1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-25 12:12:47 +01:00

[llvm-exegesis][NFC] Disassociate snippet generators from benchmark runners

The addition of `inverse_throughput` mode highlighted the disjointedness
of snippet generators and benchmark runners because it used the
`UopsSnippetGenerator` with the  `LatencyBenchmarkRunner`.
To keep the code consistent tie the snippet generators to
parallelization/serialization rather than their benchmark runners.

Renaming `LatencySnippetGenerator` -> `SerialSnippetGenerator`.
Renaming `UopsSnippetGenerator` -> `ParallelSnippetGenerator`.

Differential Revision: https://reviews.llvm.org/D72928
This commit is contained in:
Miloš Stojanović 2020-01-17 14:28:54 +01:00
parent dea61577e1
commit 5da66b880e
20 changed files with 270 additions and 190 deletions

View File

@ -6,7 +6,6 @@
//
//===----------------------------------------------------------------------===//
#include "../Target.h"
#include "../Latency.h"
#include "AArch64.h"
#include "AArch64RegisterInfo.h"

View File

@ -244,9 +244,9 @@ Analysis::makePointsPerSchedClass() const {
return Entries;
}
// Uops repeat the same opcode over again. Just show this opcode and show the
// whole snippet only on hover.
static void writeUopsSnippetHtml(raw_ostream &OS,
// Parallel benchmarks repeat the same opcode multiple times. Just show this
// opcode and show the whole snippet only on hover.
static void writeParallelSnippetHtml(raw_ostream &OS,
const std::vector<MCInst> &Instructions,
const MCInstrInfo &InstrInfo) {
if (Instructions.empty())
@ -282,7 +282,7 @@ void Analysis::printPointHtml(const InstructionBenchmark &Point,
break;
case InstructionBenchmark::Uops:
case InstructionBenchmark::InverseThroughput:
writeUopsSnippetHtml(OS, Point.Key.Instructions, *InstrInfo_);
writeParallelSnippetHtml(OS, Point.Key.Instructions, *InstrInfo_);
break;
default:
llvm_unreachable("invalid mode");

View File

@ -27,18 +27,20 @@ add_library(LLVMExegesis
BenchmarkRunner.cpp
Clustering.cpp
CodeTemplate.cpp
Latency.cpp
LatencyBenchmarkRunner.cpp
LlvmState.cpp
MCInstrDescView.cpp
ParallelSnippetGenerator.cpp
PerfHelper.cpp
RegisterAliasing.cpp
RegisterValue.cpp
SchedClassResolution.cpp
SerialSnippetGenerator.cpp
SnippetFile.cpp
SnippetGenerator.cpp
SnippetRepetitor.cpp
Target.cpp
Uops.cpp
UopsBenchmarkRunner.cpp
)
llvm_update_compile_flags(LLVMExegesis)

View File

@ -0,0 +1,58 @@
//===-- LatencyBenchmarkRunner.cpp ------------------------------*- C++ -*-===//
//
// 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
//
//===----------------------------------------------------------------------===//
#include "LatencyBenchmarkRunner.h"
#include "Target.h"
#include "BenchmarkRunner.h"
namespace llvm {
namespace exegesis {
static constexpr size_t kMaxAliasingInstructions = 10;
LatencyBenchmarkRunner::LatencyBenchmarkRunner(const LLVMState &State,
InstructionBenchmark::ModeE Mode)
: BenchmarkRunner(State, Mode) {
assert((Mode == InstructionBenchmark::Latency ||
Mode == InstructionBenchmark::InverseThroughput) &&
"invalid mode");
}
LatencyBenchmarkRunner::~LatencyBenchmarkRunner() = default;
Expected<std::vector<BenchmarkMeasure>> LatencyBenchmarkRunner::runMeasurements(
const FunctionExecutor &Executor) const {
// Cycle measurements include some overhead from the kernel. Repeat the
// measure several times and take the minimum value.
constexpr const int NumMeasurements = 30;
int64_t MinValue = std::numeric_limits<int64_t>::max();
const char *CounterName = State.getPfmCounters().CycleCounter;
for (size_t I = 0; I < NumMeasurements; ++I) {
auto ExpectedCounterValue = Executor.runAndMeasure(CounterName);
if (!ExpectedCounterValue)
return ExpectedCounterValue.takeError();
if (*ExpectedCounterValue < MinValue)
MinValue = *ExpectedCounterValue;
}
std::vector<BenchmarkMeasure> Result;
switch (Mode) {
case InstructionBenchmark::Latency:
Result = {BenchmarkMeasure::Create("latency", MinValue)};
break;
case InstructionBenchmark::InverseThroughput:
Result = {BenchmarkMeasure::Create("inverse_throughput", MinValue)};
break;
default:
break;
}
return std::move(Result);
}
} // namespace exegesis
} // namespace llvm

