Pushing everything again, accidentally deleted my remote repository
This commit is contained in:
jakob.schratter
commit
1bee3e8e5b
101 changed files with 9428 additions and 0 deletions
131
ex5/CLANG_default.mk
Normal file
131
ex5/CLANG_default.mk
Normal file
|
|
@ -0,0 +1,131 @@
|
|||
# Basic Defintions for using GNU-compiler suite sequentially
|
||||
# requires setting of COMPILER=CLANG_
|
||||
# https://llvm.org/docs/CompileCudaWithLLVM.html
|
||||
# https://llvm.org/docs/NVPTXUsage.html
|
||||
|
||||
#CLANGPATH=//usr/lib/llvm-10/bin/
|
||||
CC = ${CLANGPATH}clang
|
||||
CXX = ${CLANGPATH}clang++
|
||||
#CXX = ${CLANGPATH}clang++ -lomptarget -fopenmp-targets=nvptx64-nvidia-cuda --cuda-path=/opt/pgi/linux86-64/2017/cuda/8.0
|
||||
#F77 = gfortran
|
||||
LINKER = ${CXX}
|
||||
|
||||
#http://clang.llvm.org/docs/UsersManual.html#options-to-control-error-and-warning-messages
|
||||
WARNINGS += -pedantic -Weverything -Wdocumentation -Wconversion -Wshadow -Wfloat-conversion
|
||||
WARNINGS += -Wno-c++98-compat -Wno-sign-conversion -Wno-date-time -Wno-shorten-64-to-32 -Wno-padded -ferror-limit=1
|
||||
WARNINGS += -Wno-unsafe-buffer-usage
|
||||
#-fsyntax-only -Wdocumentation -Wconversion -Wshadow -Wfloat-conversion -pedantic
|
||||
|
||||
CXXFLAGS += -O3 -std=c++17 -ferror-limit=1 ${WARNINGS}
|
||||
# don't use -Ofast
|
||||
# -ftrapv
|
||||
LINKFLAGS += -O3
|
||||
|
||||
# different libraries in Ubuntu or manajaró
|
||||
ifndef UBUNTU
|
||||
UBUNTU=1
|
||||
endif
|
||||
|
||||
# BLAS, LAPACK
|
||||
LINKFLAGS += -llapack -lblas
|
||||
# -lopenblas
|
||||
ifeq ($(UBUNTU),1)
|
||||
# ubuntu
|
||||
else
|
||||
# on archlinux
|
||||
LINKFLAGS += -lcblas
|
||||
endif
|
||||
|
||||
# interprocedural optimization
|
||||
CXXFLAGS += -flto
|
||||
LINKFLAGS += -flto
|
||||
|
||||
#sudo apt install libomp-dev
|
||||
# OpenMP
|
||||
CXXFLAGS += -fopenmp
|
||||
LINKFLAGS += -fopenmp
|
||||
|
||||
# very good check
|
||||
# http://clang.llvm.org/extra/clang-tidy/
|
||||
# good check, see: http://llvm.org/docs/CodingStandards.html#include-style
|
||||
SWITCH_OFF=,-readability-magic-numbers,-readability-redundant-control-flow,-readability-redundant-member-init
|
||||
SWITCH_OFF+=,-readability-redundant-member-init,-readability-isolate-declaration
|
||||
#READABILITY=,readability*${SWITCH_OFF}
|
||||
#TIDYFLAGS = -checks=llvm-*,-llvm-header-guard -header-filter=.* -enable-check-profile -extra-arg="-std=c++17" -extra-arg="-fopenmp"
|
||||
TIDYFLAGS = -checks=llvm-*,-llvm-header-guard${READABILITY} -header-filter=.* -enable-check-profile -extra-arg="-std=c++17" -extra-arg="-fopenmp"
|
||||
#TIDYFLAGS += -checks='modernize*
|
||||
# ???
|
||||
#TIDYFLAGS = -checks='cert*' -header-filter=.*
|
||||
# MPI checks ??
|
||||
#TIDYFLAGS = -checks='mpi*'
