benchmark.go

Documentation: testing

     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.
     4  
     5  package testing
     6  
     7  import (
     8  	"flag"
     9  	"fmt"
    10  	"internal/race"
    11  	"os"
    12  	"runtime"
    13  	"sync"
    14  	"sync/atomic"
    15  	"time"
    16  )
    17  
    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")
    21  
    22  // Global lock to ensure only one benchmark runs at a time.
    23  var benchmarkLock sync.Mutex
    24  
    25  // Used for every benchmark for measuring memory.
    26  var memStats runtime.MemStats
    27  
    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 {
    31  	Name string
    32  	F    func(b *B)
    33  }
    34  
    35  // B is a type passed to Benchmark functions to manage benchmark
    36  // timing and to specify the number of iterations to run.
    37  //
    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.
    43  //
    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.
    48  type B struct {
    49  	common
    50  	importPath       string // import path of the package containing the benchmark
    51  	context          *benchContext
    52  	N                int
    53  	previousN        int           // number of iterations in the previous run
    54  	previousDuration time.Duration // total duration of the previous run
    55  	benchFunc        func(b *B)
    56  	benchTime        time.Duration
    57  	bytes            int64
    58  	missingBytes     bool // one of the subbenchmarks does not have bytes set.
    59  	timerOn          bool
    60  	showAllocResult  bool
    61  	result           BenchmarkResult
    62  	parallelism      int // RunParallel creates parallelism*GOMAXPROCS goroutines
    63  	// The initial states of memStats.Mallocs and memStats.TotalAlloc.
    64  	startAllocs uint64
    65  	startBytes  uint64
    66  	// The net total of this test after being run.
    67  	netAllocs uint64
    68  	netBytes  uint64
    69  }
    70  
    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() {
    75  	if !b.timerOn {
    76  		runtime.ReadMemStats(&memStats)
    77  		b.startAllocs = memStats.Mallocs
    78  		b.startBytes = memStats.TotalAlloc
    79  		b.start = time.Now()
    80  		b.timerOn = true
    81  	}
    82  }
    83  
    84  // StopTimer stops timing a test. This can be used to pause the timer
    85  // while performing complex initialization that you don't
    86  // want to measure.
    87  func (b *B) StopTimer() {
    88  	if b.timerOn {
    89  		b.duration += time.Since(b.start)
    90  		runtime.ReadMemStats(&memStats)
    91  		b.netAllocs += memStats.Mallocs - b.startAllocs
    92  		b.netBytes += memStats.TotalAlloc - b.startBytes
    93  		b.timerOn = false
    94  	}
    95  }
    96  
    97  // ResetTimer zeros the elapsed benchmark time and memory allocation counters.
    98  // It does not affect whether the timer is running.
    99  func (b *B) ResetTimer() {
   100  	if b.timerOn {
   101  		runtime.ReadMemStats(&memStats)
   102  		b.startAllocs = memStats.Mallocs
   103  		b.startBytes = memStats.TotalAlloc
   104  		b.start = time.Now()
   105  	}
   106  	b.duration = 0
   107  	b.netAllocs = 0
   108  	b.netBytes = 0
   109  }
   110  
   111  // SetBytes records the number of bytes processed in a single operation.
   112  // If this is called, the benchmark will report ns/op and MB/s.
   113  func (b *B) SetBytes(n int64) { b.bytes = n }
   114  
   115  // ReportAllocs enables malloc statistics for this benchmark.
   116  // It is equivalent to setting -test.benchmem, but it only affects the
   117  // benchmark function that calls ReportAllocs.
   118  func (b *B) ReportAllocs() {
   119  	b.showAllocResult = true
   120  }
   121  
   122  func (b *B) nsPerOp() int64 {
   123  	if b.N <= 0 {
   124  		return 0
   125  	}
   126  	return b.duration.Nanoseconds() / int64(b.N)
   127  }
   128  
   129  // runN runs a single benchmark for the specified number of iterations.
   130  func (b *B) runN(n int) {
   131  	benchmarkLock.Lock()
   132  	defer benchmarkLock.Unlock()
   133  	// Try to get a comparable environment for each run
   134  	// by clearing garbage from previous runs.