View File

@ -1,4 +1,4 @@
//===-- Latency.h -----------------------------------------------*- C++ -*-===//
//===-- LatencyBenchmarkRunner.h --------------------------------*- C++ -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
@ -15,23 +15,10 @@
#define LLVM_TOOLS_LLVM_EXEGESIS_LATENCY_H
#include "BenchmarkRunner.h"
#include "Error.h"
#include "MCInstrDescView.h"
#include "SnippetGenerator.h"
namespace llvm {
namespace exegesis {
class LatencySnippetGenerator : public SnippetGenerator {
public:
using SnippetGenerator::SnippetGenerator;
~LatencySnippetGenerator() override;
Expected<std::vector<CodeTemplate>>
generateCodeTemplates(const Instruction &Instr,
const BitVector &ForbiddenRegisters) const override;
};
class LatencyBenchmarkRunner : public BenchmarkRunner {
public:
LatencyBenchmarkRunner(const LLVMState &State,

View File

@ -5,8 +5,8 @@
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#include "../Error.h"
#include "../Target.h"
#include "../Latency.h"
#include "MCTargetDesc/MipsBaseInfo.h"
#include "Mips.h"
#include "MipsRegisterInfo.h"

View File

@ -1,4 +1,4 @@
//===-- Uops.cpp ------------------------------------------------*- C++ -*-===//
//===-- ParallelSnippetGenerator.cpp ----------------------------*- C++ -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
@ -6,9 +6,8 @@
//
//===----------------------------------------------------------------------===//
#include "Uops.h"
#include "ParallelSnippetGenerator.h"
#include "Assembler.h"
#include "BenchmarkRunner.h"
#include "MCInstrDescView.h"
#include "Target.h"
@ -16,9 +15,9 @@
// FIXME: Load constants into registers (e.g. with fld1) to not break
// instructions like x87.
// Ideally we would like the only limitation on executing uops to be the issue
// ports. Maximizing port pressure increases the likelihood that the load is
// distributed evenly across possible ports.
// Ideally we would like the only limitation on executing instructions to be the
// availability of the CPU resources (e.g. execution ports) needed to execute
// them, instead of the availability of their data dependencies.
// To achieve that, one approach is to generate instructions that do not have
// data dependencies between them.
@ -89,11 +88,9 @@ getVariablesWithTiedOperands(const Instruction &Instr) {
return Result;
}
UopsBenchmarkRunner::~UopsBenchmarkRunner() = default;
ParallelSnippetGenerator::~ParallelSnippetGenerator() = default;
UopsSnippetGenerator::~UopsSnippetGenerator() = default;
void UopsSnippetGenerator::instantiateMemoryOperands(
void ParallelSnippetGenerator::instantiateMemoryOperands(
const unsigned ScratchSpacePointerInReg,
std::vector<InstructionTemplate> &Instructions) const {
if (ScratchSpacePointerInReg == 0)
@ -157,7 +154,7 @@ static std::vector<InstructionTemplate> generateSnippetUsingStaticRenaming(
}
}
Expected<std::vector<CodeTemplate>> UopsSnippetGenerator::generateCodeTemplates(
Expected<std::vector<CodeTemplate>> ParallelSnippetGenerator::generateCodeTemplates(
const Instruction &Instr, const BitVector &ForbiddenRegisters) const {
CodeTemplate CT;
CT.ScratchSpacePointerInReg =
@ -219,34 +216,7 @@ Expected<std::vector<CodeTemplate>> UopsSnippetGenerator::generateCodeTemplates(
return getSingleton(std::move(CT));
}
Expected<std::vector<BenchmarkMeasure>>
UopsBenchmarkRunner::runMeasurements(const FunctionExecutor &Executor) const {
std::vector<BenchmarkMeasure> Result;
const PfmCountersInfo &PCI = State.getPfmCounters();
// Uops per port.
for (const auto *IssueCounter = PCI.IssueCounters,
*IssueCounterEnd = PCI.IssueCounters + PCI.NumIssueCounters;
IssueCounter != IssueCounterEnd; ++IssueCounter) {
if (!IssueCounter->Counter)
continue;
auto ExpectedCounterValue = Executor.runAndMeasure(IssueCounter->Counter);
if (!ExpectedCounterValue)
return ExpectedCounterValue.takeError();
Result.push_back(BenchmarkMeasure::Create(IssueCounter->ProcResName,
*ExpectedCounterValue));
}
// NumMicroOps.
if (const char *const UopsCounter = PCI.UopsCounter) {
auto ExpectedCounterValue = Executor.runAndMeasure(UopsCounter);
if (!ExpectedCounterValue)
return ExpectedCounterValue.takeError();
Result.push_back(
BenchmarkMeasure::Create("NumMicroOps", *ExpectedCounterValue));
}
return std::move(Result);
}
constexpr const size_t UopsSnippetGenerator::kMinNumDifferentAddresses;
constexpr const size_t ParallelSnippetGenerator::kMinNumDifferentAddresses;
} // namespace exegesis
} // namespace llvm

