diff --git a/sm3/sm3.go b/sm3/sm3.go
new file mode 100644
index 0000000..ea219b6
--- /dev/null
+++ b/sm3/sm3.go
@@ -0,0 +1,205 @@
+package sm3
+
+import (
+	"encoding/binary"
+	"errors"
+	"hash"
+)
+
+// Size the size of a SM3 checksum in bytes.
+const Size = 32
+
+// BlockSize the blocksize of SM3 in bytes.
+const BlockSize = 64
+
+const (
+	chunk = 64
+	init0 = 0x7380166f
+	init1 = 0x4914b2b9
+	init2 = 0x172442d7
+	init3 = 0xda8a0600
+	init4 = 0xa96f30bc
+	init5 = 0x163138aa
+	init6 = 0xe38dee4d
+	init7 = 0xb0fb0e4e
+)
+
+// digest represents the partial evaluation of a checksum.
+type digest struct {
+	h   [8]uint32
+	x   [chunk]byte
+	nx  int
+	len uint64
+}
+
+const (
+	magic256      = "sm3\x03"
+	marshaledSize = len(magic256) + 8*4 + chunk + 8
+)
+
+func (d *digest) MarshalBinary() ([]byte, error) {
+	b := make([]byte, 0, marshaledSize)
+	b = append(b, magic256...)
+	b = appendUint32(b, d.h[0])
+	b = appendUint32(b, d.h[1])
+	b = appendUint32(b, d.h[2])
+	b = appendUint32(b, d.h[3])
+	b = appendUint32(b, d.h[4])
+	b = appendUint32(b, d.h[5])
+	b = appendUint32(b, d.h[6])
+	b = appendUint32(b, d.h[7])
+	b = append(b, d.x[:d.nx]...)
+	b = b[:len(b)+len(d.x)-int(d.nx)] // already zero
+	b = appendUint64(b, d.len)
+	return b, nil
+}
+
+func (d *digest) UnmarshalBinary(b []byte) error {
+	if len(b) < len(magic256) || (string(b[:len(magic256)]) != magic256) {
+		return errors.New("sm3: invalid hash state identifier")
+	}
+	if len(b) != marshaledSize {
+		return errors.New("sm3: invalid hash state size")
+	}
+	b = b[len(magic256):]
+	b, d.h[0] = consumeUint32(b)
+	b, d.h[1] = consumeUint32(b)
+	b, d.h[2] = consumeUint32(b)
+	b, d.h[3] = consumeUint32(b)
+	b, d.h[4] = consumeUint32(b)
+	b, d.h[5] = consumeUint32(b)
+	b, d.h[6] = consumeUint32(b)
+	b, d.h[7] = consumeUint32(b)
+	b = b[copy(d.x[:], b):]
+	b, d.len = consumeUint64(b)
+	d.nx = int(d.len % chunk)
+	return nil
+}
+
+func appendUint64(b []byte, x uint64) []byte {
+	var a [8]byte
+	binary.BigEndian.PutUint64(a[:], x)
+	return append(b, a[:]...)
+}
+
+func appendUint32(b []byte, x uint32) []byte {
+	var a [4]byte
+	binary.BigEndian.PutUint32(a[:], x)
+	return append(b, a[:]...)
+}
+
+func consumeUint64(b []byte) ([]byte, uint64) {
+	_ = b[7]
+	x := uint64(b[7]) | uint64(b[6])<<8 | uint64(b[5])<<16 | uint64(b[4])<<24 |
+		uint64(b[3])<<32 | uint64(b[2])<<40 | uint64(b[1])<<48 | uint64(b[0])<<56
+	return b[8:], x
+}
+
+func consumeUint32(b []byte) ([]byte, uint32) {
+	_ = b[3]
+	x := uint32(b[3]) | uint32(b[2])<<8 | uint32(b[1])<<16 | uint32(b[0])<<24
+	return b[4:], x
+}
+
+// New returns a new hash.Hash computing the SM3 checksum. The Hash
+// also implements encoding.BinaryMarshaler and
+// encoding.BinaryUnmarshaler to marshal and unmarshal the internal
+// state of the hash.
