Core MPI routines

.travis.yml

@@ -7,17 +7,22 @@ addons:
         - gfortran 
         - libblas-dev 
         - liblapack-dev 
+        - openmpi-bin
+        - libopenmpi-dev
 env:
 ## Uncomment the following line if you want to run also the interface tests
 #   - W90TESTSWITHINTERFACE=true
-   - W90TESTSWITHINTERFACE=false
+   - W90TESTSWITHINTERFACE=false W90BINARYPARALLEL=false
+   - W90TESTSWITHINTERFACE=false W90BINARYPARALLEL=true
 install:
 # Possibly install QE and other interface code
     - ./test-suite/external-codes/get-external-codes.sh
 # Install Wannier
-    - cp config/make.inc.gfort.travisci make.inc
+#- cp config/make.inc.gfort.travisci make.inc
+    - ./test-suite/travis_copy_make.inc.sh
     - make -j default
 script:
     - ./test-suite/run_tests_travis.sh
 git:
   depth: 3
+

CHANGE.log

@@ -4,14 +4,20 @@
 
 	The Maximally-Localised Generalised Wannier Functions Code
 
-v2.x.y (25th January 2017)
+v2.x.y (DATE)
 * G0W0 interface implemented (A. Marrazzo (EPFL, CH) and S. Tsirkin (DIPC, Spain)).
-  Added two utilities (gw2wannier90.py and k_mapper.py) and example 23 on G0W0 bands interpolation with Yambo. 
+  Added two utilities (gw2wannier90.py and k_mapper.py) and example 23 on
+  G0W0 bands interpolation with Yambo.
+* Parallelized the disentanglement and wannierise subroutines
+  with respect to k-points. Other parts of the code have been
+  modified so that only the root process writes in the .wout file.
+  wannier90.x can now be run in parallel (see user guide)
+  (contributed by Guillaume Geranton, FZ Julich)
 v2.1.0 (13th January 2017)
 * Implementation of the symmetry-adapted Wannier functions
   (see R. Sakuma, Phys. Rev. B 87, 235109 (2013), courtesy
    of R. Sakuma (Lund University, Sweden), T. Koretsune (Riken, JP),
-   Y. Nomura (U. Tokyo, JP), Y. Nohara (Atomic-Scale Material 
+   Y. Nomura (U. Tokyo, JP), Y. Nohara (Atomic-Scale Material
    Simulations, Co., Ltd.), R. Arita (Riken, JP))
 * Streamlined the interface between wannier90 and tight-binding
   codes such as pythtb (new input variable: write_tb). Also,

config/make.inc.gfort.traviscimpi

@@ -0,0 +1,21 @@
+#===================================
+# gfortran for the travis-ci service
+# This file for MPI builds
+#===================================
+F90 = gfortran
+
+COMMS  = mpi 
+MPIF90 = mpif90 
+
+# Options for debugging  When we move to 4.9 add -fsanitize=leak
+FCOPTS =  -fstrict-aliasing  -fno-omit-frame-pointer -fno-realloc-lhs -fcheck=bounds,do,recursion,pointer -ffree-form -Wall -Waliasing -Wsurprising -Wline-truncation -Wno-tabs -Wno-uninitialized -Wno-unused-dummy-argument -Wno-unused -Wno-character-truncation -O1 -g -fbacktrace 
+LDOPTS =  -fstrict-aliasing  -fno-omit-frame-pointer -fno-realloc-lhs -fcheck=bounds,do,recursion,pointer -ffree-form -Wall -Waliasing -Wsurprising -Wline-truncation -Wno-tabs -Wno-uninitialized -Wno-unused-dummy-argument -Wno-unused -Wno-character-truncation -O1 -g -fbacktrace 
+
+#=======================
+# System LAPACK and BLAS
+# e.g. use
+#   sudo apt-get install libblas-dev liblapack-dev
+# on Ubuntu
+#=======================
+LIBS = -llapack -lblas
+

doc/tutorial/tutorial.tex

@@ -121,10 +121,8 @@
 
 \sectiontitle{Parallel execution}
 \label{sec:parallel}
-Presently, {\tt wannier90.x} is a serial-only executable, so it
-\emph{cannot} be run in parallel using MPI libraries. On the contrary,
-{\tt postw90.x} can be run in parallel to speed up the calculations,
-using the MPI libraries.
+{\tt postw90.x} and {\tt wannier90.x} can be run in parallel to speed up
+the calculations, using the MPI libraries.
 