View File

@ -1,4 +1,4 @@
//===-- Uops.h --------------------------------------------------*- C++ -*-===//
//===-- ParallelSnippetGenerator.h ------------------------------*- C++ -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
@ -7,23 +7,22 @@
//===----------------------------------------------------------------------===//
///
/// \file
/// A BenchmarkRunner implementation to measure uop decomposition.
/// A SnippetGenerator implementation to create parallel instruction snippets.
///
//===----------------------------------------------------------------------===//
#ifndef LLVM_TOOLS_LLVM_EXEGESIS_UOPS_H
#define LLVM_TOOLS_LLVM_EXEGESIS_UOPS_H
#ifndef LLVM_TOOLS_LLVM_EXEGESIS_PARALLELSNIPPETGENERATOR_H
#define LLVM_TOOLS_LLVM_EXEGESIS_PARALLELSNIPPETGENERATOR_H
#include "BenchmarkRunner.h"
#include "SnippetGenerator.h"
namespace llvm {
namespace exegesis {
class UopsSnippetGenerator : public SnippetGenerator {
class ParallelSnippetGenerator : public SnippetGenerator {
public:
using SnippetGenerator::SnippetGenerator;
~UopsSnippetGenerator() override;
~ParallelSnippetGenerator() override;
Expected<std::vector<CodeTemplate>>
generateCodeTemplates(const Instruction &Instr,
@ -60,20 +59,7 @@ private:
std::vector<InstructionTemplate> &SnippetTemplate) const;
};
class UopsBenchmarkRunner : public BenchmarkRunner {
public:
UopsBenchmarkRunner(const LLVMState &State)
: BenchmarkRunner(State, InstructionBenchmark::Uops) {}
~UopsBenchmarkRunner() override;
static constexpr const size_t kMinNumDifferentAddresses = 6;
private:
Expected<std::vector<BenchmarkMeasure>>
runMeasurements(const FunctionExecutor &Executor) const override;
};
} // namespace exegesis
} // namespace llvm
#endif // LLVM_TOOLS_LLVM_EXEGESIS_UOPS_H
#endif // LLVM_TOOLS_LLVM_EXEGESIS_PARALLELSNIPPETGENERATOR_H

View File

@ -7,7 +7,6 @@
// The PowerPC ExegesisTarget.
//===----------------------------------------------------------------------===//
#include "../Target.h"
#include "../Latency.h"
#include "PPC.h"
#include "PPCRegisterInfo.h"

View File

@ -1,4 +1,4 @@
//===-- Latency.cpp ---------------------------------------------*- C++ -*-===//
//===-- SerialSnippetGenerator.cpp ------------------------------*- C++ -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
@ -6,17 +6,13 @@
//
//===----------------------------------------------------------------------===//
#include "Latency.h"
#include "SerialSnippetGenerator.h"
#include "Assembler.h"
#include "BenchmarkRunner.h"
#include "MCInstrDescView.h"
#include "PerfHelper.h"
#include "Target.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/MC/MCInst.h"
#include "llvm/MC/MCInstBuilder.h"
#include "llvm/Support/FormatVariadic.h"
#include "CodeTemplate.h"
#include <algorithm>
#include <numeric>
#include <vector>
namespace llvm {
namespace exegesis {
@ -149,10 +145,10 @@ static void appendCodeTemplates(const LLVMState &State,
}
}
LatencySnippetGenerator::~LatencySnippetGenerator() = default;
SerialSnippetGenerator::~SerialSnippetGenerator() = default;
Expected<std::vector<CodeTemplate>>
LatencySnippetGenerator::generateCodeTemplates(
SerialSnippetGenerator::generateCodeTemplates(
const Instruction &Instr, const BitVector &ForbiddenRegisters) const {
std::vector<CodeTemplate> Results;
const ExecutionMode EM = getExecutionModes(Instr, ForbiddenRegisters);
@ -169,43 +165,5 @@ LatencySnippetGenerator::generateCodeTemplates(
return std::move(Results);
}
LatencyBenchmarkRunner::LatencyBenchmarkRunner(const LLVMState &State,
InstructionBenchmark::ModeE Mode)
: BenchmarkRunner(State, Mode) {
assert((Mode == InstructionBenchmark::Latency ||
Mode == InstructionBenchmark::InverseThroughput) &&
"invalid mode");
}
LatencyBenchmarkRunner::~LatencyBenchmarkRunner() = default;
Expected<std::vector<BenchmarkMeasure>> LatencyBenchmarkRunner::runMeasurements(
const FunctionExecutor &Executor) const {
// Cycle measurements include some overhead from the kernel. Repeat the
// measure several times and take the minimum value.
constexpr const int NumMeasurements = 30;
int64_t MinValue = std::numeric_limits<int64_t>::max();
const char *CounterName = State.getPfmCounters().CycleCounter;
for (size_t I = 0; I < NumMeasurements; ++I) {
auto ExpectedCounterValue = Executor.runAndMeasure(CounterName);
if (!ExpectedCounterValue)
return ExpectedCounterValue.takeError();
if (*ExpectedCounterValue < MinValue)
MinValue = *ExpectedCounterValue;
}
std::vector<BenchmarkMeasure> Result;
switch (Mode) {
case InstructionBenchmark::Latency:
Result = {BenchmarkMeasure::Create("latency", MinValue)};
break;
case InstructionBenchmark::InverseThroughput:
Result = {BenchmarkMeasure::Create("inverse_throughput", MinValue)};
break;
default:
break;
}
return std::move(Result);
}
} // namespace exegesis
} // namespace llvm

