OpenBLAS/Makefile.system

#
# Include user definition
#

# TO suppress recursive includes
INCLUDED = 1

ifndef TOPDIR
TOPDIR = . 
endif

ifndef RELAPACK_REPLACE
RELAPACK_REPLACE=0
endif

# we need to use the host system's architecture for getarch compile options even especially when cross-compiling
HOSTARCH := $(shell uname -m)
ifeq ($(HOSTARCH), amd64)
HOSTARCH=x86_64
endif

# Catch conflicting usage of ARCH in some BSD environments
ifeq ($(ARCH), amd64)
override ARCH=x86_64
else ifeq ($(ARCH), powerpc64)
override ARCH=power
else ifeq ($(ARCH), powerpc64le)
override ARCH=power
else ifeq ($(ARCH), powerpc)
override ARCH=power
else ifeq ($(ARCH), i386)
override ARCH=x86
else ifeq ($(ARCH), armv6)
override ARCH=arm
else ifeq ($(ARCH), armv7)
override ARCH=arm
else ifeq ($(ARCH), aarch64)
override ARCH=arm64
else ifeq ($(ARCH), mipsel)
override ARCH=mips
else ifeq ($(ARCH), mips64el)
override ARCH=mips64
else ifeq ($(ARCH), zarch)
override ARCH=zarch
endif

NETLIB_LAPACK_DIR = $(TOPDIR)/lapack-netlib

# Default C compiler
# - Only set if not specified on the command line or inherited from the environment.
# - CC is an implicit variable so neither '?=' or 'ifndef' can be used.
#   http://stackoverflow.com/questions/4029274/mingw-and-make-variables
# - Default value is 'cc' which is not always a valid command (e.g. MinGW).
ifeq ($(origin CC),default)

# Check if $(CC) refers to a valid command and set the value to gcc if not
ifneq ($(findstring cmd.exe,$(SHELL)),)
ifeq ($(shell where $(CC) 2>NUL),)
CC = gcc
endif
else # POSIX-ish
ifeq ($(shell command -v $(CC) 2>/dev/null),)
ifeq ($(shell uname -s),Darwin)
CC = clang
# EXTRALIB += -Wl,-no_compact_unwind
else
CC = gcc
endif # Darwin
endif # CC exists
endif # Shell is sane

endif # CC is set to default

# Default Fortran compiler (FC) is selected by f_check.

ifndef MAKEFILE_RULE
include $(TOPDIR)/Makefile.rule
else
include $(TOPDIR)/$(MAKEFILE_RULE)
endif

#
#  Beginning of system configuration
#
ifneq ($(BUILD_SINGLE),1)
ifneq ($(BUILD_DOUBLE),1)
ifneq ($(BUILD_COMPLEX),1)
ifneq ($(BUILD_COMPLEX16),1)
override BUILD_SINGLE=1
override BUILD_DOUBLE=1
override BUILD_COMPLEX=1
override BUILD_COMPLEX16=1
endif
endif
endif
endif

ifndef HOSTCC
HOSTCC	 = $(CC)
endif

ifdef TARGET
GETARCH_FLAGS := -DFORCE_$(TARGET)
GETARCH_FLAGS += -DUSER_TARGET
ifeq ($(TARGET), GENERIC)
ifeq ($(DYNAMIC_ARCH), 1)
override NO_EXPRECISION=1
export NO_EXPRECISION
endif
endif
endif

# Force fallbacks for 32bit

ifeq ($(BINARY), 32)
ifeq ($(TARGET), HASWELL)
GETARCH_FLAGS := -DFORCE_NEHALEM
endif
ifeq ($(TARGET), SKYLAKEX)
GETARCH_FLAGS := -DFORCE_NEHALEM
endif
ifeq ($(TARGET), COOPERLAKE)
GETARCH_FLAGS := -DFORCE_NEHALEM
endif
ifeq ($(TARGET), SAPPHIRERAPIDS)
GETARCH_FLAGS := -DFORCE_NEHALEM
endif
ifeq ($(TARGET), SANDYBRIDGE)
GETARCH_FLAGS := -DFORCE_NEHALEM
endif
ifeq ($(TARGET), BULLDOZER)
GETARCH_FLAGS := -DFORCE_BARCELONA
endif
ifeq ($(TARGET), PILEDRIVER)
GETARCH_FLAGS := -DFORCE_BARCELONA
endif
ifeq ($(TARGET), STEAMROLLER)
GETARCH_FLAGS := -DFORCE_BARCELONA
endif
ifeq ($(TARGET), EXCAVATOR)
GETARCH_FLAGS := -DFORCE_BARCELONA
endif
ifeq ($(TARGET), ZEN)
GETARCH_FLAGS := -DFORCE_BARCELONA
endif
ifeq ($(TARGET), ARMV8)
GETARCH_FLAGS := -DFORCE_ARMV7
endif
ifeq ($(TARGET), POWER8)
GETARCH_FLAGS := -DFORCE_POWER6
endif
ifeq ($(TARGET), POWER9)
GETARCH_FLAGS := -DFORCE_POWER6
endif
ifeq ($(TARGET), POWER10)
GETARCH_FLAGS := -DFORCE_POWER6
endif
endif

#TARGET_CORE will override TARGET which is used in DYNAMIC_ARCH=1.
#
ifdef TARGET_CORE
GETARCH_FLAGS := -DFORCE_$(TARGET_CORE)
endif

# Force fallbacks for 32bit

ifeq ($(BINARY), 32)
ifeq ($(TARGET_CORE), HASWELL)
GETARCH_FLAGS := -DFORCE_NEHALEM
endif
ifeq ($(TARGET_CORE), SKYLAKEX)
GETARCH_FLAGS := -DFORCE_NEHALEM
endif
ifeq ($(TARGET_CORE), COOPERLAKE)
GETARCH_FLAGS := -DFORCE_NEHALEM
endif
ifeq ($(TARGET_CORE), SAPPHIRERAPIDS)
GETARCH_FLAGS := -DFORCE_NEHALEM
endif
ifeq ($(TARGET_CORE), SANDYBRIDGE)
GETARCH_FLAGS := -DFORCE_NEHALEM
endif
ifeq ($(TARGET_CORE), BULLDOZER)
GETARCH_FLAGS := -DFORCE_BARCELONA
endif
ifeq ($(TARGET_CORE), PILEDRIVER)
GETARCH_FLAGS := -DFORCE_BARCELONA
endif
ifeq ($(TARGET_CORE), STEAMROLLER)
GETARCH_FLAGS := -DFORCE_BARCELONA
endif
ifeq ($(TARGET_CORE), EXCAVATOR)
GETARCH_FLAGS := -DFORCE_BARCELONA
endif
ifeq ($(TARGET_CORE), ZEN)
GETARCH_FLAGS := -DFORCE_BARCELONA
endif
endif


# On x86_64 build getarch with march=native unless the compiler is PGI. This is required to detect AVX512 support in getarch.
ifeq ($(HOSTARCH), x86_64)
ifeq ($(findstring pgcc,$(HOSTCC))$(findstring nvc,$(HOSTCC)),)
GETARCH_FLAGS += -march=native
endif
endif

ifdef INTERFACE64
ifneq ($(INTERFACE64), 0)
GETARCH_FLAGS	+= -DUSE64BITINT
endif
endif

ifndef GEMM_MULTITHREAD_THRESHOLD
GEMM_MULTITHREAD_THRESHOLD=4
endif
GETARCH_FLAGS	+= -DGEMM_MULTITHREAD_THRESHOLD=$(GEMM_MULTITHREAD_THRESHOLD)

ifeq ($(NO_AVX), 1)
GETARCH_FLAGS	+= -DNO_AVX
endif

ifeq ($(BINARY), 32)
GETARCH_FLAGS	+= -DNO_AVX -DNO_AVX2 -DNO_AVX512
NO_AVX512 = 1
endif

ifeq ($(NO_AVX2), 1)
GETARCH_FLAGS	+= -DNO_AVX2
endif

ifeq ($(NO_AVX512), 1)
GETARCH_FLAGS	+= -DNO_AVX512
endif

ifeq ($(DEBUG), 1)
GETARCH_FLAGS	+= -g
endif

ifeq ($(QUIET_MAKE), 1)
MAKE += -s
endif

ifndef NO_PARALLEL_MAKE
NO_PARALLEL_MAKE=0
endif
GETARCH_FLAGS	+= -DNO_PARALLEL_MAKE=$(NO_PARALLEL_MAKE)

ifdef MAKE_NB_JOBS
GETARCH_FLAGS += -DMAKE_NB_JOBS=$(MAKE_NB_JOBS)
endif

ifeq ($(HOSTCC), loongcc)
GETARCH_FLAGS  += -static
endif

#if don't use Fortran, it will only compile CBLAS.
ifeq ($(ONLY_CBLAS), 1)
NO_LAPACK = 1
else
ONLY_CBLAS = 0
endif

