CC          = gcc
CXX         = g++
CPP         =
FC          = mpifort
LD          = mpifort
NVCC        = nvcc
#
GPUVER      = K20X
#
AR          = ar -r
#
DFLAGS      = -D__FFTW3                                         \
              -D__LIBINT                                        \
              -D__parallel -D__SCALAPACK                        \
              -D__ELPA                                          \
              -D__LIBXSMM                                       \
              -D__LIBXC                                         \
              -D__SPGLIB                                        \
              -D__ACC -D__DBCSR_ACC -D__PW_CUDA
#
INCS        = -I/usr/include                                    \
              -I/usr/include/elpa_openmp-$(_elpaver)/elpa       \
              -I/usr/include/elpa_openmp-$(_elpaver)/modules    \
              -I/opt/cuda/include
#
OPTFLAGS    = -O3 -march=native -fopenmp -ftree-vectorize
#
CFLAGS      = $(DFLAGS) $(INCS) $(OPTFLAGS)
CXXFLAGS    = $(DFLAGS) $(INCS) $(OPTFLAGS) -std=c++11
CPPFLAGS    =
FCFLAGS     = $(DFLAGS) $(INCS) $(OPTFLAGS)                     \
              -ffree-form -ffree-line-length-none
NVFLAGS     = $(DFLAGS) -O3 -Xcompiler="-fopenmp"               \
              -gencode arch=compute_30,code=sm_30               \
              -gencode arch=compute_35,code=sm_35               \
              -gencode arch=compute_37,code=sm_37               \
              -gencode arch=compute_50,code=sm_50               \
              -gencode arch=compute_60,code=sm_60               \
              -gencode arch=compute_61,code=sm_61               \
              -gencode arch=compute_70,code=sm_70               \
              -gencode arch=compute_75,code=sm_75
#
LDFLAGS     = $(FCFLAGS)
#
LIBS        = -L/opt/cuda/lib64 -lcudart -lcuda                 \
              -lcublas -lcufft -lrt -lnvrtc                     \
              -lxsmmf -lxsmm                                    \
              -lelpa_openmp -lscalapack -lblas -llapack         \
              -lfftw3 -lfftw3_threads                           \
              /usr/lib/libint2.a                                \
              -lxcf03 -lxc                                      \
              -lsymspg                                          \
              -lstdc++ -lpthread -lm -ldl -lz