THUNDERX2T99: Bug Fixes in D/Z NRM2 and ZGEMM

kernel/arm64/dznrm2_thunderx2t99.c

@@ -301,7 +301,7 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
 	: "cc",
 	  "memory",
 	  "x0", "x1", "x2", "x3", "x4", "x5",
-	  "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7"
+	  "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "d8"
 	);
 
 }

kernel/arm64/zgemm_kernel_4x4_thunderx2t99.S

@@ -474,19 +474,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	ld2	{v2.2d, v3.2d}, [pA]
 	add	pA, pA, #32
 
-	OP_rr	v20.2d, v0.2d, v8.d[0]
-	OP_ii	v20.2d, v1.2d, v8.d[1]
-	OP_ri	v21.2d, v0.2d, v8.d[1]
-	OP_ir	v21.2d, v1.2d, v8.d[0]
+	OP_rr	v20.2d, v0.2d, v9.d[0]
+	OP_ii	v20.2d, v1.2d, v9.d[1]
+	OP_ri	v21.2d, v0.2d, v9.d[1]
+	OP_ir	v21.2d, v1.2d, v9.d[0]
 
 	ldr	q10, [pB]
 	ldr	q11, [pB, #16]
 	add	pB, pB, #32
 
-	OP_rr	v18.2d, v2.2d, v9.d[0]
-	OP_ii	v18.2d, v3.2d, v9.d[1]
-	OP_ri	v19.2d, v2.2d, v9.d[1]
-	OP_ir	v19.2d, v3.2d, v9.d[0]
+	OP_rr	v18.2d, v2.2d, v8.d[0]
+	OP_ii	v18.2d, v3.2d, v8.d[1]
+	OP_ri	v19.2d, v2.2d, v8.d[1]
+	OP_ir	v19.2d, v3.2d, v8.d[0]
 
 	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]