#For small matrix optimization
ifeq ($(ARCH), x86_64)
SMALL_MATRIX_OPT = 1
else ifeq ($(ARCH), power)
SMALL_MATRIX_OPT = 1
BUILD_BFLOAT16 = 1
else ifeq ($(ARCH), arm64)
SMALL_MATRIX_OPT = 1
endif
ifeq ($(ARCH), loongarch64)
SMALL_MATRIX_OPT = 1
endif
ifeq ($(ARCH), arm64)
GEMM_GEMV_FORWARD = 1
endif
ifeq ($(ARCH), riscv)
GEMM_GEMV_FORWARD = 1
endif
ifeq ($(ARCH), power)
GEMM_GEMV_FORWARD = 1
endif

ifeq ($(SMALL_MATRIX_OPT), 1)
CCOMMON_OPT += -DSMALL_MATRIX_OPT
endif
ifeq ($(GEMM_GEMV_FORWARD), 1)
ifneq ($(ONLY_CBLAS), 1)
CCOMMON_OPT += -DGEMM_GEMV_FORWARD
endif
endif

# This operation is expensive, so execution should be once.
ifndef GOTOBLAS_MAKEFILE
export GOTOBLAS_MAKEFILE = 1

# Generating Makefile.conf and config.h
DUMMY := $(shell $(MAKE) -C $(TOPDIR) -f Makefile.prebuild CC="$(CC)" FC="$(FC)" HOSTCC="$(HOSTCC)" HOST_CFLAGS="$(GETARCH_FLAGS)" CFLAGS="$(CFLAGS)" BINARY=$(BINARY) USE_OPENMP=$(USE_OPENMP) DYNAMIC_ARCH=$(DYNAMIC_ARCH) TARGET_CORE=$(TARGET_CORE) ONLY_CBLAS=$(ONLY_CBLAS) TARGET=$(TARGET) all)

endif

ifndef TARGET_CORE
-include $(TOPDIR)/Makefile.conf
else
HAVE_NEON=
HAVE_VFP=
HAVE_VFPV3=
HAVE_VFPV4=
HAVE_MMX=
HAVE_SSE=
HAVE_SSE2=
HAVE_SSE3=
HAVE_SSSE3=
HAVE_SSE4_1=
HAVE_SSE4_2=
HAVE_SSE4A=
HAVE_SSE5=
HAVE_AVX=
HAVE_AVX2=
HAVE_FMA3=
include $(TOPDIR)/Makefile_kernel.conf
endif


ifndef NUM_PARALLEL
NUM_PARALLEL = 1
endif

ifndef NUM_THREADS
NUM_THREADS = $(NUM_CORES)
endif

ifeq ($(NUM_THREADS), 1)
override USE_THREAD = 0
override USE_OPENMP = 0
endif

ifdef USE_THREAD
ifeq ($(USE_THREAD), 0)
SMP =
else
SMP = 1
endif
else
ifeq ($(NUM_THREADS), 1)
SMP =
else
SMP = 1
endif
endif

ifeq ($(SMP), 1)
USE_LOCKING = 
endif

ifndef NEED_PIC
NEED_PIC = 1
endif

ARFLAGS	=
CPP	= $(COMPILER) -E
AR	?= $(CROSS_SUFFIX)ar
AS	?= $(CROSS_SUFFIX)as
LD	?= $(CROSS_SUFFIX)ld
RANLIB	?= $(CROSS_SUFFIX)ranlib
NM	= $(CROSS_SUFFIX)nm
DLLWRAP = $(CROSS_SUFFIX)dllwrap
OBJCOPY = $(CROSS_SUFFIX)objcopy
OBJCONV = $(CROSS_SUFFIX)objconv


# When fortran support was either not detected or actively deselected, only build BLAS.
ifeq ($(NOFORTRAN), 1)
C_LAPACK = 1
override FEXTRALIB = 
endif

ifeq ($(C_COMPILER), GCC)
GCCVERSIONGTEQ4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 4)
GCCVERSIONGT4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 4)
GCCVERSIONGT5 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 5)
GCCVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 7)
GCCVERSIONGTEQ8 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 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)
# Note that the behavior of -dumpversion is compile-time-configurable for
# gcc-7.x and newer. Use -dumpfullversion there
ifeq ($(GCCVERSIONGTEQ7),1)
	GCCDUMPVERSION_PARAM := -dumpfullversion
else
	GCCDUMPVERSION_PARAM := -dumpversion
endif
GCCMINORVERSIONGTEQ1 := $(shell expr `$(CC) $(GCCDUMPVERSION_PARAM) | cut -f2 -d.` \>= 1)
GCCMINORVERSIONGTEQ2 := $(shell expr `$(CC) $(GCCDUMPVERSION_PARAM) | cut -f2 -d.` \>= 2)
GCCMINORVERSIONGTEQ4 := $(shell expr `$(CC) $(GCCDUMPVERSION_PARAM) | cut -f2 -d.` \>= 4)
GCCMINORVERSIONGTEQ7 := $(shell expr `$(CC) $(GCCDUMPVERSION_PARAM) | cut -f2 -d.` \>= 7)
endif

ifeq ($(C_COMPILER), CLANG)
CLANGVERSIONGTEQ9 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 9)
CLANGVERSIONGTEQ12 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 12)
endif

#
#  OS dependent settings
#

ifeq ($(OSNAME), Darwin)
ifndef MACOSX_DEPLOYMENT_TARGET
ifeq ($(ARCH), arm64)
export MACOSX_DEPLOYMENT_TARGET=11.0
ifeq ($(C_COMPILER), GCC)
export NO_SVE = 1
endif
else
export MACOSX_DEPLOYMENT_TARGET=10.8
endif
endif
MD5SUM = md5 -r
XCVER = $(shell pkgutil --pkg-info=com.apple.pkg.Xcode |awk '/version:/ {print $2}'|cut -d: -f2|cut -f1 -d.)
ifeq (x$(XCVER)x,xx)
XCVER = $(shell pkgutil --pkg-info=com.apple.pkg.CLTools_Executables |awk '/version:/ {print $2}'|cut -d: -f2|cut -f1 -d.)
endif
ifeq (x$(XCVER), x 15)
CCOMMON_OPT += -Wl,-ld_classic
FCOMMON_OPT += -Wl,-ld_classic
endif
endif

ifneq (,$(findstring $(OSNAME), FreeBSD OpenBSD DragonFly))
MD5SUM = md5 -r
endif

ifeq ($(OSNAME), NetBSD)
MD5SUM = md5 -n
endif

ifeq ($(OSNAME), Linux)
EXTRALIB	+= -lm
NO_EXPRECISION = 1
endif

ifeq ($(OSNAME), Android)
EXTRALIB	+= -lm
endif

ifeq ($(OSNAME), AIX)
EXTRALIB	+= -lm
endif

ifeq ($(OSNAME), $(filter $(OSNAME),FreeBSD OpenBSD NetBSD DragonFly))
ifeq ($(ARCH), $(filter $(ARCH),arm arm64))
EXTRALIB	+= -lm
endif
endif

ifeq ($(OSNAME), WINNT)
NEED_PIC = 0
NO_EXPRECISION = 1

EXTRALIB        += -defaultlib:advapi32

SUFFIX  = obj
PSUFFIX = pobj
LIBSUFFIX = a

ifeq ($(C_COMPILER), CLANG)
CCOMMON_OPT	+= -DMS_ABI
endif

#Version tests for supporting specific features (MS_ABI, POWER9 intrinsics)
ifeq ($(GCCVERSIONGT4), 1)
# GCC Major version > 4
# It is compatible with MSVC ABI.
CCOMMON_OPT	+= -DMS_ABI
endif

ifeq ($(GCCVERSIONGTEQ4), 1)
ifeq ($(GCCMINORVERSIONGTEQ7), 1)
# GCC Version >=4.7
# It is compatible with MSVC ABI.
CCOMMON_OPT	+= -DMS_ABI
endif
endif

