1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-22 02:33:06 +01:00

[llvm-exegesis] Add options to SnippetGenerator.

Summary:
This adds a `-max-configs-per-opcode` option to limit the number of
configs per opcode.

Reviewers: gchatelet

Subscribers: tschuett, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D68642

llvm-svn: 374054
This commit is contained in:
Clement Courbet 2019-10-08 14:30:24 +00:00
parent 24a2861b8f
commit e0fc2857f3
11 changed files with 86 additions and 28 deletions

View File

@ -195,11 +195,23 @@ OPTIONS
to specify at least one of the `-analysis-clusters-output-file=` and
`-analysis-inconsistencies-output-file=`.
.. option:: -num-repetitions=<Number of repetition>
.. option:: -num-repetitions=<Number of repetitions>
Specify the number of repetitions of the asm snippet.
Higher values lead to more accurate measurements but lengthen the benchmark.
.. option:: -max-configs-per-opcode=<value>
Specify the maximum configurations that can be generated for each opcode.
By default this is `1`, meaning that we assume that a single measurement is
enough to characterize an opcode. This might not be true of all instructions:
for example, the performance characteristics of the LEA instruction on X86
depends on the value of assigned registers and immediates. Setting a value of
`-max-configs-per-opcode` larger than `1` allows `llvm-exegesis` to explore
more configurations to discover if some register or immediate assignments
lead to different performance characteristics.
.. option:: -benchmarks-file=</path/to/file>
File to read (`analysis` mode) or write (`latency`/`uops`/`inverse_throughput`

View File

@ -0,0 +1,24 @@
# RUN: llvm-exegesis -mode=latency -opcode-name=SBB8rr -max-configs-per-opcode=1 | FileCheck -check-prefixes=CHECK,CHECK1 %s
# RUN: llvm-exegesis -mode=latency -opcode-name=SBB8rr -max-configs-per-opcode=2 | FileCheck -check-prefixes=CHECK,CHECK2 %s
CHECK: ---
CHECK-NEXT: mode: latency
CHECK-NEXT: key:
CHECK-NEXT: instructions:
CHECK-NEXT: SBB8rr
CHECK-NEXT: config: ''
CHECK-NEXT: register_initial_values:
CHECK-DAG: - '[[REG1:[A-Z0-9]+]]=0x0'
CHECK-LAST: ...
CHECK1-NOT: SBB8rr
CHECK2: ---
CHECK2-NEXT: mode: latency
CHECK2-NEXT: key:
CHECK2-NEXT: instructions:
CHECK2-NEXT: SBB8rr
CHECK2-NEXT: config: ''
CHECK2-NEXT: register_initial_values:
CHECK2-DAG: - '[[REG1:[A-Z0-9]+]]=0x0'
CHECK2-LAST: ...

View File

@ -24,7 +24,7 @@ namespace exegesis {
class LatencySnippetGenerator : public SnippetGenerator {
public:
LatencySnippetGenerator(const LLVMState &State) : SnippetGenerator(State) {}
using SnippetGenerator::SnippetGenerator;
~LatencySnippetGenerator() override;
llvm::Expected<std::vector<CodeTemplate>>

View File

@ -33,7 +33,8 @@ std::vector<CodeTemplate> getSingleton(CodeTemplate &&CT) {
SnippetGeneratorFailure::SnippetGeneratorFailure(const llvm::Twine &S)
: llvm::StringError(S, llvm::inconvertibleErrorCode()) {}
SnippetGenerator::SnippetGenerator(const LLVMState &State) : State(State) {}
SnippetGenerator::SnippetGenerator(const LLVMState &State, const Options &Opts)
: State(State), Opts(Opts) {}
SnippetGenerator::~SnippetGenerator() = default;
@ -81,6 +82,9 @@ SnippetGenerator::generateConfigurations(
computeRegisterInitialValues(CT.Instructions);
BC.Key.Config = CT.Config;
Output.push_back(std::move(BC));
if (Output.size() >= Opts.MaxConfigsPerOpcode)
return Output; // Early exit if we exceeded the number of allowed
// configs.
}
}
return Output;

