make.defs.ibmp9nv

#
# This file shows the IBM POWER9 + NVIDIA V100 toolchain options for PRKs using
# OpenMP, MPI and/or Fortran (sans coarrays) only.
#
# Base compilers and language options
#
# C99 is required in some implementations.
CC=xlc_r -qlanglvl=stdc99
# All of the Fortran code is written for the 2008 standard and requires preprocessing.
# You might need to modify the build system for the preprocessor options to work.
FC=xlf2008_r
XLFPP=-WF,
# C++11 may not be required but does no harm here.
CXX=xlc++_r -qlanglvl=extended1y
#CXX=g++-9 -std=gnu++17
#
# Compiler flags
#
DEFAULT_OPT_FLAGS=-O3 #-qsuppress=1500-036
#
# OpenMP flags
#
# You can also use -qopenmp.  -openmp is deprecated.
OPENMPFLAG=-qsmp=omp
OPENMPSIMDFLAG=-qsmp=omp
OFFLOADFLAG=-qoffload -qtgtarch=sm_70
#
# OpenCL flags
#
# POCL
# http://portablecl.org/docs/html/using.html#linking-your-program-directly-with-pocl is not correct...
#OPENCLFLAG=-I/opt/pocl/latest/include -L/opt/pocl/latest/lib -lpoclu -I/opt/pocl/latest/share/pocl/include -lOpenCL
# Linux
OPENCLDIR=/usr
OPENCLFLAG=-I${OPENCLDIR} -L${OPENCLDIR}/lib64 -lOpenCL -Wno-deprecated-declarations -qsuppress=1500-029 -qstrict
#
# MPI C++
#
MPIINC=-I/usr/tce/packages/spectrum-mpi/ibm/spectrum-mpi-rolling-release/include
MPILIB=-L/usr/tce/packages/spectrum-mpi/ibm/spectrum-mpi-rolling-release/lib -lmpiprofilesupport -lmpi_ibm
#
# OCCA
#
#OCCADIR=${HOME}/prk-repo/Cxx11/occa
#
# TBB
#
TBBDIR=${HOME}/TBB
TBBLIBDIR=${HOME}/TBB/build/linux_ppc64le_xl_cc4.8.5_libc2.17_kernel4.14.0_release
TBBFLAG=-I${TBBDIR}/include -L${TBBLIBDIR} -ltbb
#
# Parallel STL, Boost, etc.
#
BOOSTROOT=${HOME}/boost/libs
BOOSTFLAG=
BOOSTFLAG+=-I${BOOSTROOT}/circular_buffer/include
BOOSTFLAG+=-I${BOOSTROOT}/compute/include
BOOSTFLAG+=-I${BOOSTROOT}/algorithm/include
BOOSTFLAG+=-I${BOOSTROOT}/config/include
BOOSTFLAG+=-I${BOOSTROOT}/core/include
BOOSTFLAG+=-I${BOOSTROOT}/log/include
BOOSTFLAG+=-I${BOOSTROOT}/array/include
BOOSTFLAG+=-I${BOOSTROOT}/multi_array/include
BOOSTFLAG+=-I${BOOSTROOT}/optional/include
BOOSTFLAG+=-I${BOOSTROOT}/preprocessor/include
BOOSTFLAG+=-I${BOOSTROOT}/type_index/include
BOOSTFLAG+=-I${BOOSTROOT}/utility/include
BOOSTFLAG+=-I${BOOSTROOT}/assert/include
BOOSTFLAG+=-I${BOOSTROOT}/static_assert/include
BOOSTFLAG+=-I${BOOSTROOT}/exception/include
BOOSTFLAG+=-I${BOOSTROOT}/throw_exception/include
BOOSTFLAG+=-I${BOOSTROOT}/concept_check/include
BOOSTFLAG+=-I${BOOSTROOT}/type_traits/include
BOOSTFLAG+=-I${BOOSTROOT}/iterator/include
BOOSTFLAG+=-I${BOOSTROOT}/mpl/include
BOOSTFLAG+=-I${BOOSTROOT}/detail/include
BOOSTFLAG+=-I${BOOSTROOT}/functional/include
BOOSTFLAG+=-I${BOOSTROOT}/move/include
BOOSTFLAG+=-I${BOOSTROOT}/range/include
BOOSTFLAG+=-I${BOOSTROOT}/function/include
BOOSTFLAG+=-I${BOOSTROOT}/integer/include
BOOSTFLAG+=-I${BOOSTROOT}/container_hash/include
BOOSTFLAG+=-I${BOOSTROOT}/bind/include
BOOSTFLAG+=-I${BOOSTROOT}/chrono/include
BOOSTFLAG+=-I${BOOSTROOT}/predef/include
BOOSTFLAG+=-I${BOOSTROOT}/ratio/include
BOOSTFLAG+=-I${BOOSTROOT}/function_types/include
BOOSTFLAG+=-I${BOOSTROOT}/tuple/include
BOOSTFLAG+=-I${BOOSTROOT}/lexical_cast/include
BOOSTFLAG+=-I${BOOSTROOT}/numeric/conversion/include
BOOSTFLAG+=-I${BOOSTROOT}/container/include
BOOSTFLAG+=-I${BOOSTROOT}/math/include
BOOSTFLAG+=-I${BOOSTROOT}/fusion/include
BOOSTFLAG+=-I${BOOSTROOT}/typeof/include
BOOSTFLAG+=-I${BOOSTROOT}/uuid/include
BOOSTFLAG+=-I${BOOSTROOT}/smart_ptr/include
BOOSTFLAG+=-I${BOOSTROOT}/proto/include
BOOSTFLAG+=-DBOOST_COMPUTE_USE_CPP11
RANGEFLAG=-DUSE_BOOST_IRANGE ${BOOSTFLAG}
#RANGEFLAG=-DUSE_RANGES_TS -I./range-v3/include
PSTLFLAG=${OPENMPSIMDFLAG} ${TBBFLAG} -I./pstl/include ${RANGEFLAG} -Wno-\#pragma-messages
KOKKOSDIR=${HOME}/KOKKOS/install-cuda
KOKKOSFLAG=-I${KOKKOSDIR}/include -L${KOKKOSDIR}/lib64 -lkokkoscore
RAJADIR=
RAJAFLAG=-I${RAJADIR}/include -L${RAJADIR}/lib -lRAJA ${OPENMPFLAG} ${TBBFLAG}
THRUSTDIR=/usr/local/cuda-10.1/targets/ppc64le-linux/include
THRUSTFLAG=-I${THRUSTDIR}
#
# SYCL flags
#
# triSYCL
# https://github.com/triSYCL/triSYCL is header-only so just clone in Cxx11 directory...
SYCLDIR=./triSYCL
#SYCLCXX=${CXX} ${OPENMPFLAG} $(DEFAULT_OPT_FLAGS)
#SYCLCXX=${CXX} $(DEFAULT_OPT_FLAGS)
SYCLCXX=g++ -O3 -std=gnu++11
SYCLFLAG=-I$(SYCLDIR)/include -DTRISYCL
SYCLFLAG+=$(BOOSTFLAG)
#
# CBLAS for C++ DGEMM
#
BLASFLAG=-DESSL
CBLASFLAG=-DESSL
#
# CUDA flags
#
# Linux w/ NVIDIA CUDA
NVCC=/usr/local/cuda-10.1/bin/nvcc
CUDAFLAGS=-g -O3 -std=c++11
CUDAFLAGS+=--expt-extended-lambda
CUDAFLAGS+=-arch=sm_70
CUDAFLAGS+=-Xcudafe --diag_suppress=esa_on_defaulted_function_ignored