runtime: add BenchmarkMemclrRange

This benchmark is added to test improvements in memclr_amd64.
As it is stated in Intel Optimization Manual 15.16.3.3, AVX2-implemented
memclr can produce a skewed result with the branch predictor being
trained by the large loop iteration count.

This benchmark generates sizes between some specified range. This should
help to measure how memclr works when branch predictors may be incorrectly
trained.

Change-Id: I14d173cafe43ca47198ed920e655547a66b3909f
Reviewed-on: https://go-review.googlesource.com/c/go/+/373362
Reviewed-by: Keith Randall <khr@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Keith Randall <khr@golang.org>
Auto-Submit: Michael Pratt <mpratt@google.com>
Reviewed-by: Michael Pratt <mpratt@google.com>
Reviewed-by: Keith Randall <khr@google.com>

src/runtime/memmove_test.go

@@ -411,6 +411,63 @@ func BenchmarkGoMemclr(b *testing.B) {
 	})
 }
 
+func BenchmarkMemclrRange(b *testing.B) {
+	type RunData struct {
+		data []int
+	}
+
+	benchSizes := []RunData{
+		RunData{[]int{1043, 1078, 1894, 1582, 1044, 1165, 1467, 1100, 1919, 1562, 1932, 1645,
+			1412, 1038, 1576, 1200, 1029, 1336, 1095, 1494, 1350, 1025, 1502, 1548, 1316, 1296,
+			1868, 1639, 1546, 1626, 1642, 1308, 1726, 1665, 1678, 1187, 1515, 1598, 1353, 1237,
+			1977, 1452, 2012, 1914, 1514, 1136, 1975, 1618, 1536, 1695, 1600, 1733, 1392, 1099,
+			1358, 1996, 1224, 1783, 1197, 1838, 1460, 1556, 1554, 2020}}, // 1kb-2kb
+		RunData{[]int{3964, 5139, 6573, 7775, 6553, 2413, 3466, 5394, 2469, 7336, 7091, 6745,
+			4028, 5643, 6164, 3475, 4138, 6908, 7559, 3335, 5660, 4122, 3945, 2082, 7564, 6584,
+			5111, 2288, 6789, 2797, 4928, 7986, 5163, 5447, 2999, 4968, 3174, 3202, 7908, 8137,
+			4735, 6161, 4646, 7592, 3083, 5329, 3687, 2754, 3599, 7231, 6455, 2549, 8063, 2189,
+			7121, 5048, 4277, 6626, 6306, 2815, 7473, 3963, 7549, 7255}}, // 2kb-8kb
+		RunData{[]int{16304, 15936, 15760, 4736, 9136, 11184, 10160, 5952, 14560, 15744,
+			6624, 5872, 13088, 14656, 14192, 10304, 4112, 10384, 9344, 4496, 11392, 7024,
+			5200, 10064, 14784, 5808, 13504, 10480, 8512, 4896, 13264, 5600}}, // 4kb-16kb
+		RunData{[]int{164576, 233136, 220224, 183280, 214112, 217248, 228560, 201728}}, // 128kb-256kb
+	}
+
+	for _, t := range benchSizes {
+		total := 0
+		minLen := 0
+		maxLen := 0
+
+		for _, clrLen := range t.data {
+			if clrLen > maxLen {
+				maxLen = clrLen
+			}
+			if clrLen < minLen || minLen == 0 {
+				minLen = clrLen
+			}
+			total += clrLen
+		}
+		buffer := make([]byte, maxLen)
+
+		text := ""
+		if minLen >= (1 << 20) {
+			text = fmt.Sprint(minLen>>20, "M ", (maxLen+(1<<20-1))>>20, "M")
+		} else if minLen >= (1 << 10) {
+			text = fmt.Sprint(minLen>>10, "K ", (maxLen+(1<<10-1))>>10, "K")
+		} else {
+			text = fmt.Sprint(minLen, " ", maxLen)
+		}
+		b.Run(text, func(b *testing.B) {
+			b.SetBytes(int64(total))
+			for i := 0; i < b.N; i++ {
+				for _, clrLen := range t.data {
+					MemclrBytes(buffer[:clrLen])
+				}
+			}
+		})
+	}
+}
+
 func BenchmarkClearFat8(b *testing.B) {
 	for i := 0; i < b.N; i++ {
 		var x [8 / 4]uint32