1 // Copyright 2009 The Go Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style
3 // license that can be found in the LICENSE file.
18 var matchBenchmarks = flag.String("test.bench", "", "run only benchmarks matching `regexp`")
19 var benchTime = flag.Duration("test.benchtime", 1*time.Second, "run each benchmark for duration `d`")
20 var benchmarkMemory = flag.Bool("test.benchmem", false, "print memory allocations for benchmarks")
22 // Global lock to ensure only one benchmark runs at a time.
23 var benchmarkLock sync.Mutex
25 // Used for every benchmark for measuring memory.
26 var memStats runtime.MemStats
28 // An internal type but exported because it is cross-package; part of the implementation
29 // of the "go test" command.
30 type InternalBenchmark struct {
35 // B is a type passed to Benchmark functions to manage benchmark
36 // timing and to specify the number of iterations to run.
38 // A benchmark ends when its Benchmark function returns or calls any of the methods
39 // FailNow, Fatal, Fatalf, SkipNow, Skip, or Skipf. Those methods must be called
40 // only from the goroutine running the Benchmark function.
41 // The other reporting methods, such as the variations of Log and Error,
42 // may be called simultaneously from multiple goroutines.
44 // Like in tests, benchmark logs are accumulated during execution
45 // and dumped to standard error when done. Unlike in tests, benchmark logs
46 // are always printed, so as not to hide output whose existence may be
47 // affecting benchmark results.
50 importPath string // import path of the package containing the benchmark
53 previousN int // number of iterations in the previous run
54 previousDuration time.Duration // total duration of the previous run
56 benchTime time.Duration
58 missingBytes bool // one of the subbenchmarks does not have bytes set.
61 result BenchmarkResult
62 parallelism int // RunParallel creates parallelism*GOMAXPROCS goroutines
63 // The initial states of memStats.Mallocs and memStats.TotalAlloc.
66 // The net total of this test after being run.
71 // StartTimer starts timing a test. This function is called automatically
72 // before a benchmark starts, but it can also used to resume timing after
73 // a call to StopTimer.
74 func (b *B) StartTimer() {
76 runtime.ReadMemStats(&memStats)
77 b.startAllocs = memStats.Mallocs
78 b.startBytes = memStats.TotalAlloc
84 // StopTimer stops timing a test. This can be used to pause the timer
85 // while performing complex initialization that you don't
87 func (b *B) StopTimer() {
89 b.duration += time.Now().Sub(b.start)
90 if *benchmarkMemory || b.showAllocResult {
91 runtime.ReadMemStats(&memStats)
92 b.netAllocs += memStats.Mallocs - b.startAllocs
93 b.netBytes += memStats.TotalAlloc - b.startBytes
99 // ResetTimer zeros the elapsed benchmark time and memory allocation counters.
100 // It does not affect whether the timer is running.
101 func (b *B) ResetTimer() {
103 if *benchmarkMemory || b.showAllocResult {
104 runtime.ReadMemStats(&memStats)
105 b.startAllocs = memStats.Mallocs
106 b.startBytes = memStats.TotalAlloc
115 // SetBytes records the number of bytes processed in a single operation.
116 // If this is called, the benchmark will report ns/op and MB/s.
117 func (b *B) SetBytes(n int64) { b.bytes = n }
119 // ReportAllocs enables malloc statistics for this benchmark.
120 // It is equivalent to setting -test.benchmem, but it only affects the
121 // benchmark function that calls ReportAllocs.
122 func (b *B) ReportAllocs() {
123 b.showAllocResult = true
126 func (b *B) nsPerOp() int64 {
130 return b.duration.Nanoseconds() / int64(b.N)
133 // runN runs a single benchmark for the specified number of iterations.
134 func (b *B) runN(n int) {
136 defer benchmarkLock.Unlock()
137 // Try to get a comparable environment for each run
138 // by clearing garbage from previous runs.
140 b.raceErrors = -race.Errors()
148 b.previousDuration = b.duration
149 b.raceErrors += race.Errors()
150 if b.raceErrors > 0 {
151 b.Errorf("race detected during execution of benchmark")
155 func min(x, y int) int {
162 func max(x, y int) int {
169 // roundDown10 rounds a number down to the nearest power of 10.
170 func roundDown10(n int) int {
172 // tens = floor(log_10(n))
179 for i := 0; i < tens; i++ {
185 // roundUp rounds x up to a number of the form [1eX, 2eX, 3eX, 5eX].
