From d3727e7130318e1e73977afe655702230309ae22 Mon Sep 17 00:00:00 2001
From: Sun Yimin <emmansun@users.noreply.github.com>
Date: Wed, 28 Aug 2024 09:42:12 +0800
Subject: [PATCH] align address mode

---
 internal/subtle/xor_s390x.s | 17 ++++++++---------
 internal/subtle/xor_test.go |  2 +-
 2 files changed, 9 insertions(+), 10 deletions(-)

diff --git a/internal/subtle/xor_s390x.s b/internal/subtle/xor_s390x.s
index 5e43261..30b9efb 100644
--- a/internal/subtle/xor_s390x.s
+++ b/internal/subtle/xor_s390x.s
@@ -1,6 +1,5 @@
-// Copyright 2016 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
+// The original code (non-vector) is ported from Golang 
+// https://github.com/golang/go/blob/master/src/crypto/aes/ctr_s390x.go
 
 //go:build !purego
 
@@ -17,13 +16,13 @@ TEXT ·xorBytes(SB),NOSPLIT,$0-32
 	CMPBLT	R4, $64, less_than64
 
 loop64b:
-	VLM (R2)(R5*1), V0, V3
-	VLM (R3)(R5*1), V4, V7
+	VLM 0(R2)(R5*1), V0, V3
+	VLM 0(R3)(R5*1), V4, V7
 	VX V0, V4, V4
 	VX V1, V5, V5
 	VX V2, V6, V6
 	VX V3, V7, V7
-	VSTM V4, V7, (R1)(R5*1)
+	VSTM V4, V7, 0(R1)(R5*1)
 	LAY	64(R5), R5
 	SUB	$64, R4
 	CMPBGE	R4, $64, loop64b
@@ -31,11 +30,11 @@ loop64b:
 less_than64:
 	CMPBEQ	R4, $0, done // quick end
 	CMPBLT	R4, $32, less_than32
-	VLM (R2)(R5*1), V0, V1
-	VLM (R3)(R5*1), V2, V3
+	VLM 0(R2)(R5*1), V0, V1
+	VLM 0(R3)(R5*1), V2, V3
 	VX V0, V2, V2
 	VX V1, V3, V3
-	VSTM V2, V3, (R1)(R5*1)
+	VSTM V2, V3, 0(R1)(R5*1)
 	LAY	32(R5), R5
 	SUB	$32, R4
 
diff --git a/internal/subtle/xor_test.go b/internal/subtle/xor_test.go
index 53f9808..aa98c88 100644
--- a/internal/subtle/xor_test.go
+++ b/internal/subtle/xor_test.go
@@ -43,7 +43,7 @@ func TestXORBytes(t *testing.T) {
 					}
 
 					if subtle.XORBytes(d[alignD:], p, q); !bytes.Equal(d, want) {
-						t.Errorf("n=%d alignP=%d alignQ=%d alignD=%d:\n\tp = %x\n\tq = %x\n\td = %x\n\twant %x\n", n, alignP, alignQ, alignD, p, q, d, want)
+						t.Fatalf("n=%d alignP=%d alignQ=%d alignD=%d:\n\tp = %x\n\tq = %x\n\td = %x\n\twant %x\n", n, alignP, alignQ, alignD, p, q, d, want)
 					}
 				}
 			}