PutByteBuffer(resp)
						n++
					}
					atomic.AddInt64(&ops, int64(n))
					atomic.AddInt64(&lat, latency)
				})
				spent := time.Since(t)
				if spent > 0 && n > 0 {
					// Since we are benchmarking n parallel servers we need to multiply by n.
					// This will give an estimate of the total ops/s.
					latency := float64(atomic.LoadInt64(&lat)) / float64(time.Millisecond)

	memprofilerate    = flag.Int64("memprofilerate", 0, "set runtime.MemProfileRate to `rate`")
	benchmarkFlag     = flag.String("benchmark", "", "set to 'mem' or 'cpu' to enable phase benchmarking")
	benchmarkFileFlag = flag.String("benchmarkprofile", "", "emit phase profiles to `base`_phase.{cpu,mem}prof")

	flagW ternaryFlag
	FlagW = new(bool) // the -w flag, computed in main from flagW
)

				stop := applyGCLoad(b)
				runOne(b)
				// Make sure to stop the timer before we wait! The load created above
				// is very heavy-weight and not easy to stop, so we could end up
				// confusing the benchmarking framework for small b.N.
				b.StopTimer()
				stop()
			})
		}
	}

	// Measure the cost of counting goroutines
	b.Run("small-nil", run(func() bool {
		GoroutineProfile(nil)
		return true

        int stride = 8;
        int minLength = Math.min(left.length, right.length);
        int strideLimit = minLength & ~(stride - 1);
        int i;

        /*
         * Compare 8 bytes at a time. Benchmarking on x86 shows a stride of 8 bytes is no slower
         * than 4 bytes even on 32-bit. On the other hand, it is substantially faster on 64-bit.
         */
        for (i = 0; i < strideLimit; i += stride) {

        int stride = 8;
        int minLength = Math.min(left.length, right.length);
        int strideLimit = minLength & ~(stride - 1);
        int i;

        /*
         * Compare 8 bytes at a time. Benchmarking on x86 shows a stride of 8 bytes is no slower
         * than 4 bytes even on 32-bit. On the other hand, it is substantially faster on 64-bit.
         */
        for (i = 0; i < strideLimit; i += stride) {

   * elements will appear in the returned list in the same order they appeared in {@code elements}.
   *
   * <p><b>Performance note:</b> According to our
   * benchmarking
   * on Open JDK 7, {@link #immutableSortedCopy} generally performs better (in both time and space)
   * than this method, and this method in turn generally performs better than copying the list and

   * elements will appear in the returned list in the same order they appeared in {@code elements}.
   *
   * <p><b>Performance note:</b> According to our
   * benchmarking
   * on Open JDK 7, {@link #immutableSortedCopy} generally performs better (in both time and space)
   * than this method, and this method in turn generally performs better than copying the list and

		latencies = append(latencies, time.Since(start))
	}
	// Make sure to stop the timer before we wait! The load created above
	// is very heavy-weight and not easy to stop, so we could end up
	// confusing the benchmarking framework for small b.N.
	b.StopTimer()
	stop()

	// Disable the default */op metrics.
	// ns/op doesn't mean anything because it's an average, but we

	// Converting img to rgba is redundant for video-001.png, which is already
	// in the RGBA format, but for those copy/pasting this benchmark (but
	// changing the source image), the conversion ensures that we're still
	// benchmarking encoding an RGBA image.
	rgba := image.NewRGBA(bo)
	draw.Draw(rgba, bo, img, bo.Min, draw.Src)

	b.SetBytes(int64(bo.Dx() * bo.Dy() * 4))
	b.ReportAllocs()
	b.ResetTimer()
	for i := 0; i < b.N; i++ {

					if !ok {
						return nil, nil, fmt.Errorf("not a pod")
					}
					return pod.ObjectMeta.Labels, fields.Set{
						"metadata.name": pod.Name,
					}, nil
				},
			}

			// now we start benchmarking
			b.ResetTimer()
			for i := 0; i < b.N; i++ {
				list := tc.newListObjectFunc()
				if err := store.GetList(ctx, dir, storage.ListOptions{Predicate: pred, Recursive: true}, list); err != nil {