186 func roundUp(n int) int {
187 base := roundDown10(n)
191 case n <= (2 * base):
193 case n <= (3 * base):
195 case n <= (5 * base):
202 // run1 runs the first iteration of benchFunc. It returns whether more
203 // iterations of this benchmarks should be run.
204 func (b *B) run1() bool {
205 if ctx := b.context; ctx != nil {
206 // Extend maxLen, if needed.
207 if n := len(b.name) + ctx.extLen + 1; n > ctx.maxLen {
208 ctx.maxLen = n + 8 // Add additional slack to avoid too many jumps in size.
212 // Signal that we're done whether we return normally
213 // or by FailNow's runtime.Goexit.
222 fmt.Fprintf(b.w, "--- FAIL: %s\n%s", b.name, b.output)
225 // Only print the output if we know we are not going to proceed.
226 // Otherwise it is printed in processBench.
227 if atomic.LoadInt32(&b.hasSub) != 0 || b.finished {
232 if b.chatty && (len(b.output) > 0 || b.finished) {
234 fmt.Fprintf(b.w, "--- %s: %s\n%s", tag, b.name, b.output)
241 var labelsOnce sync.Once
243 // run executes the benchmark in a separate goroutine, including all of its
244 // subbenchmarks. b must not have subbenchmarks.
245 func (b *B) run() BenchmarkResult {
246 labelsOnce.Do(func() {
247 fmt.Fprintf(b.w, "goos: %s\n", runtime.GOOS)
248 fmt.Fprintf(b.w, "goarch: %s\n", runtime.GOARCH)
249 if b.importPath != "" {
250 fmt.Fprintf(b.w, "pkg: %s\n", b.importPath)
253 if b.context != nil {
254 // Running go test --test.bench
255 b.context.processBench(b) // Must call doBench.
257 // Running func Benchmark.
263 func (b *B) doBench() BenchmarkResult {
269 // launch launches the benchmark function. It gradually increases the number
270 // of benchmark iterations until the benchmark runs for the requested benchtime.
271 // launch is run by the doBench function as a separate goroutine.
272 // run1 must have been called on b.
273 func (b *B) launch() {
274 // Signal that we're done whether we return normally
275 // or by FailNow's runtime.Goexit.
280 // Run the benchmark for at least the specified amount of time.
282 for n := 1; !b.failed && b.duration < d && n < 1e9; {
284 // Predict required iterations.
285 n = int(d.Nanoseconds())
286 if nsop := b.nsPerOp(); nsop != 0 {
289 // Run more iterations than we think we'll need (1.2x).
290 // Don't grow too fast in case we had timing errors previously.
291 // Be sure to run at least one more than last time.
292 n = max(min(n+n/5, 100*last), last+1)
293 // Round up to something easy to read.
297 b.result = BenchmarkResult{b.N, b.duration, b.bytes, b.netAllocs, b.netBytes}
300 // The results of a benchmark run.
301 // MemAllocs and MemBytes may be zero if memory benchmarking is not requested
302 // using B.ReportAllocs or the -benchmem command line flag.
303 type BenchmarkResult struct {
304 N int // The number of iterations.
305 T time.Duration // The total time taken.
306 Bytes int64 // Bytes processed in one iteration.
307 MemAllocs uint64 // The total number of memory allocations.
308 MemBytes uint64 // The total number of bytes allocated.
311 func (r BenchmarkResult) NsPerOp() int64 {
315 return r.T.Nanoseconds() / int64(r.N)
318 func (r BenchmarkResult) mbPerSec() float64 {
319 if r.Bytes <= 0 || r.T <= 0 || r.N <= 0 {
322 return (float64(r.Bytes) * float64(r.N) / 1e6) / r.T.Seconds()
325 // AllocsPerOp returns r.MemAllocs / r.N.
326 func (r BenchmarkResult) AllocsPerOp() int64 {
330 return int64(r.MemAllocs) / int64(r.N)
333 // AllocedBytesPerOp returns r.MemBytes / r.N.
334 func (r BenchmarkResult) AllocedBytesPerOp() int64 {
338 return int64(r.MemBytes) / int64(r.N)
341 func (r BenchmarkResult) String() string {
345 mb = fmt.Sprintf("\t%7.2f MB/s", mbs)
348 ns := fmt.Sprintf("%10d ns/op", nsop)
349 if r.N > 0 && nsop < 100 {
350 // The format specifiers here make sure that
351 // the ones digits line up for all three possible formats.
