Semileptonic Data Generation (SemiLep)

This project contains semileptonic data production code for Mike's PhD.

If you wish to perform data analysis only, don't build this package. Instead, build the subpackage: Meson (aka Mike's) Lattice Utilities (MLU).

Folder structure

Item	Description
MLU	Subpackage Meson (aka Mike's) Lattice Utilities (MLU)
SemiLep	xml3pt generates the SemiLep dataset on Tursa (and previously Tesseract). Includes other utilities with dependencies on Grid and Hadrons
xml	Sample xml used for data production

Building

1. Install dependencies

You must install these mandatory dependencies

You must install the mandatory dependencies for MLU

1. HDF5
2. OpenMP
3. GNU Scientific Library (GSL), version 2.5 or later (I used 2.7 in production) or from Mike's PhD page
4. Bash with associative arrays (version >= 4.2, latest preferred)

and additional dependencies for SemiLep

5. c-lime
6. Grid Latest version (with HDF5, OpenMP and c-lime)
7. Hadrons commit 49fd078f7b8d43d5c6c2c4814e1a8987afc244f0

This commit of Hadrons requires -fpermissive to be included in CXXFLAGS in order to build successfully with g++ (-Xcompiler -fpermissive for nvcc). It also has a dependency on a Grid deflation header which has moved, requiring (assuming Grid has been installed in $Prefix)

cd ~/$Prefix/include/Grid/algorithms/iterative
ln -s ../deflation/Deflation.h

2. Bootstrap

Choose one of the following options:

2a. Build both SemiLep and MLU together

./bootstrap.sh

2b. Build SemiLep only (if you already have MLU installed)

autoreconf -fvi --no-recursive

3. Configure

Use the usual incantation

mkdir build
cd build

In configure, if you wish to build MLU as a subpackage, set --with-MLU=yes (or leave it out entirely - yes is the default). Otherwise set --with-MLU=*path* to link to a pre-installed copy of MLU in *path*, or '--with-MLU=no' if MLU can be found via CPPFLAGS etc.

Optionally set with-grid=*path*, with-hadrons=*path*, --with-hdf5=*path*, --with-gsl=*path* and --with-lime=*path*. E.g.

../configure --with-grid=*path* --with-hadrons=*path* --with-gsl=*path* --prefix=*InstallPath*

If you need to specify additional paths, use CPPFLAGS, LDFLAGS, LIBS etc. E.g. to pick up prerequisites installed in a location specified by $GridPre:

../configure CPPFLAGS=-I$GridPre/include LDFLAGS=-L$GridPre/lib --with-grid=*path* --with-hadrons=*path* --prefix=*InstallPath*

You can also specify CXX, CC, CXXLD, CCLD to choose specific c++ and c compilers and linkers. If not set, CXX and CXXLD will be taken from 'hadrons-config'. configure will also do its best to apportion hadrons-config --cxxflags between CPPFLAGS (c & c++ pre-processor flags) and CXXFLAGS (c++ compiler flags). E.g. if you also wish to use Xcode clang and tell it to use OpenMP installed in $GridPkg (download the OpenMP library separately, e.g. with MacPorts):

../configure CXX=clang++ CC=clang CPPFLAGS="-I$GridPkg/include -I$GridPkg/include/libomp -Xpreprocessor -fopenmp" LDFLAGS="-L$GridPkg/lib -L$GridPkg/lib/libomp" LIBS=-lomp --with-grid=*path* --with-hadrons=*path* --prefix=*InstallPath*

MacOS example configure scripts can be found in

• ConfigAll.sh configure SemiLep and MLU as a single package
• ConfigSemiLep.sh configure SemiLep only

If you wish to build SemiLep for gpu, it is recommended that MLU still be built for cpu so that the MLU utilities (bootstrap, MultiFit, Continuum etc) can be run from the shell. If building both at the same time, specify different compiler settings for MLU using --with-cxx, --with-cc, --with-cxxld, --with-ccld, --with-cxxflags, --with-cppflags, --with-ldflags and --with-libs.

E.g. building for gpu on Tursa (assuming Grid, Hadrons and GSL are all available in $Prefix)

../configure CXX= CPPFLAGS="-I$Prefix/include" LDFLAGS="-L$Prefix/lib" --with-grid=$Prefix --with-hadrons=$Prefix --with-cxx=g++ --with-cxxld=g++ --prefix=$Prefix

More help is available

../configure --help=recursive

4. Build and install

Use the usual incantation:

make -j 20
make install

or, using Xcode, open SemiLep.xcodeproj and build the scheme All.

Semileptonic data production

This is a very brief introduction to how data for this PhD are produced on the (Tursa) supercomputer.

The project folder for heavy-light semileptonics is:

/mnt/lustre/tursafs1/home/dp207/dp207/shared/projects/semilep

This contains 3 folders:

1. runs contains scripts to produce and post-process data
2. data contains the results of those runs
3. study contains prior studies

Within data and runs we have subdirectories for each ensemble, C1, C2, F1M, M1, M2 and M3.

runs contains a code subdirectory containing versions 2, 3 and 4 of the code used for data production (version 1 ran on Tesseract). Each version contains a sourceme.sh Bash script (for GPU) or a sourcemecpu.sh Bash script (for the CPU nodes).

The data directories for each ensemble generally contain 1 or more run directories containing the output from Grid/Hadrons, as well as an analyse directory containing post processing.

Within each data/run directory there are subdirectories for each configuration, each of which containing a log subdirectory containing the .sh script to perform the run, the .log from the run, the .xml input to the executable run from the script, etc.

The source code for the production runs is SemiLep/xml3pt.cpp with a sample .xml in Xml/xml3pt.xml

NB: The MakeBS.sh scripts from each runs/ensemble subdirectory are the definitive record of which data comprise the data set.

Post-processing and analysis

See the MLU documentation.