add v2x8 kernel + fix sve dtrmm

kernel/arm64/KERNEL.A64FX

@@ -143,34 +143,22 @@ endif
 SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
 SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
-DGEMMKERNEL    =  dgemm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N).S
-DTRMMKERNEL    =  dtrmm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N).S
 
-ifneq ($(DGEMM_UNROLL_M), $(DGEMM_UNROLL_N))
+DGEMMKERNEL    =  dgemm_kernel_sve_v2x$(DGEMM_UNROLL_N).S
+DTRMMKERNEL    =  dtrmm_kernel_sve_v1x$(DGEMM_UNROLL_N).S
 
-ifeq ($(DGEMM_UNROLL_M), 8)
-DGEMMINCOPY    =  dgemm_ncopy_$(DGEMM_UNROLL_M).S
-DGEMMITCOPY    =  dgemm_tcopy_$(DGEMM_UNROLL_M).S
-else
-DGEMMINCOPY    =  ../generic/gemm_ncopy_$(DGEMM_UNROLL_M).c
-DGEMMITCOPY    =  ../generic/gemm_tcopy_$(DGEMM_UNROLL_M).c
-endif
-
-DGEMMINCOPYOBJ =  dgemm_incopy$(TSUFFIX).$(SUFFIX)
-DGEMMITCOPYOBJ =  dgemm_itcopy$(TSUFFIX).$(SUFFIX)
-endif
-
-ifeq ($(DGEMM_UNROLL_N), 4)
+DGEMMINCOPY    =  dgemm_ncopy_sve_v1.c
+DGEMMITCOPY    =  dgemm_tcopy_sve_v1.c
 DGEMMONCOPY    =  dgemm_ncopy_$(DGEMM_UNROLL_N).S
 DGEMMOTCOPY    =  dgemm_tcopy_$(DGEMM_UNROLL_N).S
-else
-DGEMMONCOPY    =  ../generic/gemm_ncopy_$(DGEMM_UNROLL_N).c
-DGEMMOTCOPY    =  ../generic/gemm_tcopy_$(DGEMM_UNROLL_N).c
-endif
 
+DGEMMINCOPYOBJ =  dgemm_incopy$(TSUFFIX).$(SUFFIX)
+DGEMMITCOPYOBJ =  dgemm_itcopy$(TSUFFIX).$(SUFFIX)
 DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
 DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
+
+
 CGEMMKERNEL    =  cgemm_kernel_$(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N).S
 CTRMMKERNEL    =  ctrmm_kernel_$(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N).S
 ifneq ($(CGEMM_UNROLL_M), $(CGEMM_UNROLL_N))

kernel/arm64/dtrmm_kernel_sve_v1x8.S

@@ -344,21 +344,21 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
 
 	add	pCRow1, pCRow0, LDC
-    fmla z16.d, p1/m, z16.d, alphaZ
+    fmul z16.d, p1/m, z16.d, alphaZ
     st1d  z16.d, p1, [pCRow0]
 	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
 
 	add	pCRow2, pCRow1, LDC
-    fmla z17.d, p1/m, z17.d, alphaZ
+    fmul z17.d, p1/m, z17.d, alphaZ
     st1d  z17.d, p1, [pCRow1]
 	prfm	PLDL2KEEP, [pCRow2, #C_PRE_SIZE]
 
 	add	pCRow1, pCRow2, LDC
-    fmla z18.d, p1/m, z18.d, alphaZ
+    fmul z18.d, p1/m, z18.d, alphaZ
     st1d z18.d, p1, [pCRow2]
 	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
 
-    fmla z19.d, p1/m, z19.d, alphaZ
+    fmul z19.d, p1/m, z19.d, alphaZ
     st1d  z19.d, p1, [pCRow1]
 
 	add	pCRow0, pCRow0, lanes, lsl #3	// pC = pC + lanes  * 8
@@ -392,11 +392,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
 
 	add	pCRow1, pCRow0, LDC
-    fmla z16.d, p1/m, z16.d, alphaZ
+    fmul z16.d, p1/m, z16.d, alphaZ
     st1d  z16.d, p1, [pCRow0]
 	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
 
-    fmla z17.d, p1/m, z17.d, alphaZ
+    fmul z17.d, p1/m, z17.d, alphaZ
     st1d  z17.d, p1, [pCRow1]
 
 	add	pCRow0, pCRow0, lanes, lsl #3	// pC = pC + lanes  * 8
@@ -426,7 +426,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
 
-    fmla z16.d, p1/m, z16.d, alphaZ
+    fmul z16.d, p1/m, z16.d, alphaZ
     st1d  z16.d, p1, [pCRow0]
 
 

param.h

@@ -3328,8 +3328,8 @@ is a big desktop or server with abundant cache rather than a phone or embedded d
 #define SGEMM_DEFAULT_UNROLL_M  16
 #define SGEMM_DEFAULT_UNROLL_N  4
 
-#define DGEMM_DEFAULT_UNROLL_M  8
-#define DGEMM_DEFAULT_UNROLL_N  4
+#define DGEMM_DEFAULT_UNROLL_M  4
+#define DGEMM_DEFAULT_UNROLL_N  8
 
 #define CGEMM_DEFAULT_UNROLL_M  8
 #define CGEMM_DEFAULT_UNROLL_N  4