353 ns = fmt.Sprintf("%13.2f ns/op", float64(r.T.Nanoseconds())/float64(r.N))
355 ns = fmt.Sprintf("%12.1f ns/op", float64(r.T.Nanoseconds())/float64(r.N))
358 return fmt.Sprintf("%8d\t%s%s", r.N, ns, mb)
361 // MemString returns r.AllocedBytesPerOp and r.AllocsPerOp in the same format as 'go test'.
362 func (r BenchmarkResult) MemString() string {
363 return fmt.Sprintf("%8d B/op\t%8d allocs/op",
364 r.AllocedBytesPerOp(), r.AllocsPerOp())
367 // benchmarkName returns full name of benchmark including procs suffix.
368 func benchmarkName(name string, n int) string {
370 return fmt.Sprintf("%s-%d", name, n)
375 type benchContext struct {
378 maxLen int // The largest recorded benchmark name.
379 extLen int // Maximum extension length.
382 // An internal function but exported because it is cross-package; part of the implementation
383 // of the "go test" command.
384 func RunBenchmarks(matchString func(pat, str string) (bool, error), benchmarks []InternalBenchmark) {
385 runBenchmarks("", matchString, benchmarks)
388 func runBenchmarks(importPath string, matchString func(pat, str string) (bool, error), benchmarks []InternalBenchmark) bool {
389 // If no flag was specified, don't run benchmarks.
390 if len(*matchBenchmarks) == 0 {
393 // Collect matching benchmarks and determine longest name.
395 for _, procs := range cpuList {
396 if procs > maxprocs {
400 ctx := &benchContext{
401 match: newMatcher(matchString, *matchBenchmarks, "-test.bench"),
402 extLen: len(benchmarkName("", maxprocs)),
404 var bs []InternalBenchmark
405 for _, Benchmark := range benchmarks {
406 if _, matched, _ := ctx.match.fullName(nil, Benchmark.Name); matched {
407 bs = append(bs, Benchmark)
408 benchName := benchmarkName(Benchmark.Name, maxprocs)
409 if l := len(benchName) + ctx.extLen + 1; l > ctx.maxLen {
420 importPath: importPath,
421 benchFunc: func(b *B) {
422 for _, Benchmark := range bs {
423 b.Run(Benchmark.Name, Benchmark.F)
426 benchTime: *benchTime,
433 // processBench runs bench b for the configured CPU counts and prints the results.
434 func (ctx *benchContext) processBench(b *B) {
435 for i, procs := range cpuList {
436 runtime.GOMAXPROCS(procs)
437 benchName := benchmarkName(b.name, procs)
438 fmt.Fprintf(b.w, "%-*s\t", ctx.maxLen, benchName)
439 // Recompute the running time for all but the first iteration.
443 signal: make(chan bool),
448 benchFunc: b.benchFunc,
449 benchTime: b.benchTime,
455 // The output could be very long here, but probably isn't.
456 // We print it all, regardless, because we don't want to trim the reason
457 // the benchmark failed.
458 fmt.Fprintf(b.w, "--- FAIL: %s\n%s", benchName, b.output)
461 results := r.String()
462 if *benchmarkMemory || b.showAllocResult {
463 results += "\t" + r.MemString()
465 fmt.Fprintln(b.w, results)
466 // Unlike with tests, we ignore the -chatty flag and always print output for
467 // benchmarks since the output generation time will skew the results.
468 if len(b.output) > 0 {
470 fmt.Fprintf(b.w, "--- BENCH: %s\n%s", benchName, b.output)
472 if p := runtime.GOMAXPROCS(-1); p != procs {
473 fmt.Fprintf(os.Stderr, "testing: %s left GOMAXPROCS set to %d\n", benchName, p)
478 // Run benchmarks f as a subbenchmark with the given name. It reports
479 // whether there were any failures.
481 // A subbenchmark is like any other benchmark. A benchmark that calls Run at
482 // least once will not be measured itself and will be called once with N=1.
484 // Run may be called simultaneously from multiple goroutines, but all such
485 // calls must happen before the outer benchmark function for b returns.
486 func (b *B) Run(name string, f func(b *B)) bool {
487 // Since b has subbenchmarks, we will no longer run it as a benchmark itself.
488 // Release the lock and acquire it on exit to ensure locks stay paired.