View File

@ -0,0 +1,37 @@
//===-- SerialSnippetGenerator.h --------------------------------*- C++ -*-===//
//
// 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
//
//===----------------------------------------------------------------------===//
///
/// \file
/// A SnippetGenerator implementation to create serial instruction snippets.
///
//===----------------------------------------------------------------------===//
#ifndef LLVM_TOOLS_LLVM_EXEGESIS_SERIALSNIPPETGENERATOR_H
#define LLVM_TOOLS_LLVM_EXEGESIS_SERIALSNIPPETGENERATOR_H
#include "Error.h"
#include "MCInstrDescView.h"
#include "SnippetGenerator.h"
namespace llvm {
namespace exegesis {
class SerialSnippetGenerator : public SnippetGenerator {
public:
using SnippetGenerator::SnippetGenerator;
~SerialSnippetGenerator() override;
Expected<std::vector<CodeTemplate>>
generateCodeTemplates(const Instruction &Instr,
const BitVector &ForbiddenRegisters) const override;
};
} // namespace exegesis
} // namespace llvm
#endif // LLVM_TOOLS_LLVM_EXEGESIS_SERIALSNIPPETGENERATOR_H

View File

@ -7,8 +7,10 @@
//===----------------------------------------------------------------------===//
#include "Target.h"
#include "Latency.h"
#include "Uops.h"
#include "LatencyBenchmarkRunner.h"
#include "ParallelSnippetGenerator.h"
#include "SerialSnippetGenerator.h"
#include "UopsBenchmarkRunner.h"
namespace llvm {
namespace exegesis {
@ -43,10 +45,10 @@ std::unique_ptr<SnippetGenerator> ExegesisTarget::createSnippetGenerator(
case InstructionBenchmark::Unknown:
return nullptr;
case InstructionBenchmark::Latency:
return createLatencySnippetGenerator(State, Opts);
return createSerialSnippetGenerator(State, Opts);
case InstructionBenchmark::Uops:
case InstructionBenchmark::InverseThroughput:
return createUopsSnippetGenerator(State, Opts);
return createParallelSnippetGenerator(State, Opts);
}
return nullptr;
}
@ -77,14 +79,14 @@ ExegesisTarget::createBenchmarkRunner(InstructionBenchmark::ModeE Mode,
return nullptr;
}
std::unique_ptr<SnippetGenerator> ExegesisTarget::createLatencySnippetGenerator(
std::unique_ptr<SnippetGenerator> ExegesisTarget::createSerialSnippetGenerator(
const LLVMState &State, const SnippetGenerator::Options &Opts) const {
return std::make_unique<LatencySnippetGenerator>(State, Opts);
return std::make_unique<SerialSnippetGenerator>(State, Opts);
}
std::unique_ptr<SnippetGenerator> ExegesisTarget::createUopsSnippetGenerator(
std::unique_ptr<SnippetGenerator> ExegesisTarget::createParallelSnippetGenerator(
const LLVMState &State, const SnippetGenerator::Options &Opts) const {
return std::make_unique<UopsSnippetGenerator>(State, Opts);
return std::make_unique<ParallelSnippetGenerator>(State, Opts);
}
std::unique_ptr<BenchmarkRunner> ExegesisTarget::createLatencyBenchmarkRunner(

