Imported Upstream version 1.3upstream/1.3

1 files changed, 95 insertions, 7 deletions

diff --git a/src/pkg/testing/benchmark.go b/src/pkg/testing/benchmark.go
index 3473c5b2c..1fbf5c861 100644
--- a/src/pkg/testing/benchmark.go
+++ b/src/pkg/testing/benchmark.go
@@ -10,6 +10,7 @@ import (
 	"os"
 	"runtime"
 	"sync"
+	"sync/atomic"
 	"time"
 )
 
@@ -34,12 +35,15 @@ type InternalBenchmark struct {
 // timing and to specify the number of iterations to run.
 type B struct {
 	common
-	N               int
-	benchmark       InternalBenchmark
-	bytes           int64
-	timerOn         bool
-	showAllocResult bool
-	result          BenchmarkResult
+	N                int
+	previousN        int           // number of iterations in the previous run
+	previousDuration time.Duration // total duration of the previous run
+	benchmark        InternalBenchmark
+	bytes            int64
+	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
@@ -74,7 +78,7 @@ func (b *B) StopTimer() {
 	}
 }
 
-// ResetTimer sets the elapsed benchmark time to zero.
+// ResetTimer zeros the elapsed benchmark time and memory allocation counters.
 // It does not affect whether the timer is running.
 func (b *B) ResetTimer() {
 	if b.timerOn {
@@ -114,10 +118,13 @@ func (b *B) runN(n int) {
 	// by clearing garbage from previous runs.
 	runtime.GC()
 	b.N = n
+	b.parallelism = 1
 	b.ResetTimer()
 	b.StartTimer()
 	b.benchmark.F(b)
 	b.StopTimer()
+	b.previousN = n
+	b.previousDuration = b.duration
 }
 
 func min(x, y int) int {
@@ -343,6 +350,87 @@ func (b *B) trimOutput() {
 	}
 }
 
+// A PB is used by RunParallel for running parallel benchmarks.
+type PB struct {
+	globalN *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 *PB) Next() bool {
+	if pb.cache == 0 {
+		n := atomic.AddUint64(pb.globalN, pb.grain)
+		if n <= pb.bN {
+			pb.cache = pb.grain
+		} else if n < pb.bN+pb.grain {
+			pb.cache = pb.bN + pb.grain - n
+		} else {
+			return false
+		}
+	}
+	pb.cache--
+	return true
+}
+
+// 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 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 StartTimer, StopTimer, or ResetTimer functions,
+// because they have global effect.
+func (b *B) RunParallel(body func(*PB)) {
+	// Calculate grain size as number of iterations that take ~100µs.
+	// 100µs is enough to amortize the overhead and provide sufficient
+	// dynamic load balancing.
+	grain := uint64(0)
+	if b.previousN > 0 && b.previousDuration > 0 {
+		grain = 1e5 * uint64(b.previousN) / uint64(b.previousDuration)
+	}
+	if grain < 1 {
+		grain = 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 grain > 1e4 {
+		grain = 1e4
+	}
+
+	n := uint64(0)
+	numProcs := b.parallelism * runtime.GOMAXPROCS(0)
+	var wg sync.WaitGroup
+	wg.Add(numProcs)
+	for p := 0; p < numProcs; p++ {
+		go func() {
+			defer wg.Done()
+			pb := &PB{
+				globalN: &n,
+				grain:   grain,
+				bN:      uint64(b.N),
+			}
+			body(pb)
+		}()
+	}
+	wg.Wait()
+	if n <= uint64(b.N) && !b.Failed() {
+		b.Fatal("RunParallel: body exited without pb.Next() == false")
+	}
+}
+
+// SetParallelism sets the number of goroutines used by 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 *B) SetParallelism(p int) {
+	if p >= 1 {
+		b.parallelism = p
+	}
+}
+
 // Benchmark benchmarks a single function. Useful for creating
 // custom benchmarks that do not use the "go test" command.
 func Benchmark(f func(b *B)) BenchmarkResult {