# Ensure the correct stack alignment on Win32
# http://permalink.gmane.org/gmane.comp.lib.openblas.general/97
ifeq ($(ARCH), x86)
CCOMMON_OPT += -mincoming-stack-boundary=2
FCOMMON_OPT += -mincoming-stack-boundary=2
endif

endif

ifeq ($(OSNAME), Interix)
NEED_PIC = 0
NO_EXPRECISION = 1

INTERIX_TOOL_DIR = /opt/gcc.3.3/i586-pc-interix3/bin
endif

ifeq ($(OSNAME), CYGWIN_NT)
NEED_PIC = 0
NO_EXPRECISION = 1
OS_CYGWIN_NT = 1
endif

ifneq ($(OSNAME), WINNT)
ifneq ($(OSNAME), CYGWIN_NT)
ifneq ($(OSNAME), Interix)
ifneq ($(OSNAME), Android)
ifdef SMP
EXTRALIB   += -lpthread
endif
endif
endif
endif
endif

# ifeq logical or
ifeq ($(OSNAME), $(filter $(OSNAME),WINNT CYGWIN_NT Interix))
OS_WINDOWS=1
endif

ifdef QUAD_PRECISION
CCOMMON_OPT	+= -DQUAD_PRECISION
NO_EXPRECISION = 1
endif

ifneq ($(ARCH), x86)
ifneq ($(ARCH), x86_64)
NO_EXPRECISION = 1
endif
endif

ifdef UTEST_CHECK
CCOMMON_OPT	+= -DUTEST_CHECK
SANITY_CHECK = 1
endif

ifdef SANITY_CHECK
CCOMMON_OPT	+= -DSANITY_CHECK -DREFNAME=$(*F)f$(BU)
endif

MAX_STACK_ALLOC ?= 2048
ifneq ($(MAX_STACK_ALLOC), 0)
CCOMMON_OPT	+= -DMAX_STACK_ALLOC=$(MAX_STACK_ALLOC)
endif

ifdef USE_LOCKING
ifneq ($(USE_LOCKING), 0)
CCOMMON_OPT	+= -DUSE_LOCKING
endif
endif

#
#  Architecture dependent settings
#

ifeq ($(ARCH), x86)
ifndef BINARY
NO_BINARY_MODE	= 1
endif

ifeq ($(CORE), generic)
NO_EXPRECISION = 1
endif

ifndef NO_EXPRECISION
ifeq ($(F_COMPILER), GFORTRAN)
# ifeq logical or. GCC or LSB
ifeq ($(C_COMPILER), $(filter $(C_COMPILER),GCC LSB))
EXPRECISION	= 1
CCOMMON_OPT	+= -DEXPRECISION -m128bit-long-double
FCOMMON_OPT	+= -m128bit-long-double
endif
ifeq ($(C_COMPILER), CLANG)
EXPRECISION	= 1
CCOMMON_OPT	+= -DEXPRECISION
FCOMMON_OPT	+= -m128bit-long-double
endif
endif
endif
endif

ifeq ($(ARCH), x86_64)

ifeq ($(CORE), generic)
NO_EXPRECISION = 1
endif

ifndef NO_EXPRECISION
ifeq ($(F_COMPILER), GFORTRAN)
# ifeq logical or. GCC or LSB
ifeq ($(C_COMPILER), $(filter $(C_COMPILER),GCC LSB))
EXPRECISION	= 1
CCOMMON_OPT	+= -DEXPRECISION -m128bit-long-double
FCOMMON_OPT	+= -m128bit-long-double
endif
ifeq ($(C_COMPILER), CLANG)
EXPRECISION	= 1
CCOMMON_OPT	+= -DEXPRECISION
FCOMMON_OPT	+= -m128bit-long-double
endif
endif
endif
endif

ifeq ($(C_COMPILER), INTEL)
CCOMMON_OPT    += -wd981
endif


ifeq ($(USE_OPENMP), 1)

#check
ifeq ($(USE_THREAD), 0)
$(error OpenBLAS: Cannot set both USE_OPENMP=1 and USE_THREAD=0. The USE_THREAD=0 is only for building single thread version.)
endif

# ifeq logical or. GCC or LSB
ifeq ($(C_COMPILER), $(filter $(C_COMPILER),GCC LSB))
CCOMMON_OPT    += -fopenmp
endif

ifeq ($(C_COMPILER), CLANG)
CCOMMON_OPT    += -fopenmp
ifeq ($(F_COMPILER), GFORTRAN)
FEXTRALIB := $(subst -lgomp,-lomp,$(FEXTRALIB))
endif
endif

ifeq ($(C_COMPILER), INTEL)
CCOMMON_OPT    += -fopenmp
endif

ifeq ($(C_COMPILER), PGI)
CCOMMON_OPT    += -mp
endif

ifeq ($(C_COMPILER), OPEN64)
CCOMMON_OPT    += -mp
CEXTRALIB   += -lstdc++
endif

ifeq ($(C_COMPILER), PATHSCALE)
CCOMMON_OPT    += -mp
endif
endif


ifeq ($(DYNAMIC_ARCH), 1)
ifeq ($(ARCH), x86)
DYNAMIC_CORE = KATMAI COPPERMINE NORTHWOOD PRESCOTT BANIAS \
	       CORE2 PENRYN DUNNINGTON NEHALEM ATHLON OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO
endif

ifeq ($(ARCH), x86_64)
DYNAMIC_CORE = PRESCOTT CORE2 
ifeq ($(DYNAMIC_OLDER), 1)
DYNAMIC_CORE += PENRYN DUNNINGTON 
endif
DYNAMIC_CORE += NEHALEM 
ifeq ($(DYNAMIC_OLDER), 1)
DYNAMIC_CORE += OPTERON OPTERON_SSE3 
endif
DYNAMIC_CORE += BARCELONA 
ifeq ($(DYNAMIC_OLDER), 1)
DYNAMIC_CORE += BOBCAT ATOM NANO
endif
ifneq ($(NO_AVX), 1)
DYNAMIC_CORE += SANDYBRIDGE BULLDOZER PILEDRIVER STEAMROLLER EXCAVATOR
endif
ifneq ($(NO_AVX2), 1)
DYNAMIC_CORE += HASWELL ZEN
endif
ifneq ($(NO_AVX512), 1)
ifneq ($(NO_AVX2), 1)
DYNAMIC_CORE += SKYLAKEX COOPERLAKE SAPPHIRERAPIDS
endif
endif
endif

ifdef DYNAMIC_LIST
override DYNAMIC_CORE = PRESCOTT $(DYNAMIC_LIST)
XCCOMMON_OPT = -DDYNAMIC_LIST -DDYN_PRESCOTT
XCCOMMON_OPT += $(foreach dcore,$(DYNAMIC_LIST),-DDYN_$(dcore))
CCOMMON_OPT += $(XCCOMMON_OPT)
#CCOMMON_OPT	+= -DDYNAMIC_LIST='$(DYNAMIC_LIST)'
endif

ifeq ($(ARCH), arm64)
DYNAMIC_CORE =  ARMV8
DYNAMIC_CORE += CORTEXA53
DYNAMIC_CORE += CORTEXA57
DYNAMIC_CORE += NEOVERSEN1
ifneq ($(NO_SVE), 1)
DYNAMIC_CORE += NEOVERSEV1
DYNAMIC_CORE += NEOVERSEN2
DYNAMIC_CORE += ARMV8SVE
DYNAMIC_CORE += A64FX
endif
DYNAMIC_CORE += THUNDERX
DYNAMIC_CORE += THUNDERX2T99
DYNAMIC_CORE += TSV110
DYNAMIC_CORE += EMAG8180
DYNAMIC_CORE += THUNDERX3T110
ifdef DYNAMIC_LIST
override DYNAMIC_CORE = ARMV8 $(DYNAMIC_LIST)
XCCOMMON_OPT = -DDYNAMIC_LIST -DDYN_ARMV8
XCCOMMON_OPT += $(foreach dcore,$(DYNAMIC_LIST),-DDYN_$(dcore))
endif
endif

ifeq ($(ARCH), mips64)
DYNAMIC_CORE = LOONGSON3R3 LOONGSON3R4 MIPS64_GENERIC
ifdef DYNAMIC_LIST
override DYNAMIC_CORE = MIPS64_GENERIC $(DYNAMIC_LIST)
XCCOMMON_OPT = -DDYNAMIC_LIST -DDYN_MIPS64_GENERIC
XCCOMMON_OPT += $(foreach dcore,$(DYNAMIC_LIST),-DDYN_$(dcore))
endif
endif