|
||||
# ??
|
||||
#TIDYFLAGS = -checks='performance*' -header-filter=.*
|
||||
#TIDYFLAGS = -checks='portability-*' -header-filter=.*
|
||||
#TIDYFLAGS = -checks='readability-*' -header-filter=.*
|
||||
|
||||
default: ${PROGRAM}
|
||||
|
||||
${PROGRAM}: ${OBJECTS}
|
||||
$(LINKER) $^ ${LINKFLAGS} -o $@
|
||||
|
||||
clean:
|
||||
@rm -f ${PROGRAM} ${OBJECTS}
|
||||
|
||||
clean_all:: clean
|
||||
@rm -f *_ *~ *.bak *.log *.out *.tar
|
||||
|
||||
codecheck: tidy_check
|
||||
tidy_check:
|
||||
clang-tidy ${SOURCES} ${TIDYFLAGS} -- ${SOURCES}
|
||||
# see also http://clang-developers.42468.n3.nabble.com/Error-while-trying-to-load-a-compilation-database-td4049722.html
|
||||
|
||||
run: clean ${PROGRAM}
|
||||
# time ./${PROGRAM} ${PARAMS}
|
||||
./${PROGRAM} ${PARAMS}
|
||||
|
||||
# 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 *
|
||||
|
||||
doc:
|
||||
doxygen Doxyfile
|
||||
|
||||
#########################################################################
|
||||
|
||||
.cpp.o:
|
||||
$(CXX) -c $(CXXFLAGS) -o $@ $<
|
||||
|
||||
.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: ${PROGRAM}
|
||||
valgrind --tool=callgrind --simulate-cache=yes ./$^ ${PARAMS}
|
||||
# kcachegrind callgrind.out.<pid> &
|
||||
kcachegrind `ls -1tr callgrind.out.* |tail -1`
|
||||
|
||||
# Check for wrong memory accesses, memory leaks, ...
|
||||
# use smaller data sets
|
||||
mem: ${PROGRAM}
|
||||
valgrind -v --leak-check=yes --tool=memcheck --undef-value-errors=yes --track-origins=yes --log-file=$^.addr.out --show-reachable=yes ./$^ ${PARAMS}
|
||||
|
||||
# Simple run time profiling of your code
|
||||
# CXXFLAGS += -g -pg
|
||||
# LINKFLAGS += -pg
|
||||
prof: ${PROGRAM}
|
||||
perf record ./$^ ${PARAMS}
|
||||
perf report
|
||||
# gprof -b ./$^ > gp.out
|
||||
# kprof -f gp.out -p gprof &
|
||||
|
||||
codecheck: tidy_check
|
||||
212
ex5/GCCMKL_default.mk
Normal file
212
ex5/GCCMKL_default.mk
Normal file
|
|
@ -0,0 +1,212 @@
|
|||
# Basic Defintions for using GNU-compiler suite with OpenMP und MKL
|
||||
# requires setting of COMPILER=GCCMKL_
|
||||
|
||||
# install MKL in manjaro
|
||||
# https://linux-packages.com/manjaro-linux/package/intel-mkl
|
||||
# > sudo pacman -Sy
|
||||
# > sudo pacman -S intel-mkl
|
||||
|
||||
ifeq ($(ONEAPI),1)
|
||||
MKL_INCLUDE=/opt/intel/oneapi/mkl/2024.0/include
|
||||
MKL_LIB=/opt/intel/oneapi/2024.0/lib
|
||||
export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/opt/intel/oneapi/2024.0/lib
|
||||
else
|
