$Z_2-\phi^4$ code

Download and Installation

In order to fetch the code:

$ git clone git@github.com:HISKP-LQCD/Z2-phi4.git
$ cd Z2-phi4
$ git submodule update --init --recursive

Building

To build the program you may create a build directory and copy inside one of the the build script of the folder example_script. One example

  do_cmake_m100.sh

to build on marconi100 (Nvidia volta 100).

   mkdir build
   cd build
   cp ../example_script/do_cmake_m100.sh .
   bash do_cmake_m100.sh
   make

if you want to build the program somewhere else you need to change the path inside the script do_cmake.sh (src_dir=where_I_create_the_build_directory). If the compilation goes well a folder main will be created with the executable. you can add option

-DCMAKE_CXX_FLAGS=-DTIMER
-DCMAKE_CXX_FLAGS=-DDEBUG
-DFFTW=ON
-DCUFFT=ON

in the building scritp to have more extra info on the timer

Running

copy the example for the infile (example.in) and create the output directory. The output directory is specified in the example.in file ad outpath

    cp ../../example.in  .
    mkdir data
   ./main -i example.in

Tips

for the GPU pascal of Qbig you need at least 25 local metropolis step to saturate the bandwith

metropolis_local_hits = 25

Output

Configurations

The configuration are written in binary files with an header, the function to write the configurations are in

modules/write_viewer.cpp

with the functions

void write_header(FILE *f_conf, cluster::IO_params params ,int iconf)
void write_viewer(FILE *f_conf,int layout_value, cluster::IO_params params, int iconf  , const Viewphi &phi  )

The configuration will be written in the outpath folder specified in the infile with a tipical name

T4_L4_msq00.130500_msq10.131350_l00.010000_l10.010000_mu0.020000_g0.000000_rep1_conf1

Contraction

The contraction are written in binary files with an header which differs from the header of the configuration. The header is written with

@ modules/measurements.cpp
void write_header_measuraments(FILE *f_conf, cluster::IO_params params )

After the header there is an int (configuration number) and a block of data (sizeof(double) * T * number_of_correlators) corresponding to the contraction done with that configuration. The clock of data is written with the instructions

write (int) configuration_number 

for(int t=0; t<T; t++) {
write (double) correlator 1  at time t
write (double) correlator 2  at time t
write (double) correlator 3  at time t
...
}

To do

Random number generator:
- randlux serial Not used anymore
- Mersenne twister parallel Not working on GPU
- Kokkos random generator: Vigna, Sebastiano (2014). "An experimental exploration of Marsaglia's xorshift generators, scrambled" See: http://arxiv.org/abs/1402.6246
Cluster update
- add routines for $\mu=g=0$, not really helpful
Appending
Header to the confs
Measurements
parallelization
- OMP : Kokkos
- MPI

HISKP-LQCD / Z2-phi4