bytes: SSE for bytes.IndexByte on amd64

Performance on 2.8 GHz Intel Core i7:

Before:
BenchmarkIndexByte4K  1000000              2997 ns/op        1366.70 MB/s
BenchmarkIndexByte4M      500           3049772 ns/op        1375.28 MB/s
BenchmarkIndexByte64M      50          49582280 ns/op        1353.48 MB/s

After:
BenchmarkIndexByte4K 10000000               298 ns/op       13744.97 MB/s
BenchmarkIndexByte4M    10000            285993 ns/op       14665.76 MB/s
BenchmarkIndexByte64M     500           4618172 ns/op       14531.48 MB/s

R=rsc, PeterGo, r2, r
CC=golang-dev
https://golang.org/cl/2888041

1 files changed, 53 insertions, 0 deletions

diff --git a/src/pkg/bytes/bytes_test.go b/src/pkg/bytes/bytes_test.go
index 6f42338eb8..f3ca371f83 100644
--- a/src/pkg/bytes/bytes_test.go
+++ b/src/pkg/bytes/bytes_test.go
@@ -93,6 +93,9 @@ var indexTests = []BinOpTest{
 	{"abc", "b", 1},
 	{"abc", "c", 2},
 	{"abc", "x", -1},
+	{"barfoobarfooyyyzzzyyyzzzyyyzzzyyyxxxzzzyyy", "x", 33},
+	{"foofyfoobarfoobar", "y", 4},
+	{"oooooooooooooooooooooo", "r", -1},
 }
 
 var lastIndexTests = []BinOpTest{
@@ -177,6 +180,56 @@ func TestIndexByte(t *testing.T) {
 	}
 }
 
+// test a larger buffer with different sizes and alignments
+func TestIndexByteBig(t *testing.T) {
+	const n = 1024
+	b := make([]byte, n)
+	for i := 0; i < n; i++ {
+		// different start alignments
+		b1 := b[i:]
+		for j := 0; j < len(b1); j++ {
+			b1[j] = 'x'
+			pos := IndexByte(b1, 'x')
+			if pos != j {
+				t.Errorf("IndexByte(%q, 'x') = %v", b1, pos)
+			}
+			b1[j] = 0
+			pos = IndexByte(b1, 'x')
+			if pos != -1 {
+				t.Errorf("IndexByte(%q, 'x') = %v", b1, pos)
+			}
+		}
+		// different end alignments
+		b1 = b[:i]
+		for j := 0; j < len(b1); j++ {
+			b1[j] = 'x'
+			pos := IndexByte(b1, 'x')
+			if pos != j {
+				t.Errorf("IndexByte(%q, 'x') = %v", b1, pos)
+			}
+			b1[j] = 0
+			pos = IndexByte(b1, 'x')
+			if pos != -1 {
+				t.Errorf("IndexByte(%q, 'x') = %v", b1, pos)
+			}
+		}
+		// different start and end alignments
+		b1 = b[i/2 : n-(i+1)/2]
+		for j := 0; j < len(b1); j++ {
+			b1[j] = 'x'
+			pos := IndexByte(b1, 'x')
+			if pos != j {
+				t.Errorf("IndexByte(%q, 'x') = %v", b1, pos)
+			}
+			b1[j] = 0
+			pos = IndexByte(b1, 'x')
+			if pos != -1 {
+				t.Errorf("IndexByte(%q, 'x') = %v", b1, pos)
+			}
+		}
+	}
+}
+
 func TestIndexRune(t *testing.T) {
 	for _, tt := range indexRuneTests {
 		a := []byte(tt.a)