View File

@ -144,9 +144,9 @@ private:
// Targets can implement their own snippet generators/benchmarks runners by
// implementing these.
std::unique_ptr<SnippetGenerator> virtual createLatencySnippetGenerator(
std::unique_ptr<SnippetGenerator> virtual createSerialSnippetGenerator(
const LLVMState &State, const SnippetGenerator::Options &Opts) const;
std::unique_ptr<SnippetGenerator> virtual createUopsSnippetGenerator(
std::unique_ptr<SnippetGenerator> virtual createParallelSnippetGenerator(
const LLVMState &State, const SnippetGenerator::Options &Opts) const;
std::unique_ptr<BenchmarkRunner> virtual createLatencyBenchmarkRunner(
const LLVMState &State, InstructionBenchmark::ModeE Mode) const;

View File

@ -0,0 +1,46 @@
//===-- UopsBenchmarkRunner.cpp ---------------------------------*- C++ -*-===//
//
// 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
//
//===----------------------------------------------------------------------===//
#include "UopsBenchmarkRunner.h"
#include "Target.h"
namespace llvm {
namespace exegesis {
UopsBenchmarkRunner::~UopsBenchmarkRunner() = default;
Expected<std::vector<BenchmarkMeasure>>
UopsBenchmarkRunner::runMeasurements(const FunctionExecutor &Executor) const {
std::vector<BenchmarkMeasure> Result;
const PfmCountersInfo &PCI = State.getPfmCounters();
// Uops per port.
for (const auto *IssueCounter = PCI.IssueCounters,
*IssueCounterEnd = PCI.IssueCounters + PCI.NumIssueCounters;
IssueCounter != IssueCounterEnd; ++IssueCounter) {
if (!IssueCounter->Counter)
continue;
auto ExpectedCounterValue = Executor.runAndMeasure(IssueCounter->Counter);
if (!ExpectedCounterValue)
return ExpectedCounterValue.takeError();
Result.push_back(BenchmarkMeasure::Create(IssueCounter->ProcResName,
*ExpectedCounterValue));
}
// NumMicroOps.
if (const char *const UopsCounter = PCI.UopsCounter) {
auto ExpectedCounterValue = Executor.runAndMeasure(UopsCounter);
if (!ExpectedCounterValue)
return ExpectedCounterValue.takeError();
Result.push_back(
BenchmarkMeasure::Create("NumMicroOps", *ExpectedCounterValue));
}
return std::move(Result);
}
} // namespace exegesis
} // namespace llvm

View File

@ -0,0 +1,38 @@
//===-- UopsBenchmarkRunner.h -----------------------------------*- C++ -*-===//
//
// 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
//
//===----------------------------------------------------------------------===//
///
/// \file
/// A BenchmarkRunner implementation to measure uop decomposition.
///
//===----------------------------------------------------------------------===//
#ifndef LLVM_TOOLS_LLVM_EXEGESIS_UOPSBENCHMARKRUNNER_H
#define LLVM_TOOLS_LLVM_EXEGESIS_UOPSBENCHMARKRUNNER_H
#include "BenchmarkRunner.h"
namespace llvm {
namespace exegesis {
class UopsBenchmarkRunner : public BenchmarkRunner {
public:
UopsBenchmarkRunner(const LLVMState &State)
: BenchmarkRunner(State, InstructionBenchmark::Uops) {}
~UopsBenchmarkRunner() override;
static constexpr const size_t kMinNumDifferentAddresses = 6;
private:
Expected<std::vector<BenchmarkMeasure>>
runMeasurements(const FunctionExecutor &Executor) const override;
};
} // namespace exegesis
} // namespace llvm
#endif // LLVM_TOOLS_LLVM_EXEGESIS_UOPSBENCHMARKRUNNER_H

