// Copyright 2009 The Go Authors. All rights reserved.// Use of this source code is governed by a BSD-style// license that can be found in the LICENSE file.package testingimport ()func initBenchmarkFlags() {matchBenchmarks = flag.String("test.bench", "", "run only benchmarks matching `regexp`")benchmarkMemory = flag.Bool("test.benchmem", false, "print memory allocations for benchmarks")flag.Var(&benchTime, "test.benchtime", "run each benchmark for duration `d` or N times if `d` is of the form Nx")}var ( matchBenchmarks *string benchmarkMemory *bool benchTime = durationOrCountFlag{d: 1 * time.Second} // changed during test of testing package)type durationOrCountFlag struct { d time.Duration n int allowZero bool}func ( *durationOrCountFlag) () string {if .n > 0 {returnfmt.Sprintf("%dx", .n) }return .d.String()}func ( *durationOrCountFlag) ( string) error {ifstrings.HasSuffix(, "x") { , := strconv.ParseInt([:len()-1], 10, 0)if != nil || < 0 || (!.allowZero && == 0) {returnfmt.Errorf("invalid count") } * = durationOrCountFlag{n: int()}returnnil } , := time.ParseDuration()if != nil || < 0 || (!.allowZero && == 0) {returnfmt.Errorf("invalid duration") } * = durationOrCountFlag{d: }returnnil}// Global lock to ensure only one benchmark runs at a time.var benchmarkLock sync.Mutex// Used for every benchmark for measuring memory.var memStats runtime.MemStats// InternalBenchmark is an internal type but exported because it is cross-package;// it is part of the implementation of the "go test" command.typeInternalBenchmarkstruct { Name string F func(b *B)}// B is a type passed to [Benchmark] functions to manage benchmark// timing and to specify the number of iterations to run.//// A benchmark ends when its Benchmark function returns or calls any of the methods// FailNow, Fatal, Fatalf, SkipNow, Skip, or Skipf. Those methods must be called// only from the goroutine running the Benchmark function.// The other reporting methods, such as the variations of Log and Error,// may be called simultaneously from multiple goroutines.//// Like in tests, benchmark logs are accumulated during execution// and dumped to standard output when done. Unlike in tests, benchmark logs// are always printed, so as not to hide output whose existence may be// affecting benchmark results.typeBstruct {common importPath string// import path of the package containing the benchmark context *benchContext N int previousN int// number of iterations in the previous run previousDuration time.Duration// total duration of the previous run benchFunc func(b *B) benchTime durationOrCountFlag bytes int64 missingBytes bool// one of the subbenchmarks does not have bytes set. timerOn bool showAllocResult bool result BenchmarkResult parallelism int// RunParallel creates parallelism*GOMAXPROCS goroutines// The initial states of memStats.Mallocs and memStats.TotalAlloc. startAllocs uint64 startBytes uint64// The net total of this test after being run. netAllocs uint64 netBytes uint64// Extra metrics collected by ReportMetric. extra map[string]float64}// StartTimer starts timing a test. This function is called automatically// before a benchmark starts, but it can also be used to resume timing after// a call to [B.StopTimer].func ( *B) () {if !.timerOn {runtime.ReadMemStats(&memStats) .startAllocs = memStats.Mallocs .startBytes = memStats.TotalAlloc .start = highPrecisionTimeNow() .timerOn = true }}// StopTimer stops timing a test. This can be used to pause the timer// while performing complex initialization that you don't// want to measure.func ( *B) () {if .timerOn { .duration += highPrecisionTimeSince(.start)runtime.ReadMemStats(&memStats) .netAllocs += memStats.Mallocs - .startAllocs .netBytes += memStats.TotalAlloc - .startBytes .timerOn = false }}// ResetTimer zeroes the elapsed benchmark time and memory allocation counters// and deletes user-reported metrics.// It does not affect whether the timer is running.func ( *B) () {if .extra == nil {// Allocate the extra map before reading memory stats. // Pre-size it to make more allocation unlikely. .extra = make(map[string]float64, 16) } else {clear(.extra) }if .timerOn {runtime.ReadMemStats(&memStats) .startAllocs = memStats.Mallocs .startBytes = memStats.TotalAlloc .start = highPrecisionTimeNow() } .duration = 0 .netAllocs = 0 .netBytes = 0}// SetBytes records the number of bytes processed in a single operation.