diff --git a/sm3/sm3blocks_ppc64x.go b/sm3/sm3blocks_ppc64x.go
index 1058a2d..f3ee04d 100644
--- a/sm3/sm3blocks_ppc64x.go
+++ b/sm3/sm3blocks_ppc64x.go
@@ -8,3 +8,6 @@ package sm3
 
 //go:noescape
 func transposeMatrix(dig **[8]uint32)
+
+//go:noescape
+func copyResultsBy4(dig *uint32, p *byte)
diff --git a/sm3/sm3blocks_ppc64x.s b/sm3/sm3blocks_ppc64x.s
index ecf9846..47ef3cb 100644
--- a/sm3/sm3blocks_ppc64x.s
+++ b/sm3/sm3blocks_ppc64x.s
@@ -8,7 +8,7 @@
 
 // For P9 instruction emulation
 #define ESPERMW  V21 // Endian swapping permute into BE
-#define TMP2    V22  // Temporary for P8_STXVB16X/P8_STXVB16X
+#define TMP2    V22  // Temporary for STOREWORDS
 
 DATA ·mask+0x00(SB)/8, $0x0c0d0e0f08090a0b // Permute for vector doubleword endian swap
 DATA ·mask+0x08(SB)/8, $0x0405060700010203
@@ -100,3 +100,41 @@ TEXT ·transposeMatrix(SB),NOSPLIT,$0
 	STOREWORDS(V7, R4, R6)
 
 	RET
+
+// func copyResultsBy4(dig *uint32, dst *byte)
+TEXT ·copyResultsBy4(SB),NOSPLIT,$0
+	MOVD	dig+0(FP), R3
+	MOVD	dst+8(FP), R4
+	
+	LXVD2X 	(R0)(R3), V0
+	STXVD2X 	V0, (R0)(R4)
+
+	MOVD	$16, R5
+	LXVD2X 	(R5)(R3), V0
+	STXVD2X 	V0, (R5)(R4)
+	
+	ADD 	$16, R5
+	LXVD2X 	(R5)(R3), V0
+	STXVD2X 	V0, (R5)(R4)
+
+	ADD 	$16, R5
+	LXVD2X 	(R5)(R3), V0
+	STXVD2X 	V0, (R5)(R4)
+
+	ADD 	$16, R5
+	LXVD2X 	(R5)(R3), V0
+	STXVD2X 	V0, (R5)(R4)
+
+	ADD 	$16, R5
+	LXVD2X 	(R5)(R3), V0
+	STXVD2X 	V0, (R5)(R4)
+
+	ADD 	$16, R5
+	LXVD2X 	(R5)(R3), V0
+	STXVD2X 	V0, (R5)(R4)
+
+	ADD 	$16, R5
+	LXVD2X 	(R5)(R3), V0
+	STXVD2X 	V0, (R5)(R4)	
+
+	RET
diff --git a/sm3/sm3blocks_ppc64x_test.go b/sm3/sm3blocks_ppc64x_test.go
index e3246fb..5d7d89d 100644
--- a/sm3/sm3blocks_ppc64x_test.go
+++ b/sm3/sm3blocks_ppc64x_test.go
@@ -7,6 +7,7 @@
 package sm3
 
 import (
+	"encoding/binary"
 	"fmt"
 	"testing"
 )
@@ -33,3 +34,26 @@ func TestTransposeMatrix(t *testing.T) {
 		fmt.Println()
 	}
 }
+
+func TestCopyResultsBy4(t *testing.T) {
+	var m [4][8]uint32
+	var ret [128]byte
+	var k uint32 = 0
+	for i := 0; i < 4; i++ {
+		for j := 0; j < 8; j++ {
+			m[i][j] = k
+			k++
+			fmt.Printf("%08x ", m[i][j])
+		}
+		fmt.Println()
+	}
+	copyResultsBy4(&m[0][0], &ret[0])
+	fmt.Printf("got: %x\n", ret[:])
+
+	for i := 0; i < 4; i++ {
+		for j := 0; j < 8; j++ {
+			binary.BigEndian.PutUint32(ret[i*32+j*4:], m[i][j])
+		}
+	}
+	fmt.Printf("expected %x\n", ret[:])
+}