WiP CORTEX A55 support

KERNEL.CORTEXA55

@@ -0,0 +1,196 @@
+SAMINKERNEL  = ../arm/amin.c
+DAMINKERNEL  = ../arm/amin.c
+CAMINKERNEL  = ../arm/zamin.c
+ZAMINKERNEL  = ../arm/zamin.c
+
+SMAXKERNEL   = ../arm/max.c
+DMAXKERNEL   = ../arm/max.c
+
+SMINKERNEL   = ../arm/min.c
+DMINKERNEL   = ../arm/min.c
+
+ISAMINKERNEL = ../arm/iamin.c
+IDAMINKERNEL = ../arm/iamin.c
+ICAMINKERNEL = ../arm/izamin.c
+IZAMINKERNEL = ../arm/izamin.c
+
+ISMAXKERNEL  = ../arm/imax.c
+IDMAXKERNEL  = ../arm/imax.c
+
+ISMINKERNEL  = ../arm/imin.c
+IDMINKERNEL  = ../arm/imin.c
+
+STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
+STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
+STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
+STRSMKERNEL_RT	=  ../generic/trsm_kernel_RT.c
+
+DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+DTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+DTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+DTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+CTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+CTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+CTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+CTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+ZTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+ZTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+ZTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+ZTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+SAMAXKERNEL  = amax.S
+DAMAXKERNEL  = amax.S
+CAMAXKERNEL  = zamax.S
+ZAMAXKERNEL  = zamax.S
+
+SAXPYKERNEL  = axpy.S
+DAXPYKERNEL  = axpy.S
+CAXPYKERNEL  = zaxpy.S
+ZAXPYKERNEL  = zaxpy.S
+
+SROTKERNEL   = rot.S
+DROTKERNEL   = rot.S
+CROTKERNEL   = zrot.S
+ZROTKERNEL   = zrot.S
+
+SSCALKERNEL  = scal.S
+DSCALKERNEL  = scal.S
+CSCALKERNEL  = zscal.S
+ZSCALKERNEL  = zscal.S
+
+SGEMVNKERNEL = gemv_n.S
+DGEMVNKERNEL = gemv_n.S
+CGEMVNKERNEL = zgemv_n.S
+ZGEMVNKERNEL = zgemv_n.S
+
+SGEMVTKERNEL = gemv_t.S
+DGEMVTKERNEL = gemv_t.S
+CGEMVTKERNEL = zgemv_t.S
+ZGEMVTKERNEL = zgemv_t.S
+
+
+SASUMKERNEL    = asum.S
+DASUMKERNEL    = asum.S
+CASUMKERNEL    = casum.S
+ZASUMKERNEL    = zasum.S
+
+SCOPYKERNEL    = copy.S
+DCOPYKERNEL    = copy.S
+CCOPYKERNEL    = copy.S
+ZCOPYKERNEL    = copy.S
+
+SSWAPKERNEL    = swap.S
+DSWAPKERNEL    = swap.S
+CSWAPKERNEL    = swap.S
+ZSWAPKERNEL    = swap.S
+
+ISAMAXKERNEL   = iamax.S
+IDAMAXKERNEL   = iamax.S
+ICAMAXKERNEL   = izamax.S
+IZAMAXKERNEL   = izamax.S
+
+SNRM2KERNEL    = nrm2.S
+DNRM2KERNEL    = nrm2.S
+CNRM2KERNEL    = znrm2.S
+ZNRM2KERNEL    = znrm2.S
+
+ifneq ($(C_COMPILER), PGI)
+SDOTKERNEL   = ../generic/dot.c
+else
+SDOTKERNEL   = dot.S
+endif
+DDOTKERNEL   = dot.S
+ifneq ($(C_COMPILER), PGI)
+CDOTKERNEL   = zdot.S
+ZDOTKERNEL   = zdot.S
+else
+CDOTKERNEL = ../arm/zdot.c
+ZDOTKERNEL = ../arm/zdot.c
