Fix ARMV9SME target in DYNAMIC_ARCH and add SME query code for MacOS (#5222)

* Fix ARMV9SME target and add support_sme1 code for MacOS
* make sgemm_direct unconditionally available on all arm64
* build a (dummy) sgemm_direct kernel on all arm64





* Update dynamic_arm64.c

common_param.h

@@ -224,10 +224,8 @@ BLASLONG (*ismin_k) (BLASLONG, float *, BLASLONG);
   int  (*sgemm_direct_performant) (BLASLONG M, BLASLONG N, BLASLONG K);
 #endif
 #ifdef ARCH_ARM64
-#ifdef HAVE_SME
   void (*sgemm_direct) (BLASLONG, BLASLONG, BLASLONG, float *, BLASLONG , float *, BLASLONG , float * , BLASLONG);
 #endif
-#endif
 
   
   int    (*sgemm_kernel   )(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG);

driver/others/dynamic_arm64.c

@@ -43,6 +43,14 @@
 #include <sys/auxv.h>
 #endif
 
+#ifdef __APPLE__
+#include <sys/sysctl.h>
+int32_t value;
+size_t length=sizeof(value);
+int64_t value64;
+size_t length64=sizeof(value64);
+#endif
+
 extern gotoblas_t  gotoblas_ARMV8;
 #ifdef DYNAMIC_LIST
 #ifdef DYN_CORTEXA53
@@ -120,7 +128,7 @@ extern gotoblas_t gotoblas_ARMV9SME;
 #else
 #define gotoblas_ARMV9SME gotoblas_ARMV8
 #endif
-#ifdef DYN_CORTEX_A55
+#ifdef DYN_CORTEXA55
 extern gotoblas_t  gotoblas_CORTEXA55;
 #else
 #define gotoblas_CORTEXA55 gotoblas_ARMV8
@@ -147,17 +155,17 @@ extern gotoblas_t  gotoblas_NEOVERSEV1;
 extern gotoblas_t  gotoblas_NEOVERSEN2;
 extern gotoblas_t  gotoblas_ARMV8SVE;
 extern gotoblas_t  gotoblas_A64FX;
+#ifndef NO_SME
+extern gotoblas_t  gotoblas_ARMV9SME;
+#else
+#define gotoblas_ARMV9SME gotoblas_ARMV8SVE
+#endif
 #else
 #define gotoblas_NEOVERSEV1 gotoblas_ARMV8
 #define gotoblas_NEOVERSEN2 gotoblas_ARMV8
 #define gotoblas_ARMV8SVE   gotoblas_ARMV8
 #define gotoblas_A64FX      gotoblas_ARMV8
-#endif
-
-#ifndef NO_SME
-extern gotoblas_t  gotoblas_ARMV9SME;
-#else
-#define gotoblas_ARMV9SME gotoblas_ARMV8SVE
+#define gotoblas_ARMV9SME   gotoblas_ARMV8
 #endif
 
 extern gotoblas_t  gotoblas_THUNDERX3T110;
@@ -168,7 +176,7 @@ extern void openblas_warning(int verbose, const char * msg);
 #define FALLBACK_VERBOSE 1
 #define NEOVERSEN1_FALLBACK "OpenBLAS : Your OS does not support SVE instructions. OpenBLAS is using Neoverse N1 kernels as a fallback, which may give poorer performance.\n"
 
-#define NUM_CORETYPES   18
+#define NUM_CORETYPES   19
 
 /*
  * In case asm/hwcap.h is outdated on the build system, make sure
@@ -207,6 +215,7 @@ static char *corename[] = {
   "cortexa55",
   "armv8sve",
   "a64fx",
+  "armv9sme",
   "unknown"
 };
 
@@ -229,6 +238,7 @@ char *gotoblas_corename(void) {
   if (gotoblas == &gotoblas_CORTEXA55)    return corename[15];
   if (gotoblas == &gotoblas_ARMV8SVE)     return corename[16];
   if (gotoblas == &gotoblas_A64FX)        return corename[17];
+  if (gotoblas == &gotoblas_ARMV9SME)     return corename[18];
   return corename[NUM_CORETYPES];
 }
 
@@ -266,6 +276,7 @@ static gotoblas_t *force_coretype(char *coretype) {
     case 15: return (&gotoblas_CORTEXA55);
     case 16: return (&gotoblas_ARMV8SVE);
     case 17: return (&gotoblas_A64FX);
+    case 18: return (&gotoblas_ARMV9SME);
   }
   snprintf(message, 128, "Core not found: %s\n", coretype);
   openblas_warning(1, message);
@@ -277,6 +288,11 @@ static gotoblas_t *get_coretype(void) {
   char coremsg[128];
 
 #if defined (OS_DARWIN)
+//future	#if !defined(NO_SME)
+//  		if (support_sme1()) {
+//    			return &gotoblas_ARMV9SME;
+//		  }
+//		#endif
   return &gotoblas_NEOVERSEN1;
 #endif
 	