   135  	runtime.GC()
   136  	b.raceErrors = -race.Errors()
   137  	b.N = n
   138  	b.parallelism = 1
   139  	b.ResetTimer()
   140  	b.StartTimer()
   141  	b.benchFunc(b)
   142  	b.StopTimer()
   143  	b.previousN = n
   144  	b.previousDuration = b.duration
   145  	b.raceErrors += race.Errors()
   146  	if b.raceErrors > 0 {
   147  		b.Errorf("race detected during execution of benchmark")
   148  	}
   149  }
   150  
   151  func min(x, y int) int {
   152  	if x > y {
   153  		return y
   154  	}
   155  	return x
   156  }
   157  
   158  func max(x, y int) int {
   159  	if x < y {
   160  		return y
   161  	}
   162  	return x
   163  }
   164  
   165  // roundDown10 rounds a number down to the nearest power of 10.
   166  func roundDown10(n int) int {
   167  	var tens = 0
   168  	// tens = floor(log_10(n))
   169  	for n >= 10 {
   170  		n = n / 10
   171  		tens++
   172  	}
   173  	// result = 10^tens
   174  	result := 1
   175  	for i := 0; i < tens; i++ {
   176  		result *= 10
   177  	}
   178  	return result
   179  }
   180  
   181  // roundUp rounds x up to a number of the form [1eX, 2eX, 3eX, 5eX].
   182  func roundUp(n int) int {
   183  	base := roundDown10(n)
   184  	switch {
   185  	case n <= base:
   186  		return base
   187  	case n <= (2 * base):
   188  		return 2 * base
   189  	case n <= (3 * base):
   190  		return 3 * base
   191  	case n <= (5 * base):
   192  		return 5 * base
   193  	default:
   194  		return 10 * base
   195  	}
   196  }
   197  
   198  // run1 runs the first iteration of benchFunc. It returns whether more
   199  // iterations of this benchmarks should be run.
   200  func (b *B) run1() bool {
   201  	if ctx := b.context; ctx != nil {
   202  		// Extend maxLen, if needed.
   203  		if n := len(b.name) + ctx.extLen + 1; n > ctx.maxLen {
   204  			ctx.maxLen = n + 8 // Add additional slack to avoid too many jumps in size.
   205  		}
   206  	}
   207  	go func() {
   208  		// Signal that we're done whether we return normally
   209  		// or by FailNow's runtime.Goexit.
   210  		defer func() {
   211  			b.signal <- true
   212  		}()
   213  
   214  		b.runN(1)
   215  	}()
   216  	<-b.signal
   217  	if b.failed {
   218  		fmt.Fprintf(b.w, "--- FAIL: %s\n%s", b.name, b.output)
   219  		return false
   220  	}
   221  	// Only print the output if we know we are not going to proceed.
   222  	// Otherwise it is printed in processBench.
   223  	if atomic.LoadInt32(&b.hasSub) != 0 || b.finished {
   224  		tag := "BENCH"
   225  		if b.skipped {
   226  			tag = "SKIP"
   227  		}
   228  		if b.chatty && (len(b.output) > 0 || b.finished) {
   229  			b.trimOutput()
   230  			fmt.Fprintf(b.w, "--- %s: %s\n%s", tag, b.name, b.output)
   231  		}
   232  		return false
   233  	}
   234  	return true
   235  }
   236  
   237  var labelsOnce sync.Once
   238  
   239  // run executes the benchmark in a separate goroutine, including all of its
   240  // subbenchmarks. b must not have subbenchmarks.
   241  func (b *B) run() {
   242  	labelsOnce.Do(func() {
   243  		fmt.Fprintf(b.w, "goos: %s\n", runtime.GOOS)
   244  		fmt.Fprintf(b.w, "goarch: %s\n", runtime.GOARCH)
   245  		if b.importPath != "" {
   246  			fmt.Fprintf(b.w, "pkg: %s\n", b.importPath)
   247  		}
   248  	})
   249  	if b.context != nil {
   250  		// Running go test --test.bench
   251  		b.context.processBench(b) // Must call doBench.
   252  	} else {
   253  		// Running func Benchmark.
