make.inc.nvhpc.cpu

# Host only configuration file for NVHPC (NVC++ compiler)
# Uses variables from their modulefils
# https://docs.nvidia.com/hpc-sdk/compilers/hpc-compilers-user-guide/#env-vars-use-mod-linux

# NVHPC root dir
ifneq ($(NVHPC_ROOT),)
  NVHPC_DIR = $(NVHPC_ROOT)
else
  NVHPC_DIR = /opt/nvidia/hpc_sdk/Linux_x86_64/21.7
endif

# C++ compiler

# NVHPC has distributed memory support
H2OPUS_USE_MPI = 1

# Defaults to OpenMPI-3.1.5 (fine if not using CUDA_AWARE_MPI)
# use OpenMPI-4.0.5 instead if you want to run with CUDA_AWARE_MPI
# $(NVHPC_DIR)/comm_libs/openmpi4/openmpi-4.0.5
MPI_DIR = $(OPAL_PREFIX)
CXX = $(MPI_DIR)/bin/mpicxx

# mpicxx wrapper uses nvc++
CXXFLAGS = -fPIC -Wall -mp=multicore

# Enable flops reporting
CXXCPPFLAGS = -DH2OPUS_PROFILING_ENABLED

# Flag H2OPUS we are using NVOMP layer for host threading
H2OPUS_USE_NVOMP = 1

# NVHPC (PGI) provides BLAS/LAPACK
BLAS_LIBS = -Wl,-rpath,$(NVHPC_DIR)/compilers/lib -L$(NVHPC_DIR)/compilers/lib -llapack -lblas -lnvf -lrt

# Thrust include
THRUST_INCDIR = $(NVHPC_DIR)/cuda/include