ifeq ($(ARCH), loongarch64)
DYNAMIC_CORE = LA64_GENERIC LA264 LA464
endif

ifeq ($(ARCH), riscv64)
DYNAMIC_CORE = RISCV64_GENERIC
DYNAMIC_CORE += RISCV64_ZVL128B
DYNAMIC_CORE += RISCV64_ZVL256B
ifdef DYNAMIC_LIST
override DYNAMIC_CORE = RISCV64_GENERIC $(DYNAMIC_LIST)
XCCOMMON_OPT = -DDYNAMIC_LIST -DDYN_RISCV64_GENERIC
XCCOMMON_OPT += $(foreach dcore,$(DYNAMIC_LIST),-DDYN_$(dcore))
endif
endif

ifeq ($(ARCH), zarch)
DYNAMIC_CORE = ZARCH_GENERIC

# if the compiler accepts -march=arch11 or -march=z13 and can compile a file
# with z13-specific inline assembly, then we can include support for Z13.
# note: -march=z13 is equivalent to -march=arch11 yet some compiler releases
# only support one or the other.
# note: LLVM version 6.x supported -march=z13 yet could not handle vector
# registers in inline assembly, so the check for supporting the -march flag is
# not enough.
ZARCH_TEST_COMPILE=-c $(TOPDIR)/kernel/zarch/damin_z13.c -I$(TOPDIR) -o /dev/null > /dev/null 2> /dev/null
ZARCH_CC_SUPPORTS_ARCH11=$(shell $(CC) -march=arch11 $(ZARCH_TEST_COMPILE) && echo 1)
ZARCH_CC_SUPPORTS_Z13=$(shell $(CC) -march=z13 $(ZARCH_TEST_COMPILE) && echo 1)

ifeq ($(or $(ZARCH_CC_SUPPORTS_ARCH11), $(ZARCH_CC_SUPPORTS_Z13)), 1)
DYNAMIC_CORE += Z13
CCOMMON_OPT += -DDYN_Z13
else
$(info OpenBLAS: Not building Z13 kernels because the compiler $(CC) does not support it)
endif

# as above for z13, check for -march=arch12 and z14 support in the compiler.
ZARCH_CC_SUPPORTS_ARCH12=$(shell $(CC) -march=arch12 $(ZARCH_TEST_COMPILE) && echo 1)
ZARCH_CC_SUPPORTS_Z14=$(shell $(CC) -march=z14 $(ZARCH_TEST_COMPILE) && echo 1)
ifeq ($(or $(ZARCH_CC_SUPPORTS_ARCH12), $(ZARCH_CC_SUPPORTS_Z14)), 1)
DYNAMIC_CORE += Z14
CCOMMON_OPT += -DDYN_Z14
else
$(info OpenBLAS: Not building Z14 kernels because the compiler $(CC) does not support it)
endif

endif # ARCH zarch

ifeq ($(ARCH), power)
ifneq ($(C_COMPILER), PGI)
DYNAMIC_CORE = POWER6
DYNAMIC_CORE += POWER8
ifneq ($(C_COMPILER), GCC)
DYNAMIC_CORE += POWER9
DYNAMIC_CORE += POWER10
CCOMMON_OPT += -DHAVE_P10_SUPPORT
endif
ifeq ($(C_COMPILER), GCC)
ifeq ($(GCCVERSIONGT5), 1)
DYNAMIC_CORE += POWER9
else
$(info, OpenBLAS: Your gcc version is too old to build the POWER9 kernels.)
endif
ifeq ($(OSNAME), AIX)
LDVERSIONGTEQ35 := 1
else
LDVERSIONGTEQ35 := $(shell expr `$(CC) -Wl,--version 2> /dev/null | head -1 | cut -f2 -d "." | cut -f1 -d "-"` \>= 35)
endif
ifeq ($(GCCVERSIONGTEQ11)$(LDVERSIONGTEQ35), 11)
DYNAMIC_CORE += POWER10
CCOMMON_OPT += -DHAVE_P10_SUPPORT
else ifeq ($(GCCVERSIONGTEQ10), 1)
ifeq ($(GCCMINORVERSIONGTEQ2)$(LDVERSIONGTEQ35), 11)
DYNAMIC_CORE += POWER10
CCOMMON_OPT += -DHAVE_P10_SUPPORT
endif
else
$(info, OpenBLAS: Your gcc version is too old to build the POWER10 kernels.)
endif
endif
else
DYNAMIC_CORE = POWER8
DYNAMIC_CORE += POWER9
endif
endif

# If DYNAMIC_CORE is not set, DYNAMIC_ARCH cannot do anything, so force it to empty
ifndef DYNAMIC_CORE
override DYNAMIC_ARCH=
endif
endif

ifeq ($(ARCH), ia64)
NO_BINARY_MODE	= 1
BINARY_DEFINED	= 1

ifeq ($(F_COMPILER), GFORTRAN)
ifeq ($(C_COMPILER), GCC)
# EXPRECISION	= 1
# CCOMMON_OPT	+= -DEXPRECISION
endif
endif
endif

ifeq ($(ARCH), $(filter $(ARCH),mips64 mips))
NO_BINARY_MODE	= 1
endif

ifeq ($(ARCH), alpha)
NO_BINARY_MODE	= 1
BINARY_DEFINED	= 1
endif

ifeq ($(ARCH), arm)
NO_BINARY_MODE  = 1
BINARY_DEFINED  = 1

ifneq ($(EMBEDDED), 1)
CCOMMON_OPT += -marm
FCOMMON_OPT += -marm
else
CCOMMON_OPT += -DOS_EMBEDDED -mthumb -mcpu=cortex-m4 -mfloat-abi=hard -mfpu=fpv4-sp-d16
endif

# If softfp abi is mentioned on the command line, force it.
ifeq ($(ARM_SOFTFP_ABI), 1)
CCOMMON_OPT += -mfloat-abi=softfp
FCOMMON_OPT += -mfloat-abi=softfp
endif

ifeq ($(OSNAME), Android)
ifeq ($(ARM_SOFTFP_ABI), 1)
EXTRALIB	+= -lm
else
EXTRALIB	+= -Wl,-lm_hard
endif
endif
endif

ifeq ($(ARCH), arm64)
NO_BINARY_MODE  = 1
BINARY_DEFINED  = 1
ifdef INTERFACE64
ifneq ($(INTERFACE64), 0)
ifeq ($(F_COMPILER), GFORTRAN)
FCOMMON_OPT +=  -fdefault-integer-8
endif
ifeq ($(F_COMPILER), FLANG)
FCOMMON_OPT += -i8
endif
endif
endif
endif

ifeq ($(ARCH), riscv64)
NO_BINARY_MODE  = 1
BINARY_DEFINED  = 1
ifdef INTERFACE64
ifneq ($(INTERFACE64), 0)
ifeq ($(F_COMPILER), GFORTRAN)
FCOMMON_OPT +=  -fdefault-integer-8
endif
ifeq ($(F_COMPILER), FLANG)
FCOMMON_OPT += -i8
endif
endif
endif
endif

ifeq ($(ARCH), loongarch64)
NO_BINARY_MODE  = 1
BINARY_DEFINED  = 1
ifdef INTERFACE64
ifneq ($(INTERFACE64), 0)
ifeq ($(F_COMPILER), GFORTRAN)
FCOMMON_OPT +=  -fdefault-integer-8
endif
ifeq ($(F_COMPILER), FLANG)
FCOMMON_OPT += -i8
endif
endif
endif
endif

ifeq ($(ARCH), csky)
NO_BINARY_MODE	= 1
BINARY_DEFINED	= 1
endif

#
#  C Compiler dependent settings
#


# ifeq logical or. GCC or CLANG or LSB
# http://stackoverflow.com/questions/7656425/makefile-ifeq-logical-or
ifeq ($(C_COMPILER), $(filter $(C_COMPILER),GCC CLANG LSB))
CCOMMON_OPT += -Wall
COMMON_PROF += -fno-inline
NO_UNINITIALIZED_WARN =  -Wno-uninitialized

ifeq ($(QUIET_MAKE), 1)
CCOMMON_OPT += $(NO_UNINITIALIZED_WARN) -Wno-unused
endif

ifdef NO_BINARY_MODE

ifeq ($(ARCH), $(filter $(ARCH),mips64))
ifdef BINARY64
CCOMMON_OPT += -mabi=64
else
CCOMMON_OPT += -mabi=n32
endif
BINARY_DEFINED = 1
else ifeq ($(ARCH), $(filter $(ARCH),mips))
CCOMMON_OPT += -mabi=32
BINARY_DEFINED = 1
endif

ifneq (, $(filter $(CORE), MIPS64_GENERIC))
CCOMMON_OPT += -DNO_MSA
FCOMMON_OPT += -DNO_MSA
endif