   254  		b.doBench()
   255  	}
   256  }
   257  
   258  func (b *B) doBench() BenchmarkResult {
   259  	go b.launch()
   260  	<-b.signal
   261  	return b.result
   262  }
   263  
   264  // launch launches the benchmark function. It gradually increases the number
   265  // of benchmark iterations until the benchmark runs for the requested benchtime.
   266  // launch is run by the doBench function as a separate goroutine.
   267  // run1 must have been called on b.
   268  func (b *B) launch() {
   269  	// Signal that we're done whether we return normally
   270  	// or by FailNow's runtime.Goexit.
   271  	defer func() {
   272  		b.signal <- true
   273  	}()
   274  
   275  	// Run the benchmark for at least the specified amount of time.
   276  	d := b.benchTime
   277  	for n := 1; !b.failed && b.duration < d && n < 1e9; {
   278  		last := n
   279  		// Predict required iterations.
   280  		n = int(d.Nanoseconds())
   281  		if nsop := b.nsPerOp(); nsop != 0 {
   282  			n /= int(nsop)
   283  		}
   284  		// Run more iterations than we think we'll need (1.2x).
   285  		// Don't grow too fast in case we had timing errors previously.
   286  		// Be sure to run at least one more than last time.
   287  		n = max(min(n+n/5, 100*last), last+1)
   288  		// Round up to something easy to read.
   289  		n = roundUp(n)
   290  		b.runN(n)
   291  	}
   292  	b.result = BenchmarkResult{b.N, b.duration, b.bytes, b.netAllocs, b.netBytes}
   293  }
   294  
   295  // The results of a benchmark run.
   296  type BenchmarkResult struct {
   297  	N         int           // The number of iterations.
   298  	T         time.Duration // The total time taken.
   299  	Bytes     int64         // Bytes processed in one iteration.
   300  	MemAllocs uint64        // The total number of memory allocations.
   301  	MemBytes  uint64        // The total number of bytes allocated.
   302  }
   303  
   304  func (r BenchmarkResult) NsPerOp() int64 {
   305  	if r.N <= 0 {
   306  		return 0
   307  	}
   308  	return r.T.Nanoseconds() / int64(r.N)
   309  }
   310  
   311  func (r BenchmarkResult) mbPerSec() float64 {
   312  	if r.Bytes <= 0 || r.T <= 0 || r.N <= 0 {
   313  		return 0
   314  	}
   315  	return (float64(r.Bytes) * float64(r.N) / 1e6) / r.T.Seconds()
   316  }
   317  
   318  // AllocsPerOp returns r.MemAllocs / r.N.
   319  func (r BenchmarkResult) AllocsPerOp() int64 {
   320  	if r.N <= 0 {
   321  		return 0
   322  	}
   323  	return int64(r.MemAllocs) / int64(r.N)
   324  }
   325  
   326  // AllocedBytesPerOp returns r.MemBytes / r.N.
   327  func (r BenchmarkResult) AllocedBytesPerOp() int64 {
   328  	if r.N <= 0 {
   329  		return 0
   330  	}
   331  	return int64(r.MemBytes) / int64(r.N)
   332  }
   333  
   334  func (r BenchmarkResult) String() string {
   335  	mbs := r.mbPerSec()
   336  	mb := ""
   337  	if mbs != 0 {
   338  		mb = fmt.Sprintf("\t%7.2f MB/s", mbs)
   339  	}
   340  	nsop := r.NsPerOp()
   341  	ns := fmt.Sprintf("%10d ns/op", nsop)
   342  	if r.N > 0 && nsop < 100 {
   343  		// The format specifiers here make sure that
   344  		// the ones digits line up for all three possible formats.
   345  		if nsop < 10 {
   346  			ns = fmt.Sprintf("%13.2f ns/op", float64(r.T.Nanoseconds())/float64(r.N))
   347  		} else {
   348  			ns = fmt.Sprintf("%12.1f ns/op", float64(r.T.Nanoseconds())/float64(r.N))
   349  		}
   350  	}
   351  	return fmt.Sprintf("%8d\t%s%s", r.N, ns, mb)
   352  }
   353  
   354  // MemString returns r.AllocedBytesPerOp and r.AllocsPerOp in the same format as 'go test'.
