::readYaml(const LLVMState &State, MemoryBufferRef Buffer) {
- yaml::Input Yin(Buffer);
- YamlContext Context(State);
- InstructionBenchmark Benchmark;
- if (Yin.setCurrentDocument())
- yaml::yamlize(Yin, Benchmark, /*unused*/ true, Context);
- if (!Context.getLastError().empty())
- return make_error<Failure>(Context.getLastError());
- return Benchmark;
+InstructionBenchmark::readYaml(const LLVMState &State, StringRef Filename) {
+ if (auto ExpectedMemoryBuffer =
+ errorOrToExpected(MemoryBuffer::getFile(Filename, /*IsText=*/true))) {
+ yaml::Input Yin(*ExpectedMemoryBuffer.get());
+ YamlContext Context(State);
+ InstructionBenchmark Benchmark;
+ if (Yin.setCurrentDocument())
+ yaml::yamlize(Yin, Benchmark, /*unused*/ true, Context);
+ if (!Context.getLastError().empty())
+ return make_error<Failure>(Context.getLastError());
+ return Benchmark;
+ } else {
+ return ExpectedMemoryBuffer.takeError();
+ }
}
Expected<std::vector<InstructionBenchmark>>
-InstructionBenchmark::readYamls(const LLVMState &State,
- MemoryBufferRef Buffer) {
- yaml::Input Yin(Buffer);
- YamlContext Context(State);
- std::vector<InstructionBenchmark> Benchmarks;
- while (Yin.setCurrentDocument()) {
- Benchmarks.emplace_back();
- yamlize(Yin, Benchmarks.back(), /*unused*/ true, Context);
- if (Yin.error())
- return errorCodeToError(Yin.error());
- if (!Context.getLastError().empty())
- return make_error<Failure>(Context.getLastError());
- Yin.nextDocument();
+InstructionBenchmark::readYamls(const LLVMState &State, StringRef Filename) {
+ if (auto ExpectedMemoryBuffer =
+ errorOrToExpected(MemoryBuffer::getFile(Filename, /*IsText=*/true))) {
+ yaml::Input Yin(*ExpectedMemoryBuffer.get());
+ YamlContext Context(State);
+ std::vector<InstructionBenchmark> Benchmarks;
+ while (Yin.setCurrentDocument()) {
+ Benchmarks.emplace_back();
+ yamlize(Yin, Benchmarks.back(), /*unused*/ true, Context);
+ if (Yin.error())
+ return errorCodeToError(Yin.error());
+ if (!Context.getLastError().empty())
+ return make_error<Failure>(Context.getLastError());
+ Yin.nextDocument();
+ }
+ return Benchmarks;
+ } else {
+ return ExpectedMemoryBuffer.takeError();
}
- return Benchmarks;
}
Error InstructionBenchmark::writeYamlTo(const LLVMState &State,
#include "RegisterValue.h"
#include "llvm/ADT/StringMap.h"
#include "llvm/ADT/StringRef.h"
-#include "llvm/ADT/StringSet.h"
#include "llvm/MC/MCInst.h"
#include "llvm/MC/MCInstBuilder.h"
#include "llvm/Support/YAMLTraits.h"
#include <limits>
-#include <set>
#include <string>
#include <unordered_map>
#include <vector>
enum ResultAggregationModeE { Min, Max, Mean, MinVariance };
// Read functions.
static Expected<InstructionBenchmark> readYaml(const LLVMState &State,
- MemoryBufferRef Buffer);
+ StringRef Filename);
static Expected<std::vector<InstructionBenchmark>>
- readYamls(const LLVMState &State, MemoryBufferRef Buffer);
-
- // Given a set of serialized instruction benchmarks, returns the set of
- // triples and CPUs that appear in the list of benchmarks.
- struct TripleAndCpu {
- std::string LLVMTriple;
- std::string CpuName;
- bool operator<(const TripleAndCpu &O) const {
- return std::tie(LLVMTriple, CpuName) < std::tie(O.LLVMTriple, O.CpuName);
- }
- };
- static Expected<std::set<TripleAndCpu>>
- readTriplesAndCpusFromYamls(MemoryBufferRef Buffer);
+ readYamls(const LLVMState &State, StringRef Filename);
class Error readYamlFrom(const LLVMState &State, StringRef InputContent);
InitializeNativeTargetDisassembler();
InitializeNativeExegesisTarget();
- auto MemoryBuffer = ExitOnFileError(
- BenchmarkFile,
- errorOrToExpected(MemoryBuffer::getFile(BenchmarkFile, /*IsText=*/true)));
-
- const auto TriplesAndCpus = ExitOnFileError(
- BenchmarkFile,
- InstructionBenchmark::readTriplesAndCpusFromYamls(*MemoryBuffer));
- if (TriplesAndCpus.empty()) {
- errs() << "no benchmarks to analyze\n";
- return;
- }
- if (TriplesAndCpus.size() > 1) {
- ExitWithError("analysis file contains benchmarks from several CPUs. This "
- "is unsupported.");
- }
- auto TripleAndCpu = *TriplesAndCpus.begin();
- if (!CpuName.empty()) {
- llvm::errs() << "overridding file CPU name (" << TripleAndCpu.CpuName
- << ") with provided CPU name (" << CpuName << ")\n";
- TripleAndCpu.CpuName = CpuName;
- }
- llvm::errs() << "using Triple '" << TripleAndCpu.LLVMTriple << "' and CPU '"
- << TripleAndCpu.CpuName << "'\n";
-
// Read benchmarks.
