From e7299cda751f7742091a67c932ea6057397a8533 Mon Sep 17 00:00:00 2001
From: Sun Yimin <emmansun@users.noreply.github.com>
Date: Thu, 5 Sep 2024 14:54:46 +0800
Subject: [PATCH] debug ppc64le byte order

---
 sm3/sm3blocks_ppc64x.s       | 12 +++++++++++-
 sm3/sm3blocks_ppc64x_test.go |  4 ++--
 2 files changed, 13 insertions(+), 3 deletions(-)

diff --git a/sm3/sm3blocks_ppc64x.s b/sm3/sm3blocks_ppc64x.s
index 8eb7db9..53eb11d 100644
--- a/sm3/sm3blocks_ppc64x.s
+++ b/sm3/sm3blocks_ppc64x.s
@@ -70,7 +70,7 @@ TEXT ·transposeMatrix(SB),NOSPLIT,$0
 #else
 	MOVD	$·mask+0x10(SB), R4
 #endif
-
+/*
 	LVX 	(R0)(R4), V8
 	LVX 	(R6)(R4), V9
 	LVX 	(R8)(R4), V10
@@ -105,5 +105,15 @@ TEXT ·transposeMatrix(SB),NOSPLIT,$0
 	MOVD 	(R7)(R3), R4
 	P8_STXVB16X(V3, R4, R0)
 	P8_STXVB16X(V7, R4, R6)
+*/
+	MOVD 	(R0)(R3), R4
+	LXVD2X (R0)(R4), V0
+	VSPLTW $1, V0, V2
+	STXVD2X V2, (R0)(R4)
+
+	LXVD2X (R6)(R4), V1
+	VPERM	V1, V1, ESPERM, V1
+	VSPLTW $1, V1, V2
+	STXVD2X V2, (R6)(R4)
 
 	RET
diff --git a/sm3/sm3blocks_ppc64x_test.go b/sm3/sm3blocks_ppc64x_test.go
index 7f708cd..e3246fb 100644
--- a/sm3/sm3blocks_ppc64x_test.go
+++ b/sm3/sm3blocks_ppc64x_test.go
@@ -18,7 +18,7 @@ func TestTransposeMatrix(t *testing.T) {
 		for j := 0; j < 8; j++ {
 			m[i][j] = k
 			k++
-			fmt.Printf("%04x ", m[i][j])
+			fmt.Printf("%08x ", m[i][j])
 		}
 		fmt.Println()
 	}
@@ -28,7 +28,7 @@ func TestTransposeMatrix(t *testing.T) {
 	fmt.Println()
 	for i := 0; i < 4; i++ {
 		for j := 0; j < 8; j++ {
-			fmt.Printf("%04x ", m[i][j])
+			fmt.Printf("%08x ", m[i][j])
 		}
 		fmt.Println()
 	}