ifneq (, $(filter $(CORE),LOONGSON3R3 LOONGSON3R4))
CCOMMON_OPT += -march=loongson3a
FCOMMON_OPT += -march=loongson3a
endif

ifeq ($(CORE), MIPS24K)
CCOMMON_OPT += -mips32r2 -mtune=24kc  $(MSA_FLAGS)
FCOMMON_OPT += -mips32r2 -mtune=24kc  $(MSA_FLAGS)
endif

ifeq ($(CORE), MIPS1004K)
CCOMMON_OPT += -mips32r2   $(MSA_FLAGS)
FCOMMON_OPT += -mips32r2   $(MSA_FLAGS)
endif

ifeq ($(CORE), P5600)
CCOMMON_OPT += -mips32r5 -mnan=2008 -mtune=p5600  $(MSA_FLAGS)
FCOMMON_OPT += -mips32r5 -mnan=2008 -mtune=p5600  $(MSA_FLAGS)
endif

ifeq ($(CORE), I6400)
CCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=i6400 $(MSA_FLAGS)
FCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=i6400 $(MSA_FLAGS)
endif

ifeq ($(CORE), P6600)
CCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=p6600 $(MSA_FLAGS)
FCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=p6600 $(MSA_FLAGS)
endif

ifeq ($(CORE), I6500)
CCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=i6400 $(MSA_FLAGS)
FCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=i6400 $(MSA_FLAGS)
endif

ifeq ($(OSNAME), AIX)
BINARY_DEFINED = 1
endif

ifeq ($(ARCH), loongarch64)
LA64_ABI=$(shell $(CC) -mabi=lp64d -c $(TOPDIR)/cpuid_loongarch64.c -o /dev/null > /dev/null 2> /dev/null && echo lp64d)
LA64_ARCH=$(shell $(CC) -march=loongarch64 -c $(TOPDIR)/cpuid_loongarch64.c -o /dev/null > /dev/null 2> /dev/null && echo loongarch64)
ifneq ($(LA64_ABI), lp64d)
LA64_ABI=lp64
endif
ifneq ($(LA64_ARCH), loongarch64)
CCOMMON_OPT += -mabi=$(LA64_ABI)
FCOMMON_OPT += -mabi=$(LA64_ABI)
else
CCOMMON_OPT += -march=loongarch64 -mabi=$(LA64_ABI)
FCOMMON_OPT += -march=loongarch64 -mabi=$(LA64_ABI)
endif
endif

endif

ifndef BINARY_DEFINED
ifneq ($(OSNAME), AIX)
ifdef BINARY64
ifneq ($(ARCH), riscv64)
CCOMMON_OPT += -m64
endif
else
CCOMMON_OPT += -m32
endif
endif
endif

endif

ifeq ($(C_COMPILER), PGI)
PGCVERSIONGT20 := $(shell expr `$(CC) --version|sed -n "2p" |sed -e "s/[^0-9.]//g" |cut -d "." -f 1` \> 20)
PGCVERSIONEQ20 := $(shell expr `$(CC) --version|sed -n "2p" |sed -e "s/[^0-9.]//g" |cut -d "." -f 1` == 20)
PGCMINORVERSIONGE11 := $(shell expr `$(CC) --version|sed -n "2p" |cut -d "-" -f 1 |sed -e "s/[^0-9.]//g" |cut -c 4-5` \>= 11)
PGCVERSIONCHECK := $(PGCVERSIONGT20)$(PGCVERSIONEQ20)$(PGCMINORVERSIONGE11)
ifeq ($(PGCVERSIONCHECK), $(filter $(PGCVERSIONCHECK), 100 101 011))
NEWPGI := 1
PGCVERSIONGT21 := $(shell expr `$(CC) --version|sed -n "2p" |sed -e "s/[^0-9.]//g" |cut -d "." -f 1` \> 21)
PGCVERSIONEQ21 := $(shell expr `$(CC) --version|sed -n "2p" |sed -e "s/[^0-9.]//g" |cut -d "." -f 1` == 21)
PGCVERSIONCHECK2 := $(PGCVERSIONGT21)$(PGCVERSIONEQ21)$(PGCMINORVERSIONGE11)
ifeq ($(PGCVERSIONCHECK2), $(filter $(PGCVERSIONCHECK2), 100 101 011))
NEWPGI2 := 1
endif
endif
ifdef BINARY64
ifeq ($(ARCH), x86_64)
ifeq (,$(findstring tp,$(CFLAGS)))
ifneq ($(NEWPGI2),1)
CCOMMON_OPT += -tp p7-64
else
CCOMMON_OPT += -tp px
endif
endif
ifneq ($(NEWPGI),1)
CCOMMON_OPT +=  -D__MMX__ -Mnollvm
endif
else
ifeq ($(ARCH), power)
ifeq (,$(findstring tp,$(CFLAGS)))
ifeq ($(CORE), POWER8)
CCOMMON_OPT += -tp pwr8
endif
ifeq ($(CORE), POWER9)
CCOMMON_OPT += -tp pwr9
endif
endif
endif
endif
else
ifneq ($(NEWPGI2),1)
ifeq (,$(findstring tp,$(CFLAGS)))
CCOMMON_OPT += -tp p7
else
CCOMMON_OPT += -tp px
endif
endif
endif
endif

ifeq ($(C_COMPILER), PATHSCALE)
ifdef BINARY64
CCOMMON_OPT += -m64
else
CCOMMON_OPT += -m32
endif
endif

#
#  Fortran Compiler dependent settings
#

ifeq ($(F_COMPILER), NAG)
FCOMMON_OPT += -dcfuns -recursive -ieee=full -w=obs -thread_safe
ifdef INTERFACE64
ifneq ($(INTERFACE64), 0)
FCOMMON_OPT += -i8
endif
endif
ifeq ($(USE_OPENMP), 1)
FCOMMON_OPT += -openmp
endif
endif

ifeq ($(F_COMPILER), FLANG)
CCOMMON_OPT += -DF_INTERFACE_FLANG
FCOMMON_OPT += -Mrecursive -Kieee
ifeq ($(OSNAME), Linux)
ifeq ($(ARCH), x86_64)
FLANG_VENDOR := $(shell $(FC) --version|head -1 |cut -f 1 -d " ")
ifeq ($(FLANG_VENDOR), AMD)
FCOMMON_OPT += -fno-unroll-loops
endif
endif
endif
ifdef BINARY64
ifdef INTERFACE64
ifneq ($(INTERFACE64), 0)
FCOMMON_OPT += -i8
endif
endif
FCOMMON_OPT += -Wall
else
FCOMMON_OPT += -Wall
endif
ifeq ($(USE_OPENMP), 1)
FCOMMON_OPT += -fopenmp
endif
endif

ifeq ($(F_COMPILER), G77)
CCOMMON_OPT += -DF_INTERFACE_G77
FCOMMON_OPT += -Wall
ifndef NO_BINARY_MODE
ifneq ($(OSNAME), AIX)
ifdef BINARY64
FCOMMON_OPT += -m64
else
FCOMMON_OPT += -m32
endif
endif
endif
endif

ifeq ($(F_COMPILER), G95)
CCOMMON_OPT += -DF_INTERFACE_G95
FCOMMON_OPT += -Wall
ifneq ($(OSNAME), AIX)
ifndef NO_BINARY_MODE
ifdef BINARY64
FCOMMON_OPT += -m64
else
FCOMMON_OPT += -m32
endif
endif
ifneq ($(NO_LAPACKE), 1)
FCOMMON_OPT += -fno-second-underscore
endif
endif
endif

ifeq ($(F_COMPILER), $(filter $(F_COMPILER),GFORTRAN FLANGNEW))
CCOMMON_OPT += -DF_INTERFACE_GFORT
ifeq ($(F_COMPILER), GFORTRAN)
FCOMMON_OPT += -Wall
# make single-threaded LAPACK calls thread-safe #1847
FCOMMON_OPT += -frecursive
# work around ABI problem with passing single-character arguments
FCOMMON_OPT += -fno-optimize-sibling-calls
#Don't include -lgfortran, when NO_LAPACK=1 or lsbcc
ifneq ($(NOFORTRAN), 1)
ifneq ($(NOFORTRAN), 2)
ifneq ($(NO_LAPACK), 1)
EXTRALIB += -lgfortran
endif
endif
endif
endif
ifdef NO_BINARY_MODE
ifeq ($(ARCH), $(filter $(ARCH),mips64))
ifdef BINARY64
FCOMMON_OPT += -mabi=64
else
FCOMMON_OPT += -mabi=n32
endif
else ifeq ($(ARCH), $(filter $(ARCH),mips))
FCOMMON_OPT += -mabi=32
endif
else
ifdef BINARY64
ifneq ($(OSNAME), AIX)
ifneq ($(ARCH), riscv64)
FCOMMON_OPT += -m64
endif
endif
ifdef INTERFACE64
ifneq ($(INTERFACE64), 0)
FCOMMON_OPT +=  -fdefault-integer-8
endif
endif
else
ifneq ($(OSNAME), AIX)
FCOMMON_OPT += -m32
endif
endif
endif
ifeq ($(USE_OPENMP), 1)
FCOMMON_OPT += -fopenmp
endif
endif

