Added scaffolding for Oryon arch as in SD X Elite

Using NeoverseN1 Kernels for now with cache info
taken from official specs.

Author: theAeon

Makefile.arm64

@@ -224,6 +224,21 @@ endif
 endif
 endif
 
+# Detect Qualcomm Oryon.
+ifeq ($(CORE), ORYON)
+ifeq (1, $(filter 1,$(GCCVERSIONGTEQ15) $(ISCLANG)))
+CCOMMON_OPT += -mcpu=oryon-1
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -mcpu=oryon-1
+endif
+else
+CCOMMON_OPT += -march=armv8.7-a -mtune=neoverse-n1
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8.7-a -mtune=neoverse-n1
+endif
+endif
+endif
+
 # Detect Ampere AmpereOne(ampere1,ampere1a) processors.
 ifeq ($(CORE), AMPERE1)
 ifeq (1, $(filter 1,$(GCCVERSIONGTEQ12) $(ISCLANG)))

Makefile.system

@@ -713,6 +713,7 @@ DYNAMIC_CORE =  ARMV8
 DYNAMIC_CORE += CORTEXA53
 DYNAMIC_CORE += CORTEXA57
 DYNAMIC_CORE += NEOVERSEN1
+DYNAMIC_CORE += ORYON
 ifneq ($(NO_SVE), 1)
 DYNAMIC_CORE += NEOVERSEV1
 DYNAMIC_CORE += NEOVERSEN2

TargetList.txt

@@ -115,6 +115,7 @@ A64FX
 ARMV8SVE
 ARMV9SME
 FT2000
+ORYON
 
 9.System Z:
 ZARCH_GENERIC

cmake/arch.cmake

@@ -33,7 +33,7 @@ endif ()
 
 if (DYNAMIC_ARCH)
   if (ARM64)
-    set(DYNAMIC_CORE ARMV8 CORTEXA53 CORTEXA57 THUNDERX THUNDERX2T99 TSV110 EMAG8180 NEOVERSEN1 THUNDERX3T110)
+    set(DYNAMIC_CORE ARMV8 CORTEXA53 CORTEXA57 THUNDERX THUNDERX2T99 TSV110 EMAG8180 NEOVERSEN1 THUNDERX3T110 ORYON)
     if (${CMAKE_C_COMPILER_ID} STREQUAL "GNU")
       if (${CMAKE_C_COMPILER_VERSION} VERSION_GREATER_EQUAL 10) # SVE ACLE supported in GCC >= 10
         set(DYNAMIC_CORE ${DYNAMIC_CORE} NEOVERSEV1 NEOVERSEN2 ARMV8SVE A64FX)

cmake/cc.cmake

@@ -285,6 +285,18 @@ if (${CORE} STREQUAL NEOVERSEN1)
   endif ()
 endif ()
 
+if (${CORE} STREQUAL ORYON)
+  if (NOT DYNAMIC_ARCH)
+    if (${GCC_VERSION} VERSION_GREATER 15.0 OR ${GCC_VERSION} VERSION_EQUAL 15.0)
+        set (CCOMMON_OPT  "${CCOMMON_OPT} -march=armv8.7-a -mtune=oryon-1")
+    elseif (${GCC_VERSION} VERSION_GREATER 9.4 OR ${GCC_VERSION} VERSION_EQUAL 9.4)
+      set (CCOMMON_OPT  "${CCOMMON_OPT} -march=armv8.7-a -mtune=neoverse-n1")
+    else ()
+      set (CCOMMON_OPT "${CCOMMON_OPT} -march=armv8.7-a")
+    endif ()
+  endif ()
+endif ()
+
 if (${CORE} STREQUAL AMPEREONE)
   if (NOT DYNAMIC_ARCH)
     if (${CMAKE_C_COMPILER_ID} STREQUAL "NVHPC")

