Improvements to GEMV kernels

kernel/arm64/gemv_n.S

@@ -68,6 +68,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define SHZ	3
 #endif
 
+#define A_PRE_SIZE 768
+#define Y_PRE_SIZE 768
+
 /******************************************************************************/
 
 .macro SAVE_REGS
@@ -105,36 +108,42 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	ld1	{v2.4s, v3.4s}, [A_PTR], #32
 	ld1	{v4.4s, v5.4s}, [Y_IPTR], #32
 	fmla	v4.4s, v1.4s, v2.4s
+	prfm	PLDL1KEEP, [A_PTR, #A_PRE_SIZE]
 	fmla	v5.4s, v1.4s, v3.4s
 	st1	{v4.4s, v5.4s}, [Y_OPTR], #32
 
 	ld1	{v6.4s, v7.4s}, [A_PTR], #32
 	ld1	{v8.4s, v9.4s}, [Y_IPTR], #32
 	fmla	v8.4s, v1.4s, v6.4s
+	prfm	PLDL1KEEP, [Y_IPTR, #Y_PRE_SIZE]
 	fmla	v9.4s, v1.4s, v7.4s
 	st1	{v8.4s, v9.4s}, [Y_OPTR], #32
 #else //DOUBLE
 	ld1	{v2.2d, v3.2d}, [A_PTR], #32
 	ld1	{v4.2d, v5.2d}, [Y_IPTR], #32
 	fmla	v4.2d, v1.2d, v2.2d
+	prfm	PLDL1KEEP, [A_PTR, #A_PRE_SIZE]
 	fmla	v5.2d, v1.2d, v3.2d
 	st1	{v4.2d, v5.2d}, [Y_OPTR], #32
 
 	ld1	{v6.2d, v7.2d}, [A_PTR], #32
 	ld1	{v8.2d, v9.2d}, [Y_IPTR], #32
 	fmla	v8.2d, v1.2d, v6.2d
+	prfm	PLDL1KEEP, [Y_IPTR, #Y_PRE_SIZE]
 	fmla	v9.2d, v1.2d, v7.2d
 	st1	{v8.2d, v9.2d}, [Y_OPTR], #32
 
 	ld1	{v10.2d, v11.2d}, [A_PTR], #32
 	ld1	{v12.2d, v13.2d}, [Y_IPTR], #32
 	fmla	v12.2d, v1.2d, v10.2d
+	prfm	PLDL1KEEP, [A_PTR, #A_PRE_SIZE]
 	fmla	v13.2d, v1.2d, v11.2d
 	st1	{v12.2d, v13.2d}, [Y_OPTR], #32
 
 	ld1	{v14.2d, v15.2d}, [A_PTR], #32
 	ld1	{v16.2d, v17.2d}, [Y_IPTR], #32
 	fmla	v16.2d, v1.2d, v14.2d
+	prfm	PLDL1KEEP, [Y_IPTR, #Y_PRE_SIZE]
 	fmla	v17.2d, v1.2d, v15.2d
 	st1	{v16.2d, v17.2d}, [Y_OPTR], #32
 #endif