ifeq ($(F_COMPILER), INTEL)
CCOMMON_OPT += -DF_INTERFACE_INTEL
ifdef INTERFACE64
ifneq ($(INTERFACE64), 0)
FCOMMON_OPT += -i8
endif
endif
FCOMMON_OPT += -recursive -fp-model strict -assume protect-parens
ifeq ($(USE_OPENMP), 1)
FCOMMON_OPT += -fopenmp
endif
endif

ifeq ($(F_COMPILER), FUJITSU)
CCOMMON_OPT += -DF_INTERFACE_FUJITSU
ifeq ($(USE_OPENMP), 1)
FCOMMON_OPT += -openmp
endif
endif

ifeq ($(F_COMPILER), IBM)
CCOMMON_OPT += -DF_INTERFACE_IBM
FEXTRALIB += -lxlf90
ifeq ($(C_COMPILER), $(filter $(C_COMPILER),GCC CLANG))
FCOMMON_OPT += -qextname -qzerosize
endif
# FCOMMON_OPT	+= -qarch=440
ifdef BINARY64
FCOMMON_OPT += -q64
ifdef INTERFACE64
ifneq ($(INTERFACE64), 0)
FCOMMON_OPT += -qintsize=8
endif
endif
else
FCOMMON_OPT += -q32
endif
endif

ifeq ($(F_COMPILER), PGI)
CCOMMON_OPT  += -DF_INTERFACE_PGI
COMMON_PROF +=  -DPGICOMPILER
ifdef BINARY64
ifdef INTERFACE64
ifneq ($(INTERFACE64), 0)
FCOMMON_OPT += -i8
endif
endif
ifeq ($(ARCH), x86_64)
ifneq ($(NEWPGI2),1)
FCOMMON_OPT += -tp p7-64
else
FCOMMON_OPT += -tp px
endif
else
ifeq ($(ARCH), power)
ifeq ($(CORE), POWER6)
$(warning NVIDIA HPC compilers do not support POWER6.)
endif
ifeq ($(CORE), POWER8)
FCOMMON_OPT += -tp pwr8
endif
ifeq ($(CORE), POWER9)
FCOMMON_OPT += -tp pwr9
endif
ifeq ($(CORE), POWER10)
$(warning NVIDIA HPC compilers do not support POWER10.)
endif
endif
endif
else
FCOMMON_OPT += -tp p7
endif
FCOMMON_OPT += -Mrecursive -Kieee
ifeq ($(USE_OPENMP), 1)
FCOMMON_OPT += -mp
endif
endif

ifeq ($(F_COMPILER), PATHSCALE)
CCOMMON_OPT  += -DF_INTERFACE_PATHSCALE
ifdef BINARY64
ifdef INTERFACE64
ifneq ($(INTERFACE64), 0)
FCOMMON_OPT += -i8
endif
endif
endif
 
ifeq ($(USE_OPENMP), 1)
FCOMMON_OPT += -mp
endif
endif

ifeq ($(F_COMPILER), OPEN64)
CCOMMON_OPT  += -DF_INTERFACE_OPEN64
ifdef BINARY64
ifdef INTERFACE64
ifneq ($(INTERFACE64), 0)
FCOMMON_OPT += -i8
endif
endif
endif
ifeq ($(ARCH), $(filter $(ARCH),mips64 mips))
ifndef BINARY64
FCOMMON_OPT += -n32
else
FCOMMON_OPT += -n64
endif
ifeq ($(CORE), LOONGSON3R3)
FCOMMON_OPT += -loongson3 -static
endif
ifeq ($(CORE), LOONGSON3R4)
FCOMMON_OPT += -loongson3 -static
endif
else
ifndef BINARY64
FCOMMON_OPT += -m32
else
FCOMMON_OPT += -m64
endif
endif
ifeq ($(USE_OPENMP), 1)
FEXTRALIB   += -lstdc++
FCOMMON_OPT += -mp
endif
endif

ifeq ($(C_COMPILER), OPEN64)
ifeq ($(ARCH), $(filter $(ARCH),mips64 mips))
ifndef BINARY64
CCOMMON_OPT += -n32
else
CCOMMON_OPT += -n64
endif
ifeq ($(CORE), LOONGSON3R3)
CCOMMON_OPT += -loongson3 -static
endif
ifeq ($(CORE), LOONGSON3R4)
CCOMMON_OPT += -loongson3 -static
endif
else
ifndef BINARY64
CCOMMON_OPT += -m32
else
CCOMMON_OPT += -m64
endif
endif
endif

ifeq ($(C_COMPILER), SUN)
CCOMMON_OPT  += -w
ifeq ($(ARCH), x86)
CCOMMON_OPT  += -m32
else
ifdef BINARY64
CCOMMON_OPT  += -m64
else
CCOMMON_OPT  += -m32
endif
endif
endif

ifeq ($(F_COMPILER), SUN)
CCOMMON_OPT  += -DF_INTERFACE_SUN
FCOMMON_OPT  += -ftrap=%none -xrecursive
ifeq ($(ARCH), x86)
FCOMMON_OPT  += -m32
else
ifdef BINARY64
FCOMMON_OPT  += -m64
else
FCOMMON_OPT  += -m32
endif
endif
ifeq ($(USE_OPENMP), 1)
FCOMMON_OPT += -xopenmp=parallel
endif
endif

ifeq ($(F_COMPILER), COMPAQ)
CCOMMON_OPT  += -DF_INTERFACE_COMPAQ
ifeq ($(USE_OPENMP), 1)
FCOMMON_OPT += -openmp
endif
endif

ifeq ($(F_COMPILER), CRAY)
CCOMMON_OPT += -DF_INTERFACE_INTEL
FCOMMON_OPT += -hnopattern
ifdef INTERFACE64
ifneq ($(INTERFACE64), 0)
FCOMMON_OPT += -s integer64
endif
endif
ifneq ($(USE_OPENMP), 1)
FCOMMON_OPT += -O noomp
endif
endif

ifdef BINARY64
ifdef INTERFACE64
ifneq ($(INTERFACE64), 0)
CCOMMON_OPT	+=
#-DUSE64BITINT
endif
endif
endif

ifeq ($(NEED_PIC), 1)
ifeq ($(C_COMPILER), IBM)
CCOMMON_OPT += -qpic=large
else
CCOMMON_OPT += -fPIC
endif
ifeq ($(F_COMPILER), SUN)
FCOMMON_OPT  += -pic
else ifeq ($(F_COMPILER), NAG)
FCOMMON_OPT += -PIC
else ifeq ($(F_COMPILER), IBM)
FCOMMON_OPT += -qpic=large
else
FCOMMON_OPT += -fPIC
endif
endif

ifeq ($(DYNAMIC_ARCH), 1)
CCOMMON_OPT	+= -DDYNAMIC_ARCH
endif

ifeq ($(DYNAMIC_OLDER), 1)
CCOMMON_OPT	+= -DDYNAMIC_OLDER
endif

ifeq ($(C_LAPACK), 1)
CCOMMON_OPT	+= -DC_LAPACK
endif

ifeq ($(NO_LAPACK), 1)
CCOMMON_OPT	+= -DNO_LAPACK
#Disable LAPACK C interface
NO_LAPACKE = 1
endif