@@ -439,6 +455,7 @@ static gotoblas_t *get_coretype(void) {
       }
       break;
     case 0x61: // Apple
+//future	if (support_sme1()) return &gotoblas_ARMV9SME;
 	return &gotoblas_NEOVERSEN1;
       break;
     default:
@@ -446,8 +463,8 @@ static gotoblas_t *get_coretype(void) {
       openblas_warning(1, coremsg);
   }
 
-#if !defined(NO_SME) && defined(HWCAP2_SME)
-  if ((getauxval(AT_HWCAP2) & HWCAP2_SME)) {
+#if !defined(NO_SME)
+  if (support_sme1()) {
     return &gotoblas_ARMV9SME;
   }
 #endif
@@ -511,6 +528,10 @@ int support_sme1(void) {
         if(getauxval(AT_HWCAP2) & HWCAP2_SME){
                 ret = 1;
         }
+#endif
+#if defined(__APPLE__)
+	sysctlbyname("hw.optional.arm.FEAT_SME",&value64,&length64,NULL,0);
+	ret = value64;
 #endif
        return ret;
 }

kernel/CMakeLists.txt

@@ -208,7 +208,7 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
       set(USE_TRMM true)
     endif ()
     set(USE_DIRECT_SGEMM false)
-    if (X86_64 OR (ARM64 AND (UC_TARGET_CORE MATCHES ARMV9SME)))
+    if (X86_64 OR ARM64)
        set(USE_DIRECT_SGEMM true)
     endif()
 
@@ -225,9 +225,11 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
 	  set (SGEMMDIRECTSMEKERNEL sgemm_direct_sme1.S)
 	  set (SGEMMDIRECTPREKERNEL sgemm_direct_sme1_preprocess.S)
 	  GenerateNamedObjects("${KERNELDIR}/${SGEMMDIRECTKERNEL}" "" "gemm_direct" false "" "" false SINGLE)
+   	  if (HAVE_SME)
 	  GenerateNamedObjects("${KERNELDIR}/${SGEMMDIRECTSMEKERNEL}" "" "gemm_direct_sme1" false "" "" false SINGLE)
 	  GenerateNamedObjects("${KERNELDIR}/${SGEMMDIRECTPREKERNEL}" "" "gemm_direct_sme1_preprocess" false "" "" false SINGLE)
 	  endif ()
+   	  endif ()
     endif()
 
     foreach (float_type SINGLE DOUBLE)

kernel/Makefile.L3

@@ -103,8 +103,8 @@ endif
 ifeq ($(ARCH), arm64)
 ifeq ($(TARGET_CORE), ARMV9SME)
 HAVE_SME = 1
-SGEMMDIRECTKERNEL = sgemm_direct_arm64_sme1.c
 endif
+SGEMMDIRECTKERNEL = sgemm_direct_arm64_sme1.c
 endif
 endif
 endif
@@ -143,9 +143,10 @@ SKERNELOBJS += \
 	sgemm_direct_performant$(TSUFFIX).$(SUFFIX)
 endif
 ifeq ($(ARCH), arm64)
+SKERNELOBJS += \
+	sgemm_direct$(TSUFFIX).$(SUFFIX)
 ifdef HAVE_SME
 SKERNELOBJS += \
-	sgemm_direct$(TSUFFIX).$(SUFFIX) \
 	sgemm_direct_sme1$(TSUFFIX).$(SUFFIX) \
 	sgemm_direct_sme1_preprocess$(TSUFFIX).$(SUFFIX)
 endif
@@ -835,9 +836,9 @@ $(KDIR)sgemm_direct$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMMDIRECTKERNEL)
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
 endif
 ifeq ($(ARCH), arm64)
-ifdef HAVE_SME
 $(KDIR)sgemm_direct$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMMDIRECTKERNEL)
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+ifdef HAVE_SME
 $(KDIR)sgemm_direct_sme1$(TSUFFIX).$(SUFFIX) : 
 	$(CC) $(CFLAGS) -c $(KERNELDIR)/sgemm_direct_sme1.S -UDOUBLE -UCOMPLEX  -o $@
 $(KDIR)sgemm_direct_sme1_preprocess$(TSUFFIX).$(SUFFIX) : 

kernel/arm64/sgemm_direct_arm64_sme1.c

@@ -71,4 +71,10 @@ void CNAME (BLASLONG M, BLASLONG N, BLASLONG K, float * __restrict A,\
         free(A_mod);
 }
 
+#else
+
+void CNAME (BLASLONG M, BLASLONG N, BLASLONG K, float * __restrict A,\
+            BLASLONG strideA, float * __restrict B, BLASLONG strideB ,\
+            float * __restrict R, BLASLONG strideR){}
+ 
 #endif

kernel/setparam-ref.c

@@ -180,9 +180,7 @@ gotoblas_t TABLE_NAME = {
   sgemm_direct_performantTS,	
 #endif
 #ifdef ARCH_ARM64
-#ifdef HAVE_SME
   sgemm_directTS,
-#endif
 #endif
 
   sgemm_kernelTS, sgemm_betaTS,