testing: enable better loop time measurement for benchmarking.

With b.Loop() in place, the time measurement of loop scaling could be improved to be tighter. By identifying the first call to b.Loop(), we can avoid measuring the expensive ramp-up time by reset the timer tightly before the loop starts. The remaining loop scaling logic of b.N style loop is largely reused.

For #61515.

Change-Id: Ia7b8f0a8838f57c00ac6c5ef779d86f8d713c9b6
Reviewed-on: https://go-review.googlesource.com/c/go/+/612835
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: Junyang Shao <shaojunyang@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

1 files changed, 6 insertions, 0 deletions

diff --git a/src/testing/benchmark.go b/src/testing/benchmark.go
index 0271308346..2c7083db02 100644
--- a/src/testing/benchmark.go
+++ b/src/testing/benchmark.go
@@ -114,6 +114,7 @@ type B struct {
 	// Extra metrics collected by ReportMetric.
 	extra map[string]float64
 	// Remaining iterations of Loop() to be executed in benchFunc.
+	// See issue #61515.
 	loopN int
 }
 
@@ -358,6 +359,11 @@ func (b *B) ReportMetric(n float64, unit string) {
 // After the benchmark finishes, b.N will contain the total number of calls to op, so the benchmark
 // may use b.N to compute other average metrics.
 func (b *B) Loop() bool {
+	if b.loopN == b.N {
+		// If it's the first call to b.Loop() in the benchmark function.
+		// Allows more precise measurement of benchmark loop cost counts.
+		b.ResetTimer()
+	}
 	b.loopN--
 	return b.loopN >= 0
 }