Modified aligned size. Added additional prefetch instruction because of cache line is 32 bytes in Loongson 3A.

kernel/mips64/axpy_loongson3a.S

@@ -72,7 +72,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 		
-#define	PREFETCH_DISTANCE 48
+#define PREFETCH_DISTANCE 48
 		
 #define N	$4
 
@@ -153,7 +153,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	blez	I, .L13
 	NOP
-	.align 3
+	.align 5
 
 .L12:
 	PREFETCHD(PREFETCH_DISTANCE*SIZE(X))
@@ -180,6 +180,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	ST	t3,  2 * SIZE(Y)
 	ST	t4,  3 * SIZE(Y)
 
+	PREFETCHD((PREFETCH_DISTANCE+4)*SIZE(X))
+	PREFETCHD((PREFETCH_DISTANCE+4)*SIZE(Y))
+
 	MADD	t1, b5, ALPHA, a5
 	LD	a5, 12 * SIZE(X)
 	LD	b5, 12 * SIZE(Y)
@@ -206,7 +209,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	bgtz	I, .L12
 	daddiu	X, X, 8 * SIZE
-	.align 3
+	.align 5
 
 .L13:
 	MADD	t1, b1, ALPHA, a1
@@ -230,7 +233,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	daddiu	X, X, 8 * SIZE
 	daddiu	Y, Y, 8 * SIZE
-	.align 3
+	.align 5
 
 .L15:
 	andi	I,  N, 7
@@ -260,7 +263,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	j	$31
 	NOP
-	.align 3
+	.align 5
 
 .L20:
 	dsra	I, N, 3
@@ -304,7 +307,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	blez	I, .L23
 	NOP
-	.align 3
+	.align 5
 
 .L22:
 	MADD	t1, b1, ALPHA, a1
@@ -379,7 +382,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	bgtz	I, .L22
 	daddu	YY, YY, INCY
-	.align 3
+	.align 5
 
 .L23:
 	MADD	t1, b1, ALPHA, a1
@@ -411,7 +414,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	daddu	YY, YY, INCY
 	ST	t4,  0 * SIZE(YY)
 	daddu	YY, YY, INCY
-	.align 3
+	.align 5
 
 .L25:
 	andi	I,  N, 7
@@ -432,7 +435,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	bgtz	I, .L26
 	daddu	Y, Y, INCY
-	.align 3
+	.align 5
 
 .L999:
 #ifndef __64BIT__