View File

@ -51,7 +51,11 @@ public:
// Common code for all benchmark modes.
class SnippetGenerator {
public:
explicit SnippetGenerator(const LLVMState &State);
struct Options {
unsigned MaxConfigsPerOpcode = 1;
};
explicit SnippetGenerator(const LLVMState &State, const Options &Opts);
virtual ~SnippetGenerator();
@ -66,6 +70,7 @@ public:
protected:
const LLVMState &State;
const Options Opts;
private:
// API to be implemented by subclasses.

View File

@ -36,17 +36,17 @@ void ExegesisTarget::registerTarget(ExegesisTarget *Target) {
FirstTarget = Target;
}
std::unique_ptr<SnippetGenerator>
ExegesisTarget::createSnippetGenerator(InstructionBenchmark::ModeE Mode,
const LLVMState &State) const {
std::unique_ptr<SnippetGenerator> ExegesisTarget::createSnippetGenerator(
InstructionBenchmark::ModeE Mode, const LLVMState &State,
const SnippetGenerator::Options &Opts) const {
switch (Mode) {
case InstructionBenchmark::Unknown:
return nullptr;
case InstructionBenchmark::Latency:
return createLatencySnippetGenerator(State);
return createLatencySnippetGenerator(State, Opts);
case InstructionBenchmark::Uops:
case InstructionBenchmark::InverseThroughput:
return createUopsSnippetGenerator(State);
return createUopsSnippetGenerator(State, Opts);
}
return nullptr;
}
@ -66,14 +66,14 @@ ExegesisTarget::createBenchmarkRunner(InstructionBenchmark::ModeE Mode,
return nullptr;
}
std::unique_ptr<SnippetGenerator>
ExegesisTarget::createLatencySnippetGenerator(const LLVMState &State) const {
return std::make_unique<LatencySnippetGenerator>(State);
std::unique_ptr<SnippetGenerator> ExegesisTarget::createLatencySnippetGenerator(
const LLVMState &State, const SnippetGenerator::Options &Opts) const {
return std::make_unique<LatencySnippetGenerator>(State, Opts);
}
std::unique_ptr<SnippetGenerator>
ExegesisTarget::createUopsSnippetGenerator(const LLVMState &State) const {
return std::make_unique<UopsSnippetGenerator>(State);
std::unique_ptr<SnippetGenerator> ExegesisTarget::createUopsSnippetGenerator(
const LLVMState &State, const SnippetGenerator::Options &Opts) const {
return std::make_unique<UopsSnippetGenerator>(State, Opts);
}
std::unique_ptr<BenchmarkRunner> ExegesisTarget::createLatencyBenchmarkRunner(

View File

@ -125,7 +125,8 @@ public:
// Creates a snippet generator for the given mode.
std::unique_ptr<SnippetGenerator>
createSnippetGenerator(InstructionBenchmark::ModeE Mode,
const LLVMState &State) const;
const LLVMState &State,
const SnippetGenerator::Options &Opts) const;
// Creates a benchmark runner for the given mode.
std::unique_ptr<BenchmarkRunner>
createBenchmarkRunner(InstructionBenchmark::ModeE Mode,
@ -151,9 +152,9 @@ private:
// Targets can implement their own snippet generators/benchmarks runners by
// implementing these.
std::unique_ptr<SnippetGenerator> virtual createLatencySnippetGenerator(
const LLVMState &State) const;
const LLVMState &State, const SnippetGenerator::Options &Opts) const;
std::unique_ptr<SnippetGenerator> virtual createUopsSnippetGenerator(
const LLVMState &State) const;
const LLVMState &State, const SnippetGenerator::Options &Opts) const;
std::unique_ptr<BenchmarkRunner> virtual createLatencyBenchmarkRunner(
const LLVMState &State, InstructionBenchmark::ModeE Mode) const;
std::unique_ptr<BenchmarkRunner> virtual createUopsBenchmarkRunner(

