Response Ex 7

Sheet7/E14/GH_GCC_default.mk

@@ -0,0 +1,165 @@
+# Basic Defintions for using GNU-compiler suite sequentially
+# requires setting of COMPILER=GCC_
+
+#startmake as follows to avoid warnings caused by OpenMPI code
+#  make 2>&1 | grep -v openmpi
+
+
+#MPI_ROOT=/opt/homebrew/bin/
+
+
+CC	= ${MPI_ROOT}mpicc
+CXX     = ${MPI_ROOT}mpicxx
+F77	= ${MPI_ROOT}mpif77
+LINKER  = ${CXX}
+
+# If you 'mpirun ...' reports some error "... not enough slots .." then use the option '--oversubscribe'
+MPIRUN  = ${MPI_ROOT}mpirun --oversubscribe -display-map
+#MPIRUN  = ${MPI_ROOT}mpiexec
+
+# 2023, Oct 23:  ""WARNING: There is at least non-excluded one OpenFabrics device found,"
+# solution according to https://github.com/open-mpi/ompi/issues/11063
+#MPIRUN += -mca btl ^openib 
+
+# KFU:sauron
+#CXXFLAGS += -I/software/boost/1_72_0/include
+
+WARNINGS = -Wall -pedantic -Woverloaded-virtual -Wfloat-equal -Wshadow \
+           -Wredundant-decls -Wunreachable-code -Winline -fmax-errors=1
+
+# WARNINGS += -Weffc++ -Wextra
+# -Wno-pragmas
+CXXFLAGS += -std=c++17 -ffast-math -O3 -march=native ${WARNINGS}
+# -ftree-vectorizer-verbose=5  -DNDEBUG
+#          -ftree-vectorizer-verbose=2
+# CFLAGS	= -ffast-math -O3 -DNDEBUG -msse3 -fopenmp -fdump-tree-vect-details
+# CFLAGS	= -ffast-math -O3 -funroll-loops -DNDEBUG -msse3 -fopenmp -ftree-vectorizer-verbose=2
+
+# info on vectorization
+#VECTORIZE = -ftree-vectorize -fdump-tree-vect-blocks=foo.dump
+#-fdump-tree-pre=stderr
+VECTORIZE = -ftree-vectorize -fopt-info -ftree-vectorizer-verbose=5
+#CXXFLAGS += ${VECTORIZE}
+
+# -funroll-all-loops   -msse3
+#GCC  -march=knl -march=broadwell -march=haswell
+
+# for debugging purpose (save code)
+# -fsanitize=leak         # only one out the trhee can be used
+# -fsanitize=address
+# -fsanitize=thread
+SANITARY =  -fsanitize=address  -fsanitize=undefined -fsanitize=null -fsanitize=return \
+ -fsanitize=bounds -fsanitize=alignment -fsanitize=float-divide-by-zero -fsanitize=float-cast-overflow \
+ -fsanitize=bool -fsanitize=enum -fsanitize=vptr
+#CXXFLAGS  += ${SANITARY}
+#LINKFLAGS +=${SANITARY}
+
+# OpenMP
+CXXFLAGS += -fopenmp
+LINKFLAGS += -fopenmp
+
+#LISA, to run on a Mac
+# OpenMP (macOS + clang)
+OMPFLAGS = -Xpreprocessor -fopenmp
+OMPINC   = -I/opt/homebrew/opt/libomp/include
+OMPLIBS  = -L/opt/homebrew/opt/libomp/lib -lomp
+
+#CXXFLAGS += $(OMPFLAGS) $(OMPINC)
+#LINKFLAGS += $(OMPLIBS)
+
+
+default: ${PROGRAM}
+
+${PROGRAM}:	${OBJECTS}
+	$(LINKER)  $^  ${LINKFLAGS} -o $@
+	@echo
+	@echo "Start with :  $(MPIRUN) -np num_proc $(MPIFLAGS) $(PROGRAM)"
+	@echo
+
+clean:
+	@rm -f ${PROGRAM} ${OBJECTS} gmon.out
+
+clean_all:: clean
+	@rm -f *_ *~ *.bak *.log *.out *.tar *.orig
+	@rm -rf html latex
+
+run: ${PROGRAM}
+	${MPIRUN} -np 4 ./$^
+
+# tar the current directory
+MY_DIR = `basename ${PWD}`
+tar: clean_all
+	@echo "Tar the directory: " ${MY_DIR}
+	@cd .. ;\
+	tar cf ${MY_DIR}.tar ${MY_DIR} *default.mk ;\
+	cd ${MY_DIR}
+# 	tar cf `basename ${PWD}`.tar *
+
+zip: clean
+	@echo "Zip the directory: " ${MY_DIR}
+	@cd .. ;\
+	zip -r ${MY_DIR}.zip ${MY_DIR} *default.mk ;\
+	cd ${MY_DIR}
+
+doc:
+	doxygen Doxyfile
+
+#########################################################################
+
+.cpp.o:
+	$(CXX) -c $(CXXFLAGS) -o $@ $<
+#	 2>&1 | grep -v openmpi
+
+# special: get rid of compiler warnings genereate by openmpi-files
+#.cpp.o:
+# 	@$(CXX) -c $(CXXFLAGS) $< 2>/tmp/t.txt || grep -sv openmpi /tmp/t.txt
+# 	|grep -sv openmpi
+
+.c.o:
+	$(CC) -c $(CFLAGS) -o $@ $<
+
+.f.o:
+	$(F77) -c $(FFLAGS) -o $@ $<
+
+##################################################################################################
+#    some tools
+# Cache behaviour (CXXFLAGS += -g  tracks down to source lines; no -pg in linkflags)
+cache: ${PROGRAM}
+	valgrind --tool=callgrind --simulate-cache=yes ./$^
+#	kcachegrind callgrind.out.<pid> &
+	kcachegrind `ls -1tr  callgrind.out.* |tail -1`
+
+# Check for wrong memory accesses, memory leaks, ...
+# use smaller data sets
+# no "-pg"  in compile/link options
+mem: ${PROGRAM}
+	valgrind -v --leak-check=yes --tool=memcheck --undef-value-errors=yes --track-origins=yes --log-file=$^.addr.out --show-reachable=yes mpirun -np 4 ./$^
+	# Graphical interface
+	# valkyrie
+
+#  Simple run time profiling of your code
+#  CXXFLAGS += -g -pg
+#  LINKFLAGS += -pg
+prof: ${PROGRAM}
+	perf record ./$^
+	perf report
+#	gprof -b ./$^ > gp.out
+#	kprof -f gp.out -p gprof &
+
+#Trace your heap:
+#> heaptrack ./main.GCC_
+#> heaptrack_gui heaptrack.main.GCC_.<pid>.gz
+heap: ${PROGRAM}
+	heaptrack ./$^ 11
+	heaptrack_gui  `ls -1tr  heaptrack.$^.* |tail -1` &
+
+codecheck: $(SOURCES)
+	cppcheck --enable=all --inconclusive --std=c++17 --suppress=missingIncludeSystem $^
+
+
+########################################################################
+#  get the detailed  status of all optimization flags
+info:
+	echo "detailed  status of all optimization flags"
+	$(CXX) --version
+	$(CXX) -Q $(CXXFLAGS) --help=optimizers