View File

@ -8,9 +8,9 @@
#include "../Target.h"
#include "../Error.h"
#include "../Latency.h"
#include "../SerialSnippetGenerator.h"
#include "../SnippetGenerator.h"
#include "../Uops.h"
#include "../ParallelSnippetGenerator.h"
#include "MCTargetDesc/X86BaseInfo.h"
#include "MCTargetDesc/X86MCTargetDesc.h"
#include "X86.h"
@ -242,9 +242,9 @@ static Expected<std::vector<CodeTemplate>> generateLEATemplatesCommon(
}
namespace {
class X86LatencySnippetGenerator : public LatencySnippetGenerator {
class X86SerialSnippetGenerator : public SerialSnippetGenerator {
public:
using LatencySnippetGenerator::LatencySnippetGenerator;
using SerialSnippetGenerator::SerialSnippetGenerator;
Expected<std::vector<CodeTemplate>>
generateCodeTemplates(const Instruction &Instr,
@ -253,7 +253,7 @@ public:
} // namespace
Expected<std::vector<CodeTemplate>>
X86LatencySnippetGenerator::generateCodeTemplates(
X86SerialSnippetGenerator::generateCodeTemplates(
const Instruction &Instr, const BitVector &ForbiddenRegisters) const {
if (auto E = IsInvalidOpcode(Instr))
return std::move(E);
@ -271,7 +271,7 @@ X86LatencySnippetGenerator::generateCodeTemplates(
switch (getX86FPFlags(Instr)) {
case X86II::NotFP:
return LatencySnippetGenerator::generateCodeTemplates(Instr,
return SerialSnippetGenerator::generateCodeTemplates(Instr,
ForbiddenRegisters);
case X86II::ZeroArgFP:
case X86II::OneArgFP:
@ -292,9 +292,9 @@ X86LatencySnippetGenerator::generateCodeTemplates(
}
namespace {
class X86UopsSnippetGenerator : public UopsSnippetGenerator {
class X86ParallelSnippetGenerator : public ParallelSnippetGenerator {
public:
using UopsSnippetGenerator::UopsSnippetGenerator;
using ParallelSnippetGenerator::ParallelSnippetGenerator;
Expected<std::vector<CodeTemplate>>
generateCodeTemplates(const Instruction &Instr,
@ -304,7 +304,7 @@ public:
} // namespace
Expected<std::vector<CodeTemplate>>
X86UopsSnippetGenerator::generateCodeTemplates(
X86ParallelSnippetGenerator::generateCodeTemplates(
const Instruction &Instr, const BitVector &ForbiddenRegisters) const {
if (auto E = IsInvalidOpcode(Instr))
return std::move(E);
@ -333,7 +333,7 @@ X86UopsSnippetGenerator::generateCodeTemplates(
switch (getX86FPFlags(Instr)) {
case X86II::NotFP:
return UopsSnippetGenerator::generateCodeTemplates(Instr,
return ParallelSnippetGenerator::generateCodeTemplates(Instr,
ForbiddenRegisters);
case X86II::ZeroArgFP:
case X86II::OneArgFP:
@ -577,16 +577,16 @@ private:
sizeof(kUnavailableRegisters[0]));
}
std::unique_ptr<SnippetGenerator> createLatencySnippetGenerator(
std::unique_ptr<SnippetGenerator> createSerialSnippetGenerator(
const LLVMState &State,
const SnippetGenerator::Options &Opts) const override {
return std::make_unique<X86LatencySnippetGenerator>(State, Opts);
return std::make_unique<X86SerialSnippetGenerator>(State, Opts);
}
std::unique_ptr<SnippetGenerator> createUopsSnippetGenerator(
std::unique_ptr<SnippetGenerator> createParallelSnippetGenerator(
const LLVMState &State,
const SnippetGenerator::Options &Opts) const override {
return std::make_unique<X86UopsSnippetGenerator>(State, Opts);
return std::make_unique<X86ParallelSnippetGenerator>(State, Opts);
}
bool matchesArch(Triple::ArchType Arch) const override {

View File

@ -7,13 +7,13 @@
//===----------------------------------------------------------------------===//
#include "../Common/AssemblerUtils.h"
#include "Latency.h"
#include "LlvmState.h"
#include "MCInstrDescView.h"
#include "MipsInstrInfo.h"
#include "ParallelSnippetGenerator.h"
#include "RegisterAliasing.h"
#include "SerialSnippetGenerator.h"
#include "TestBase.h"
#include "Uops.h"
#include <unordered_set>
@ -48,12 +48,12 @@ protected:
SnippetGeneratorT Generator;
};
using LatencySnippetGeneratorTest =
SnippetGeneratorTest<LatencySnippetGenerator>;
using SerialSnippetGeneratorTest = SnippetGeneratorTest<SerialSnippetGenerator>;
using UopsSnippetGeneratorTest = SnippetGeneratorTest<UopsSnippetGenerator>;
using ParallelSnippetGeneratorTest =
SnippetGeneratorTest<ParallelSnippetGenerator>;
TEST_F(LatencySnippetGeneratorTest, ImplicitSelfDependencyThroughExplicitRegs) {
TEST_F(SerialSnippetGeneratorTest, ImplicitSelfDependencyThroughExplicitRegs) {
// - ADD
// - Op0 Explicit Def RegClass(GPR32)
// - Op1 Explicit Use RegClass(GPR32)
@ -77,7 +77,7 @@ TEST_F(LatencySnippetGeneratorTest, ImplicitSelfDependencyThroughExplicitRegs) {
<< "Op0 is either set to Op1 or to Op2";
}
TEST_F(LatencySnippetGeneratorTest,
TEST_F(SerialSnippetGeneratorTest,
ImplicitSelfDependencyThroughExplicitRegsForbidAll) {
// - XOR
// - Op0 Explicit Def RegClass(GPR32)
@ -96,7 +96,7 @@ TEST_F(LatencySnippetGeneratorTest,
consumeError(std::move(Error));
}
TEST_F(UopsSnippetGeneratorTest, MemoryUse) {
TEST_F(ParallelSnippetGeneratorTest, MemoryUse) {
// LB reads from memory.
// - LB
// - Op0 Explicit Def RegClass(GPR32)
@ -110,10 +110,11 @@ TEST_F(UopsSnippetGeneratorTest, MemoryUse) {
const auto CodeTemplates = checkAndGetCodeTemplates(Opcode);
ASSERT_THAT(CodeTemplates, SizeIs(1));
const auto &CT = CodeTemplates[0];
EXPECT_THAT(CT.Info, HasSubstr("instruction is parallel, repeating a random one."));
EXPECT_THAT(CT.Info,
HasSubstr("instruction is parallel, repeating a random one."));
EXPECT_THAT(CT.Execution, ExecutionMode::UNKNOWN);
ASSERT_THAT(CT.Instructions,
SizeIs(UopsSnippetGenerator::kMinNumDifferentAddresses));
SizeIs(ParallelSnippetGenerator::kMinNumDifferentAddresses));
const InstructionTemplate &IT = CT.Instructions[0];
EXPECT_THAT(IT.getOpcode(), Opcode);
ASSERT_THAT(IT.getVariableValues(), SizeIs(3));