||||
MKL_INCLUDE=/usr/include/mkl
|
||||
MKL_LIB=/usr/lib/x86_64-linux-gnu/mkl
|
||||
endif
|
||||
|
||||
CC = gcc
|
||||
CXX = g++
|
||||
F77 = gfortran
|
||||
LINKER = ${CXX}
|
||||
|
||||
WARNINGS = -Wall -pedantic -Wextra -Weffc++ -Woverloaded-virtual -Wfloat-equal -Wshadow \
|
||||
-Wredundant-decls -Winline -fmax-errors=1
|
||||
# -Wunreachable-code
|
||||
CXXFLAGS += -ffast-math -O3 -march=native -std=c++17 ${WARNINGS}
|
||||
#CXXFLAGS += -Ofast -funroll-all-loops -std=c++17 ${WARNINGS}
|
||||
#-msse3
|
||||
# -ftree-vectorizer-verbose=2 -DNDEBUG
|
||||
# -ftree-vectorizer-verbose=5
|
||||
# -ftree-vectorize -fdump-tree-vect-blocks=foo.dump -fdump-tree-pre=stderr
|
||||
|
||||
# CFLAGS = -ffast-math -O3 -DNDEBUG -msse3 -fopenmp -fdump-tree-vect-details
|
||||
# CFLAGS = -ffast-math -O3 -funroll-loops -DNDEBUG -msse3 -fopenmp -ftree-vectorizer-verbose=2
|
||||
# #CFLAGS = -ffast-math -O3 -DNDEBUG -msse3 -fopenmp
|
||||
# FFLAGS = -ffast-math -O3 -DNDEBUG -msse3 -fopenmp
|
||||
# LFLAGS = -ffast-math -O3 -DNDEBUG -msse3 -fopenmp
|
||||
LINKFLAGS += -O3
|
||||
|
||||
#architecture
|
||||
CPU = -march=znver2
|
||||
#CPU = -march=core-avx2
|
||||
CXXFLAGS += ${CPU}
|
||||
LINKFLAGS += ${CPU}
|
||||
|
||||
# MKL
|
||||
#CXXFLAGS += -I/usr/include/mkl -DUSE_MKL -Wno-redundant-decls
|
||||
CXXFLAGS += -I${MKL_INCLUDE} -DUSE_MKL -Wno-redundant-decls
|
||||
#LINKFLAGS += -lmkl_intel_lp64 -lmkl_tbb_thread -ltbb -lmkl_core
|
||||
#LINKFLAGS += -lmkl_intel_lp64 -lmkl_gnu_thread -lmkl_core -L/usr/lib/x86_64-linux-gnu/mkl
|
||||
LINKFLAGS += -L${MKL_LIB} -lmkl_intel_lp64 -lmkl_gnu_thread -lmkl_core
|
||||
#LINKFLAGS += -lmkl_intel_lp64 -lmkl_sequential -lmkl_core
|
||||
|
||||
# workaround for MKL slow down on AMD hardware
|
||||
# https://danieldk.eu/Posts/2020-08-31-MKL-Zen.html
|
||||
default: run
|
||||
libfakeintel.so:
|
||||
gcc -shared -fPIC -o libfakeintel.so fakeintel.c
|
||||
echo "call: export LD_PRELOAD=./libfakeintel.so "
|
||||
|
||||
# different libraries in Ubuntu or manajaro
|
||||
#ifndef UBUNTU
|
||||
#UBUNTU=1
|
||||
#endif
|
||||
|
||||
## BLAS, LAPACK
|
||||
#ifeq ($(UBUNTU),1)
|
||||
#LINKFLAGS += -llapack -lblas
|
||||
## -lopenblas
|
||||
#else
|
||||
## on archlinux
|
||||
#LINKFLAGS += -llapack -lopenblas -lcblas
|
||||
#endif
|
||||
|
||||
# interprocedural optimization
|
||||
CXXFLAGS += -flto
|
||||
LINKFLAGS += -flto
|
||||
|
||||
# for debugging purpose (save code)
|
||||
# -fsanitize=leak # only one out the three 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
|
