Fixed overflow internal buffer bug of (s/d/c/z)gemv on x86.

kernel/x86/gemv_t_sse2.S

@@ -76,7 +76,7 @@
 #endif
 
 #define STACKSIZE	16
-#define ARGS	16
+#define ARGS	20
 
 #define M		 4 + STACKSIZE+ARGS(%esp)
 #define N		 8 + STACKSIZE+ARGS(%esp)
@@ -89,10 +89,9 @@
 #define STACK_INCY	44 + STACKSIZE+ARGS(%esp)
 #define BUFFER		48 + STACKSIZE+ARGS(%esp)
 
-#define MMM	0+STACKSIZE(%esp)
-#define AA	4+STACKSIZE(%esp)
-#define LDAX 8+STACKSIZE(%esp)
-#define NN	12+STACKSIZE(%esp)
+#define MMM	0+ARGS(%esp)
+#define AA	4+ARGS(%esp)
+#define XX 8+ARGS(%esp)
 
 #define I	%eax
 #define J	%ebx
@@ -117,10 +116,8 @@
 	PROFCODE
 
 
-	movl	STACK_LDA,  LDA
-	movl	LDA,LDAX			# backup LDA
-	movl	N,J
-	movl	J,NN				# backup N
+	movl	STACK_X,  X
+	movl	X,XX
 	movl	A,J
 	movl	J,AA				# backup A
     movl	M,J
@@ -130,7 +127,6 @@
 	addl	$1,J
 	sall    $21,J                           # J=2^21*sizeof(double)=buffer size(16MB)
 	subl    $4, J                           # Don't use last 4 double in the buffer.
-	                                        # Now, split M by block J
 	subl	J,MMM				# MMM=MMM-J
 	movl	J,M		
 	jge		.L00t
@@ -142,15 +138,13 @@
 	movl	%eax,M
 
 .L00t:
+	movl	XX,%eax
+	movl	%eax, X
+
 	movl	AA,%eax
 	movl	%eax,A			 	# mov AA to A
 
-	movl	NN,%eax
-	movl	%eax,N				# reset N
-
-
-	movl	LDAX,  LDA			# reset LDA
-	movl	STACK_X,    X
+	movl	STACK_LDA,  LDA
 	movl	STACK_INCX, INCX
 	movl	STACK_INCY, INCY
 
@@ -605,6 +599,9 @@
 	movl 	M,J
 	leal 	(,J,SIZE),%eax
 	addl	%eax,AA
+	movl	STACK_INCX,INCX
+	imull	INCX,%eax
+	addl	%eax,XX
 	jmp		.L0t
 	ALIGN_4