Change DGEMM tiling size from 8x8 to 4x8 for the RISC-V ZVL256B architecture

Testing has shown that 4x8 or 4x4 performs better than the original 8x8 tiling
size for this kernel.

As we do not wish to perturb the behaviour of the DTRMM kernel at this point,
the DTRMM tiling size is explicitly set to the original 8x8.

Author: Kwok Cheung Yeung

cmake/prebuild.cmake

@@ -1475,7 +1475,7 @@ endif ()
       "#define L2_ASSOCIATIVE 4\n")
     set(SGEMM_UNROLL_M 16)
     set(SGEMM_UNROLL_N 8)
-    set(DGEMM_UNROLL_M 8)
+    set(DGEMM_UNROLL_M 4)
     set(DGEMM_UNROLL_N 8)
     set(CGEMM_UNROLL_M 8)
     set(CGEMM_UNROLL_N 8)

param.h

@@ -3231,7 +3231,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define SGEMM_DEFAULT_UNROLL_M  16
 #define SGEMM_DEFAULT_UNROLL_N  8
 
-#define DGEMM_DEFAULT_UNROLL_M  8
+#define DGEMM_DEFAULT_UNROLL_M  4
 #define DGEMM_DEFAULT_UNROLL_N  8
 
 #define CGEMM_DEFAULT_UNROLL_M  8
@@ -3240,6 +3240,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ZGEMM_DEFAULT_UNROLL_M  8
 #define ZGEMM_DEFAULT_UNROLL_N  4
 
+#define DTRMM_DEFAULT_UNROLL_M  8
+#define DTRMM_DEFAULT_UNROLL_N  8
+
 #undef SHGEMM_DEFAULT_P
 #define SHGEMM_DEFAULT_P 128
 #undef SBGEMM_DEFAULT_P