||||
|
||||
default: ${PROGRAM}
|
||||
|
||||
${PROGRAM}: ${OBJECTS}
|
||||
$(LINKER) $^ ${LINKFLAGS} -o $@
|
||||
|
||||
clean:
|
||||
@rm -f ${PROGRAM} ${OBJECTS}
|
||||
|
||||
clean_all:: clean
|
||||
-@rm -f *_ *~ *.bak *.log *.out *.tar *.orig *.optrpt
|
||||
-@rm -rf html
|
||||
|
||||
run: clean ${PROGRAM} libfakeintel.so
|
||||
#run: ${PROGRAM}
|
||||
# time ./${PROGRAM} ${PARAMS}
|
||||
./${PROGRAM} ${PARAMS}
|
||||
|
||||
# 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 *
|
||||
#find . -size +10M > large_files
|
||||
#--exclude-from ${MY_DIR}/large_files
|
||||
|
||||
zip: clean
|
||||
@echo "Zip the directory: " ${MY_DIR}
|
||||
@cd .. ;\
|
||||
zip -r ${MY_DIR}.zip ${MY_DIR} *default.mk ;\
|
||||
cd ${MY_DIR}
|
||||
|
||||
doc:
|
||||
doxygen Doxyfile
|
||||
|
||||
#########################################################################
|
||||
.SUFFIXES: .f90
|
||||
|
||||
.cpp.o:
|
||||
$(CXX) -c $(CXXFLAGS) -o $@ $<
|
||||
# $(CXX) -c $(CXXFLAGS) -o $@ $< 2>&1 | tee -a $<.log
|
||||
# $(CXX) -c $(CXXFLAGS) -o $@ $< 2>&1 | tee -a $(<:.cpp=.log)
|
||||
|
||||
.c.o:
|
||||
$(CC) -c $(CFLAGS) -o $@ $<
|
||||
|
||||
.f.o:
|
||||
$(F77) -c $(FFLAGS) -o $@ $<
|
||||
|
||||
.f90.o:
|
||||
$(F77) -c $(FFLAGS) -o $@ $<
|
||||
|
||||
##################################################################################################
|
||||
# some tools
|
||||
# Cache behaviour (CXXFLAGS += -g tracks down to source lines; no -pg in linkflags)
|
||||
cache: ${PROGRAM}
|
||||
valgrind --tool=callgrind --simulate-cache=yes ./$^ ${PARAMS}
|
||||
# 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 ./$^ ${PARAMS}
|
||||
# Graphical interface
|
||||
# valkyrie
|
||||
|
||||
# Simple run time profiling of your code
|
||||
# CXXFLAGS += -g -pg
|
||||
# LINKFLAGS += -pg
|
||||
prof: ${PROGRAM}
|
||||
perf record ./$^ ${PARAMS}
|
||||
perf report
|
||||
# gprof -b ./$^ > gp.out
|
||||
# kprof -f gp.out -p gprof &
|
||||
|
||||
# perf in Ubuntu 20.04: https://www.howtoforge.com/how-to-install-perf-performance-analysis-tool-on-ubuntu-20-04/
|
||||
# * install
|
||||
# * sudo vi /etc/sysctl.conf
|
||||
# add kernel.perf_event_paranoid = 0
|
||||
|
||||
#Trace your heap:
|
||||
#> heaptrack ./main.GCC_
|
||||
#> heaptrack_gui heaptrack.main.GCC_.<pid>.gz
|
||||
heap: ${PROGRAM}
|
||||
heaptrack ./$^ ${PARAMS} 11
|
||||
heaptrack_gui `ls -1tr heaptrack.$^.* |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
