sw64: fix head loop in deep-copy_template.S

Sunway inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I5PN9S

--------------------------------

Fix conditional branch of head loop in deep-copy_template.S, so it will
go to simd loop properly when dst is 32 bytes aligned.
Signed-off-by: NMao Minkai <maominkai@wxiat.com>
Signed-off-by: NGu Zitao <guzitao@wxiat.com>

arch/sw_64/lib/deep-copy_template.S

@@ -83,9 +83,9 @@ $quad_loop_head:
 	addl	$17, 8, $17
 	FIXUP_LDST( stl $2, 0($16) )
 	addl	$16, 8, $16
-	and	$16, 31, $1
 	blt	$18, $simd_end
-	beq	$16, $dest_aligned_32
+	and	$16, 31, $1
+	beq	$1, $dest_aligned_32
 	br	$31, $quad_loop_head
 
 $dest_aligned_32:
@@ -191,7 +191,8 @@ $quad_u_loop_head:
 	FIXUP_LDST( stl $2, 0($16) )
 	addl	$16, 8, $16
 	blt	$18, $simd_end
-	beq	$16, $dest_aligned_32
+	and	$16, 31, $1
+	beq	$1, $dest_aligned_32
 	br	$31, $quad_u_loop_head
 
 $prep_simd_u_loop: