diff --git a/src/pkg/bytes/asm_amd64.s b/src/pkg/bytes/asm_amd64.s
index 7e78700ecfb..c6793cbdcc7 100644
--- a/src/pkg/bytes/asm_amd64.s
+++ b/src/pkg/bytes/asm_amd64.s
@@ -3,15 +3,90 @@
 // license that can be found in the LICENSE file.
 
 TEXT ·IndexByte(SB),7,$0
-	MOVQ	p+0(FP), SI
-	MOVL	len+8(FP), CX
-	MOVB	b+16(FP), AL
-	MOVQ	SI, DI
+	MOVQ p+0(FP), SI
+	MOVL len+8(FP), BX
+	MOVB b+16(FP), AL
+	MOVQ SI, DI
+
+	CMPL BX, $16
+	JLT small
+
+	// round up to first 16-byte boundary
+	TESTQ $15, SI
+	JZ aligned
+	MOVQ SI, CX
+	ANDQ $~15, CX
+	ADDQ $16, CX
+
+	// search the beginning
+	SUBQ SI, CX
 	REPN; SCASB
-	JZ 3(PC)
-	MOVL	$-1, ret+24(FP)
+	JZ success
+
+// DI is 16-byte aligned; get ready to search using SSE instructions
+aligned:
+	// round down to last 16-byte boundary
+	MOVQ BX, R11
+	ADDQ SI, R11
+	ANDQ $~15, R11
+
+	// shuffle X0 around so that each byte contains c
+	MOVD AX, X0
+	PUNPCKLBW X0, X0
+	PUNPCKLBW X0, X0
+	PSHUFL $0, X0, X0
+	JMP condition
+
+sse:
+	// move the next 16-byte chunk of the buffer into X1
+	MOVO (DI), X1
+	// compare bytes in X0 to X1
+	PCMPEQB X0, X1
+	// take the top bit of each byte in X1 and put the result in DX
+	PMOVMSKB X1, DX
+	TESTL DX, DX
+	JNZ ssesuccess
+	ADDQ $16, DI
+
+condition:
+	CMPQ DI, R11
+	JLT sse
+
+	// search the end
+	MOVQ SI, CX
+	ADDQ BX, CX
+	SUBQ R11, CX
+	// if CX == 0, the zero flag will be set and we'll end up
+	// returning a false success
+	JZ failure
+	REPN; SCASB
+	JZ success
+
+failure:
+	MOVL $-1, ret+24(FP)
 	RET
-	SUBQ	SI, DI
-	SUBL	$1, DI
-	MOVL	DI, ret+24(FP)
+
+// handle for lengths < 16
+small:
+	MOVL BX, CX
+	REPN; SCASB
+	JZ success
+	MOVL $-1, ret+24(FP)
 	RET
+
+// we've found the chunk containing the byte
+// now just figure out which specific byte it is
+ssesuccess:
+	// get the index of the least significant set bit
+	BSFW DX, DX
+	SUBQ SI, DI
+	ADDQ DI, DX
+	MOVL DX, ret+24(FP)
+	RET
+
+success:
+	SUBQ SI, DI
+	SUBL $1, DI
+	MOVL DI, ret+24(FP)
+	RET
+
diff --git a/src/pkg/bytes/bytes_test.go b/src/pkg/bytes/bytes_test.go
index 6f42338eb8e..f3ca371f83e 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)