// If this is called, the benchmark will report ns/op and MB/s.func ( *B) ( int64) { .bytes = }// ReportAllocs enables malloc statistics for this benchmark.// It is equivalent to setting -test.benchmem, but it only affects the// benchmark function that calls ReportAllocs.func ( *B) () { .showAllocResult = true}// runN runs a single benchmark for the specified number of iterations.func ( *B) ( int) {benchmarkLock.Lock()deferbenchmarkLock.Unlock()deferfunc() { .runCleanup(normalPanic) .checkRaces() }()// Try to get a comparable environment for each run // by clearing garbage from previous runs.runtime.GC() .resetRaces() .N = .parallelism = 1 .ResetTimer() .StartTimer() .benchFunc() .StopTimer() .previousN = .previousDuration = .duration}// run1 runs the first iteration of benchFunc. It reports whether more// iterations of this benchmarks should be run.func ( *B) () bool {if := .context; != nil {// Extend maxLen, if needed.if := len(.name) + .extLen + 1; > .maxLen { .maxLen = + 8// Add additional slack to avoid too many jumps in size. } }gofunc() {// Signal that we're done whether we return normally // or by FailNow's runtime.Goexit.deferfunc() { .signal <- true }() .runN(1) }() <-.signalif .failed {fmt.Fprintf(.w, "%s--- FAIL: %s\n%s", .chatty.prefix(), .name, .output)returnfalse }// Only print the output if we know we are not going to proceed. // Otherwise it is printed in processBench. .mu.RLock() := .finished .mu.RUnlock()if .hasSub.Load() || { := "BENCH"if .skipped { = "SKIP" }if .chatty != nil && (len(.output) > 0 || ) { .trimOutput()fmt.Fprintf(.w, "%s--- %s: %s\n%s", .chatty.prefix(), , .name, .output) }returnfalse }returntrue}var labelsOnce sync.Once// run executes the benchmark in a separate goroutine, including all of its// subbenchmarks. b must not have subbenchmarks.func ( *B) () {labelsOnce.Do(func() {fmt.Fprintf(.w, "goos: %s\n", runtime.GOOS)fmt.Fprintf(.w, "goarch: %s\n", runtime.GOARCH)if .importPath != "" {fmt.Fprintf(.w, "pkg: %s\n", .importPath) }if := sysinfo.CPUName(); != "" {fmt.Fprintf(.w, "cpu: %s\n", ) } })if .context != nil {// Running go test --test.bench .context.processBench() // Must call doBench. } else {// Running func Benchmark. .doBench() }}func ( *B) () BenchmarkResult {go .launch() <-.signalreturn .result}// launch launches the benchmark function. It gradually increases the number// of benchmark iterations until the benchmark runs for the requested benchtime.// launch is run by the doBench function as a separate goroutine.// run1 must have been called on b.func ( *B) () {// Signal that we're done whether we return normally // or by FailNow's runtime.Goexit.deferfunc() { .signal <- true }()// Run the benchmark for at least the specified amount of time.if .benchTime.n > 0 {// We already ran a single iteration in run1. // If -benchtime=1x was requested, use that result. // See https://golang.org/issue/32051.if .benchTime.n > 1 { .runN(.benchTime.n) } } else { := .benchTime.dfor := int64(1); !.failed && .duration < && < 1e9; { := // Predict required iterations. := .Nanoseconds() := int64(.N) := .duration.Nanoseconds()if <= 0 {// Round up, to avoid div by zero. = 1 }// Order of operations matters. // For very fast benchmarks, prevIters ~= prevns. // If you divide first, you get 0 or 1, // which can hide an order of magnitude in execution time. // So multiply first, then divide. = * / // Run more iterations than we think we'll need (1.2x). += / 5// Don't grow too fast in case we had timing errors previously. = min(, 100*)// Be sure to run at least one more than last time. = max(, +1)// Don't run more than 1e9 times. (This also keeps n in int range on 32 bit platforms.) = min(, 1e9) .runN(int()) } } .result = BenchmarkResult{.N, .duration, .bytes, .netAllocs, .netBytes, .extra}}// Elapsed returns the measured elapsed time of the benchmark.// The duration reported by Elapsed matches the one measured by// [B.StartTimer], [B.StopTimer], and [B.ResetTimer].func ( *B) () time.Duration { := .durationif .timerOn { += highPrecisionTimeSince(.start) }return}// ReportMetric adds "n unit" to the reported benchmark results.