View File

@ -22,7 +22,7 @@ namespace exegesis {
class UopsSnippetGenerator : public SnippetGenerator {
public:
UopsSnippetGenerator(const LLVMState &State) : SnippetGenerator(State) {}
using SnippetGenerator::SnippetGenerator;
~UopsSnippetGenerator() override;
llvm::Expected<std::vector<CodeTemplate>>

View File

@ -462,14 +462,16 @@ private:
sizeof(kUnavailableRegisters[0]));
}
std::unique_ptr<SnippetGenerator>
createLatencySnippetGenerator(const LLVMState &State) const override {
return std::make_unique<X86LatencySnippetGenerator>(State);
std::unique_ptr<SnippetGenerator> createLatencySnippetGenerator(
const LLVMState &State,
const SnippetGenerator::Options &Opts) const override {
return std::make_unique<X86LatencySnippetGenerator>(State, Opts);
}
std::unique_ptr<SnippetGenerator>
createUopsSnippetGenerator(const LLVMState &State) const override {
return std::make_unique<X86UopsSnippetGenerator>(State);
std::unique_ptr<SnippetGenerator> createUopsSnippetGenerator(
const LLVMState &State,
const SnippetGenerator::Options &Opts) const override {
return std::make_unique<X86UopsSnippetGenerator>(State, Opts);
}
bool matchesArch(llvm::Triple::ArchType Arch) const override {

View File

@ -95,6 +95,12 @@ static cl::opt<unsigned>
cl::desc("number of time to repeat the asm snippet"),
cl::cat(BenchmarkOptions), cl::init(10000));
static cl::opt<unsigned> MaxConfigsPerOpcode(
"max-configs-per-opcode",
cl::desc(
"allow to snippet generator to generate at most that many configs"),
cl::cat(BenchmarkOptions), cl::init(1));
static cl::opt<bool> IgnoreInvalidSchedClass(
"ignore-invalid-sched-class",
cl::desc("ignore instructions that do not define a sched class"),
@ -214,8 +220,11 @@ generateSnippets(const LLVMState &State, unsigned Opcode,
if (InstrDesc.isCall() || InstrDesc.isReturn())
return make_error<Failure>("Unsupported opcode: isCall/isReturn");
SnippetGenerator::Options Options;
Options.MaxConfigsPerOpcode = MaxConfigsPerOpcode;
const std::unique_ptr<SnippetGenerator> Generator =
State.getExegesisTarget().createSnippetGenerator(BenchmarkMode, State);
State.getExegesisTarget().createSnippetGenerator(BenchmarkMode, State,
Options);
if (!Generator)
llvm::report_fatal_error("cannot create snippet generator");
return Generator->generateConfigurations(Instr, ForbiddenRegs);

View File

@ -45,7 +45,7 @@ protected:
template <typename SnippetGeneratorT>
class SnippetGeneratorTest : public X86SnippetGeneratorTest {
protected:
SnippetGeneratorTest() : Generator(State) {}
SnippetGeneratorTest() : Generator(State, SnippetGenerator::Options()) {}
std::vector<CodeTemplate> checkAndGetCodeTemplates(unsigned Opcode) {
randomGenerator().seed(0); // Initialize seed.
@ -335,7 +335,8 @@ TEST_F(UopsSnippetGeneratorTest, MemoryUse) {
class FakeSnippetGenerator : public SnippetGenerator {
public:
FakeSnippetGenerator(const LLVMState &State) : SnippetGenerator(State) {}
FakeSnippetGenerator(const LLVMState &State, const Options &Opts)
: SnippetGenerator(State, Opts) {}
Instruction createInstruction(unsigned Opcode) {
return State.getIC().getInstr(Opcode);