Amber20和以前版本安装有很大差别,且自动生成安装目录amber20,安装文件采用cmake_run文件安装,每部安装要设置其中的参数,这两天刚装好,采用官方提供的benchmark验证成功:
tar xvf Amber20.tar.bz2
tar xvf AmberTools20.tar.bz2
export AMBERHOME=/home/user/amber2020/amber20_src
一、串行安装
cd amber20_src/build
# optional: edit the run_cmake script to make any needed changes;
# most users should not need to do this
./run_cmake
# Next, build and install the code:
make install
source /home/xzhfood/amber2020/amber20/amber.sh
cd $AMBERHOME
make test.serial
二、并行安装
#compile the parallel (MPI)
cd $AMBERHOME/AmberTools/src
./configure_mpich gnu
You must download MPICH and extract it here
(for example, type 'tar xvfz mpich-3.3.2.tar.gz')
tar xvf mpich-3.3.tar.gz
cd /home/user/amber2020/amber20_src/build
# edit the run_cmake script to set -DMPI=TRUE
./run_cmake
make install
source /home/user/amber2020/amber20/amber.sh
# To run tests: Note the value below may depend on your MPI implementation
export DO_PARALLEL="mpirun -np 28"
cd $AMBERHOME
source amber.sh
make test.parallel
# Note, some tests, like the replica exchange tests, require more
# than 2 threads, so we suggest that you test with either 4 or 8
# threads as well
export DO_PARALLEL="mpirun -np 4"
make test.parallel
三、单GPU安装
Building and Testing the GPU code
export CUDA_HOME=/usr/local/cuda-9.2
cd build
# edit the run_cmake script to set -DCUDA=TRUE (you should also be using the GNU compilers here)
./run_cmake
make install
cd $AMBERHOME
export CUDA_VISIBLE_DEVICES=0
make test.cuda.serial
(choose the GPU id you wish to test)
四、多GPU安装
Building pmemd.cuda.MPI
cd build
# edit the run_cmake script to have -DMPI=TRUE -DCUDA=TRUE
./run_cmake
make install
cd $AMBERHOME
export DO_PARALLEL="mpirun -np 28" # for bash/sh
make test.cuda.parallel