// If the metric is per-iteration, the caller should divide by b.N,// and by convention units should end in "/op".// ReportMetric overrides any previously reported value for the same unit.// ReportMetric panics if unit is the empty string or if unit contains// any whitespace.// If unit is a unit normally reported by the benchmark framework itself// (such as "allocs/op"), ReportMetric will override that metric.// Setting "ns/op" to 0 will suppress that built-in metric.func ( *B) ( float64, string) {if == "" {panic("metric unit must not be empty") }ifstrings.IndexFunc(, unicode.IsSpace) >= 0 {panic("metric unit must not contain whitespace") } .extra[] = }// BenchmarkResult contains the results of a benchmark run.typeBenchmarkResultstruct { N int// The number of iterations. T time.Duration// The total time taken. Bytes int64// Bytes processed in one iteration. MemAllocs uint64// The total number of memory allocations. MemBytes uint64// The total number of bytes allocated.// Extra records additional metrics reported by ReportMetric. Extra map[string]float64}// NsPerOp returns the "ns/op" metric.func ( BenchmarkResult) () int64 {if , := .Extra["ns/op"]; {returnint64() }if .N <= 0 {return0 }return .T.Nanoseconds() / int64(.N)}// mbPerSec returns the "MB/s" metric.func ( BenchmarkResult) () float64 {if , := .Extra["MB/s"]; {return }if .Bytes <= 0 || .T <= 0 || .N <= 0 {return0 }return (float64(.Bytes) * float64(.N) / 1e6) / .T.Seconds()}// AllocsPerOp returns the "allocs/op" metric,// which is calculated as r.MemAllocs / r.N.func ( BenchmarkResult) () int64 {if , := .Extra["allocs/op"]; {returnint64() }if .N <= 0 {return0 }returnint64(.MemAllocs) / int64(.N)}// AllocedBytesPerOp returns the "B/op" metric,// which is calculated as r.MemBytes / r.N.func ( BenchmarkResult) () int64 {if , := .Extra["B/op"]; {returnint64() }if .N <= 0 {return0 }returnint64(.MemBytes) / int64(.N)}// String returns a summary of the benchmark results.// It follows the benchmark result line format from// https://golang.org/design/14313-benchmark-format, not including the// benchmark name.// Extra metrics override built-in metrics of the same name.// String does not include allocs/op or B/op, since those are reported// by [BenchmarkResult.MemString].func ( BenchmarkResult) () string { := new(strings.Builder)fmt.Fprintf(, "%8d", .N)// Get ns/op as a float. , := .Extra["ns/op"]if ! { = float64(.T.Nanoseconds()) / float64(.N) }if != 0 { .WriteByte('\t')prettyPrint(, , "ns/op") }if := .mbPerSec(); != 0 {fmt.Fprintf(, "\t%7.2f MB/s", ) }// Print extra metrics that aren't represented in the standard // metrics.var []stringfor := range .Extra {switch {case"ns/op", "MB/s", "B/op", "allocs/op":// Built-in metrics reported elsewhere.continue } = append(, ) }slices.Sort()for , := range { .WriteByte('\t')prettyPrint(, .Extra[], ) }return .String()}func prettyPrint( io.Writer, float64, string) {// Print all numbers with 10 places before the decimal point // and small numbers with four sig figs. Field widths are // chosen to fit the whole part in 10 places while aligning // the decimal point of all fractional formats.varstringswitch := math.Abs(); {case == 0 || >= 999.95: = "%10.0f %s"case >= 99.995: = "%12.1f %s"case >= 9.9995: = "%13.2f %s"case >= 0.99995: = "%14.3f %s"case >= 0.099995: = "%15.4f %s"case >= 0.0099995: = "%16.5f %s"case >= 0.00099995: = "%17.6f %s"default: = "%18.7f %s" }fmt.Fprintf(, , , )}// MemString returns r.AllocedBytesPerOp and r.AllocsPerOp in the same format as 'go test'.func ( BenchmarkResult) () string {returnfmt.Sprintf("%8d B/op\t%8d allocs/op", .AllocedBytesPerOp(), .AllocsPerOp())}// benchmarkName returns full name of benchmark including procs suffix.func benchmarkName( string, int) string {if != 1 {returnfmt.Sprintf("%s-%d", , ) }return}type benchContext struct { match *matcher maxLen int// The largest recorded benchmark name. extLen int// Maximum extension length.