 To enable the parallel version to be built, you must specify some
 flags in the {\tt make.inc} file of {\tt wannier90} and {\tt postw90};

doc/user_guide/parameters.tex

@@ -1,10 +1,11 @@
 \chapter{Parameters}\label{chap:parameters}
 
 \section{Usage}
-{\tt
-\begin{quote}
-wannier90.x [-pp] [seedname]
-\end{quote} }
+{\tt wannier90.x} can be run in parallel using MPI libraries to
+reduce the computation time.
+
+For serial execution use: {\tt wannier90.x [-pp] [seedname]}
+
 \begin{itemize}
 \item{ {\tt seedname}: If a seedname string is given the code will read its input
 from a file {\tt seedname.win}. The default value is {\tt wannier}. One can also equivalently provide the string
@@ -14,6 +15,25 @@ \section{Usage}
 This information is written to the file {\tt seedname.nnkp}.}
 \end{itemize}
 
+For parallel execution use: {\tt mpirun -np NUMPROCS wannier90.x [-pp] [seedname]}
+
+\begin{itemize} \item 
+{\tt NUMPROCS}: substitute with the number of processors that you want
+to use.
+\end{itemize}
+
+Note that the {\tt mpirun} command and command-line flags may be
+different in your MPI implementation: read your MPI manual or ask your
+computer administrator.
+
+Note also that this requires that the {\tt wannier90.x} executable has been
+compiled in its parallel version (follow the instructions in the file
+{\tt README.install} in the main directory of the wannier90
+distribution) and
+that the MPI libraries and binaries are installed and correctly
+configured on your machine.
+
+
 \section{{\tt seedname.win} File\label{sec:seednamefile}}
 The \wannier\ input file {\tt seedname.win} has a flexible free-form
 structure. 

doc/user_guide/postw90params.tex

@@ -3,9 +3,9 @@ \chapter{Parameters}
 \section{Introduction}
 
 The \texttt{wannier90.x} code described in Part~\ref{part:w90}
-calculates the maximally-localized Wannier functions. The \texttt{wannier90.x} code is a
-serial executable (i.e., it cannot be executed in parallel on different
-CPUs).
+calculates the maximally-localized Wannier functions. %The \texttt{wannier90.x} code is a
+%serial executable (i.e., it cannot be executed in parallel on different
+%CPUs).
 
 The \texttt{postw90.x} executable contains instead a series of modules
 that take the Wannier functions calculated by \texttt{wannier90.x} and

src/Makefile.2

@@ -5,7 +5,8 @@
 include ../../make.inc
 
 OBJS  =  constants.o io.o utility.o parameters.o hamiltonian.o overlap.o \
-	 kmesh.o disentangle.o ws_distance.o wannierise.o plot.o transport.o sitesym.o
+	 kmesh.o disentangle.o ws_distance.o wannierise.o plot.o transport.o sitesym.o comms.o
+
 
 OBJS2  =  wannier_lib.o
 
@@ -31,8 +32,8 @@ TEMP2 = $(F90)
 endif
 
 
-wannier libs w90chk2chk serialobjs: POSTOPTS = 
-wannier libs w90chk2chk serialobjs: COMPILER = $(F90)
+wannier libs w90chk2chk serialobjs: POSTOPTS = $(TEMP1) 
+wannier libs w90chk2chk serialobjs: COMPILER = $(TEMP2)
 wannier: ../../wannier90.x 
 w90chk2chk: ../../w90chk2chk.x
 
@@ -42,7 +43,7 @@ serialobjs: $(OBJS)
 	$(COMPILER) ../w90chk2chk.F90 $(LDOPTS) $(OBJS) $(LIBS) -o ../../w90chk2chk.x
 
 ../../wannier90.x: $(OBJS) ../wannier_prog.F90
-	$(COMPILER) ../wannier_prog.F90 $(LDOPTS) $(OBJS) $(LIBS) -o ../../wannier90.x
+	$(COMPILER) ../wannier_prog.F90 $(POSTOPTS) $(LDOPTS) $(OBJS) $(LIBS) -o ../../wannier90.x
 
 post: POSTOPTS = $(TEMP1)
 post: COMPILER = $(TEMP2)
@@ -88,7 +89,7 @@ io.o: ../io.F90 constants.o
 utility.o: ../utility.F90  constants.o io.o
 	 $(COMPILER) $(POSTOPTS) $(FCOPTS) -c ../utility.F90
 
-parameters.o: ../parameters.F90 constants.o io.o utility.o
+parameters.o: ../parameters.F90 constants.o io.o utility.o comms.o
 	 $(COMPILER) $(POSTOPTS) $(FCOPTS) -c ../parameters.F90
 
 hamiltonian.o: ../hamiltonian.F90 ws_distance.o constants.o io.o utility.o parameters.o
@@ -100,10 +101,10 @@ overlap.o: ../overlap.F90 constants.o io.o utility.o parameters.o sitesym.o
 kmesh.o: ../kmesh.F90 constants.o io.o utility.o parameters.o 
 	 $(COMPILER) $(POSTOPTS) $(FCOPTS) -c ../kmesh.F90
 
-disentangle.o: ../disentangle.F90 constants.o io.o parameters.o sitesym.o
+disentangle.o: ../disentangle.F90 constants.o io.o parameters.o sitesym.o comms.o
 	 $(COMPILER) $(POSTOPTS) $(FCOPTS) -c ../disentangle.F90
 
-wannierise.o: ../wannierise.F90 hamiltonian.o constants.o io.o utility.o parameters.o sitesym.o
+wannierise.o: ../wannierise.F90 hamiltonian.o constants.o io.o utility.o parameters.o sitesym.o comms.o
 	 $(COMPILER) $(POSTOPTS) $(FCOPTS) -c ../wannierise.F90
 
 plot.o: ../plot.F90 constants.o io.o utility.o parameters.o hamiltonian.o ws_distance.o