- const LLVMState State = ExitOnErr(
- LLVMState::Create(TripleAndCpu.LLVMTriple, TripleAndCpu.CpuName));
+ const LLVMState State = ExitOnErr(LLVMState::Create("", ""));
const std::vector<InstructionBenchmark> Points = ExitOnFileError(
- BenchmarkFile, InstructionBenchmark::readYamls(State, *MemoryBuffer));
+ BenchmarkFile, InstructionBenchmark::readYamls(State, BenchmarkFile));
outs() << "Parsed " << Points.size() << " benchmark points\n";
if (Points.empty()) {
errs() << "no benchmarks to analyze\n";
return;
}
+ // FIXME: Check that all points have the same triple/cpu.
// FIXME: Merge points from several runs (latency and uops).
+ std::string Error;
+ const auto *TheTarget =
+ TargetRegistry::lookupTarget(Points[0].LLVMTriple, Error);
+ if (!TheTarget) {
+ errs() << "unknown target '" << Points[0].LLVMTriple << "'\n";
+ return;
+ }
+
+ std::unique_ptr<MCSubtargetInfo> SubtargetInfo(
+ TheTarget->createMCSubtargetInfo(Points[0].LLVMTriple, CpuName, ""));
+
+ std::unique_ptr<MCInstrInfo> InstrInfo(TheTarget->createMCInstrInfo());
+ assert(InstrInfo && "Unable to create instruction info!");
+
const auto Clustering = ExitOnErr(InstructionBenchmarkClustering::create(
Points, AnalysisClusteringAlgorithm, AnalysisDbscanNumPoints,
- AnalysisClusteringEpsilon, &State.getSubtargetInfo(),
- &State.getInstrInfo()));
+ AnalysisClusteringEpsilon, SubtargetInfo.get(), InstrInfo.get()));
- const Analysis Analyzer(State, Clustering, AnalysisInconsistencyEpsilon,
- AnalysisDisplayUnstableOpcodes);
+ const Analysis Analyzer(
+ *TheTarget, std::move(SubtargetInfo), std::move(InstrInfo), Clustering,
+ AnalysisInconsistencyEpsilon, AnalysisDisplayUnstableOpcodes, CpuName);
maybeRunAnalysis<Analysis::PrintClusters>(Analyzer, "analysis clusters",
AnalysisClustersOutputFile);
#include "gtest/gtest.h"
using ::testing::AllOf;
-using ::testing::ElementsAre;
using ::testing::Eq;
-using ::testing::Field;
using ::testing::get;
using ::testing::Pointwise;
using ::testing::Property;
errs() << Filename << "-------\n";
ExitOnErr(ToDisk.writeYaml(State, Filename));
- const std::unique_ptr<MemoryBuffer> Buffer =
- std::move(*MemoryBuffer::getFile(Filename));
-
- {
- // Read Triples/Cpu only.
- const auto TriplesAndCpus =
- ExitOnErr(InstructionBenchmark::readTriplesAndCpusFromYamls(*Buffer));
-
- ASSERT_THAT(TriplesAndCpus,
- testing::ElementsAre(
- AllOf(Field(&InstructionBenchmark::TripleAndCpu::LLVMTriple,
- Eq("llvm_triple")),
- Field(&InstructionBenchmark::TripleAndCpu::CpuName,
- Eq("cpu_name")))));
- }
{
// One-element version.
const auto FromDisk =
- ExitOnErr(InstructionBenchmark::readYaml(State, *Buffer));
+ ExitOnErr(InstructionBenchmark::readYaml(State, Filename));
EXPECT_THAT(FromDisk.Key.Instructions,
Pointwise(EqMCInst(), ToDisk.Key.Instructions));
{
// Vector version.
const auto FromDiskVector =
- ExitOnErr(InstructionBenchmark::readYamls(State, *Buffer));
+ ExitOnErr(InstructionBenchmark::readYamls(State, Filename));
ASSERT_EQ(FromDiskVector.size(), size_t{1});
const auto FromDisk = FromDiskVector[0];
EXPECT_THAT(FromDisk.Key.Instructions,