}// RunBenchmarks is an internal function but exported because it is cross-package;// it is part of the implementation of the "go test" command.func ( func(, string) (bool, error), []InternalBenchmark) {runBenchmarks("", , )}func runBenchmarks( string, func(, string) (bool, error), []InternalBenchmark) bool {// If no flag was specified, don't run benchmarks.iflen(*matchBenchmarks) == 0 {returntrue }// Collect matching benchmarks and determine longest name. := 1for , := rangecpuList {if > { = } } := &benchContext{match: newMatcher(, *matchBenchmarks, "-test.bench", *skip),extLen: len(benchmarkName("", )), }var []InternalBenchmarkfor , := range {if , , := .match.fullName(nil, .Name); { = append(, ) := benchmarkName(.Name, )if := len() + .extLen + 1; > .maxLen { .maxLen = } } } := &B{common: common{name: "Main",w: os.Stdout,bench: true, },importPath: ,benchFunc: func( *B) {for , := range { .Run(.Name, .F) } },benchTime: benchTime,context: , }ifVerbose() { .chatty = newChattyPrinter(.w) } .runN(1)return !.failed}// processBench runs bench b for the configured CPU counts and prints the results.func ( *benchContext) ( *B) {for , := rangecpuList {for := uint(0); < *count; ++ {runtime.GOMAXPROCS() := benchmarkName(.name, )// If it's chatty, we've already printed this information.if .chatty == nil {fmt.Fprintf(.w, "%-*s\t", .maxLen, ) }// Recompute the running time for all but the first iteration.if > 0 || > 0 { = &B{common: common{signal: make(chanbool),name: .name,w: .w,chatty: .chatty,bench: true, },benchFunc: .benchFunc,benchTime: .benchTime, } .run1() } := .doBench()if .failed {// The output could be very long here, but probably isn't. // We print it all, regardless, because we don't want to trim the reason // the benchmark failed.fmt.Fprintf(.w, "%s--- FAIL: %s\n%s", .chatty.prefix(), , .output)continue } := .String()if .chatty != nil {fmt.Fprintf(.w, "%-*s\t", .maxLen, ) }if *benchmarkMemory || .showAllocResult { += "\t" + .MemString() }fmt.Fprintln(.w, )// Unlike with tests, we ignore the -chatty flag and always print output for // benchmarks since the output generation time will skew the results.iflen(.output) > 0 { .trimOutput()fmt.Fprintf(.w, "%s--- BENCH: %s\n%s", .chatty.prefix(), , .output) }if := runtime.GOMAXPROCS(-1); != {fmt.Fprintf(os.Stderr, "testing: %s left GOMAXPROCS set to %d\n", , ) }if .chatty != nil && .chatty.json { .chatty.Updatef("", "=== NAME %s\n", "") } } }}// If hideStdoutForTesting is true, Run does not print the benchName.// This avoids a spurious print during 'go test' on package testing itself,// which invokes b.Run in its own tests (see sub_test.go).var hideStdoutForTesting = false// Run benchmarks f as a subbenchmark with the given name. It reports// whether there were any failures.//// A subbenchmark is like any other benchmark. A benchmark that calls Run at// least once will not be measured itself and will be called once with N=1.func ( *B) ( string, func( *B)) bool {// Since b has subbenchmarks, we will no longer run it as a benchmark itself. // Release the lock and acquire it on exit to ensure locks stay paired. .hasSub.Store(true)benchmarkLock.Unlock()deferbenchmarkLock.Lock() , , := .name, true, falseif .context != nil { , , = .context.match.fullName(&.common, ) }if ! {returntrue }var [maxStackLen]uintptr := runtime.Callers(2, [:]) := &B{common: common{signal: make(chanbool),name: ,parent: &.common,level: .level + 1,creator: [:],w: .w,chatty: .chatty,bench: true, },importPath: .importPath,benchFunc: ,benchTime: .benchTime,context: .context, }if {// Partial name match, like -bench=X/Y matching BenchmarkX. // Only process sub-benchmarks, if any. .hasSub.Store(true) }if .chatty != nil {labelsOnce.Do(func() {fmt.Printf("goos: %s\n", runtime.GOOS)fmt.Printf("goarch: %s\n", runtime.GOARCH)if .importPath != "" {fmt.Printf("pkg: %s\n", .importPath) }if := sysinfo.CPUName(); != "" {fmt.Printf("cpu: %s\n", ) } })if !hideStdoutForTesting {if .chatty.json { .chatty.Updatef(, "=== RUN %s\n", ) }fmt.Println() } }if .run1() { .run() } .add(.result)return !.failed}// add simulates running benchmarks in sequence in a single iteration. It is// used to give some meaningful results in case func Benchmark is used in// combination with Run.func ( *B) ( BenchmarkResult) { := &.result// The aggregated BenchmarkResults