Merge pull request OpenMathLib#5353 from nakagawa-fj/feature/gemm_divide_rate_for_A64FX

Multi-thread Performance Improvement of GEMM with DIVIDE_RATE=1 for A64FX

Author: martin-frbg

driver/level3/gemm.c

@@ -59,6 +59,10 @@
 #define GEMM_Q 128
 #endif
 
+#ifdef GEMM_DIVIDE_RATE
+#define DIVIDE_RATE GEMM_DIVIDE_RATE
+#endif
+
 #ifdef THREADED_LEVEL3
 #include "level3_thread.c"
 #else

param.h

@@ -3701,6 +3701,8 @@ is a big desktop or server with abundant cache rather than a phone or embedded d
 
 #elif defined(A64FX) // 512-bit SVE
 
+#define GEMM_DIVIDE_RATE  1
+
 #if defined(XDOUBLE) || defined(DOUBLE)
 #define GEMM_PREFERED_SIZE  8
 #else