+endif
+DSDOTKERNEL  = dot.S
+
+DGEMM_BETA     = dgemm_beta.S
+SGEMM_BETA     = sgemm_beta.S
+
+ifeq ($(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N), 8x8)
+SGEMMKERNEL    =  sgemm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N)_cortexa53.S
+STRMMKERNEL    =  strmm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N)_cortexa53.S
+else
+SGEMMKERNEL    =  sgemm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N).S
+STRMMKERNEL    =  strmm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N).S
+endif
+ifneq ($(SGEMM_UNROLL_M), $(SGEMM_UNROLL_N))
+ifeq ($(SGEMM_UNROLL_M), 16)
+SGEMMITCOPY    =  sgemm_tcopy_$(SGEMM_UNROLL_M).S
+else
+SGEMMITCOPY    =  ../generic/gemm_tcopy_$(SGEMM_UNROLL_M).c
+endif
+ifeq ($(SGEMM_UNROLL_M), 4)
+SGEMMINCOPY    =  sgemm_ncopy_$(SGEMM_UNROLL_M).S
+else
+SGEMMINCOPY    =  ../generic/gemm_ncopy_$(SGEMM_UNROLL_M).c
+endif
+SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
+SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+
+SGEMMOTCOPY    =  sgemm_tcopy_$(SGEMM_UNROLL_N).S
+SGEMMONCOPY    =  sgemm_ncopy_$(SGEMM_UNROLL_N).S
+SGEMMONCOPYOBJ =  sgemm_oncopy$(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), $(DGEMM_UNROLL_N))
+
+ifeq ($(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 ($(DGEMM_UNROLL_N), 4)
+DGEMMONCOPY    =  dgemm_ncopy_$(DGEMM_UNROLL_N).S
+DGEMMOTCOPY    =  dgemm_tcopy_$(DGEMM_UNROLL_N).S
+else
+DGEMMONCOPY    =  ../generic/gemm_ncopy_$(DGEMM_UNROLL_N).c
+DGEMMOTCOPY    =  ../generic/gemm_tcopy_$(DGEMM_UNROLL_N).c
+endif
+
+DGEMMONCOPYOBJ =  dgemm_oncopy$(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), $(CGEMM_UNROLL_N))
+CGEMMINCOPY    =  ../generic/zgemm_ncopy_$(CGEMM_UNROLL_M).c
+CGEMMITCOPY    =  ../generic/zgemm_tcopy_$(CGEMM_UNROLL_M).c
+CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
+CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+CGEMMONCOPY    =  ../generic/zgemm_ncopy_$(CGEMM_UNROLL_N).c
+CGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(CGEMM_UNROLL_N).c
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+ZGEMMKERNEL    =  zgemm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N).S
+ZTRMMKERNEL    =  ztrmm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N).S
+ifneq ($(ZGEMM_UNROLL_M), $(ZGEMM_UNROLL_N))
+ZGEMMINCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_M).c
+ZGEMMITCOPY    =  ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_M).c
+ZGEMMINCOPYOBJ =  zgemm_incopy$(TSUFFIX).$(SUFFIX)
+ZGEMMITCOPYOBJ =  zgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+ZGEMMONCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_N).c
+ZGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_N).c
+ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
+ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)

