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.
52 previousN int // number of iterations in the previous run
53 previousDuration time.Duration // total duration of the previous run
55 benchTime time.Duration
57 missingBytes bool // one of the subbenchmarks does not have bytes set.
60 result BenchmarkResult
61 parallelism int // RunParallel creates parallelism*GOMAXPROCS goroutines
62 // The initial states of memStats.Mallocs and memStats.TotalAlloc.
65 // The net total of this test after being run.
70 // StartTimer starts timing a test. This function is called automatically
71 // before a benchmark starts, but it can also used to resume timing after
72 // a call to StopTimer.
73 func (b *B) StartTimer() {
75 runtime.ReadMemStats(&memStats)
76 b.startAllocs = memStats.Mallocs
77 b.startBytes = memStats.TotalAlloc
83 // StopTimer stops timing a test. This can be used to pause the timer
84 // while performing complex initialization that you don't
86 func (b *B) StopTimer() {
88 b.duration += time.Now().Sub(b.start)
89 runtime.ReadMemStats(&memStats)
90 b.netAllocs += memStats.Mallocs - b.startAllocs
91 b.netBytes += memStats.TotalAlloc - b.startBytes
96 // ResetTimer zeros the elapsed benchmark time and memory allocation counters.
97 // It does not affect whether the timer is running.
98 func (b *B) ResetTimer() {
100 runtime.ReadMemStats(&memStats)
101 b.startAllocs = memStats.Mallocs
102 b.startBytes = memStats.TotalAlloc
110 // SetBytes records the number of bytes processed in a single operation.
111 // If this is called, the benchmark will report ns/op and MB/s.
112 func (b *B) SetBytes(n int64) { b.bytes = n }
114 // ReportAllocs enables malloc statistics for this benchmark.
115 // It is equivalent to setting -test.benchmem, but it only affects the
116 // benchmark function that calls ReportAllocs.
117 func (b *B) ReportAllocs() {
118 b.showAllocResult = true
121 func (b *B) nsPerOp() int64 {
125 return b.duration.Nanoseconds() / int64(b.N)
128 // runN runs a single benchmark for the specified number of iterations.
129 func (b *B) runN(n int) {
131 defer benchmarkLock.Unlock()
132 // Try to get a comparable environment for each run
133 // by clearing garbage from previous runs.
135 b.raceErrors = -race.Errors()
143 b.previousDuration = b.duration
144 b.raceErrors += race.Errors()
145 if b.raceErrors > 0 {
146 b.Errorf("race detected during execution of benchmark")
150 func min(x, y int) int {
157 func max(x, y int) int {
164 // roundDown10 rounds a number down to the nearest power of 10.
165 func roundDown10(n int) int {
167 // tens = floor(log_10(n))
174 for i := 0; i < tens; i++ {
180 // roundUp rounds x up to a number of the form [1eX, 2eX, 3eX, 5eX].
181 func roundUp(n int) int {
182 base := roundDown10(n)
186 case n <= (2 * base):
188 case n <= (3 * base):
190 case n <= (5 * base):
197 // run1 runs the first iteration of benchFunc. It returns whether more
198 // iterations of this benchmarks should be run.
199 func (b *B) run1() bool {
200 if ctx := b.context; ctx != nil {
201 // Extend maxLen, if needed.
202 if n := len(b.name) + ctx.extLen + 1; n > ctx.maxLen {
203 ctx.maxLen = n + 8 // Add additional slack to avoid too many jumps in size.
207 // Signal that we're done whether we return normally
208 // or by FailNow's runtime.Goexit.
217 fmt.Fprintf(b.w, "--- FAIL: %s\n%s", b.name, b.output)
220 // Only print the output if we know we are not going to proceed.
221 // Otherwise it is printed in processBench.
222 if atomic.LoadInt32(&b.hasSub) != 0 || b.finished {
227 if b.chatty && (len(b.output) > 0 || b.finished) {
229 fmt.Fprintf(b.w, "--- %s: %s\n%s", tag, b.name, b.output)
236 // run executes the benchmark in a separate goroutine, including all of its
237 // subbenchmarks. b must not have subbenchmarks.
238 func (b *B) run() BenchmarkResult {
239 if b.context != nil {
240 // Running go test --test.bench
241 b.context.processBench(b) // Must call doBench.
243 // Running func Benchmark.
249 func (b *B) doBench() BenchmarkResult {
255 // launch launches the benchmark function. It gradually increases the number
256 // of benchmark iterations until the benchmark runs for the requested benchtime.
257 // launch is run by the doBench function as a separate goroutine.
258 // run1 must have been called on b.
259 func (b *B) launch() {
260 // Signal that we're done whether we return normally
261 // or by FailNow's runtime.Goexit.
266 // Run the benchmark for at least the specified amount of time.
268 for n := 1; !b.failed && b.duration < d && n < 1e9; {
270 // Predict required iterations.
271 n = int(d.Nanoseconds())
272 if nsop := b.nsPerOp(); nsop != 0 {
275 // Run more iterations than we think we'll need (1.2x).
