[llvm] r352632 - [llvm-exegesis] Add throughput mode.
Clement Courbet via llvm-commits
llvm-commits at lists.llvm.org
Wed Jan 30 08:02:20 PST 2019
Author: courbet
Date: Wed Jan 30 08:02:20 2019
New Revision: 352632
URL: http://llvm.org/viewvc/llvm-project?rev=352632&view=rev
Log:
[llvm-exegesis] Add throughput mode.
Summary:
This just uses the latency benchmark runner on the parallel uops snippet
generator.
Fixes PR37698.
Reviewers: gchatelet
Subscribers: tschuett, RKSimon, llvm-commits
Differential Revision: https://reviews.llvm.org/D57000
Added:
llvm/trunk/test/tools/llvm-exegesis/X86/inverse_throughput-by-opcode-name.s
Modified:
llvm/trunk/docs/CommandGuide/llvm-exegesis.rst
llvm/trunk/tools/llvm-exegesis/lib/BenchmarkResult.cpp
llvm/trunk/tools/llvm-exegesis/lib/BenchmarkResult.h
llvm/trunk/tools/llvm-exegesis/lib/BenchmarkRunner.h
llvm/trunk/tools/llvm-exegesis/lib/Latency.cpp
llvm/trunk/tools/llvm-exegesis/lib/Latency.h
llvm/trunk/tools/llvm-exegesis/lib/Target.cpp
llvm/trunk/tools/llvm-exegesis/lib/Target.h
llvm/trunk/tools/llvm-exegesis/llvm-exegesis.cpp
Modified: llvm/trunk/docs/CommandGuide/llvm-exegesis.rst
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/docs/CommandGuide/llvm-exegesis.rst?rev=352632&r1=352631&r2=352632&view=diff
==============================================================================
--- llvm/trunk/docs/CommandGuide/llvm-exegesis.rst (original)
+++ llvm/trunk/docs/CommandGuide/llvm-exegesis.rst Wed Jan 30 08:02:20 2019
@@ -10,13 +10,13 @@ DESCRIPTION
-----------
:program:`llvm-exegesis` is a benchmarking tool that uses information available
-in LLVM to measure host machine instruction characteristics like latency or port
-decomposition.
+in LLVM to measure host machine instruction characteristics like latency,
+throughput, or port decomposition.
Given an LLVM opcode name and a benchmarking mode, :program:`llvm-exegesis`
generates a code snippet that makes execution as serial (resp. as parallel) as
-possible so that we can measure the latency (resp. uop decomposition) of the
-instruction.
+possible so that we can measure the latency (resp. inverse throughput/uop decomposition)
+of the instruction.
The code snippet is jitted and executed on the host subtarget. The time taken
(resp. resource usage) is measured using hardware performance counters. The
result is printed out as YAML to the standard output.
@@ -37,11 +37,13 @@ instruction, run:
$ llvm-exegesis -mode=latency -opcode-name=ADD64rr
-Measuring the uop decomposition of an instruction works similarly:
+Measuring the uop decomposition or inverse throughput of an instruction works similarly:
.. code-block:: bash
$ llvm-exegesis -mode=uops -opcode-name=ADD64rr
+ $ llvm-exegesis -mode=inverse_throughput -opcode-name=ADD64rr
+
The output is a YAML document (the default is to write to stdout, but you can
redirect the output to a file using `-benchmarks-file`):
@@ -186,7 +188,7 @@ OPTIONS
Specify the custom code snippet to measure. See example 2 for details.
Either `opcode-index`, `opcode-name` or `snippets-file` must be set.
-.. option:: -mode=[latency|uops|analysis]
+.. option:: -mode=[latency|uops|inverse_throughput|analysis]
Specify the run mode.
@@ -197,8 +199,8 @@ OPTIONS
.. option:: -benchmarks-file=</path/to/file>
- File to read (`analysis` mode) or write (`latency`/`uops` modes) benchmark
- results. "-" uses stdin/stdout.
+ File to read (`analysis` mode) or write (`latency`/`uops`/`inverse_throughput`
+ modes) benchmark results. "-" uses stdin/stdout.
