OpenBLAS/Makefile.x86_64

# CCOMMON_OPT	+= -DFASTCPU

ifeq ($(OSNAME), SunOS)
ifdef BINARY64
ifeq ($(F_COMPILER), SUN)
FCOMMON_OPT += -m64
endif
endif
endif


ifneq ($(DYNAMIC_ARCH),1)
ADD_CPUFLAGS = 1
else
ifdef TARGET_CORE
ADD_CPUFLAGS = 1
endif
endif

ifdef ADD_CPUFLAGS
ifdef HAVE_SSE3
CCOMMON_OPT += -msse3
ifneq ($(F_COMPILER), NAG)
FCOMMON_OPT += -msse3
endif
endif
ifdef HAVE_SSSE3
CCOMMON_OPT += -mssse3
ifneq ($(F_COMPILER), NAG)
FCOMMON_OPT += -mssse3
endif
endif
ifdef HAVE_SSE4_1
CCOMMON_OPT += -msse4.1
ifneq ($(F_COMPILER), NAG)
FCOMMON_OPT += -msse4.1
endif
endif
ifndef OLDGCC
ifdef HAVE_AVX
CCOMMON_OPT += -mavx
ifneq ($(F_COMPILER), NAG)
FCOMMON_OPT += -mavx
endif
endif
endif
ifndef NO_AVX2
ifdef HAVE_AVX2
CCOMMON_OPT += -mavx2
ifneq ($(F_COMPILER), NAG)
FCOMMON_OPT += -mavx2
endif
endif
endif

ifeq ($(CORE), SKYLAKEX)
ifndef NO_AVX512
CCOMMON_OPT += -march=skylake-avx512
ifneq ($(F_COMPILER), NAG)
FCOMMON_OPT += -march=skylake-avx512
endif
ifeq ($(OSNAME), CYGWIN_NT)
CCOMMON_OPT += -fno-asynchronous-unwind-tables
FCOMMON_OPT += -fno-asynchronous-unwind-tables
endif
ifeq ($(OSNAME), WINNT)
ifeq ($(C_COMPILER), GCC)
CCOMMON_OPT += -fno-asynchronous-unwind-tables
FCOMMON_OPT += -fno-asynchronous-unwind-tables
endif
endif
endif
endif

ifeq ($(CORE), COOPERLAKE)
ifndef NO_AVX512
ifeq ($(C_COMPILER), GCC)
 # cooperlake support was added in 10.1
 ifeq ($(GCCVERSIONGTEQ10)$(GCCMINORVERSIONGTEQ1), 11)
  CCOMMON_OPT += -march=cooperlake
  ifneq ($(F_COMPILER), NAG)
   FCOMMON_OPT += -march=cooperlake
  endif
 else  # gcc not support, fallback to avx512
  CCOMMON_OPT += -march=skylake-avx512
  ifneq ($(F_COMPILER), NAG)
   FCOMMON_OPT += -march=skylake-avx512
  endif
 endif
else ifeq ($(C_COMPILER), CLANG)
 # cooperlake support was added in clang 9
 ifeq ($(CLANGVERSIONGTEQ9), 1)
  CCOMMON_OPT += -march=cooperlake
  ifneq ($(F_COMPILER), NAG)
   FCOMMON_OPT += -march=cooperlake
  endif
 else  # not supported in clang, fallback to avx512
  CCOMMON_OPT += -march=skylake-avx512
  ifneq ($(F_COMPILER), NAG)
   FCOMMON_OPT += -march=skylake-avx512
  endif
 endif
endif
ifeq ($(OSNAME), CYGWIN_NT)
CCOMMON_OPT += -fno-asynchronous-unwind-tables
FCOMMON_OPT += -fno-asynchronous-unwind-tables
endif
ifeq ($(OSNAME), WINNT)
ifeq ($(C_COMPILER), GCC)
CCOMMON_OPT += -fno-asynchronous-unwind-tables
FCOMMON_OPT += -fno-asynchronous-unwind-tables
endif
endif
endif
endif