View File

@ -7,12 +7,12 @@
//===----------------------------------------------------------------------===//
#include "../Common/AssemblerUtils.h"
#include "Latency.h"
#include "LlvmState.h"
#include "MCInstrDescView.h"
#include "ParallelSnippetGenerator.h"
#include "RegisterAliasing.h"
#include "SerialSnippetGenerator.h"
#include "TestBase.h"
#include "Uops.h"
#include "X86InstrInfo.h"
#include <unordered_set>
@ -59,12 +59,12 @@ protected:
SnippetGeneratorT Generator;
};
using LatencySnippetGeneratorTest =
SnippetGeneratorTest<LatencySnippetGenerator>;
using SerialSnippetGeneratorTest = SnippetGeneratorTest<SerialSnippetGenerator>;
using UopsSnippetGeneratorTest = SnippetGeneratorTest<UopsSnippetGenerator>;
using ParallelSnippetGeneratorTest =
SnippetGeneratorTest<ParallelSnippetGenerator>;
TEST_F(LatencySnippetGeneratorTest, ImplicitSelfDependencyThroughImplicitReg) {
TEST_F(SerialSnippetGeneratorTest, ImplicitSelfDependencyThroughImplicitReg) {
// - ADC16i16
// - Op0 Explicit Use Immediate
// - Op1 Implicit Def Reg(AX)
@ -90,7 +90,7 @@ TEST_F(LatencySnippetGeneratorTest, ImplicitSelfDependencyThroughImplicitReg) {
EXPECT_THAT(IT.getVariableValues()[0], IsInvalid()) << "Immediate is not set";
}
TEST_F(LatencySnippetGeneratorTest, ImplicitSelfDependencyThroughTiedRegs) {
TEST_F(SerialSnippetGeneratorTest, ImplicitSelfDependencyThroughTiedRegs) {
// - ADD16ri
// - Op0 Explicit Def RegClass(GR16)
// - Op1 Explicit Use RegClass(GR16) TiedToOp0
@ -114,7 +114,7 @@ TEST_F(LatencySnippetGeneratorTest, ImplicitSelfDependencyThroughTiedRegs) {
EXPECT_THAT(IT.getVariableValues()[1], IsInvalid()) << "Operand 2 is not set";
}
TEST_F(LatencySnippetGeneratorTest, ImplicitSelfDependencyThroughExplicitRegs) {
TEST_F(SerialSnippetGeneratorTest, ImplicitSelfDependencyThroughExplicitRegs) {
// - VXORPSrr
// - Op0 Explicit Def RegClass(VR128)
// - Op1 Explicit Use RegClass(VR128)
@ -138,7 +138,7 @@ TEST_F(LatencySnippetGeneratorTest, ImplicitSelfDependencyThroughExplicitRegs) {
<< "Op0 is either set to Op1 or to Op2";
}
TEST_F(LatencySnippetGeneratorTest,
TEST_F(SerialSnippetGeneratorTest,
ImplicitSelfDependencyThroughExplicitRegsForbidAll) {
// - VXORPSrr
// - Op0 Explicit Def RegClass(VR128)
@ -158,7 +158,7 @@ TEST_F(LatencySnippetGeneratorTest,
consumeError(std::move(Error));
}
TEST_F(LatencySnippetGeneratorTest, DependencyThroughOtherOpcode) {
TEST_F(SerialSnippetGeneratorTest, DependencyThroughOtherOpcode) {
// - CMP64rr
// - Op0 Explicit Use RegClass(GR64)
// - Op1 Explicit Use RegClass(GR64)
@ -182,7 +182,7 @@ TEST_F(LatencySnippetGeneratorTest, DependencyThroughOtherOpcode) {
}
}
TEST_F(LatencySnippetGeneratorTest, LAHF) {
TEST_F(SerialSnippetGeneratorTest, LAHF) {
// - LAHF
// - Op0 Implicit Def Reg(AH)
// - Op1 Implicit Use Reg(EFLAGS)
@ -198,7 +198,7 @@ TEST_F(LatencySnippetGeneratorTest, LAHF) {
}
}
TEST_F(UopsSnippetGeneratorTest, ParallelInstruction) {
