input and output for basic hmc-nf211-tmclover CI job on 8^4 lattice

[kostrzewa]

steps towards extending github CI runs to nf=2+1+1 tmclover

basic (tmlqcd-native solvers)

• add sample input and (unverified) output for basic nf=2+1+1 tmclover using tmlqcd-native rgmixedcg and cgmmsnd
• add .github/workflows/nf211-basic-tmclover.yaml

ddalphaamg

• add sample input and output for the same using DDalphaAMG in the light sector (and perhaps also in heavy sector)
• add .github/workflows/nf211-ddalphaamg-tmclover.yaml

qphix

• add sample input and output for the same using QPhiX in the light and heavy sectors
• add .github/workflows/nf211-qphix-tmclover.yaml

after completion and testing

• remove nf2 workflows

[kostrzewa]

The columns in output.data to be excluded (-X) for the first numdiff command in the workflow are

18 and 5-16

[kostrzewa]

So it will be something like:

numdiff -r 1e-4 -X 1:18 -X 1:5-16 -X 2:18 -X 2:5-16 output.data ${refpath}/output.data

e.g.:

                                                      X  X   X   X   X   X   X   X  X   X   X   X        X
00000018 0.376141674326 -0.084303039260 1.087959e+00 58 273 223 750 350 840 3215 0 338 481 403 556 1 2.472512e+01 1.305379e-01    

This is because the integer numbers in the middle are iteration counters (which can vary a bit from machine to machine especially with mixed-precision solvers and multigrid) and the second-to-last column is the trajectory time, which is bound to be very different on different machines.

[kostrzewa]

To compile tmLQCD locally you need

c-lime: https://github.com/usqcd-software/c-lime which can be compiled as follows:

cmake \
  -DCMAKE_C_FLAGS='-mtune=native -march=native -O2' \
 -DCMAKE_INSTALL_PREFIX=$(pwd)/install_dir \
 -S ${PATH_TO_C_LIME_SOURCE_CODE}

make -j
make install

and (of course) tmLQCD, which can be compiled as follows with just the tmLQCD-native solvers:

cmake \    
  -DCMAKE_C_FLAGS='-mtune=native -march=native -O3 -fopenmp' \    
  -DCMAKE_CXX_FLAGS='-mtune=native -march=native -O3 -fopenmp' \    
  -DCMAKE_BUILD_TYPE=RELEASE \    
  -DCMAKE_INSTALL_PREFIX=$(pwd)/install_dir \    
  -DCMAKE_PREFIX_PATH=${PATH_TO_C_LIME_BUILD_DIR}/install_dir \    
  -DTM_USE_HALFSPINOR=ON \    
  -DTM_USE_GAUGE_COPY=ON \    
  -DTM_USE_MPI=ON \    
  -DTM_USE_OMP=ON \    
  -DTM_USE_LEMON=ON \    
  -S ${PATH_TO_TMLQCD_SOURCE_CODE}

make -j

where for the second step you don't really need to install. The binaries will be in src/bin in the build directory.