.. option:: -analysis-clusters-output-file=</path/to/file>
Added: llvm/trunk/test/tools/llvm-exegesis/X86/inverse_throughput-by-opcode-name.s
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/tools/llvm-exegesis/X86/inverse_throughput-by-opcode-name.s?rev=352632&view=auto
==============================================================================
--- llvm/trunk/test/tools/llvm-exegesis/X86/inverse_throughput-by-opcode-name.s (added)
+++ llvm/trunk/test/tools/llvm-exegesis/X86/inverse_throughput-by-opcode-name.s Wed Jan 30 08:02:20 2019
@@ -0,0 +1,8 @@
+# RUN: llvm-exegesis -mode=inverse_throughput -opcode-name=ADD32rr | FileCheck %s
+
+CHECK: ---
+CHECK-NEXT: mode: inverse_throughput
+CHECK-NEXT: key:
+CHECK-NEXT: instructions:
+CHECK-NEXT: ADD32rr
+CHECK: key: inverse_throughput
Modified: llvm/trunk/tools/llvm-exegesis/lib/BenchmarkResult.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvm-exegesis/lib/BenchmarkResult.cpp?rev=352632&r1=352631&r2=352632&view=diff
==============================================================================
--- llvm/trunk/tools/llvm-exegesis/lib/BenchmarkResult.cpp (original)
+++ llvm/trunk/tools/llvm-exegesis/lib/BenchmarkResult.cpp Wed Jan 30 08:02:20 2019
@@ -209,6 +209,8 @@ struct ScalarEnumerationTraits<exegesis:
Io.enumCase(Value, "", exegesis::InstructionBenchmark::Unknown);
Io.enumCase(Value, "latency", exegesis::InstructionBenchmark::Latency);
Io.enumCase(Value, "uops", exegesis::InstructionBenchmark::Uops);
+ Io.enumCase(Value, "inverse_throughput",
+ exegesis::InstructionBenchmark::InverseThroughput);
}
};
Modified: llvm/trunk/tools/llvm-exegesis/lib/BenchmarkResult.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvm-exegesis/lib/BenchmarkResult.h?rev=352632&r1=352631&r2=352632&view=diff
==============================================================================
--- llvm/trunk/tools/llvm-exegesis/lib/BenchmarkResult.h (original)
+++ llvm/trunk/tools/llvm-exegesis/lib/BenchmarkResult.h Wed Jan 30 08:02:20 2019
@@ -57,7 +57,7 @@ struct BenchmarkMeasure {
// The result of an instruction benchmark.
struct InstructionBenchmark {
InstructionBenchmarkKey Key;
- enum ModeE { Unknown, Latency, Uops };
+ enum ModeE { Unknown, Latency, Uops, InverseThroughput };
ModeE Mode;
std::string CpuName;
std::string LLVMTriple;
Modified: llvm/trunk/tools/llvm-exegesis/lib/BenchmarkRunner.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvm-exegesis/lib/BenchmarkRunner.h?rev=352632&r1=352631&r2=352632&view=diff
==============================================================================
--- llvm/trunk/tools/llvm-exegesis/lib/BenchmarkRunner.h (original)
+++ llvm/trunk/tools/llvm-exegesis/lib/BenchmarkRunner.h Wed Jan 30 08:02:20 2019
@@ -75,6 +75,7 @@ public:
protected:
const LLVMState &State;
+ const InstructionBenchmark::ModeE Mode;
private:
virtual llvm::Expected<std::vector<BenchmarkMeasure>>
@@ -84,7 +85,6 @@ private:
writeObjectFile(const BenchmarkCode &Configuration,
llvm::ArrayRef<llvm::MCInst> Code) const;
- const InstructionBenchmark::ModeE Mode;
const std::unique_ptr<ScratchSpace> Scratch;
};
Modified: llvm/trunk/tools/llvm-exegesis/lib/Latency.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvm-exegesis/lib/Latency.cpp?rev=352632&r1=352631&r2=352632&view=diff
==============================================================================
--- llvm/trunk/tools/llvm-exegesis/lib/Latency.cpp (original)
+++ llvm/trunk/tools/llvm-exegesis/lib/Latency.cpp Wed Jan 30 08:02:20 2019
@@ -165,6 +165,14 @@ LatencySnippetGenerator::generateCodeTem
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;
llvm::Expected<std::vector<BenchmarkMeasure>>
@@ -184,8 +192,17 @@ LatencyBenchmarkRunner::runMeasurements(
if (*ExpectedCounterValue < MinValue)
MinValue = *ExpectedCounterValue;
}
- std::vector<BenchmarkMeasure> Result = {
- BenchmarkMeasure::Create("latency", MinValue)};
+ 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);
}
Modified: llvm/trunk/tools/llvm-exegesis/lib/Latency.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvm-exegesis/lib/Latency.h?rev=352632&r1=352631&r2=352632&view=diff
==============================================================================
--- llvm/trunk/tools/llvm-exegesis/lib/Latency.h (original)
+++ llvm/trunk/tools/llvm-exegesis/lib/Latency.h Wed Jan 30 08:02:20 2019
@@ -32,8 +32,8 @@ public:
