diff --git a/.github/workflows/test_ppc64.yaml b/.github/workflows/test_ppc64.yaml
index 6402247..2efe1de 100644
--- a/.github/workflows/test_ppc64.yaml
+++ b/.github/workflows/test_ppc64.yaml
@@ -41,3 +41,9 @@ jobs:
       env:
         GOARCH: ${{ matrix.arch }}
         GOPPC64: ${{ matrix.ppc64 }}
+
+    - name: Test SM4
+      run: go test -v -short ./sm4/...
+      env:
+        GOARCH: ${{ matrix.arch }}
+        GOPPC64: ${{ matrix.ppc64 }}        
diff --git a/sm4/asm_ppc64x.s b/sm4/asm_ppc64x.s
new file mode 100644
index 0000000..c9e8077
--- /dev/null
+++ b/sm4/asm_ppc64x.s
@@ -0,0 +1,210 @@
+// Copyright 2024 Sun Yimin. All rights reserved.
+// Use of this source code is governed by a MIT-style
+// license that can be found in the LICENSE file.
+
+//go:build (ppc64 || ppc64le) && !purego
+
+#include "textflag.h"
+
+#define REVERSE_WORDS V23
+#define V_FOUR V24
+#define M0 V25
+#define M1 V26
+#define M2 V27
+#define M3 V28
+#define M1L V20
+#define M1H V21
+#define M2L V22
+#define M2H V23
+#define NIBBLE_MASK V29
+#define INVERSE_SHIFT_ROWS V30
+// For instruction emulation
+#define ESPERMW  V31 // Endian swapping permute into BE
+
+DATA ·rcon+0x00(SB)/8, $0x0b0a09080f0e0d0c // byte swap per word
+DATA ·rcon+0x08(SB)/8, $0x0302010007060504
+DATA ·rcon+0x10(SB)/8, $0x0001020310111213 // Permute for transpose matrix
+DATA ·rcon+0x18(SB)/8, $0x0405060714151617
+DATA ·rcon+0x20(SB)/8, $0x08090a0b18191a1b
+DATA ·rcon+0x28(SB)/8, $0x0c0d0e0f1c1d1e1f
+DATA ·rcon+0x30(SB)/8, $0x0001020304050607
+DATA ·rcon+0x38(SB)/8, $0x1011121314151617
+DATA ·rcon+0x40(SB)/8, $0x08090a0b0c0d0e0f
+DATA ·rcon+0x48(SB)/8, $0x18191a1b1c1d1e1f
+DATA ·rcon+0x50(SB)/8, $0x0c0d0e0f08090a0b // reverse words
+DATA ·rcon+0x58(SB)/8, $0x0405060700010203
+DATA ·rcon+0x60(SB)/8, $0x0F0F0F0F0F0F0F0F // nibble mask
+DATA ·rcon+0x68(SB)/8, $0x0F0F0F0F0F0F0F0F
+DATA ·rcon+0x70(SB)/8, $0x0B0E0104070A0D00 // inverse shift rows
+DATA ·rcon+0x78(SB)/8, $0x0306090C0F020508
+DATA ·rcon+0x80(SB)/8, $0x0A7FC3B6D5A01C69 // affine transform matrix m1 low
+DATA ·rcon+0x88(SB)/8, $0x3045F98CEF9A2653
+DATA ·rcon+0x90(SB)/8, $0xC35BF46CAF379800 // affine transform matrix m1 high
+DATA ·rcon+0x98(SB)/8, $0x68F05FC7049C33AB
+DATA ·rcon+0xa0(SB)/8, $0x9A950A05FEF16E61 // affine transform matrix m2 low
+DATA ·rcon+0xa8(SB)/8, $0x0E019E916A65FAF5
+DATA ·rcon+0xb0(SB)/8, $0x892D69CD44E0A400 // affine transform matrix m2 high
+DATA ·rcon+0xb8(SB)/8, $0x2C88CC68E14501A5
+
+GLOBL ·rcon(SB), RODATA, $192
+
+#ifdef GOARCH_ppc64le
+#define NEEDS_PERMW
+
+#define PPC64X_LXVW4X(RA,RB,VT) \
+	LXVW4X	(RA+RB), VT \
+	VPERM	VT, VT, ESPERMW, VT
+
+#else
+#define PPC64X_LXVW4X(RA,RB,VT)  LXVW4X	(RA+RB), VT
+#endif // defined(GOARCH_ppc64le)
+
+// r = s <<< n
+// Due to VSPLTISW's limitation, the n MUST be [0, 15],
+// If n > 15, we have to call it multiple times.
