diff --git a/base14.go b/base14.go
index 8f58382..6c1ab0c 100644
--- a/base14.go
+++ b/base14.go
@@ -40,7 +40,7 @@ func Encode(b []byte) (encd []byte) {
 		outlen += 10
 	}
 	encd = make([]byte, outlen)
-	encode(offset, outlen, append(b, 0), encd)
+	encode(offset, outlen, b, encd)
 	return
 }
 
@@ -62,7 +62,7 @@ func EncodeTo(b, encd []byte) error {
 	if len(encd) < outlen {
 		return errors.New("encd too small")
 	}
-	encode(offset, outlen, append(b, 0), encd)
+	encode(offset, outlen, b, encd)
 	return nil
 }
 
@@ -105,7 +105,7 @@ func Decode(b []byte) (decd []byte) {
 		}
 	}
 	outlen = outlen/8*7 + offset
-	decd = make([]byte, outlen+1)
+	decd = make([]byte, outlen+8)
 	decode(offset, outlen, b, decd)
 	return decd[:outlen]
 }
diff --git a/base14_amd64.s b/base14_amd64.s
index fb2beb0..685b3c6 100644
--- a/base14_amd64.s
+++ b/base14_amd64.s
@@ -3,8 +3,8 @@
 
 #include "textflag.h"
 
-// func encode(offset, outlen int, b, encd []byte)
-TEXT ·encode(SB), NOSPLIT, $0-64
+// func _encode(offset, outlen int, b, encd []byte) (sum uint64, n uint64)
+TEXT ·_encode(SB), NOSPLIT, $0-81
     MOVQ ·offset+0(FP), R10
     MOVQ ·outlen+8(FP), AX
     MOVQ ·data+16(FP), DI
@@ -116,16 +116,16 @@ encrem:
 encsav:
     MOVQ $21955383195992142, CX
     ADDQ CX, DX
-    MOVQ DX, 0(R9)(SI*8)
-    MOVB $61, -2(R9)(AX*1)
-    MOVB R10, -1(R9)(AX*1)
+    SHLQ $3, SI
+    MOVQ DX, ·sum+64(FP)
+    MOVQ SI, ·sum+72(FP)
 
 encend:
     RET
 
 
-// func decode(offset, outlen int, b, decd []byte)
-TEXT ·decode(SB), NOSPLIT, $0-64
+// func _decode(offset, outlen int, b, decd []byte)
+TEXT ·_decode(SB), NOSPLIT, $0-64
     MOVQ ·offset+0(FP), BX
     MOVQ ·outlen+8(FP), R8
     MOVQ ·data+16(FP), DI
diff --git a/base14_asm.go b/base14_asm.go
index 4ba6ec1..ae64d06 100644
--- a/base14_asm.go
+++ b/base14_asm.go
@@ -3,10 +3,36 @@
 
 package base14
 
-//go:noescape
-//go:nosplit
-func encode(offset, outlen int, b, encd []byte)
+import (
+	"encoding/binary"
+)
 
 //go:noescape
 //go:nosplit
-func decode(offset, outlen int, b, decd []byte)
+func _encode(offset, outlen int, b, encd []byte) (sum uint64, n uint64)
+
+//go:noescape
+//go:nosplit
+func _decode(offset, outlen int, b, decd []byte)
+
+func encode(offset, outlen int, b, encd []byte) {
+	if len(b) == 7 {
+		b = append(b, 0)
+	}
+	sum, n := _encode(offset, outlen, b, encd)
+	if offset == 0 {
+		return
+	}
+	var tmp [8]byte
+	binary.LittleEndian.PutUint64(tmp[:], sum)
+	copy(encd[n:], tmp[:])
+	encd[outlen-2] = '='
+	encd[outlen-1] = byte(offset)
+}
+
+func decode(offset, outlen int, b, decd []byte) {
+	if offset != 0 && cap(b) == len(b) {
+		b = append(b, make([]byte, 8)...)
+	}
+	_decode(offset, outlen, b, decd)
+}