class LatencyBenchmarkRunner : public BenchmarkRunner {
public:
- LatencyBenchmarkRunner(const LLVMState &State)
- : BenchmarkRunner(State, InstructionBenchmark::Latency) {}
+ LatencyBenchmarkRunner(const LLVMState &State,
+ InstructionBenchmark::ModeE Mode);
~LatencyBenchmarkRunner() override;
private:
Modified: llvm/trunk/tools/llvm-exegesis/lib/Target.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvm-exegesis/lib/Target.cpp?rev=352632&r1=352631&r2=352632&view=diff
==============================================================================
--- llvm/trunk/tools/llvm-exegesis/lib/Target.cpp (original)
+++ llvm/trunk/tools/llvm-exegesis/lib/Target.cpp Wed Jan 30 08:02:20 2019
@@ -45,6 +45,7 @@ ExegesisTarget::createSnippetGenerator(I
case InstructionBenchmark::Latency:
return createLatencySnippetGenerator(State);
case InstructionBenchmark::Uops:
+ case InstructionBenchmark::InverseThroughput:
return createUopsSnippetGenerator(State);
}
return nullptr;
@@ -57,7 +58,8 @@ ExegesisTarget::createBenchmarkRunner(In
case InstructionBenchmark::Unknown:
return nullptr;
case InstructionBenchmark::Latency:
- return createLatencyBenchmarkRunner(State);
+ case InstructionBenchmark::InverseThroughput:
+ return createLatencyBenchmarkRunner(State, Mode);
case InstructionBenchmark::Uops:
return createUopsBenchmarkRunner(State);
}
@@ -74,9 +76,9 @@ ExegesisTarget::createUopsSnippetGenerat
return llvm::make_unique<UopsSnippetGenerator>(State);
}
-std::unique_ptr<BenchmarkRunner>
-ExegesisTarget::createLatencyBenchmarkRunner(const LLVMState &State) const {
- return llvm::make_unique<LatencyBenchmarkRunner>(State);
+std::unique_ptr<BenchmarkRunner> ExegesisTarget::createLatencyBenchmarkRunner(
+ const LLVMState &State, InstructionBenchmark::ModeE Mode) const {
+ return llvm::make_unique<LatencyBenchmarkRunner>(State, Mode);
}
std::unique_ptr<BenchmarkRunner>
Modified: llvm/trunk/tools/llvm-exegesis/lib/Target.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvm-exegesis/lib/Target.h?rev=352632&r1=352631&r2=352632&view=diff
==============================================================================
--- llvm/trunk/tools/llvm-exegesis/lib/Target.h (original)
+++ llvm/trunk/tools/llvm-exegesis/lib/Target.h Wed Jan 30 08:02:20 2019
@@ -130,7 +130,7 @@ private:
std::unique_ptr<SnippetGenerator> virtual createUopsSnippetGenerator(
const LLVMState &State) const;
std::unique_ptr<BenchmarkRunner> virtual createLatencyBenchmarkRunner(
- const LLVMState &State) const;
+ const LLVMState &State, InstructionBenchmark::ModeE Mode) const;
std::unique_ptr<BenchmarkRunner> virtual createUopsBenchmarkRunner(
const LLVMState &State) const;
Modified: llvm/trunk/tools/llvm-exegesis/llvm-exegesis.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvm-exegesis/llvm-exegesis.cpp?rev=352632&r1=352631&r2=352632&view=diff
==============================================================================
--- llvm/trunk/tools/llvm-exegesis/llvm-exegesis.cpp (original)
+++ llvm/trunk/tools/llvm-exegesis/llvm-exegesis.cpp Wed Jan 30 08:02:20 2019
@@ -56,16 +56,19 @@ static cl::opt<std::string> SnippetsFile
static cl::opt<std::string> BenchmarkFile("benchmarks-file", cl::desc(""),
cl::init(""));
-static cl::opt<exegesis::InstructionBenchmark::ModeE>
- BenchmarkMode("mode", cl::desc("the mode to run"),
- cl::values(clEnumValN(exegesis::InstructionBenchmark::Latency,
- "latency", "Instruction Latency"),
- clEnumValN(exegesis::InstructionBenchmark::Uops,
- "uops", "Uop Decomposition"),
- // When not asking for a specific benchmark mode,
- // we'll analyse the results.
- clEnumValN(exegesis::InstructionBenchmark::Unknown,
- "analysis", "Analysis")));
+static cl::opt<exegesis::InstructionBenchmark::ModeE> BenchmarkMode(
+ "mode", cl::desc("the mode to run"),
+ cl::values(clEnumValN(exegesis::InstructionBenchmark::Latency, "latency",
+ "Instruction Latency"),
+ clEnumValN(exegesis::InstructionBenchmark::InverseThroughput,
+ "inverse_throughput",
+ "Instruction Inverse Throughput"),
+ clEnumValN(exegesis::InstructionBenchmark::Uops, "uops",
+ "Uop Decomposition"),
+ // When not asking for a specific benchmark mode,
+ // we'll analyse the results.
+ clEnumValN(exegesis::InstructionBenchmark::Unknown, "analysis",
+ "Analysis")));
static cl::opt<unsigned>
NumRepetitions("num-repetitions",
More information about the llvm-commits
mailing list