+// VSPLTISW takes a 5-bit immediate value as an operand.
+// I also did NOT find one vector instruction to use immediate value for ROTL.
+#define PROLD(s, r, tmp, n) \
+	VSPLTISW $n, tmp \
+	VRLW	s, tmp, r
+
+#define TRANSPOSE_MATRIX(T0, T1, T2, T3) \
+	VPERM T0, T1, M0, TMP0; \
+	VPERM T2, T3, M0, TMP1; \
+	VPERM T0, T1, M1, TMP2; \
+	VPERM T2, T3, M1, TMP3; \
+	VPERM TMP0, TMP1, M2, T0; \
+	VPERM TMP0, TMP1, M3, T1; \
+	VPERM TMP2, TMP3, M2, T2; \
+	VPERM TMP2, TMP3, M3, T3
+
+// Affine Transform
+// parameters:
+// -  L: table low nibbles
+// -  H: table high nibbles
+// -  x: 128 bits register as sbox input/output data
+// -  y: 128 bits temp register
+// -  z: 128 bits temp register
+#define AFFINE_TRANSFORM(L, H, V_FOUR, x, y, z)  \
+	VAND NIBBLE_MASK, x, z;              \
+	VPERM L, L, z, y;                    \
+	VSRW x, V_FOUR, x;                   \
+	VAND NIBBLE_MASK, x, z;              \
+	VPERM H, H, z, x;                    \
+	VXOR y, x, x
+
+// Affine Transform
+// parameters:
+// -  L: table low nibbles
+// -  H: table high nibbles
+// -  x: 128 bits register as sbox input/output data
+// -  y: 128 bits temp register
+// -  z: 128 bits temp register
+#define AFFINE_TRANSFORM_N(L, H, V_FOUR, x, y, z)  \
+	VNAND NIBBLE_MASK, x, z;             \
+	VPERM L, L, z, y;                    \
+	VSRW x, V_FOUR, x;                   \
+	VAND NIBBLE_MASK, x, z;              \
+	VPERM H, H, z, x;                    \
+	VXOR y, x, x
+
+// SM4 sbox function
+// parameters:
+// -  x: 128 bits register as sbox input/output data
+// -  y: 128 bits temp register
+// -  z: 128 bits temp register
+#define SM4_SBOX(x, y, z) \
+	; \
+	AFFINE_TRANSFORM(M1L, M1H, V_FOUR, x, y, z); \
+	; \
+	VPERM x, x, INVERSE_SHIFT_ROWS, x;           \
+	VCIPHERLAST x, NIBBLE_MASK, x;               \
+	; \
+	AFFINE_TRANSFORM_N(M2L, M2H, V_FOUR, x, y, z)
+
+#define SM4_TAO_L2(x, y, z)         \
+	SM4_SBOX(x, y, z);                      \
+	;                                       \ //####################  4 parallel L2 linear transforms ##################//
+	VSPLTISW $13, z;                        \
+	VRLW	x, z, y;                        \ // y = x <<< 13
+	VXOR x, y, y;                           \ 
+	VRLW x, z, x;                           \ // x = x <<< 13
+	VSPLTISW $10, z;                        \
+	VRLW x, z, x;                           \ // x = x <<< 10
+	VXOR x, y, x
+
+#define SM4_EXPANDKEY_ROUND(CK, x, y, z, t0, t1, t2, t3, target) \
+	VXOR t1, CK, x;                      \
+	VXOR t2, x, x;                       \
+	VXOR t3, x, x;                       \
+	SM4_TAO_L2(x, y, z);                 \
+	VXOR x, t0, t0;                      \
+	VSLDOI $4, target, t0, target
+
+
+// func expandKeyAsm(key *byte, ck, enc, dec *uint32, inst int)
+TEXT ·expandKeyAsm(SB),NOSPLIT,$0
+	// prepare/load constants
+	VSPLTISW $4, V_FOUR;