276 // Don't grow too fast in case we had timing errors previously.
277 // Be sure to run at least one more than last time.
278 n = max(min(n+n/5, 100*last), last+1)
279 // Round up to something easy to read.
283 b.result = BenchmarkResult{b.N, b.duration, b.bytes, b.netAllocs, b.netBytes}
286 // The results of a benchmark run.
287 type BenchmarkResult struct {
288 N int // The number of iterations.
289 T time.Duration // The total time taken.
290 Bytes int64 // Bytes processed in one iteration.
291 MemAllocs uint64 // The total number of memory allocations.
292 MemBytes uint64 // The total number of bytes allocated.
295 func (r BenchmarkResult) NsPerOp() int64 {
299 return r.T.Nanoseconds() / int64(r.N)
302 func (r BenchmarkResult) mbPerSec() float64 {
303 if r.Bytes <= 0 || r.T <= 0 || r.N <= 0 {
306 return (float64(r.Bytes) * float64(r.N) / 1e6) / r.T.Seconds()
309 func (r BenchmarkResult) AllocsPerOp() int64 {
313 return int64(r.MemAllocs) / int64(r.N)
316 func (r BenchmarkResult) AllocedBytesPerOp() int64 {
320 return int64(r.MemBytes) / int64(r.N)
323 func (r BenchmarkResult) String() string {
327 mb = fmt.Sprintf("\t%7.2f MB/s", mbs)
330 ns := fmt.Sprintf("%10d ns/op", nsop)
331 if r.N > 0 && nsop < 100 {
332 // The format specifiers here make sure that
333 // the ones digits line up for all three possible formats.
335 ns = fmt.Sprintf("%13.2f ns/op", float64(r.T.Nanoseconds())/float64(r.N))
337 ns = fmt.Sprintf("%12.1f ns/op", float64(r.T.Nanoseconds())/float64(r.N))
340 return fmt.Sprintf("%8d\t%s%s", r.N, ns, mb)
343 func (r BenchmarkResult) MemString() string {
344 return fmt.Sprintf("%8d B/op\t%8d allocs/op",
345 r.AllocedBytesPerOp(), r.AllocsPerOp())
348 // benchmarkName returns full name of benchmark including procs suffix.
349 func benchmarkName(name string, n int) string {
351 return fmt.Sprintf("%s-%d", name, n)
356 type benchContext struct {
359 maxLen int // The largest recorded benchmark name.
360 extLen int // Maximum extension length.
363 // An internal function but exported because it is cross-package; part of the implementation
364 // of the "go test" command.
365 func RunBenchmarks(matchString func(pat, str string) (bool, error), benchmarks []InternalBenchmark) {
366 runBenchmarks(matchString, benchmarks)
369 func runBenchmarks(matchString func(pat, str string) (bool, error), benchmarks []InternalBenchmark) bool {
370 // If no flag was specified, don't run benchmarks.
371 if len(*matchBenchmarks) == 0 {
374 // Collect matching benchmarks and determine longest name.
376 for _, procs := range cpuList {
377 if procs > maxprocs {
381 ctx := &benchContext{
382 match: newMatcher(matchString, *matchBenchmarks, "-test.bench"),
383 extLen: len(benchmarkName("", maxprocs)),
385 var bs []InternalBenchmark
386 for _, Benchmark := range benchmarks {
387 if _, matched := ctx.match.fullName(nil, Benchmark.Name); matched {
388 bs = append(bs, Benchmark)
389 benchName := benchmarkName(Benchmark.Name, maxprocs)
390 if l := len(benchName) + ctx.extLen + 1; l > ctx.maxLen {
401 benchFunc: func(b *B) {
402 for _, Benchmark := range bs {
403 b.Run(Benchmark.Name, Benchmark.F)
406 benchTime: *benchTime,
413 // processBench runs bench b for the configured CPU counts and prints the results.
414 func (ctx *benchContext) processBench(b *B) {
415 for i, procs := range cpuList {
416 runtime.GOMAXPROCS(procs)
417 benchName := benchmarkName(b.name, procs)
418 fmt.Fprintf(b.w, "%-*s\t", ctx.maxLen, benchName)
419 // Recompute the running time for all but the first iteration.
423 signal: make(chan bool),
428 benchFunc: b.benchFunc,
429 benchTime: b.benchTime,
435 // The output could be very long here, but probably isn't.
436 // We print it all, regardless, because we don't want to trim the reason
437 // the benchmark failed.
438 fmt.Fprintf(b.w, "--- FAIL: %s\n%s", benchName, b.output)
441 results := r.String()
442 if *benchmarkMemory || b.showAllocResult {
443 results += "\t" + r.MemString()
445 fmt.Fprintln(b.w, results)
446 // Unlike with tests, we ignore the -chatty flag and always print output for
447 // benchmarks since the output generation time will skew the results.
448 if len(b.output) > 0 {
450 fmt.Fprintf(b.w, "--- BENCH: %s\n%s", benchName, b.output)
452 if p := runtime.GOMAXPROCS(-1); p != procs {
453 fmt.Fprintf(os.Stderr, "testing: %s left GOMAXPROCS set to %d\n", benchName, p)
458 // Run benchmarks f as a subbenchmark with the given name. It reports
459 // whether there were any failures.