ifeq ($(NO_LAPACKE), 1)
CCOMMON_OPT	+= -DNO_LAPACKE
endif

ifeq ($(NO_AVX), 1)
CCOMMON_OPT	+= -DNO_AVX
endif

ifeq ($(ARCH), x86)
CCOMMON_OPT	+= -DNO_AVX
endif

ifeq ($(NO_AVX2), 1)
CCOMMON_OPT	+= -DNO_AVX2
endif

ifeq ($(NO_AVX512), 1)
CCOMMON_OPT	+= -DNO_AVX512
endif

ifeq ($(NO_SVE), 1)
CCOMMON_OPT     += -DNO_SVE
endif

ifdef SMP
CCOMMON_OPT	+= -DSMP_SERVER

ifeq ($(ARCH), mips64)
USE_SIMPLE_THREADED_LEVEL3 = 1
endif

ifeq ($(USE_OPENMP), 1)
# USE_SIMPLE_THREADED_LEVEL3 = 1
# NO_AFFINITY = 1
CCOMMON_OPT	+= -DUSE_OPENMP
endif

ifeq ($(BIGNUMA), 1)
CCOMMON_OPT	+= -DBIGNUMA
endif

endif

ifeq ($(NO_WARMUP), 1)
CCOMMON_OPT	+= -DNO_WARMUP
endif

ifeq ($(CONSISTENT_FPCSR), 1)
CCOMMON_OPT	+= -DCONSISTENT_FPCSR
endif

# Only for development
# CCOMMON_OPT	 += -DPARAMTEST
# CCOMMON_OPT	 += -DPREFETCHTEST
# CCOMMON_OPT	 += -DNO_SWITCHING
# USE_PAPI = 1

ifdef USE_PAPI
CCOMMON_OPT	 += -DUSE_PAPI
EXTRALIB	 += -lpapi -lperfctr
endif

ifdef BUFFERSIZE
CCOMMON_OPT	 += -DBUFFERSIZE=$(BUFFERSIZE)
endif

ifdef DYNAMIC_THREADS
CCOMMON_OPT	 += -DDYNAMIC_THREADS
endif

CCOMMON_OPT	+= -DMAX_CPU_NUMBER=$(NUM_THREADS)

CCOMMON_OPT	+= -DMAX_PARALLEL_NUMBER=$(NUM_PARALLEL)

ifdef USE_SIMPLE_THREADED_LEVEL3
CCOMMON_OPT	+= -DUSE_SIMPLE_THREADED_LEVEL3
endif

ifeq ($(USE_TLS), 1)
CCOMMON_OPT += -DUSE_TLS
endif

ifeq ($(BUILD_BFLOAT16), 1)
CCOMMON_OPT += -DBUILD_BFLOAT16
endif
ifeq ($(BUILD_SINGLE), 1)
CCOMMON_OPT += -DBUILD_SINGLE=1
endif
ifeq ($(BUILD_DOUBLE), 1)
CCOMMON_OPT += -DBUILD_DOUBLE=1
endif
ifeq ($(BUILD_COMPLEX), 1)
CCOMMON_OPT += -DBUILD_COMPLEX=1
endif
ifeq ($(BUILD_COMPLEX16), 1)
CCOMMON_OPT += -DBUILD_COMPLEX16=1
endif

CCOMMON_OPT += -DVERSION=\"$(VERSION)\"

ifndef SYMBOLPREFIX
SYMBOLPREFIX =
endif

ifndef SYMBOLSUFFIX
SYMBOLSUFFIX =
endif

ifndef LIBSONAMEBASE
LIBSONAMEBASE = openblas
endif

ifndef LIBNAMEPREFIX
LIBNAMEPREFIX =
endif

SYMPREFIX=$(SYMBOLPREFIX)
ifeq ($(SYMBOLPREFIX),$(LIBNAMEPREFIX))
SYMPREFIX=
endif
SYMSUFFIX=$(SYMBOLSUFFIX)
ifeq ($(SYMBOLSUFFIX),$(LIBNAMESUFFIX))
SYMSUFFIX=
endif
ifndef LIBNAMESUFFIX
LIBNAMEBASE = $(SYMPREFIX)$(LIBSONAMEBASE)$(SYMSUFFIX)
else
LIBNAMEBASE = $(SYMPREFIX)$(LIBSONAMEBASE)$(SYMSUFFIX)$(LIBNAMESUFFIX)
endif

ifeq ($(OSNAME), CYGWIN_NT)
LIBPREFIX = cyg$(LIBNAMEPREFIX)$(LIBNAMEBASE)
else
LIBPREFIX = lib$(LIBNAMEPREFIX)$(LIBNAMEBASE)
endif

KERNELDIR	= $(TOPDIR)/kernel/$(ARCH)

include $(TOPDIR)/Makefile.$(ARCH)

ifneq ($(C_COMPILER), PGI)
ifneq ($(C_COMPILER), SUN)
CCOMMON_OPT     += -UASMNAME -UASMFNAME -UNAME -UCNAME -UCHAR_NAME -UCHAR_CNAME
endif
endif
CCOMMON_OPT	+= -DASMNAME=$(FU)$(*F) -DASMFNAME=$(FU)$(*F)$(BU) -DNAME=$(*F)$(BU) -DCNAME=$(*F) -DCHAR_NAME=\"$(*F)$(BU)\" -DCHAR_CNAME=\"$(*F)\"

ifeq ($(CORE), PPC440)
CCOMMON_OPT	+= -DALLOC_QALLOC
endif

ifeq ($(CORE), PPC440FP2)
STATIC_ALLOCATION = 1
endif

ifneq ($(OSNAME), Linux)
NO_AFFINITY = 1
endif

ifneq ($(ARCH), x86_64)
ifneq ($(ARCH), x86)
NO_AFFINITY = 1
endif
endif

ifdef NO_AFFINITY
ifeq ($(NO_AFFINITY), 0)
override undefine NO_AFFINITY
else
CCOMMON_OPT	+= -DNO_AFFINITY
endif
endif

ifdef FUNCTION_PROFILE
CCOMMON_OPT	+= -DFUNCTION_PROFILE
endif

ifdef SHMEM_ALLOCATION
ifneq ($(SHMEM_ALLOCATION), 0)
CCOMMON_OPT	+= -DALLOC_SHM
endif
endif

ifdef HUGETLB_ALLOCATION
ifneq ($(HUGETLB_ALLOCATION), 0)
CCOMMON_OPT	+= -DALLOC_HUGETLB
endif
endif

ifdef HUGETLBFILE_ALLOCATION
ifneq ($(HUGETLBFILE_ALLOCATION), 0)
CCOMMON_OPT	+= -DALLOC_HUGETLBFILE -DHUGETLB_FILE_NAME=$(HUGETLBFILE_ALLOCATION)
endif
endif

ifdef STATIC_ALLOCATION
CCOMMON_OPT	+= -DALLOC_STATIC
endif

ifdef DEVICEDRIVER_ALLOCATION
CCOMMON_OPT	+= -DALLOC_DEVICEDRIVER -DDEVICEDRIVER_NAME=\"/dev/mapper\"
endif

ifdef MIXED_MEMORY_ALLOCATION
CCOMMON_OPT	+= -DMIXED_MEMORY_ALLOCATION
endif

ifeq ($(OSNAME), SunOS)
TAR	= gtar
PATCH	= gpatch
GREP	= ggrep
AWK	= nawk
else
TAR	= tar
PATCH	= patch
GREP	= grep
AWK	= awk
endif

ifndef MD5SUM
MD5SUM	= md5sum
endif


REVISION = -r$(VERSION)
MAJOR_VERSION = $(word 1,$(subst ., ,$(VERSION)))

ifeq ($(DEBUG), 1)
COMMON_OPT += -g
endif

ifeq ($(DEBUG), 1)
FCOMMON_OPT += -g
endif

ifndef COMMON_OPT
COMMON_OPT = -O2
endif

ifndef FCOMMON_OPT
FCOMMON_OPT = -O2 -frecursive
endif

override CFLAGS     += $(COMMON_OPT) $(CCOMMON_OPT) -I$(TOPDIR)
override PFLAGS     += $(COMMON_OPT) $(CCOMMON_OPT) -I$(TOPDIR) -DPROFILE $(COMMON_PROF)
override FFLAGS     += $(COMMON_OPT) $(FCOMMON_OPT)
override FPFLAGS    += $(FCOMMON_OPT) $(COMMON_PROF)
#MAKEOVERRIDES =

ifeq ($(NEED_PIC), 1)
ifeq (,$(findstring PIC,$(FFLAGS)))
ifneq ($(F_COMPILER),IBM)
override FFLAGS += -fPIC
endif
endif
endif

