make.inc

#
# @copyright (c) 2012- King Abdullah University of Science and
#                      Technology (KAUST). All rights reserved.
#


#
# @file make.inc
#
# KBLAS is a high performance CUDA library for subset of BLAS
#    and LAPACK routines optimized for NVIDIA GPUs.
# KBLAS is provided by KAUST.
#
# @version 2.0.0
# @author Ali Charara
# @author Ahmad Abdelfattah
# @date 2017-11-13
#

#-----------------------------------------
_SUPPORT_BLAS2_=TRUE
_SUPPORT_BLAS3_=TRUE
_SUPPORT_BATCH_TR_=TRUE
_SUPPORT_SVD_=TRUE


#-----------------------------------------
_USE_MKL_=TRUE
_USE_MAGMA_=TRUE


#-----------------------------------------
#specify cuda directory
_CUDA_ROOT_=$(CUDA_ROOT)
_CUDA_ARCH_ ?= 35

# specify compilers
CXX ?= g++
CC ?= gcc
NVCC=$(_CUDA_ROOT_)/bin/nvcc

#customize the name of the static kblas library
LIB_KBLAS_NAME=kblas-gpu

#-----------------------------------------
NVOPTS =-O3 --compiler-options -fno-strict-aliasing
COPTS = -std=c99 -openmp

NVOPTS_3 = -DSM=$(_CUDA_ARCH_) -arch sm_$(_CUDA_ARCH_) -Xcompiler -fopenmp

#-----------------------------------------
ifdef _SUPPORT_SVD_
  _CUB_DIR_?=$(HOME)/cub
  #COPTS += -DSUPPORT_SVD -D__NO_NCCL__
  #NVOPTS += -DSUPPORT_SVD -D__NO_NCCL__
endif

#-----------------------------------------
ifdef _USE_MAGMA_
  COPTS += -DUSE_MAGMA
  _MAGMA_ROOT_?=$(HOME)/magma
  NVOPTS += -DUSE_MAGMA
endif

#-----------------------------------------
ifdef _SUPPORT_BLAS2_
  # specify the gpu architecture for BLAS2 routines (fermi, kepler)
  GPU_ARCH_2=kepler

  ifeq (${GPU_ARCH_2}, kepler)
    NVOPTS_2 = -DSM=35 -arch sm_35
  else ifeq (${GPU_ARCH_2}, fermi)
    NVOPTS_2 = -DSM=20 -arch sm_20
  else
    $(error GPU_ARCH_2, currently ${GPU_TARGET_2}, must be either fermi or kepler for BLAS2 routines. Please edit your make.inc file)
  endif
endif #_SUPPORT_BLAS2_

#-----------------------------------------
ifdef _USE_MKL_
  COPTS += -DUSE_MKL
  NVOPTS += -DUSE_MKL
  _MKL_ROOT_=${MKLROOT}
endif