|
||||
lscpu
|
||||
inxi -C
|
||||
lstopo
|
||||
|
||||
# Excellent hardware info
|
||||
# hardinfo
|
||||
# Life monitoring of CPU frequency etc.
|
||||
# sudo i7z
|
||||
|
||||
# Memory consumption
|
||||
# vmstat -at -SM 3
|
||||
# xfce4-taskmanager
|
||||
|
||||
|
||||
# https://www.tecmint.com/check-linux-cpu-information/
|
||||
#https://www.tecmint.com/monitor-cpu-and-gpu-temperature-in-ubuntu/
|
||||
|
||||
# Debugging:
|
||||
# https://wiki.archlinux.org/index.php/Debugging
|
||||
111
ex5/GCC_SINGLE_default.mk
Normal file
111
ex5/GCC_SINGLE_default.mk
Normal file
|
|
@ -0,0 +1,111 @@
|
|||
# Basic Defintions for using GNU-compiler suite sequentially
|
||||
# requires setting of COMPILER=GCC_
|
||||
|
||||
CC = gcc
|
||||
CXX = g++
|
||||
F77 = gfortran
|
||||
LINKER = ${CXX}
|
||||
|
||||
# on mephisto:
|
||||
#CXXFLAGS += -I/share/apps/atlas/include
|
||||
#LINKFLAGS += -L/share/apps/atlas/lib -L/usr/lib64/atlas
|
||||
#LINKFLAGS += -latlas -lcblas
|
||||
#LINKFLAGS += -lblas
|
||||
# Der <cblas.h> Header muss mit extern "C" versehen werden, damit g++ alles findet.
|
||||
|
||||
|
||||
WARNINGS = -Wall -pedantic -Wextra -Weffc++ -Woverloaded-virtual -Wfloat-equal -Wshadow \
|
||||
-Wredundant-decls -Winline -fmax-errors=1
|
||||
# -Wunreachable-code
|
||||
#CXXFLAGS += -std=c++17 -ffast-math -O3 -march=native -DNDEBUG ${WARNINGS}
|
||||
CXXFLAGS += -std=c++17 -ffast-math -O3 -march=native ${WARNINGS}
|
||||
|
||||
# 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
|
||||
|
||||
# interprocedural optimization
|
||||
#CXXFLAGS += -flto
|
||||
LINKFLAGS += -flto
|
||||
|
||||
# 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 but no OpenMP in Single mode
|
||||
#CXXFLAGS += -fopenmp
|
||||
LINKFLAGS += -fopenmp
|
||||
|
||||
default: ${PROGRAM}
|
||||
|
||||
${PROGRAM}: ${OBJECTS}
|
||||
$(LINKER) $^ ${LINKFLAGS} -o $@
|
||||
|
||||
clean:
|
||||
@rm -f ${PROGRAM} ${OBJECTS}
|
||||
|
||||
clean_all:: clean
|
||||
@rm -f *_ *~ *.bak *.log *.out *.tar *.orig *.optrpt
|
||||
@rm -rf html
|
||||
|
||||
run: clean ${PROGRAM}
|
||||
# time ./${PROGRAM}
|
||||
./${PROGRAM}
|
||||
|
||||
# tar the current directory
|
||||
MY_DIR = `basename ${PWD}`
|
||||
tar:
|
||||
@echo "Tar the directory: " ${MY_DIR}
|
||||
@cd .. ;\
|
||||
tar cf ${MY_DIR}.tar ${MY_DIR} *default.mk ;\
|
||||
cd ${MY_DIR}
|
||||
# tar cf `basename ${PWD}`.tar *
|
||||
|
||||
doc:
|
||||
doxygen Doxyfile
|
||||
|
||||
#########################################################################
|
||||
|
||||
.cpp.o:
|
||||
$(CXX) -c $(CXXFLAGS) -o $@ $<
|
||||
|
||||
.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
|
||||
mem: ${PROGRAM}
|
||||
valgrind -v --leak-check=yes --tool=memcheck --undef-value-errors=yes --track-origins=yes --log-file=$^.addr.out --show-reachable=yes ./$^
|
||||
|
||||
|
||||
thread:${PROGRAM}
|
||||
valgrind -v --tool=helgrind --log-file=$^.thread.out ./$^
|
||||
|
||||
# Simple run time profiling of your code
|
||||
# CXXFLAGS += -g -pg
|
||||
# LINKFLAGS += -pg
|
||||
prof: ${PROGRAM}
|
||||
./$^
|
||||
gprof -b ./$^ > gp.out
|
||||
# kprof -f gp.out -p gprof &
|
||||
182
ex5/GCC_default.mk
Normal file
182
ex5/GCC_default.mk
Normal file
|
|
@ -0,0 +1,182 @@
|
|||
# Basic Defintions for using GNU-compiler suite sequentially
|
||||
# requires setting of COMPILER=GCC_
|
||||
|
||||
CC = gcc
|
||||
CXX = g++
|
||||
F77 = gfortran
|
||||
LINKER = ${CXX}
|
||||
|
||||
WARNINGS = -Wall -pedantic -Wextra -Weffc++ -Woverloaded-virtual -Wfloat-equal -Wshadow \
|
||||
-Wredundant-decls -Winline -fmax-errors=1
|
||||
# -Wunreachable-code
|