#For LAPACK Fortran codes.
#Disable -fopenmp for LAPACK Fortran codes on Windows.
ifdef OS_WINDOWS
LAPACK_FFLAGS := $(filter-out -fopenmp -mp -openmp -xopenmp=parallel,$(FFLAGS))
LAPACK_FPFLAGS := $(filter-out -fopenmp -mp -openmp -xopenmp=parallel,$(FPFLAGS))
else
LAPACK_FFLAGS := $(FFLAGS)
LAPACK_FPFLAGS := $(FPFLAGS)
endif

ifeq ($(F_COMPILER),NAG)
LAPACK_FFLAGS := $(filter-out -msse3 -mssse3 -msse4.1 -mavx -mavx2 -mskylake-avx512 ,$(FFLAGS))
override FFLAGS := $(filter-out -msse3 -mssse3 -msse4.1 -mavx -mavx2 -mskylake-avx512 ,$(FFLAGS))
endif
ifeq ($(F_COMPILER),CRAY)
LAPACK_FFLAGS := $(filter-out -msse3 -mssse3 -msse4.1 -mavx -mavx2 -mskylake-avx512 ,$(FFLAGS))
override FFLAGS := $(filter-out -msse3 -mssse3 -msse4.1 -mavx -mavx2 -mskylake-avx512 ,$(FFLAGS))
endif
ifeq ($(F_COMPILER),FLANGNEW)
LAPACK_FFLAGS := $(filter-out -m32 -m64 -msse3 -mssse3 -msse4.1 -mavx -mavx2 -mskylake-avx512 -mtune=% -mabi=% ,$(FFLAGS))
override FFLAGS := $(filter-out -m32 -m64 -msse3 -mssse3 -msse4.1 -mavx -mavx2 -mskylake-avx512 -mtune=% -mabi=% ,$(FFLAGS))
endif

LAPACK_CFLAGS = $(CFLAGS)
LAPACK_CFLAGS += -DHAVE_LAPACK_CONFIG_H
ifdef INTERFACE64
ifneq ($(INTERFACE64), 0)
LAPACK_CFLAGS +=  -DLAPACK_ILP64
endif
endif

ifdef OS_WINDOWS
LAPACK_CFLAGS +=  -DOPENBLAS_OS_WINDOWS
LAPACK_CFLAGS +=  -DLAPACK_COMPLEX_STRUCTURE
endif
ifeq ($(C_COMPILER), LSB)
LAPACK_CFLAGS +=  -DLAPACK_COMPLEX_STRUCTURE
endif

ifndef SUFFIX
SUFFIX  = o
endif

ifndef PSUFFIX
PSUFFIX = po
endif

ifndef LIBSUFFIX
LIBSUFFIX = a
endif

ifneq ($(DYNAMIC_ARCH), 1)
ifndef SMP
LIBNAME		= $(LIBPREFIX)_$(LIBCORE)$(REVISION).$(LIBSUFFIX)
LIBNAME_P	= $(LIBPREFIX)_$(LIBCORE)$(REVISION)_p.$(LIBSUFFIX)
else
LIBNAME		= $(LIBPREFIX)_$(LIBCORE)p$(REVISION).$(LIBSUFFIX)
LIBNAME_P	= $(LIBPREFIX)_$(LIBCORE)p$(REVISION)_p.$(LIBSUFFIX)
endif
else
ifndef SMP
LIBNAME		= $(LIBPREFIX)$(REVISION).$(LIBSUFFIX)
LIBNAME_P	= $(LIBPREFIX)$(REVISION)_p.$(LIBSUFFIX)
else
LIBNAME		= $(LIBPREFIX)p$(REVISION).$(LIBSUFFIX)
LIBNAME_P	= $(LIBPREFIX)p$(REVISION)_p.$(LIBSUFFIX)
endif
endif

ifeq ($(FIXED_LIBNAME),1)
	LIBNAME = lib$(LIBNAMEPREFIX)$(LIBSONAMEBASE)$(LIBNAMESUFFIX).$(LIBSUFFIX)
	LIBNAME_P = lib$(LIBNAMEPREFIX)$(LISOBNAMEBASE)$(LIBNAMESUFFIX)_p.$(LIBSUFFIX)
endif

LIBDLLNAME   = $(LIBPREFIX).dll
IMPLIBNAME   = lib$(LIBNAMEBASE).dll.a
LIBSONAME    = $(LIBNAME:.$(LIBSUFFIX)=.so)
LIBDYNNAME   = $(LIBNAME:.$(LIBSUFFIX)=.dylib)
LIBDEFNAME   = $(LIBNAME:.$(LIBSUFFIX)=.def)
LIBEXPNAME   = $(LIBNAME:.$(LIBSUFFIX)=.exp)
LIBZIPNAME   = $(LIBNAME:.$(LIBSUFFIX)=.zip)

LIBS		= $(TOPDIR)/$(LIBNAME)
LIBS_P		= $(TOPDIR)/$(LIBNAME_P)


LIB_COMPONENTS = BLAS
ifneq ($(NO_CBLAS), 1)
LIB_COMPONENTS += CBLAS
endif

ifneq ($(NO_LAPACK), 1)
LIB_COMPONENTS += LAPACK
ifneq ($(NO_LAPACKE), 1)
LIB_COMPONENTS += LAPACKE
endif
ifeq ($(BUILD_RELAPACK), 1)
LIB_COMPONENTS += ReLAPACK
endif
endif

ifeq ($(ONLY_CBLAS), 1)
LIB_COMPONENTS = CBLAS
endif

export OSNAME
export ARCH
export CORE
export LIBCORE
export __BYTE_ORDER__
export ELF_VERSION
export PGCPATH
export CONFIG
export CC
export FC
export BU
export FU
export NEED2UNDERSCORES
export USE_THREAD
export NUM_THREADS
export NUM_CORES
export SMP
export MAKEFILE_RULE
export NEED_PIC
export BINARY
export BINARY32
export BINARY64
export F_COMPILER
export C_COMPILER
export USE_OPENMP
export CROSS
export CROSS_SUFFIX
export NOFORTRAN
export C_LAPACK
export NO_FBLAS
export EXTRALIB
export CEXTRALIB
export FEXTRALIB
export HAVE_SSE
export HAVE_SSE2
export HAVE_SSE3
export HAVE_SSSE3
export HAVE_SSE4_1
export HAVE_SSE4_2
export HAVE_SSE4A
export HAVE_SSE5
export HAVE_AVX
export HAVE_AVX2
export HAVE_FMA3
export HAVE_VFP
export HAVE_VFPV3
export HAVE_VFPV4
export HAVE_NEON
ifndef NO_MSA
  export HAVE_MSA
  export MSA_FLAGS
endif
export KERNELDIR
export FUNCTION_PROFILE
export TARGET_CORE
export NO_AVX512
export NO_AVX2
export BUILD_BFLOAT16
export NO_LSX
export NO_LASX

export SBGEMM_UNROLL_M
export SBGEMM_UNROLL_N
export SGEMM_UNROLL_M
export SGEMM_UNROLL_N
export DGEMM_UNROLL_M
export DGEMM_UNROLL_N
export QGEMM_UNROLL_M
export QGEMM_UNROLL_N
export CGEMM_UNROLL_M
export CGEMM_UNROLL_N
export ZGEMM_UNROLL_M
export ZGEMM_UNROLL_N
export XGEMM_UNROLL_M
export XGEMM_UNROLL_N
export CGEMM3M_UNROLL_M
export CGEMM3M_UNROLL_N
export ZGEMM3M_UNROLL_M
export ZGEMM3M_UNROLL_N
export XGEMM3M_UNROLL_M
export XGEMM3M_UNROLL_N


ifdef USE_CUDA
export CUDADIR
export CUCC
export CUFLAGS
export CULIB
endif

.SUFFIXES: .$(PSUFFIX) .$(SUFFIX) .f

.f.$(SUFFIX):
	$(FC) $(FFLAGS) -c $<  -o $(@F)

.f.$(PSUFFIX):
	$(FC) $(FPFLAGS) -pg -c $<  -o $(@F)


ifdef BINARY64
PATHSCALEPATH	= /opt/pathscale/lib/3.1
PGIPATH		= /opt/pgi/linux86-64/7.1-5/lib
else
PATHSCALEPATH	= /opt/pathscale/lib/3.1/32
PGIPATH		= /opt/pgi/linux86/7.1-5/lib
endif

ACMLPATH	= /opt/acml/4.3.0
ifneq ($(OSNAME), Darwin)
MKLPATH         = /opt/intel/mkl/10.2.2.025/lib
else
MKLPATH         = /Library/Frameworks/Intel_MKL.framework/Versions/10.0.1.014/lib
endif
ATLASPATH	= /opt/atlas/3.9.17/opteron
FLAMEPATH	= $(HOME)/flame/lib
ifneq ($(OSNAME), SunOS)
SUNPATH		= /opt/sunstudio12.1
else
SUNPATH		= /opt/SUNWspro
endif