489 atomic.StoreInt32(&b.hasSub, 1)
490 benchmarkLock.Unlock()
491 defer benchmarkLock.Lock()
493 benchName, ok, partial := b.name, true, false
494 if b.context != nil {
495 benchName, ok, partial = b.context.match.fullName(&b.common, name)
502 signal: make(chan bool),
509 importPath: b.importPath,
511 benchTime: b.benchTime,
515 // Partial name match, like -bench=X/Y matching BenchmarkX.
516 // Only process sub-benchmarks, if any.
517 atomic.StoreInt32(&sub.hasSub, 1)
526 // add simulates running benchmarks in sequence in a single iteration. It is
527 // used to give some meaningful results in case func Benchmark is used in
528 // combination with Run.
529 func (b *B) add(other BenchmarkResult) {
531 // The aggregated BenchmarkResults resemble running all subbenchmarks as
532 // in sequence in a single benchmark.
534 r.T += time.Duration(other.NsPerOp())
535 if other.Bytes == 0 {
536 // Summing Bytes is meaningless in aggregate if not all subbenchmarks
538 b.missingBytes = true
542 r.Bytes += other.Bytes
544 r.MemAllocs += uint64(other.AllocsPerOp())
545 r.MemBytes += uint64(other.AllocedBytesPerOp())
548 // trimOutput shortens the output from a benchmark, which can be very long.
549 func (b *B) trimOutput() {
550 // The output is likely to appear multiple times because the benchmark
551 // is run multiple times, but at least it will be seen. This is not a big deal
552 // because benchmarks rarely print, but just in case, we trim it if it's too long.
553 const maxNewlines = 10
554 for nlCount, j := 0, 0; j < len(b.output); j++ {
555 if b.output[j] == '\n' {
557 if nlCount >= maxNewlines {
558 b.output = append(b.output[:j], "\n\t... [output truncated]\n"...)
565 // A PB is used by RunParallel for running parallel benchmarks.
567 globalN *uint64 // shared between all worker goroutines iteration counter
568 grain uint64 // acquire that many iterations from globalN at once
569 cache uint64 // local cache of acquired iterations
570 bN uint64 // total number of iterations to execute (b.N)
573 // Next reports whether there are more iterations to execute.
574 func (pb *PB) Next() bool {
576 n := atomic.AddUint64(pb.globalN, pb.grain)
579 } else if n < pb.bN+pb.grain {
580 pb.cache = pb.bN + pb.grain - n
589 // RunParallel runs a benchmark in parallel.
590 // It creates multiple goroutines and distributes b.N iterations among them.
591 // The number of goroutines defaults to GOMAXPROCS. To increase parallelism for
592 // non-CPU-bound benchmarks, call SetParallelism before RunParallel.
593 // RunParallel is usually used with the go test -cpu flag.
595 // The body function will be run in each goroutine. It should set up any
596 // goroutine-local state and then iterate until pb.Next returns false.
597 // It should not use the StartTimer, StopTimer, or ResetTimer functions,
598 // because they have global effect. It should also not call Run.
599 func (b *B) RunParallel(body func(*PB)) {
601 return // Nothing to do when probing.
603 // Calculate grain size as number of iterations that take ~100µs.
604 // 100µs is enough to amortize the overhead and provide sufficient
605 // dynamic load balancing.
607 if b.previousN > 0 && b.previousDuration > 0 {
608 grain = 1e5 * uint64(b.previousN) / uint64(b.previousDuration)
613 // We expect the inner loop and function call to take at least 10ns,
614 // so do not do more than 100µs/10ns=1e4 iterations.
620 numProcs := b.parallelism * runtime.GOMAXPROCS(0)
621 var wg sync.WaitGroup
623 for p := 0; p < numProcs; p++ {
635 if n <= uint64(b.N) && !b.Failed() {
636 b.Fatal("RunParallel: body exited without pb.Next() == false")
640 // SetParallelism sets the number of goroutines used by RunParallel to p*GOMAXPROCS.
641 // There is usually no need to call SetParallelism for CPU-bound benchmarks.
642 // If p is less than 1, this call will have no effect.
643 func (b *B) SetParallelism(p int) {
649 // Benchmark benchmarks a single function. Useful for creating
650 // custom benchmarks that do not use the "go test" command.
652 // If f calls Run, the result will be an estimate of running all its
653 // subbenchmarks that don't call Run in sequence in a single benchmark.
654 func Benchmark(f func(b *B)) BenchmarkResult {
657 signal: make(chan bool),
661 benchTime: *benchTime,
669 type discard struct{}
671 func (discard) Write(b []byte) (n int, err error) { return len(b), nil }