resemble running all subbenchmarks as // in sequence in a single benchmark. .N = 1 .T += time.Duration(.NsPerOp())if .Bytes == 0 {// Summing Bytes is meaningless in aggregate if not all subbenchmarks // set it. .missingBytes = true .Bytes = 0 }if !.missingBytes { .Bytes += .Bytes } .MemAllocs += uint64(.AllocsPerOp()) .MemBytes += uint64(.AllocedBytesPerOp())}// trimOutput shortens the output from a benchmark, which can be very long.func ( *B) () {// The output is likely to appear multiple times because the benchmark // is run multiple times, but at least it will be seen. This is not a big deal // because benchmarks rarely print, but just in case, we trim it if it's too long.const = 10for , := 0, 0; < len(.output); ++ {if .output[] == '\n' { ++if >= { .output = append(.output[:], "\n\t... [output truncated]\n"...)break } } }}// A PB is used by RunParallel for running parallel benchmarks.typePBstruct { globalN *atomic.Uint64// shared between all worker goroutines iteration counter grain uint64// acquire that many iterations from globalN at once cache uint64// local cache of acquired iterations bN uint64// total number of iterations to execute (b.N)}// Next reports whether there are more iterations to execute.func ( *PB) () bool {if .cache == 0 { := .globalN.Add(.grain)if <= .bN { .cache = .grain } elseif < .bN+.grain { .cache = .bN + .grain - } else {returnfalse } } .cache--returntrue}// RunParallel runs a benchmark in parallel.// It creates multiple goroutines and distributes b.N iterations among them.// The number of goroutines defaults to GOMAXPROCS. To increase parallelism for// non-CPU-bound benchmarks, call [B.SetParallelism] before RunParallel.// RunParallel is usually used with the go test -cpu flag.//// The body function will be run in each goroutine. It should set up any// goroutine-local state and then iterate until pb.Next returns false.// It should not use the [B.StartTimer], [B.StopTimer], or [B.ResetTimer] functions,// because they have global effect. It should also not call [B.Run].//// RunParallel reports ns/op values as wall time for the benchmark as a whole,// not the sum of wall time or CPU time over each parallel goroutine.func ( *B) ( func(*PB)) {if .N == 0 {return// Nothing to do when probing. }// Calculate grain size as number of iterations that take ~100µs. // 100µs is enough to amortize the overhead and provide sufficient // dynamic load balancing. := uint64(0)if .previousN > 0 && .previousDuration > 0 { = 1e5 * uint64(.previousN) / uint64(.previousDuration) }if < 1 { = 1 }// We expect the inner loop and function call to take at least 10ns, // so do not do more than 100µs/10ns=1e4 iterations.if > 1e4 { = 1e4 }varatomic.Uint64 := .parallelism * runtime.GOMAXPROCS(0)varsync.WaitGroup .Add()for := 0; < ; ++ {gofunc() {defer .Done() := &PB{globalN: &,grain: ,bN: uint64(.N), } () }() } .Wait()if .Load() <= uint64(.N) && !.Failed() { .Fatal("RunParallel: body exited without pb.Next() == false") }}// SetParallelism sets the number of goroutines used by [B.RunParallel] to p*GOMAXPROCS.// There is usually no need to call SetParallelism for CPU-bound benchmarks.// If p is less than 1, this call will have no effect.func ( *B) ( int) {if >= 1 { .parallelism = }}// Benchmark benchmarks a single function. It is useful for creating// custom benchmarks that do not use the "go test" command.//// If f depends on testing flags, then [Init] must be used to register// those flags before calling Benchmark and before calling [flag.Parse].//// If f calls Run, the result will be an estimate of running all its// subbenchmarks that don't call Run in sequence in a single benchmark.func ( func( *B)) BenchmarkResult { := &B{common: common{signal: make(chanbool),w: discard{}, },benchFunc: ,benchTime: benchTime, }if .run1() { .run() }return .result}type discard struct{}func (discard) ( []byte) ( int, error) { returnlen(), nil }
The pages are generated with Goldsv0.6.9-preview. (GOOS=linux GOARCH=amd64)
Golds is a Go 101 project developed by Tapir Liu.
PR and bug reports are welcome and can be submitted to the issue list.
Please follow @Go100and1 (reachable from the left QR code) to get the latest news of Golds.