Merge pull request #3455 from cenewcombe/develop

Fix unsafe read during final iteration of zsymv_L_sse2.S

kernel/x86_64/zsymv_L_sse2.S

@@ -451,7 +451,6 @@
 #endif
 
 	MOVDDUP(4 * SIZE, A1, a1)
-	MOVDDUP(6 * SIZE, A2, a2)
 
 	movsd	 0 * SIZE(YY), yy1
 	movhpd	 1 * SIZE(YY), yy1
@@ -471,7 +470,9 @@
 	subq	IS, I
 	subq	$2, I
 	sarq	$2, I
-	jle	.L15
+	jle	.L14
+
+	MOVDDUP(6 * SIZE - (4 * SIZE), A2, a2)
 	ALIGN_3
 
 .L12:
@@ -632,6 +633,16 @@
 	jg	 .L12
 	ALIGN_3
 
+.L14:
+	movq     M, I
+	subq    IS, I
+	subq    $2, I
+	testq   $2, I
+	jle     .L16
+
+	MOVDDUP(6 * SIZE - (4 * SIZE), A2, a2)
+	jmp     .L15_pastcheck
+
 .L15:
 	movq	 M, I
 	subq	IS, I
@@ -639,6 +650,7 @@
 	testq	$2, I
 	jle	.L16
 
+.L15_pastcheck:
 	movapd	  xtemp1, xt1
 	mulpd	  a1,     xt1
 	mulpd	  atemp1, a1