Sheet7/E14/jacob_template/main.cpp

@@ -79,15 +79,15 @@ int main(int argc , char **argv )
 
     double tstart = MPI_Wtime();                  // Wall clock
 
-    JacobiSolve(SK, fv, uv );          // solve the system of equations
+    //GH  JacobiSolve(SK, fv, uv );          // solve the system of equations
     JacobiSolveMPI(mesh, SK, fv, uv );          // MPI: solve the system of equations
 
     double t1 =  MPI_Wtime() - tstart;             // Wall clock
     cout << "JacobiSolve: timing in sec. : " << t1 << endl;
 
     //if (2==myrank || (1==numprocs && 0==myrank) ) mesh.Mesh::Visualize(uv);  // Visualize only one subdomain
-    //mesh.Visualize(uv);       // Visualize all subdomains
+    mesh.Visualize(uv);       // Visualize all subdomains
 
     MPI_Finalize(); 
     return 0;
-}
+}

Sheet7/gh_response.txt

@@ -0,0 +1,38 @@
+2: OK; rank  used
+
+3:
+E5: DebugVector(): numprocs iterations (cumbersome)
+
+E6: OK
+
+E7: MinMax: incl. global exchange
+
+E8: Allroallv: two versions realized
+
+4:
+E9: only the initialization from template has been used, not the coordinates [-1]
+
+E10: _sendbuf is vector<double>
+VecAccu(...int..)  correct
+
+
+E11: ??
+Global_Nodes() : Please explain in the lecture 
+
+
+E12: Ok
+
+E13: not available
+
+E14:
+make run COMPILER=GH_GCC_#
+Iteration counts are different for all 4 processes
+Code doesn't stop
+
+See annotated E14.pdf
+
+GH:   commented main.cpp:82
+GH: uncommented main.cpp:89
+
+at least, you tried it!  [2 pts]
+