   355  func (r BenchmarkResult) MemString() string {
   356  	return fmt.Sprintf("%8d B/op\t%8d allocs/op",
   357  		r.AllocedBytesPerOp(), r.AllocsPerOp())
   358  }
   359  
   360  // benchmarkName returns full name of benchmark including procs suffix.
   361  func benchmarkName(name string, n int) string {
   362  	if n != 1 {
   363  		return fmt.Sprintf("%s-%d", name, n)
   364  	}
   365  	return name
   366  }
   367  
   368  type benchContext struct {
   369  	match *matcher
   370  
   371  	maxLen int // The largest recorded benchmark name.
   372  	extLen int // Maximum extension length.
   373  }
   374  
   375  // An internal function but exported because it is cross-package; part of the implementation
   376  // of the "go test" command.
   377  func RunBenchmarks(matchString func(pat, str string) (bool, error), benchmarks []InternalBenchmark) {
   378  	runBenchmarks("", matchString, benchmarks)
   379  }
   380  
   381  func runBenchmarks(importPath string, matchString func(pat, str string) (bool, error), benchmarks []InternalBenchmark) bool {
   382  	// If no flag was specified, don't run benchmarks.
   383  	if len(*matchBenchmarks) == 0 {
   384  		return true
   385  	}
   386  	// Collect matching benchmarks and determine longest name.
   387  	maxprocs := 1
   388  	for _, procs := range cpuList {
   389  		if procs > maxprocs {
   390  			maxprocs = procs
   391  		}
   392  	}
   393  	ctx := &benchContext{
   394  		match:  newMatcher(matchString, *matchBenchmarks, "-test.bench"),
   395  		extLen: len(benchmarkName("", maxprocs)),
   396  	}
   397  	var bs []InternalBenchmark
   398  	for _, Benchmark := range benchmarks {
   399  		if _, matched, _ := ctx.match.fullName(nil, Benchmark.Name); matched {
   400  			bs = append(bs, Benchmark)
   401  			benchName := benchmarkName(Benchmark.Name, maxprocs)
   402  			if l := len(benchName) + ctx.extLen + 1; l > ctx.maxLen {
   403  				ctx.maxLen = l
   404  			}
   405  		}
   406  	}
   407  	main := &B{
   408  		common: common{
   409  			name:   "Main",
   410  			w:      os.Stdout,
   411  			chatty: *chatty,
   412  		},
   413  		importPath: importPath,
   414  		benchFunc: func(b *B) {
   415  			for _, Benchmark := range bs {
   416  				b.Run(Benchmark.Name, Benchmark.F)
   417  			}
   418  		},
   419  		benchTime: *benchTime,
   420  		context:   ctx,
   421  	}
   422  	main.runN(1)
   423  	return !main.failed
   424  }
   425  
   426  // processBench runs bench b for the configured CPU counts and prints the results.
   427  func (ctx *benchContext) processBench(b *B) {
   428  	for i, procs := range cpuList {
   429  		for j := uint(0); j < *count; j++ {
   430  			runtime.GOMAXPROCS(procs)
   431  			benchName := benchmarkName(b.name, procs)
   432  			fmt.Fprintf(b.w, "%-*s\t", ctx.maxLen, benchName)
   433  			// Recompute the running time for all but the first iteration.
   434  			if i > 0 || j > 0 {
   435  				b = &B{
   436  					common: common{
   437  						signal: make(chan bool),
   438  						name:   b.name,
   439  						w:      b.w,
   440  						chatty: b.chatty,
   441  					},
   442  					benchFunc: b.benchFunc,
   443  					benchTime: b.benchTime,
   444  				}
   445  				b.run1()
   446  			}
   447  			r := b.doBench()
   448  			if b.failed {
   449  				// The output could be very long here, but probably isn't.
   450  				// We print it all, regardless, because we don't want to trim the reason
   451  				// the benchmark failed.
   452  				fmt.Fprintf(b.w, "--- FAIL: %s\n%s", benchName, b.output)
   453  				continue
   454  			}
   455  			results := r.String()
   456  			if *benchmarkMemory || b.showAllocResult {
   457  				results += "\t" + r.MemString()
   458  			}
   459  			fmt.Fprintln(b.w, results)
   460  			// Unlike with tests, we ignore the -chatty flag and always print output for
   461  			// benchmarks since the output generation time will skew the results.