461 // A subbenchmark is like any other benchmark. A benchmark that calls Run at
462 // least once will not be measured itself and will be called once with N=1.
464 // Run may be called simultaneously from multiple goroutines, but all such
465 // calls must happen before the outer benchmark function for b returns.
466 func (b *B) Run(name string, f func(b *B)) bool {
467 // Since b has subbenchmarks, we will no longer run it as a benchmark itself.
468 // Release the lock and acquire it on exit to ensure locks stay paired.
469 atomic.StoreInt32(&b.hasSub, 1)
470 benchmarkLock.Unlock()
471 defer benchmarkLock.Lock()
473 benchName, ok := b.name, true
474 if b.context != nil {
475 benchName, ok = b.context.match.fullName(&b.common, name)
482 signal: make(chan bool),
490 benchTime: b.benchTime,
500 // add simulates running benchmarks in sequence in a single iteration. It is
501 // used to give some meaningful results in case func Benchmark is used in
502 // combination with Run.
503 func (b *B) add(other BenchmarkResult) {
505 // The aggregated BenchmarkResults resemble running all subbenchmarks as
506 // in sequence in a single benchmark.
508 r.T += time.Duration(other.NsPerOp())
509 if other.Bytes == 0 {
510 // Summing Bytes is meaningless in aggregate if not all subbenchmarks
512 b.missingBytes = true
516 r.Bytes += other.Bytes
518 r.MemAllocs += uint64(other.AllocsPerOp())
519 r.MemBytes += uint64(other.AllocedBytesPerOp())
522 // trimOutput shortens the output from a benchmark, which can be very long.
523 func (b *B) trimOutput() {
524 // The output is likely to appear multiple times because the benchmark
525 // is run multiple times, but at least it will be seen. This is not a big deal
526 // because benchmarks rarely print, but just in case, we trim it if it's too long.
527 const maxNewlines = 10
528 for nlCount, j := 0, 0; j < len(b.output); j++ {
529 if b.output[j] == '\n' {
531 if nlCount >= maxNewlines {
532 b.output = append(b.output[:j], "\n\t... [output truncated]\n"...)
539 // A PB is used by RunParallel for running parallel benchmarks.
541 globalN *uint64 // shared between all worker goroutines iteration counter
542 grain uint64 // acquire that many iterations from globalN at once
543 cache uint64 // local cache of acquired iterations
544 bN uint64 // total number of iterations to execute (b.N)
547 // Next reports whether there are more iterations to execute.
548 func (pb *PB) Next() bool {
550 n := atomic.AddUint64(pb.globalN, pb.grain)
553 } else if n < pb.bN+pb.grain {
554 pb.cache = pb.bN + pb.grain - n
563 // RunParallel runs a benchmark in parallel.
564 // It creates multiple goroutines and distributes b.N iterations among them.
565 // The number of goroutines defaults to GOMAXPROCS. To increase parallelism for
566 // non-CPU-bound benchmarks, call SetParallelism before RunParallel.
567 // RunParallel is usually used with the go test -cpu flag.
569 // The body function will be run in each goroutine. It should set up any
570 // goroutine-local state and then iterate until pb.Next returns false.
571 // It should not use the StartTimer, StopTimer, or ResetTimer functions,
572 // because they have global effect. It should also not call Run.
573 func (b *B) RunParallel(body func(*PB)) {
575 return // Nothing to do when probing.
577 // Calculate grain size as number of iterations that take ~100µs.
578 // 100µs is enough to amortize the overhead and provide sufficient
579 // dynamic load balancing.
581 if b.previousN > 0 && b.previousDuration > 0 {
582 grain = 1e5 * uint64(b.previousN) / uint64(b.previousDuration)
587 // We expect the inner loop and function call to take at least 10ns,
588 // so do not do more than 100µs/10ns=1e4 iterations.
594 numProcs := b.parallelism * runtime.GOMAXPROCS(0)
595 var wg sync.WaitGroup
597 for p := 0; p < numProcs; p++ {
609 if n <= uint64(b.N) && !b.Failed() {
610 b.Fatal("RunParallel: body exited without pb.Next() == false")
614 // SetParallelism sets the number of goroutines used by RunParallel to p*GOMAXPROCS.
615 // There is usually no need to call SetParallelism for CPU-bound benchmarks.
616 // If p is less than 1, this call will have no effect.
617 func (b *B) SetParallelism(p int) {
623 // Benchmark benchmarks a single function. Useful for creating
624 // custom benchmarks that do not use the "go test" command.
626 // If f calls Run, the result will be an estimate of running all its
627 // subbenchmarks that don't call Run in sequence in a single benchmark.
628 func Benchmark(f func(b *B)) BenchmarkResult {
631 signal: make(chan bool),
635 benchTime: *benchTime,
638 return BenchmarkResult{}
643 type discard struct{}
645 func (discard) Write(b []byte) (n int, err error) { return len(b), nil }