TEST_F(ParallelSnippetGeneratorTest, ParallelInstruction) {
// - BNDCL32rr
// - Op0 Explicit Use RegClass(BNDR)
// - Op1 Explicit Use RegClass(GR32)
@ -218,7 +218,7 @@ TEST_F(UopsSnippetGeneratorTest, ParallelInstruction) {
EXPECT_THAT(IT.getVariableValues()[1], IsInvalid());
}
TEST_F(UopsSnippetGeneratorTest, SerialInstruction) {
TEST_F(ParallelSnippetGeneratorTest, SerialInstruction) {
// - CDQ
// - Op0 Implicit Def Reg(EAX)
// - Op1 Implicit Def Reg(EDX)
@ -237,7 +237,7 @@ TEST_F(UopsSnippetGeneratorTest, SerialInstruction) {
ASSERT_THAT(IT.getVariableValues(), SizeIs(0));
}
TEST_F(UopsSnippetGeneratorTest, StaticRenaming) {
TEST_F(ParallelSnippetGeneratorTest, StaticRenaming) {
// CMOV32rr has tied variables, we enumerate the possible values to execute
// as many in parallel as possible.
@ -268,7 +268,7 @@ TEST_F(UopsSnippetGeneratorTest, StaticRenaming) {
<< "Each instruction writes to a different register";
}
TEST_F(UopsSnippetGeneratorTest, NoTiedVariables) {
TEST_F(ParallelSnippetGeneratorTest, NoTiedVariables) {
// CMOV_GR32 has no tied variables, we make sure def and use are different
// from each other.
@ -302,7 +302,7 @@ TEST_F(UopsSnippetGeneratorTest, NoTiedVariables) {
EXPECT_THAT(IT.getVariableValues()[3], IsInvalid());
}
TEST_F(UopsSnippetGeneratorTest, MemoryUse) {
TEST_F(ParallelSnippetGeneratorTest, MemoryUse) {
// Mov32rm reads from memory.
// - MOV32rm
// - Op0 Explicit Def RegClass(GR32)
@ -326,7 +326,7 @@ TEST_F(UopsSnippetGeneratorTest, MemoryUse) {
EXPECT_THAT(CT.Info, HasSubstr("no tied variables"));
EXPECT_THAT(CT.Execution, ExecutionMode::UNKNOWN);
ASSERT_THAT(CT.Instructions,
SizeIs(UopsSnippetGenerator::kMinNumDifferentAddresses));
SizeIs(ParallelSnippetGenerator::kMinNumDifferentAddresses));
const InstructionTemplate &IT = CT.Instructions[0];
EXPECT_THAT(IT.getOpcode(), Opcode);
ASSERT_THAT(IT.getVariableValues(), SizeIs(6));

View File

@ -7,12 +7,10 @@
//===----------------------------------------------------------------------===//
#include "../Common/AssemblerUtils.h"
#include "Latency.h"
#include "LlvmState.h"
#include "MCInstrDescView.h"
#include "RegisterAliasing.h"
#include "TestBase.h"
#include "Uops.h"
#include "X86InstrInfo.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
@ -34,8 +32,7 @@ protected:
void SetUp() {
TM = State.createTargetMachine();
Context = std::make_unique<LLVMContext>();
Mod =
std::make_unique<Module>("X86SnippetRepetitorTest", *Context);
Mod = std::make_unique<Module>("X86SnippetRepetitorTest", *Context);
Mod->setDataLayout(TM->createDataLayout());
MMI = std::make_unique<MachineModuleInfo>(TM.get());
MF = &createVoidVoidPtrMachineFunction("TestFn", Mod.get(), MMI.get());

View File

@ -1,4 +1,4 @@
//===-- TestBase.cpp --------------------------------------------*- C++ -*-===//
//===-- TestBase.h ----------------------------------------------*- C++ -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.