ifeq ($(CORE), SAPPHIRERAPIDS)
ifndef NO_AVX512
ifeq ($(C_COMPILER), GCC)
 # sapphire rapids support was added in 11
 ifeq ($(GCCVERSIONGTEQ11), 1)
  CCOMMON_OPT += -march=sapphirerapids
  ifneq ($(F_COMPILER), NAG)
   FCOMMON_OPT += -march=sapphirerapids
  endif
 else  # gcc not support, fallback to avx512
  CCOMMON_OPT += -march=skylake-avx512
  ifneq ($(F_COMPILER), NAG)
   FCOMMON_OPT += -march=skylake-avx512
  endif
 endif
else ifeq ($(C_COMPILER), CLANG)
 # sapphire rapids support was added in clang 12
 ifeq ($(CLANGVERSIONGTEQ12), 1)
  CCOMMON_OPT += -march=sapphirerapids
  ifneq ($(F_COMPILER), NAG)
   FCOMMON_OPT += -march=sapphirerapids
  endif
 else  # not supported in clang, fallback to avx512
  CCOMMON_OPT += -march=skylake-avx512
  ifneq ($(F_COMPILER), NAG)
   FCOMMON_OPT += -march=skylake-avx512
  endif
 endif
endif
ifeq ($(OSNAME), CYGWIN_NT)
CCOMMON_OPT += -fno-asynchronous-unwind-tables
FCOMMON_OPT += -fno-asynchronous-unwind-tables
endif
ifeq ($(OSNAME), WINNT)
ifeq ($(C_COMPILER), GCC)
CCOMMON_OPT += -fno-asynchronous-unwind-tables
FCOMMON_OPT += -fno-asynchronous-unwind-tables
endif
endif
endif
endif

ifeq ($(CORE), ZEN)
ifdef HAVE_AVX512VL	
ifndef NO_AVX512
CCOMMON_OPT += -march=skylake-avx512
ifneq ($(F_COMPILER), NAG)
FCOMMON_OPT += -march=skylake-avx512
endif
ifeq ($(OSNAME), CYGWIN_NT)
CCOMMON_OPT += -fno-asynchronous-unwind-tables
FCOMMON_OPT += -fno-asynchronous-unwind-tables
endif
ifeq ($(OSNAME), WINNT)
ifeq ($(C_COMPILER), GCC)
CCOMMON_OPT += -fno-asynchronous-unwind-tables
FCOMMON_OPT += -fno-asynchronous-unwind-tables
endif
endif
endif
endif
endif


ifdef HAVE_AVX2
ifndef NO_AVX2
ifeq ($(C_COMPILER), GCC)
# AVX2 support was added in 4.7.0
GCCVERSIONCHECK := $(GCCVERSIONGT4)$(GCCVERSIONGTEQ4)$(GCCMINORVERSIONGTEQ7)
ifeq ($(GCCVERSIONCHECK), $(filter $(GCCVERSIONCHECK), 011 110 111))
CCOMMON_OPT += -mavx2
endif
else 
ifeq ($(C_COMPILER), CLANG)
CCOMMON_OPT += -mavx2
endif
endif
ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
ifeq ($(F_COMPILER), GFORTRAN)
# AVX2 support was added in 4.7.0
GCCVERSIONGTEQ4 := $(shell expr `$(FC) -dumpversion | cut -f1 -d.` \>= 4)
GCCVERSIONGTEQ5 := $(shell expr `$(FC) -dumpversion | cut -f1 -d.` \>= 5)
GCCMINORVERSIONGTEQ7 := $(shell expr `$(FC) -dumpversion | cut -f2 -d.` \>= 7)
GCCVERSIONCHECK := $(GCCVERSIONGTEQ5)$(GCCVERSIONGTEQ4)$(GCCMINORVERSIONGTEQ7)
ifeq ($(GCCVERSIONCHECK), $(filter $(GCCVERSIONCHECK), 011 110 111))
FCOMMON_OPT += -mavx2
endif
else
ifeq ($(F_COMPILER), FLANG)
FCOMMON_OPT += -mavx2
endif
endif
endif
endif
endif

endif


ifeq ($(OSNAME), Interix)
ARFLAGS		= -m x64
endif

ifeq ($(OSNAME), Darwin)
ifndef SMP
LIBMKL		= -L$(MKLPATH)/em64t -Wl,-rpath,$(MKLPATH)/em64t -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lguide -lpthread -lm
else
LIBMKL		= -L$(MKLPATH)/em64t -Wl,-rpath,$(MKLPATH)/em64t -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -lguide -lpthread -lm
endif
else
ifndef SMP
LIBMKL		= -L$(MKLPATH)/em64t -Wl,-rpath,$(MKLPATH)/em64t -Wl,--start-group -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -Wl,--end-group -lguide -lpthread -lm
else
LIBMKL		= -L$(MKLPATH)/em64t -Wl,-rpath,$(MKLPATH)/em64t -Wl,--start-group -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -Wl,--end-group -lguide -lpthread -lm
endif
endif


ifndef SMP
LIBATLAS	= -L$(ATLASPATH)64 -llapack -lcblas -lf77blas -latlas -lm
else
LIBATLAS	= -L$(ATLASPATH)64 -llapack -lptcblas -lptf77blas -latlas -lpthread -lm
endif