+func New() hash.Hash {
+	d := new(digest)
+	d.Reset()
+	return d
+}
+
+// Sum appends the current hash to b and returns the resulting slice.
+// It does not change the underlying hash state.
+func (d *digest) Sum(in []byte) []byte {
+	// Make a copy of d so that caller can keep writing and summing.
+	d0 := *d
+	hash := d0.checkSum()
+	return append(in, hash[:]...)
+}
+
+func (d *digest) checkSum() [Size]byte {
+	len := d.len
+	// Padding. Add a 1 bit and 0 bits until 56 bytes mod 64.
+	var tmp [64]byte
+	tmp[0] = 0x80
+	if len%64 < 56 {
+		d.Write(tmp[0 : 56-len%64])
+	} else {
+		d.Write(tmp[0 : 64+56-len%64])
+	}
+	// Length in bits.
+	len <<= 3
+	binary.BigEndian.PutUint64(tmp[:], len)
+	d.Write(tmp[0:8])
+
+	if d.nx != 0 {
+		panic("d.nx != 0")
+	}
+
+	var digest [Size]byte
+
+	binary.BigEndian.PutUint32(digest[0:], d.h[0])
+	binary.BigEndian.PutUint32(digest[4:], d.h[1])
+	binary.BigEndian.PutUint32(digest[8:], d.h[2])
+	binary.BigEndian.PutUint32(digest[12:], d.h[3])
+	binary.BigEndian.PutUint32(digest[16:], d.h[4])
+	binary.BigEndian.PutUint32(digest[20:], d.h[5])
+	binary.BigEndian.PutUint32(digest[24:], d.h[6])
+	binary.BigEndian.PutUint32(digest[28:], d.h[7])
+
+	return digest
+}
+
+func (d *digest) Write(p []byte) (nn int, err error) {
+	nn = len(p)
+	d.len += uint64(nn)
+	if d.nx > 0 {
+		n := copy(d.x[d.nx:], p)
+		d.nx += n
+		if d.nx == chunk {
+			block(d, d.x[:])
+			d.nx = 0
+		}
+		p = p[n:]
+	}
+	if len(p) >= chunk {
+		n := len(p) &^ (chunk - 1)
+		block(d, p[:n])
+		p = p[n:]
+	}
+	if len(p) > 0 {
+		d.nx = copy(d.x[:], p)
+	}
+	return
+}
+
+func (d *digest) Size() int {
+	return Size
+}
+
+func (d *digest) BlockSize() int { return BlockSize }
+
+// Reset resets the Hash to its initial state.
+func (d *digest) Reset() {
+	d.h[0] = init0
+	d.h[1] = init1
+	d.h[2] = init2
+	d.h[3] = init3
+	d.h[4] = init4
+	d.h[5] = init5
+	d.h[6] = init6
+	d.h[7] = init7
+	d.nx = 0
+	d.len = 0
+}
+
+// Sum returns the SM3 checksum of the data.