cmake/prebuild.cmake

@@ -992,6 +992,37 @@ endif ()
     set(ZGEMM_UNROLL_M 4)
     set(ZGEMM_UNROLL_N 4)
     set(SYMV_P 16)
+  elseif ("${TCORE}" STREQUAL "ORYON")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define L1_CODE_SIZE\t196608\n"
+      "#define L1_CODE_LINESIZE\t64\n"
+      "#define L1_CODE_ASSOCIATIVE\t6\n"
+      "#define L1_DATA_SIZE\t98304\n"
+      "#define L1_DATA_LINESIZE\t64\n"
+      "#define L1_DATA_ASSOCIATIVE\t6\n"
+      "#define L2_SIZE\t12582912\n\n"
+      "#define L2_LINESIZE\t32\n"
+      "#define L2_ASSOCIATIVE\t12\n"
+      "#define ITB_SIZE\t4096\n"
+      "#define ITB_DEFAULT_ENTRIES\t256\n"
+      "#define ITB_ASSOCIATIVE\t8\n"
+      "#define DTB_ASSOCIATIVE\t7\n"
+      "#define DTB_DEFAULT_ENTRIES\t48\n"
+      "#define DTB_SIZE\t4096\n"
+      "#define HAVE_VFPV4\n"
+      "#define HAVE_VFPV3\n"
+      "#define HAVE_VFP\n"
+      "#define HAVE_NEON\n"
+      "#define ARMV8\n")
+    set(SGEMM_UNROLL_M 16)
+    set(SGEMM_UNROLL_N 4)
+    set(DGEMM_UNROLL_M 8)
+    set(DGEMM_UNROLL_N 4)
+    set(CGEMM_UNROLL_M 8)
+    set(CGEMM_UNROLL_N 4)
+    set(ZGEMM_UNROLL_M 4)
+    set(ZGEMM_UNROLL_N 4)
+    set(SYMV_P 16)
   elseif ("${TCORE}" STREQUAL "NEOVERSEN1")
     file(APPEND ${TARGET_CONF_TEMP}
       "#define L1_CODE_SIZE\t65536\n"

common_arm64.h

@@ -175,7 +175,7 @@ static inline int blas_quickdivide(blasint x, blasint y){
 #define HUGE_PAGESIZE   ( 4 << 20)
 
 #ifndef BUFFERSIZE
-#if defined(NEOVERSEN1) || defined(NEOVERSEN2) || defined(NEOVERSEV1) || defined(A64FX) || defined(ARMV8SVE) || defined(ARMV9SME)
+#if defined(NEOVERSEN1) || defined(NEOVERSEN2) || defined(NEOVERSEV1) || defined(A64FX) || defined(ORYON) || defined(ARMV8SVE) || defined(ARMV9SME)
 #define BUFFER_SIZE     (32 << 22)
 #else
 #define BUFFER_SIZE     (32 << 20)

cpuid_arm64.c

@@ -71,6 +71,7 @@ size_t length64=sizeof(value64);
 #define CPU_CORTEXA710    21
 // Qualcomm
 #define CPU_FALKOR        6
+#define CPU_ORYON         26
 // Cavium
 #define CPU_THUNDERX      7
 #define CPU_THUNDERX2T99  8
@@ -113,7 +114,8 @@ static char *cpuname[] = {
   "FT2000",
   "CORTEXA76",
   "NEOVERSEV2",
-  "AMPERE1"
+  "AMPERE1",
+  "ORYON"
 };
 
 static char *cpuname_lower[] = {
@@ -143,7 +145,7 @@ static char *cpuname_lower[] = {
   "cortexa76",
   "neoversev2",
   "ampere1",
-  "ampere1a"
+  "oryon"
 };
 