LIBFLAME	= -L$(FLAMEPATH) -llapack2flame -lflame $(TOPDIR)/$(LIBNAME) -lgfortran -lpthread -lm


ifeq ($(F_COMPILER), g77)
LIBACML		= -L$(ACMLPATH)/gnu64/lib -Wl,-rpath,$(ACMLPATH)/gnu64/lib -lacml -lacml_mv -lg2c -lm
endif

ifeq ($(F_COMPILER), GFORTRAN)
ifndef SMP
LIBACML		= -L$(ACMLPATH)/gfortran64/lib -Wl,-rpath,$(ACMLPATH)/gfortran64/lib -lacml -lacml_mv -lgfortran -lm
else
LIBACML		= -L$(ACMLPATH)/gfortran64_mp/lib -Wl,-rpath,$(ACMLPATH)/gfortran64_mp/lib -lacml_mp -lacml_mv -lgfortran -lgomp -lm
endif
endif

ifeq ($(F_COMPILER), INTEL)
ifndef SMP
LIBACML		= -L$(ACMLPATH)/ifort64/lib -Wl,-rpath,$(ACMLPATH)/ifort64/lib -lacml -lacml_mv -lifcoremt_pic -lirc -lm -lpthread -ldl
else
LIBACML		= -L$(ACMLPATH)/ifort64_mp/lib -Wl,-rpath,$(ACMLPATH)/ifort64_mp/lib -lacml_mp -lacml_mv -lifcoremt_pic -liomp5 -lirc -lm -lpthread -ldl
endif
endif

ifeq ($(F_COMPILER), OPEN64)
ifndef SMP
LIBACML		= -L$(ACMLPATH)/open64/lib -Wl,-rpath,$(ACMLPATH)/open64/lib -lacml -lacml_mv -lm
else
LIBACML		= -L$(ACMLPATH)/open64_mp/lib -Wl,-rpath,$(ACMLPATH)/open64_mp/lib -lacml_mp -lacml_mv -lm -lpthread
endif
endif

ifeq ($(F_COMPILER), pgf77)
ifndef SMP
LIBACML		= -L$(ACMLPATH)/pgi64/lib    -Wl,-rpath,$(ACMLPATH)/pgi64/lib    -lacml -lacml_mv -L$(PGIPATH) -Wl,-rpath,$(PGIPATH) -lpgftnrtl -lnspgc -lpgmp -lpgc
else
LIBACML		= -L$(ACMLPATH)/pgi64_mp/lib -Wl,-rpath,$(ACMLPATH)/pgi64_mp/lib -lacml -lacml_mv -L$(PGIPATH) -Wl,-rpath,$(PGIPATH) -lpgftnrtl -lnspgc -lpgmp -lpgc
endif
endif

ifeq ($(F_COMPILER), PATHSCALE)
ifndef SMP
LIBACML		= -L$(ACMLPATH)/pathscale64/lib -Wl,-rpath,$(ACMLPATH)/pathscale64/lib -lacml -lacml_mv -Wl,-rpath,$(PATHSCALEPATH) -L$(PATHSCALEPATH) -lpathfortran -lm
else
LIBACML		= -L$(ACMLPATH)/pathscale64_mp/lib -Wl,-rpath,$(ACMLPATH)/pathscale64_mp/lib -lacml_mp -lacml_mv -Wl,-rpath,$(PATHSCALEPATH) -L$(PATHSCALEPATH) -lopenmp -lpathfortran -lm
endif
endif

ifeq ($(F_COMPILER), f90)
LIBACML		= -L$(ACMLPATH)/sun64/lib -Wl,-R,$(ACMLPATH)/sun64/lib -L$(SUNPATH)/lib/amd64 -Wl,-R,$(SUNPATH)/lib/amd64 -lacml -lacml_mv -lfsu
endif

LIBSUNPERF	= -L$(SUNPATH)/lib/amd64 -L$(SUNPATH)/rtlibs/amd64 -Wl,-R,$(SUNPATH)/lib/amd64  -Wl,-R,$(SUNPATH)/rtlibs/amd64 -lsunperf -lfui -lfsu -lmtsk

LIBVECLIB	= /System/Library/Frameworks/vecLib.framework/Versions/Current/vecLib