• Install Amber16 using gfortran/gcc
• Configuration of GPU workstations ( red and blue)
  • Slackware64-14.1
  • CUDA7.5
  • GCC-4.8.2
• Ref: http://jswails.wikidot.com/installing-amber14-and-ambertools14

• Unpack the package into some folder, e.g. /home/usr/share

cd /home/usr/share
sudo
tar xvf ~/build/amber/AmberTools16.tar.bz2
tar xvf ~/build/amber/Amber16.tar.bz2
chown -R panying:xue amber16
exit #install and test under the user panying rather than root!
cd amber16

• Setup environment variable and run configure script

export AMBERHOME=/home/usr/share/amber16 (on bash shell)
setenv AMBERHOME /home/usr/share/amber16 (on csh shell)

./configure gnu  # choose 'y' to apply bugfix/patches
sudo apt-get install libxt-dev # if "Error: The X11 libraries are not in the usual location !"
sudo apt-get install flex # if "Error: not find flex "
sudo apt-get install libzip-dev zlib1g-dev # if "Error: zlib.h not found "
sudo apt-get install libbz2-dev # if "Error: bzlib.h not found "

source <the csh or bash conf file>

• Install and test it

make install
sudo apt-get install libxext-dev # if "Error:shape.h not found!" 
####Installation of Amber16 (serial) is complete at Thu Dec  1 22:07:48 EST 2016.####
make test

• Result of test

# AmberTools16
1870 file comparisons passed
0 file comparisons failed
0 tests experienced errors
# Amber16
150 file comparisons passed
0 file comparisons failed
0 tests experienced errors

• Install openmpi (see slackbuild.org). The version I installed on spring/summer is 1.6.5.

• Run configuration script

cd $AMBERHOME
sudo apt-get install openmpi-bin
sudo apt-get install libopenmpi-dev

./configure -mpi gnu
#esp checking
#testing [C++ / fortran] cross-compile with MPI libs
#OK

• Install Amber and test it

make clean
make install
    please make sure mpi part is ok while installing.

#### 2 cores ####
export DO_PARALLEL='mpirun -np 2'
or
setenv DO_PARALLEL 'mpirun -np 2' #tcsh-mode

make test
#pls noticed that test in non-root user

#### 8 cores ####
export DO_PARALLEL='mpirun -np 8'
or
setenv DO_PARALLEL 'mpirun -np 8' #tcsh-mode

make test

• Result of test

### 2 cores
# Amber
147 file comparisons passed
0 file comparisons failed
0 tests experienced an error
869 file comparisons passed
0 file comparisons failed
0 tests experienced errors

### 8 cores
#Amber
167 file comparisons passed
0 file comparisons failed
0 tests experienced an error
# AmberTools
778 file comparisons passed
2 file comparisons failed
0 tests experienced errors

• Run configuration script

tcsh      
echo $CUDA_HOME    #check CUDA_HOME
cd $AMBERHOME
./configure -cuda gnu

• Install Amber and test it

make clean
make install
    please make sure that cuda part is ok!
make test

• Result of test

### Amber16
138 file comparisons passed
1 file comparison failed
0 tests experienced errors

• Run configuration script

cd $AMBERHOME
./configure -cuda -mpi gnu

then do as recommended

• Modify config.h ( Ref) from

...
PMEMD_CU_LIBS=./cuda/cuda.a -L$(CUDA_HOME)/lib64 -L$(CUDA_HOME)/lib -lcurand -lcufft -lcudart -lstdc++
...

to

...
PMEMD_CU_LIBS=./cuda/cuda.a -L$(CUDA_HOME)/lib64 -L$(CUDA_HOME)/lib -lcurand -lcufft -lcudart -lstdc++ -lmpi_cxx
...

• Install Amber and test it

make clean
make install
    please make sure the mpi/cuda part is ok!
export DO_PARALLEL='mpirun -np 2'
or
setenv DO_PARALLEL 'mpirun -np 2'  #tcsh-mode
make test

• Result of test

103 file comparisons passed
2 file comparisons failed
0 tests experienced errors

• Note: compiling might end up with errors like “undefined reference to MPI::Comm::Comm()”. The solution is: before running “make cuda_parallel”, add “-lmpi_cxx” to variable “PMEMD_CU_LIBS” in config.h.

• Note: don't forget to add $AMBERHOME and $CUDA_HOME into ~/.profile
  • turns out ~/.profile link to ~/.bashrc, check if ~/.bashrc or ~/.cshrc have already been set up.

• Note - gpuID

  0: 92.5, 92.7, 92.8
  1: 96.4, 96.5, 97.0
  2: 97.4, 96.9, 96.6
  3: 95.8, 95.8, 95.8
  #temp test