+func Sum(data []byte) [Size]byte {
+	var d digest
+	d.Reset()
+	d.Write(data)
+	return d.checkSum()
+}
diff --git a/sm3/sm3_test.go b/sm3/sm3_test.go
new file mode 100644
index 0000000..6b7afcf
--- /dev/null
+++ b/sm3/sm3_test.go
@@ -0,0 +1,129 @@
+package sm3
+
+import (
+	"bytes"
+	"encoding"
+	"fmt"
+	"hash"
+	"io"
+	"testing"
+)
+
+type sm3Test struct {
+	out       string
+	in        string
+	halfState string // marshaled hash state after first half of in written, used by TestGoldenMarshal
+}
+
+var golden = []sm3Test{
+	{"66c7f0f462eeedd9d1f2d46bdc10e4e24167c4875cf2f7a2297da02b8f4ba8e0", "abc", "sm3\x03s\x80\x16oI\x14\xb2\xb9\x17$B\xd7ڊ\x06\x00\xa9o0\xbc\x1618\xaa\xe3\x8d\xeeM\xb0\xfb\x0eNa\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01"},
+	{"debe9ff92275b8a138604889c18e5a4d6fdb70e5387e5765293dcba39c0c5732", "abcdabcdabcdabcdabcdabcdabcdabcdabcdabcdabcdabcdabcdabcdabcdabcd", "sm3\x03s\x80\x16oI\x14\xb2\xb9\x17$B\xd7ڊ\x06\x00\xa9o0\xbc\x1618\xaa\xe3\x8d\xeeM\xb0\xfb\x0eNabcdabcdabcdabcdabcdabcdabcdabcd\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00 "},
+}
+
+func TestGolden(t *testing.T) {
+	for i := 0; i < len(golden); i++ {
+		g := golden[i]
+		s := fmt.Sprintf("%x", Sum([]byte(g.in)))
+		if s != g.out {
+			t.Fatalf("SM3 function: sm3(%s) = %s want %s", g.in, s, g.out)
+		}
+		c := New()
+		for j := 0; j < 3; j++ {
+			if j < 2 {
+				io.WriteString(c, g.in)
+			} else {
+				io.WriteString(c, g.in[0:len(g.in)/2])
+				c.Sum(nil)
+				io.WriteString(c, g.in[len(g.in)/2:])
+			}
+			s := fmt.Sprintf("%x", c.Sum(nil))
+			if s != g.out {
+				t.Fatalf("sm3[%d](%s) = %s want %s", j, g.in, s, g.out)
+			}
+			c.Reset()
+		}
+	}
+}
+
+func TestGoldenMarshal(t *testing.T) {
+	tests := []struct {
+		name    string
+		newHash func() hash.Hash
+		gold    []sm3Test
+	}{
+		{"", New, golden},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			for _, g := range tt.gold {
+				h := tt.newHash()
+				h2 := tt.newHash()
+
+				io.WriteString(h, g.in[:len(g.in)/2])
+
+				state, err := h.(encoding.BinaryMarshaler).MarshalBinary()
+				if err != nil {
+					t.Errorf("could not marshal: %v", err)
+					continue
+				}
+
+				if string(state) != g.halfState {
+					t.Errorf("sm3%s(%q) state = %q, want %q", tt.name, g.in, state, g.halfState)
+					continue
+				}
+
+				if err := h2.(encoding.BinaryUnmarshaler).UnmarshalBinary(state); err != nil {
+					t.Errorf("could not unmarshal: %v", err)
+					continue
+				}
+
+				io.WriteString(h, g.in[len(g.in)/2:])
+				io.WriteString(h2, g.in[len(g.in)/2:])
+
+				if actual, actual2 := h.Sum(nil), h2.Sum(nil); !bytes.Equal(actual, actual2) {
+					t.Errorf("sm3%s(%q) = 0x%x != marshaled 0x%x", tt.name, g.in, actual, actual2)
+				}
+			}
+		})