 static int cpulowperf=0;
@@ -325,6 +327,8 @@ int detect(void)
     // Qualcomm
     else if (strstr(cpu_implementer, "0x51") && strstr(cpu_part, "0xc00"))
       return CPU_FALKOR;
+    else if (strstr(cpu_implementer, "0x51") && strstr(cpu_part, "0x001"))
+      return CPU_ORYON;
     // Cavium
     else if (strstr(cpu_implementer, "0x43") && strstr(cpu_part, "0x0a1"))
 			return CPU_THUNDERX;
@@ -418,7 +422,7 @@ int detect(void)
 	if (errcode != ERROR_SUCCESS) wprintf(L"Error reading cpuname from registry:%x\n",errcode);
 //wprintf(stderr,L"%s\n",(PWSTR)valstring);
 	RegCloseKey(reghandle);
-	if (strstr(valstring, "Snapdragon(R) X Elite")) return CPU_NEOVERSEN1;
+	if (strstr(valstring, "Snapdragon(R) X Elite")) return CPU_ORYON;
 	if (strstr(valstring, "Ampere(R) Altra")) return CPU_NEOVERSEN1;
 	if (strstr(valstring, "Snapdragon (TM) 8cx Gen 3")) return CPU_CORTEXX1;
 	if (strstr(valstring, "Snapdragon Compute Platform")) return CPU_CORTEXX1;
@@ -765,6 +769,24 @@ void get_cpuconfig(void)
 	    	printf("#define DTB_DEFAULT_ENTRIES 64\n");
 	    	printf("#define DTB_SIZE 4096\n");
 		break;
+	    case CPU_ORYON:
+		printf("#define ORYON\n");
+		printf("#define L1_CODE_SIZE 196608\n");
+		printf("#define L1_CODE_LINESIZE 64\n");
+		printf("#define L1_CODE_ASSOCIATIVE 6\n");
+		printf("#define L1_DATA_SIZE 98304\n");
+		printf("#define L1_DATA_LINESIZE 64\n");
+		printf("#define L1_DATA_ASSOCIATIVE 6\n");
+		printf("#define L2_SIZE 12582912\n");
+		printf("#define L2_LINESIZE 32\n");
+		printf("#define L2_ASSOCIATIVE 12\n");
+		printf("#define ITB_SIZE 4096\n");
+		printf("#define ITB_ASSOCIATIVE 8\n");
+		printf("#define ITB_DEFAULT_ENTRIES 256\n");
+		printf("#define DTB_DEFAULT_ENTRIES 224\n");
+		printf("#define DTB_ASSOCIATIVE 7\n");
+		printf("#define DTB_SIZE 4096\n");
+		break;
 	}
 	get_cpucount();
 }

getarch.c

@@ -1449,20 +1449,21 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CORENAME  "CORTEXA710"
 #endif
 
-#ifdef FORCE_NEOVERSEN1
+#ifdef FORCE_ORYON
 #define FORCE
 #define ARCHITECTURE    "ARM64"
-#define SUBARCHITECTURE "NEOVERSEN1"
+#define SUBARCHITECTURE "ORYON"
 #define SUBDIRNAME      "arm64"
-#define ARCHCONFIG   "-DNEOVERSEN1 " \
-       "-DL1_CODE_SIZE=65536 -DL1_CODE_LINESIZE=64 -DL1_CODE_ASSOCIATIVE=4 " \
-       "-DL1_DATA_SIZE=65536 -DL1_DATA_LINESIZE=64 -DL1_DATA_ASSOCIATIVE=4 " \
-       "-DL2_SIZE=1048576 -DL2_LINESIZE=64 -DL2_ASSOCIATIVE=16 " \
-       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+#define ARCHCONFIG   "-DORYON " \
+       "-DL1_CODE_SIZE=196608 -DL1_CODE_LINESIZE=64 -DL1_CODE_ASSOCIATIVE=6 " \
+       "-DL1_DATA_SIZE=98304 -DL1_DATA_LINESIZE=64 -DL1_DATA_ASSOCIATIVE=6 " \
+       "-DL2_SIZE=12582912 -DL2_LINESIZE=32 -DL2_ASSOCIATIVE=12 " \
+       "-DDTB_DEFAULT_ENTRIES=224 -DDTB_ASSOCIATIVE=7 -DDTB_SIZE=4096 " \
+       "-DITB_DEFAULT_ENTRIES=256 -DITB_ASSOCIATIVE=8 -DDTB_SIZE=4096 " \
        "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8 " \
-       "-march=armv8.2-a -mtune=neoverse-n1"
-#define LIBNAME   "neoversen1"
-#define CORENAME  "NEOVERSEN1"
+       "-march=armv8.7-a -mtune=oryon-1"
+#define LIBNAME   "oryon"
+#define CORENAME  "ORYON"
 #endif
 
 #ifdef FORCE_NEOVERSEV1
@@ -1481,7 +1482,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CORENAME  "NEOVERSEV1"
 #endif
 
-
 #ifdef FORCE_NEOVERSEN2
 #define FORCE
 #define ARCHITECTURE    "ARM64"

kernel/arm64/KERNEL.ORYON

@@ -0,0 +1 @@
+include $(KERNELDIR)/KERNEL.NEOVERSEN1