   462  			if len(b.output) > 0 {
   463  				b.trimOutput()
   464  				fmt.Fprintf(b.w, "--- BENCH: %s\n%s", benchName, b.output)
   465  			}
   466  			if p := runtime.GOMAXPROCS(-1); p != procs {
   467  				fmt.Fprintf(os.Stderr, "testing: %s left GOMAXPROCS set to %d\n", benchName, p)
   468  			}
   469  		}
   470  	}
   471  }
   472  
   473  // Run benchmarks f as a subbenchmark with the given name. It reports
   474  // whether there were any failures.
   475  //
   476  // A subbenchmark is like any other benchmark. A benchmark that calls Run at
   477  // least once will not be measured itself and will be called once with N=1.
   478  func (b *B) Run(name string, f func(b *B)) bool {
   479  	// Since b has subbenchmarks, we will no longer run it as a benchmark itself.
   480  	// Release the lock and acquire it on exit to ensure locks stay paired.
   481  	atomic.StoreInt32(&b.hasSub, 1)
   482  	benchmarkLock.Unlock()
   483  	defer benchmarkLock.Lock()
   484  
   485  	benchName, ok, partial := b.name, true, false
   486  	if b.context != nil {
   487  		benchName, ok, partial = b.context.match.fullName(&b.common, name)
   488  	}
   489  	if !ok {
   490  		return true
   491  	}
   492  	var pc [maxStackLen]uintptr
   493  	n := runtime.Callers(2, pc[:])
   494  	sub := &B{
   495  		common: common{
   496  			signal:  make(chan bool),
   497  			name:    benchName,
   498  			parent:  &b.common,
   499  			level:   b.level + 1,
   500  			creator: pc[:n],
   501  			w:       b.w,
   502  			chatty:  b.chatty,
   503  		},
   504  		importPath: b.importPath,
   505  		benchFunc:  f,
   506  		benchTime:  b.benchTime,
   507  		context:    b.context,
   508  	}
   509  	if partial {
   510  		// Partial name match, like -bench=X/Y matching BenchmarkX.
   511  		// Only process sub-benchmarks, if any.
   512  		atomic.StoreInt32(&sub.hasSub, 1)
   513  	}
   514  	if sub.run1() {
   515  		sub.run()
   516  	}
   517  	b.add(sub.result)
   518  	return !sub.failed
   519  }
   520  
   521  // add simulates running benchmarks in sequence in a single iteration. It is
   522  // used to give some meaningful results in case func Benchmark is used in
   523  // combination with Run.
   524  func (b *B) add(other BenchmarkResult) {
   525  	r := &b.result
   526  	// The aggregated BenchmarkResults resemble running all subbenchmarks as
   527  	// in sequence in a single benchmark.
   528  	r.N = 1
   529  	r.T += time.Duration(other.NsPerOp())
   530  	if other.Bytes == 0 {
   531  		// Summing Bytes is meaningless in aggregate if not all subbenchmarks
   532  		// set it.
   533  		b.missingBytes = true
   534  		r.Bytes = 0
   535  	}
   536  	if !b.missingBytes {
   537  		r.Bytes += other.Bytes
   538  	}
   539  	r.MemAllocs += uint64(other.AllocsPerOp())
   540  	r.MemBytes += uint64(other.AllocedBytesPerOp())
   541  }
   542  
   543  // trimOutput shortens the output from a benchmark, which can be very long.
   544  func (b *B) trimOutput() {
   545  	// The output is likely to appear multiple times because the benchmark
   546  	// is run multiple times, but at least it will be seen. This is not a big deal
   547  	// because benchmarks rarely print, but just in case, we trim it if it's too long.
   548  	const maxNewlines = 10
   549  	for nlCount, j := 0, 0; j < len(b.output); j++ {
   550  		if b.output[j] == '\n' {
   551  			nlCount++
   552  			if nlCount >= maxNewlines {
   553  				b.output = append(b.output[:j], "\n\t... [output truncated]\n"...)
   554  				break
   555  			}
   556  		}
   557  	}
   558  }
   559  
   560  // A PB is used by RunParallel for running parallel benchmarks.