+	}
+}
+
+func TestSize(t *testing.T) {
+	c := New()
+	if got := c.Size(); got != Size {
+		t.Errorf("Size = %d; want %d", got, Size)
+	}
+}
+
+func TestBlockSize(t *testing.T) {
+	c := New()
+	if got := c.BlockSize(); got != BlockSize {
+		t.Errorf("BlockSize = %d want %d", got, BlockSize)
+	}
+}
+
+var bench = New()
+var buf = make([]byte, 8192)
+
+func benchmarkSize(b *testing.B, size int) {
+	b.SetBytes(int64(size))
+	sum := make([]byte, bench.Size())
+	for i := 0; i < b.N; i++ {
+		bench.Reset()
+		bench.Write(buf[:size])
+		bench.Sum(sum[:0])
+	}
+}
+
+func BenchmarkHash8Bytes(b *testing.B) {
+	benchmarkSize(b, 8)
+}
+
+func BenchmarkHash1K(b *testing.B) {
+	benchmarkSize(b, 1024)
+}
+
+func BenchmarkHash8K(b *testing.B) {
+	benchmarkSize(b, 8192)
+}
diff --git a/sm3/sm3block.go b/sm3/sm3block.go
new file mode 100644
index 0000000..115116e
--- /dev/null
+++ b/sm3/sm3block.go
@@ -0,0 +1,80 @@
+package sm3
+
+import "math/bits"
+
+var _T = []uint32{
+	0x79cc4519,
+	0x7a879d8a,
+}
+
+func t(j uint8) uint32 {
+	if j < 16 {
+		return _T[0]
+	}
+	return _T[1]
+}
+
+func p0(x uint32) uint32 {
+	return x ^ bits.RotateLeft32(x, 9) ^ bits.RotateLeft32(x, 17)
+}
+
+func p1(x uint32) uint32 {
+	return x ^ bits.RotateLeft32(x, 15) ^ bits.RotateLeft32(x, 23)
+}
+
+func ff(j uint8, x, y, z uint32) uint32 {
+	if j < 16 {
+		return x ^ y ^ z
+	}
+	return (x & y) | (x & z) | (y & z)
+}
+
+func gg(j uint8, x, y, z uint32) uint32 {
+	if j < 16 {
+		return x ^ y ^ z
+	}
+	return (x & y) | (^x & z)
+}
+
+func block(dig *digest, p []byte) {
+	var w [68]uint32
+	var w1 [64]uint32
+	h0, h1, h2, h3, h4, h5, h6, h7 := dig.h[0], dig.h[1], dig.h[2], dig.h[3], dig.h[4], dig.h[5], dig.h[6], dig.h[7]
+	for len(p) >= chunk {
+		for i := 0; i < 16; i++ {
+			j := i * 4
+			w[i] = uint32(p[j])<<24 | uint32(p[j+1])<<16 | uint32(p[j+2])<<8 | uint32(p[j+3])
+		}
+		for i := 16; i < 68; i++ {
+			w[i] = p1(w[i-16]^w[i-9]^bits.RotateLeft32(w[i-3], 15)) ^ bits.RotateLeft32(w[i-13], 7) ^ w[i-6]
+		}
+		for i := 0; i < 64; i++ {
+			w1[i] = w[i] ^ w[i+4]
+		}
+		a, b, c, d, e, f, g, h := h0, h1, h2, h3, h4, h5, h6, h7
+		for i := 0; i < 64; i++ {
+			ss1 := bits.RotateLeft32(bits.RotateLeft32(a, 12)+e+bits.RotateLeft32(t(uint8(i)), i), 7)
+			ss2 := ss1 ^ bits.RotateLeft32(a, 12)
+			tt1 := ff(uint8(i), a, b, c) + d + ss2 + w1[i]
+			tt2 := gg(uint8(i), e, f, g) + h + ss1 + w[i]
+			d = c
+			c = bits.RotateLeft32(b, 9)
+			b = a
+			a = tt1
+			h = g
+			g = bits.RotateLeft32(f, 19)
+			f = e
+			e = p0(tt2)
+		}
+		h0 ^= a
+		h1 ^= b
+		h2 ^= c
+		h3 ^= d
+		h4 ^= e
+		h5 ^= f
+		h6 ^= g
+		h7 ^= h
+		p = p[chunk:]
+	}
+	dig.h[0], dig.h[1], dig.h[2], dig.h[3], dig.h[4], dig.h[5], dig.h[6], dig.h[7] = h0, h1, h2, h3, h4, h5, h6, h7
+}