Linux-x86-64-cuda-ma.psmp

# For the PCI/UZH quake machine using CUDA (DBCSR only) and HWLOC
# I. Using modules
# . ~urban/site/modules.bashrc
# module load urban/pci-cuda
# II. explicit
# Set LIBS to
# LIBS = -L/data/urban/sw/gcc-4.4/openmpi-1.4.3/BLACS-1.1/scalapack-1.8.0/scalapack-1.8.0-exe0/lib -lscalapack -L/data/urban/sw/gcc-4.4/openmpi-1.4.3/BLACS-1.1/BLACS-1.1-exe0/lib -lblacsF77init -lblacs -L/data/urban/sw/gcc-4.6.1/compile-libsmm-mkl/lib -lsmm_dnn -L/opt/intel/parallel_studio/composerxe-2011.4.191/mkl/lib/intel64 -lmkl_gf_lp64 -lmkl_sequential -lmkl_core -L/usr/local/cuda/lib64 -lcudart -lrt -lpthread

NVCC     = nvcc
HWTOPO   = yes

CC       = gcc
CXX      = CC
CPP      =
FC       = mpif90
LD       = $(FC)
AR       = ar -r
GPUVER   = K20X

CPPFLAGS =
CXXFLAGS = -O3 -I${CUDA_PATH}/include -std=c++11
DFLAGS   = -D__GFORTRAN  -D__parallel -D__SCALAPACK -D__BLACS -D__FFTSG -D__LIBINT -D__ACC -D__DBCSR_ACC -D__HAS_smm_dnn -D__HWLOC
GFLAGS   = -g -fopenmp -O3 $(DFLAGS)
FCFLAGS  = -ffree-form -fcray-pointer $(GFLAGS)
CFLAGS   = $(GFLAGS) -I$(HWLOC_BASE)/include
LDFLAGS  = $(FCFLAGS) -mkl=sequential
NVFLAGS  = $(DFLAGS) -g -O3 -arch sm_20 -Xcompiler='-fopenmp' --std=c++11
LIBS     = -L$(SCALAPACK_LIB) $(SCALAPACK_LIBS) \
	   -L$(BLACS_LIB) $(BLACS_LIBS) \
           -L$(SMM_LIB) $(SMM_LIBS) \
	   -L$(MKL_LIB) $(MKL_LIBS) \
           -L$(CUDA_LIB) $(CUDA_LIBS) \
           -L$(DERIV_LIB) $(DERIV_LIBS) -L$(INT_LIB) $(INT_LIBS)\
           -L$(HWLOC_LIB) $(HWLOC_LIBS) \
           -lpthread
LIBS    += -lstdc++ -lcudart -lnvrtc -lcuda