Merge pull request #5309 from davidz-ampere/dev-ampereone

Add support for Ampere AmpereOne processors

Makefile.arm64

@@ -191,6 +191,16 @@ endif
 endif
 endif
 
+# Detect Ampere AmpereOne(ampere1,ampere1a) processors.
+ifeq ($(CORE), AMPERE1)
+ifeq (1, $(filter 1,$(GCCVERSIONGTEQ12) $(ISCLANG)))
+CCOMMON_OPT += -march=armv8.6-a+crypto+crc+fp16+sha3+rng
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8.6-a+crypto+crc+fp16+sha3+rng
+endif
+endif
+endif
+
 # Use a53 tunings because a55 is only available in GCC>=8.1
 ifeq ($(CORE), CORTEXA55)
 ifeq (1, $(filter 1,$(GCCVERSIONGTEQ7) $(ISCLANG)))

Makefile.system

@@ -393,6 +393,8 @@ GCCVERSIONGTEQ9 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 9)
 GCCVERSIONGTEQ10 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 10)
 GCCVERSIONGTEQ11 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 11)
 GCCVERSIONGTEQ12 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 12)
+GCCVERSIONGTEQ13 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 13)
+GCCVERSIONGTEQ14 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 14)
 # Note that the behavior of -dumpversion is compile-time-configurable for
 # gcc-7.x and newer. Use -dumpfullversion there
 ifeq ($(GCCVERSIONGTEQ7),1)

cpuid_arm64.c

@@ -79,6 +79,7 @@ size_t length64=sizeof(value64);
 #define CPU_TSV110        9
 // Ampere
 #define CPU_EMAG8180	 10
+#define CPU_AMPERE1      25
 // Apple
 #define CPU_VORTEX       13
 // Fujitsu
@@ -111,7 +112,8 @@ static char *cpuname[] = {
   "CORTEXA710",
   "FT2000",
   "CORTEXA76",
-  "NEOVERSEV2"
+  "NEOVERSEV2",
+  "AMPERE1"
 };
 
 static char *cpuname_lower[] = {
@@ -139,7 +141,9 @@ static char *cpuname_lower[] = {
   "cortexa710",
   "ft2000",
   "cortexa76",
-  "neoversev2"
+  "neoversev2",
+  "ampere1",
+  "ampere1a"
 };
 
 static int cpulowperf=0;
@@ -334,6 +338,10 @@ int detect(void)
     // Ampere
     else if (strstr(cpu_implementer, "0x50") && strstr(cpu_part, "0x000"))
                         return CPU_EMAG8180;
+    else if (strstr(cpu_implementer, "0xc0")) {
+        if (strstr(cpu_part, "0xac3") || strstr(cpu_part, "0xac4"))
+            return CPU_AMPERE1;
+    }
     // Fujitsu
     else if (strstr(cpu_implementer, "0x46") && strstr(cpu_part, "0x001"))
                         return CPU_A64FX;
@@ -684,6 +692,21 @@ void get_cpuconfig(void)
 	    	printf("#define DTB_SIZE 4096\n");
 		break;
 
+	    case CPU_AMPERE1:
+		printf("#define %s\n", cpuname[d]);
+		printf("#define L1_CODE_SIZE 16384\n");
+		printf("#define L1_CODE_LINESIZE 64\n");
+		printf("#define L1_CODE_ASSOCIATIVE 4\n");
+		printf("#define L1_DATA_SIZE 65536\n");
+		printf("#define L1_DATA_LINESIZE 64\n");
+		printf("#define L1_DATA_ASSOCIATIVE 4\n");
+		printf("#define L2_SIZE 2097152\n");
+		printf("#define L2_LINESIZE 64\n");
+		printf("#define L2_ASSOCIATIVE 8\n");
+		printf("#define DTB_DEFAULT_ENTRIES 64\n");
+		printf("#define DTB_SIZE 4096\n");
+		break;
+
 	    case CPU_THUNDERX3T110:
 		printf("#define THUNDERX3T110                 \n");
 		printf("#define L1_CODE_SIZE         65536    \n");

getarch.c

@@ -158,6 +158,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /* #define FORCE_CSKY		*/
 /* #define FORCE_CK860FV        */
 /* #define FORCE_GENERIC	*/
+/* #define FORCE_AMPERE1	*/
 
 #ifdef FORCE_P2
 #define FORCE
@@ -1590,6 +1591,22 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CORENAME  "EMAG8180"
 #endif
 
+#ifdef FORCE_AMPERE1
+#define FORCE
+#define ARCHITECTURE    "ARM64"
+#define SUBARCHITECTURE "AMPERE1"
+#define SUBDIRNAME      "arm64"
+#define ARCHCONFIG   "-DAMPERE1 " \
+       "-DL1_CODE_SIZE=16384 -DL1_CODE_LINESIZE=64 -DL1_CODE_ASSOCIATIVE=4 " \
+       "-DL1_DATA_SIZE=65536 -DL1_DATA_LINESIZE=64 -DL1_DATA_ASSOCIATIVE=4 " \
+       "-DL2_SIZE=2097152 -DL2_LINESIZE=64 -DL2_ASSOCIATIVE=16 " \
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8 " \
+       "-march=armv8.6-a+crypto+crc+fp16+sha3+rng"
+#define LIBNAME   "ampere1"
+#define CORENAME  "AMPERE1"
+#endif
+
 #ifdef FORCE_THUNDERX3T110
 #define ARMV8
 #define FORCE
@@ -1820,7 +1837,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CORENAME  "CK860FV"
 #endif
 
-
 #ifndef FORCE
 
 #ifdef USER_TARGET

kernel/arm64/KERNEL.AMPERE1

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

param.h

@@ -3635,6 +3635,41 @@ is a big desktop or server with abundant cache rather than a phone or embedded d
 #define CGEMM_DEFAULT_R 4096
 #define ZGEMM_DEFAULT_R 4096
 
+#elif defined(AMPERE1)
+
+#if defined(XDOUBLE) || defined(DOUBLE)
+#define SWITCH_RATIO            8
+#else
+#define SWITCH_RATIO            16
+#endif
+
+#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 CGEMM_DEFAULT_UNROLL_M  8
+#define CGEMM_DEFAULT_UNROLL_N  4
+
+#define ZGEMM_DEFAULT_UNROLL_M  4
+#define ZGEMM_DEFAULT_UNROLL_N  4
+
+#define SGEMM_DEFAULT_P	240
+#define DGEMM_DEFAULT_P	240
+#define CGEMM_DEFAULT_P 128
+#define ZGEMM_DEFAULT_P 128
+
+#define SGEMM_DEFAULT_Q 640
+#define DGEMM_DEFAULT_Q 320
+#define CGEMM_DEFAULT_Q 224
+#define ZGEMM_DEFAULT_Q 112
+
+#define SGEMM_DEFAULT_R 4096
+#define DGEMM_DEFAULT_R 4096
+#define CGEMM_DEFAULT_R 4096
+#define ZGEMM_DEFAULT_R 4096
+
 #elif defined(A64FX) // 512-bit SVE
 
 /* When all BLAS3 routines are implemeted with SVE, SGEMM_DEFAULT_UNROLL_M should be "sve_vl".