+#ifdef NEEDS_PERMW
+	MOVD	$·rcon(SB), R4
+	LVX	(R4), ESPERMW
+#endif
+	MOVD	$·rcon+0x50(SB), R4
+	LXVD2X	(R4)(R0), REVERSE_WORDS
+	MOVD $16, R3
+	LXVD2X (R4)(R3), NIBBLE_MASK
+	MOVD $32, R3
+	LXVD2X (R4)(R3), INVERSE_SHIFT_ROWS
+	MOVD $48, R3
+	LXVD2X (R4)(R3), M1L
+	MOVD $64, R3
+	LXVD2X (R4)(R3), M1H
+	MOVD $80, R3
+	LXVD2X (R4)(R3), M2L
+	MOVD $96, R3
+	LXVD2X (R4)(R3), M2H
+
+	MOVD key+0(FP), R3
+	MOVD ck+8(FP), R4
+	MOVD enc+16(FP), R5
+	MOVD dec+24(FP), R6
+
+	ADD $112, R6
+
+	// prepare counter
+	MOVD $8, R7
+	MOVD R7, CTR
+
+	// load key
+	PPC64X_LXVW4X(R3, R0, V0)
+	VSLDOI $4, V0, V0, V1
+	VSLDOI $4, V1, V1, V2
+	VSLDOI $4, V2, V2, V3
+
+ksLoop:
+	LXVW4X (R4), V4
+	SM4_EXPANDKEY_ROUND(V4, x, y, z, V0, V1, V2, V3, V5)
+	VSLDOI $4, V4, V4, V4
+	SM4_EXPANDKEY_ROUND(V4, x, y, z, V1, V2, V3, V0, V5)
+	VSLDOI $4, V4, V4, V4
+	SM4_EXPANDKEY_ROUND(V4, x, y, z, V2, V3, V0, V1, V5)
+	VSLDOI $4, V4, V4, V4
+	SM4_EXPANDKEY_ROUND(V4, x, y, z, V3, V0, V1, V2, V5)
+	STXVW4X V5, (R5)
+	VPERM V5, V5, REVERSE_WORDS, V5
+	STXVW4X V5, (R6)
+	ADD $16, R5
+	ADD $16, R4
+	ADD $-16, R6
+	BDNZ	ksLoop
+
+    RET
+
+// func encryptBlocksAsm(xk *uint32, dst, src []byte, inst int)
+TEXT ·encryptBlocksAsm(SB),NOSPLIT,$0
+	RET
+
+// func encryptBlockAsm(xk *uint32, dst, src *byte, inst int)
+TEXT ·encryptBlockAsm(SB),NOSPLIT,$0
+	RET
diff --git a/sm4/sm4_ppc64x.go b/sm4/sm4_ppc64x.go
new file mode 100644
index 0000000..261af48
--- /dev/null
+++ b/sm4/sm4_ppc64x.go
@@ -0,0 +1,12 @@
+//go:build (ppc64 || ppc64le) && !purego
+
+package sm4
+
+//go:noescape
+func encryptBlocksAsm(xk *uint32, dst, src []byte, inst int)
+
+//go:noescape
+func encryptBlockAsm(xk *uint32, dst, src *byte, inst int)
+
+//go:noescape
+func expandKeyAsm(key *byte, ck, enc, dec *uint32, inst int)
diff --git a/sm4/sm4_ppc64x_test.go b/sm4/sm4_ppc64x_test.go
new file mode 100644
index 0000000..b018492
--- /dev/null
+++ b/sm4/sm4_ppc64x_test.go
@@ -0,0 +1,44 @@
+//go:build (ppc64 || ppc64le) && !purego
+
+package sm4
+
+import (
+	"crypto/rand"
+	"io"
+	"reflect"
+	"testing"
+	"time"
+)
+
+func TestExpandKey(t *testing.T) {
+	key := make([]byte, 16)
+
+	var encRes1 [rounds]uint32
+	var decRes1 [rounds]uint32
+	encRes2 := make([]uint32, 32)
+	decRes2 := make([]uint32, 32)
+	var timeout *time.Timer
+
+	if testing.Short() {
+		timeout = time.NewTimer(10 * time.Millisecond)
+	} else {
+		timeout = time.NewTimer(2 * time.Second)
+	}
+
+	for {
+		select {
+		case <-timeout.C:
+			return
+		default:
+		}
+		io.ReadFull(rand.Reader, key)
+		expandKeyGo(key, &encRes1, &decRes1)
+		expandKey(key, encRes2, decRes2)
+		if !reflect.DeepEqual(encRes1[:], encRes2) {
+			t.Errorf("expected=%x, result=%x\n", encRes1[:], encRes2)
+		}
+		if !reflect.DeepEqual(decRes1[:], decRes2) {
+			t.Errorf("expected=%x, result=%x\n", decRes1[:], decRes2)
+		}
+	}
+}