Makefile.arm64

@@ -57,6 +57,28 @@ endif
 endif
 endif
 
+# Use a53 tunings because a55 is only available in GCC>=8.1
+ifeq ($(CORE), CORTEXA55)
+ifeq ($(GCCVERSIONGTEQ7), 1)
+ifeq ($(GCCVERSIONGTEQ8), 1)
+CCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a55
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a55
+endif
+else
+CCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a53
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a53
+endif
+endif
+else
+CCOMMON_OPT += -march=armv8-a -mtune=cortex-a53
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8-a -mtune=cortex-a53
+endif
+endif
+endif
+
 ifeq ($(CORE), THUNDERX)
 CCOMMON_OPT += -march=armv8-a -mtune=thunderx
 ifneq ($(F_COMPILER), NAG)

Makefile.system

@@ -625,6 +625,7 @@ DYNAMIC_CORE += CORTEXA57
 DYNAMIC_CORE += CORTEXA72
 DYNAMIC_CORE += CORTEXA73
 DYNAMIC_CORE += NEOVERSEN1
+DYNAMIC_CORE += CORTEXA55
 DYNAMIC_CORE += FALKOR
 DYNAMIC_CORE += THUNDERX
 DYNAMIC_CORE += THUNDERX2T99

TargetList.txt

@@ -92,6 +92,7 @@ CORTEXA57
 CORTEXA72
 CORTEXA73
 NEOVERSEN1
+CORTEXA55
 EMAG8180
 FALKOR
 THUNDERX

cpuid_arm64.c

@@ -36,6 +36,7 @@ size_t length=sizeof(value);
 #define CPU_ARMV8       	1
 // Arm
 #define CPU_CORTEXA53     2
+#define CPU_CORTEXA55     14
 #define CPU_CORTEXA57     3
 #define CPU_CORTEXA72     4
 #define CPU_CORTEXA73     5
@@ -67,7 +68,8 @@ static char *cpuname[] = {
   "EMAG8180",
   "NEOVERSEN1",
   "THUNDERX3T110",
-  "VORTEX"	
+  "VORTEX",
+  "CORTEXA55"
 };
 
 static char *cpuname_lower[] = {
@@ -84,7 +86,8 @@ static char *cpuname_lower[] = {
   "emag8180",
   "neoversen1",
   "thunderx3t110",
-  "vortex"	
+  "vortex",
+  "cortexa55"
 };
 
 int get_feature(char *search)
@@ -161,6 +164,8 @@ int detect(void)
         return CPU_CORTEXA73;
       else if (strstr(cpu_part, "0xd0c"))
         return CPU_NEOVERSEN1;
+      else if (strstr(cpu_part, "0xd05"))
+	return CPU_CORTEXA55;
     }
     // Qualcomm
     else if (strstr(cpu_implementer, "0x51") && strstr(cpu_part, "0xc00"))
@@ -284,6 +289,8 @@ void get_cpuconfig(void)
       printf("#define %s\n", cpuname[d]);
       // Fall-through
 		case CPU_ARMV8:
+		// case CPU_CORTEXA53;
+		// case CPU_CORTEXA55;
       // Minimum parameters for ARMv8 (based on A53)
     	printf("#define L1_DATA_SIZE 32768\n");
     	printf("#define L1_DATA_LINESIZE 64\n");

getarch.c

@@ -1159,6 +1159,21 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #else
 #endif
 
+#ifdef FORCE_CORTEXA55
+#define FORCE
+#define ARCHITECTURE    "ARM64"
+#define SUBARCHITECTURE "CORTEXA55"
+#define SUBDIRNAME      "arm64"
+#define ARCHCONFIG   "-DCORTEXA55 " \
+       "-DL1_CODE_SIZE=16384 -DL1_CODE_LINESIZE=64 -DL1_CODE_ASSOCIATIVE=3 " \
+       "-DL1_DATA_SIZE=16384 -DL1_DATA_LINESIZE=64 -DL1_DATA_ASSOCIATIVE=2 " \
+       "-DL2_SIZE=65536 -DL2_LINESIZE=64 -DL2_ASSOCIATIVE=16 " \
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8"
+#define LIBNAME   "cortexa55"
+#define CORENAME  "CORTEXA55"
+#else
+#endif
 
 #ifdef FORCE_FALKOR
 #define FORCE

param.h

@@ -2959,7 +2959,7 @@ 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 2048
 
-#elif defined(CORTEXA53)
+#elif defined(CORTEXA53) || defined(CORTEXA55)
 
 #define SGEMM_DEFAULT_UNROLL_M  8
 #define SGEMM_DEFAULT_UNROLL_N  8