   561  type PB struct {
   562  	globalN *uint64 // shared between all worker goroutines iteration counter
   563  	grain   uint64  // acquire that many iterations from globalN at once
   564  	cache   uint64  // local cache of acquired iterations
   565  	bN      uint64  // total number of iterations to execute (b.N)
   566  }
   567  
   568  // Next reports whether there are more iterations to execute.
   569  func (pb *PB) Next() bool {
   570  	if pb.cache == 0 {
   571  		n := atomic.AddUint64(pb.globalN, pb.grain)
   572  		if n <= pb.bN {
   573  			pb.cache = pb.grain
   574  		} else if n < pb.bN+pb.grain {
   575  			pb.cache = pb.bN + pb.grain - n
   576  		} else {
   577  			return false
   578  		}
   579  	}
   580  	pb.cache--
   581  	return true
   582  }
   583  
   584  // RunParallel runs a benchmark in parallel.
   585  // It creates multiple goroutines and distributes b.N iterations among them.
   586  // The number of goroutines defaults to GOMAXPROCS. To increase parallelism for
   587  // non-CPU-bound benchmarks, call SetParallelism before RunParallel.
   588  // RunParallel is usually used with the go test -cpu flag.
   589  //
   590  // The body function will be run in each goroutine. It should set up any
   591  // goroutine-local state and then iterate until pb.Next returns false.
   592  // It should not use the StartTimer, StopTimer, or ResetTimer functions,
   593  // because they have global effect. It should also not call Run.
   594  func (b *B) RunParallel(body func(*PB)) {
   595  	if b.N == 0 {
   596  		return // Nothing to do when probing.
   597  	}
   598  	// Calculate grain size as number of iterations that take ~100µs.
   599  	// 100µs is enough to amortize the overhead and provide sufficient
   600  	// dynamic load balancing.
   601  	grain := uint64(0)
   602  	if b.previousN > 0 && b.previousDuration > 0 {
   603  		grain = 1e5 * uint64(b.previousN) / uint64(b.previousDuration)
   604  	}
   605  	if grain < 1 {
   606  		grain = 1
   607  	}
   608  	// We expect the inner loop and function call to take at least 10ns,
   609  	// so do not do more than 100µs/10ns=1e4 iterations.
   610  	if grain > 1e4 {
   611  		grain = 1e4
   612  	}
   613  
   614  	n := uint64(0)
   615  	numProcs := b.parallelism * runtime.GOMAXPROCS(0)
   616  	var wg sync.WaitGroup
   617  	wg.Add(numProcs)
   618  	for p := 0; p < numProcs; p++ {
   619  		go func() {
   620  			defer wg.Done()
   621  			pb := &PB{
   622  				globalN: &n,
   623  				grain:   grain,
   624  				bN:      uint64(b.N),
   625  			}
   626  			body(pb)
   627  		}()
   628  	}
   629  	wg.Wait()
   630  	if n <= uint64(b.N) && !b.Failed() {
   631  		b.Fatal("RunParallel: body exited without pb.Next() == false")
   632  	}
   633  }
   634  
   635  // SetParallelism sets the number of goroutines used by RunParallel to p*GOMAXPROCS.
   636  // There is usually no need to call SetParallelism for CPU-bound benchmarks.
   637  // If p is less than 1, this call will have no effect.
   638  func (b *B) SetParallelism(p int) {
   639  	if p >= 1 {
   640  		b.parallelism = p
   641  	}
   642  }
   643  
   644  // Benchmark benchmarks a single function. Useful for creating
   645  // custom benchmarks that do not use the "go test" command.
   646  //
   647  // If f calls Run, the result will be an estimate of running all its
   648  // subbenchmarks that don't call Run in sequence in a single benchmark.
   649  func Benchmark(f func(b *B)) BenchmarkResult {
   650  	b := &B{
   651  		common: common{
   652  			signal: make(chan bool),
   653  			w:      discard{},
   654  		},
   655  		benchFunc: f,
   656  		benchTime: *benchTime,
   657  	}
   658  	if b.run1() {
   659  		b.run()
   660  	}
   661  	return b.result
   662  }
   663  
   664  type discard struct{}
   665  
   666  func (discard) Write(b []byte) (n int, err error) { return len(b), nil }
   667
View as plain text