||||
CXXFLAGS += -ffast-math -O3 -march=native -std=c++20 ${WARNINGS}
|
||||
#CXXFLAGS += -Ofast -funroll-all-loops -std=c++17 ${WARNINGS}
|
||||
#-msse3
|
||||
# -ftree-vectorizer-verbose=2 -DNDEBUG
|
||||
# -ftree-vectorizer-verbose=5
|
||||
# -ftree-vectorize -fdump-tree-vect-blocks=foo.dump -fdump-tree-pre=stderr
|
||||
|
||||
# CFLAGS = -ffast-math -O3 -DNDEBUG -msse3 -fopenmp -fdump-tree-vect-details
|
||||
# CFLAGS = -ffast-math -O3 -funroll-loops -DNDEBUG -msse3 -fopenmp -ftree-vectorizer-verbose=2
|
||||
# #CFLAGS = -ffast-math -O3 -DNDEBUG -msse3 -fopenmp
|
||||
# FFLAGS = -ffast-math -O3 -DNDEBUG -msse3 -fopenmp
|
||||
# LFLAGS = -ffast-math -O3 -DNDEBUG -msse3 -fopenmp
|
||||
LINKFLAGS += -O3
|
||||
|
||||
#architecture
|
||||
#CPU = -march=znver2
|
||||
CXXFLAGS += ${CPU}
|
||||
LINKFLAGS += ${CPU} -ltbb
|
||||
|
||||
# different libraries in Ubuntu or manajaró
|
||||
ifndef UBUNTU
|
||||
UBUNTU=1
|
||||
endif
|
||||
|
||||
# BLAS, LAPACK
|
||||
ifeq ($(UBUNTU),1)
|
||||
LINKFLAGS += -llapack -lblas
|
||||
# -lopenblas
|
||||
else
|
||||
# on archlinux
|
||||
LINKFLAGS += -llapack -lopenblas -lcblas
|
||||
endif
|
||||
|
||||
# interprocedural optimization
|
||||
CXXFLAGS += -flto
|
||||
LINKFLAGS += -flto
|
||||
|
||||
# for debugging purpose (save code)
|
||||
# -fsanitize=leak # only one out the three 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
|
||||
|
||||
default: ${PROGRAM}
|
||||
|
||||
${PROGRAM}: ${OBJECTS}
|
||||
$(LINKER) $^ ${LINKFLAGS} -o $@
|
||||
|
||||
clean:
|
||||
@rm -f ${PROGRAM} ${OBJECTS}
|
||||
|
||||
clean_all:: clean
|
||||
-@rm -f *_ *~ *.bak *.log *.out *.tar *.orig *.optrpt
|
||||
-@rm -rf html
|
||||
|
||||
run: clean ${PROGRAM}
|
||||
#run: ${PROGRAM}
|
||||
# time ./${PROGRAM} ${PARAMS}
|
||||
./${PROGRAM} ${PARAMS}
|
||||
|
||||
# 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 *
|
||||
#find . -size +10M > large_files
|
||||
#--exclude-from ${MY_DIR}/large_files
|
||||
|
||||
zip: clean
|
||||
@echo "Zip the directory: " ${MY_DIR}
|
||||
@cd .. ;\
|
||||
zip -r ${MY_DIR}.zip ${MY_DIR} *default.mk ;\
|
||||
cd ${MY_DIR}
|
||||
|
||||
doc:
|
||||
doxygen Doxyfile
|
||||
|
||||
#########################################################################
|
||||
.SUFFIXES: .f90
|
||||
|
||||
.cpp.o:
|
||||
$(CXX) -c $(CXXFLAGS) -o $@ $<
|
||||
# $(CXX) -c $(CXXFLAGS) -o $@ $< 2>&1 | tee -a $<.log
|
||||
# $(CXX) -c $(CXXFLAGS) -o $@ $< 2>&1 | tee -a $(<:.cpp=.log)
|
||||
|
||||
.c.o:
|
||||
$(CC) -c $(CFLAGS) -o $@ $<
|
||||
|
||||
.f.o:
|
||||
$(F77) -c $(FFLAGS) -o $@ $<
|
||||
|
||||
.f90.o:
|
||||
$(F77) -c $(FFLAGS) -o $@ $<
|
||||
|
||||
##################################################################################################
|
||||
# some tools
|
||||
# Cache behaviour (CXXFLAGS += -g tracks down to source lines; no -pg in linkflags)
|
||||
cache: ${PROGRAM}
|
||||
valgrind --tool=callgrind --simulate-cache=yes ./$^ ${PARAMS}
|
||||
# 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 ./$^ ${PARAMS}
|
||||
# Graphical interface
|
||||
# valkyrie
|
||||
|
||||
# Simple run time profiling of your code
|
||||
# CXXFLAGS += -g -pg
|
||||
# LINKFLAGS += -pg
|
||||
prof: ${PROGRAM}
|
||||
perf record ./$^ ${PARAMS}
|
||||
perf report
|
||||
# gprof -b ./$^ > gp.out
|
||||
# kprof -f gp.out -p gprof &
|
||||
|
||||
# perf in Ubuntu 20.04: https://www.howtoforge.com/how-to-install-perf-performance-analysis-tool-on-ubuntu-20-04/
|
||||
# * install
|
||||
# * sudo vi /etc/sysctl.conf
|
||||
# add kernel.perf_event_paranoid = 0
|
||||
|
||||
#Trace your heap:
|
||||
#> heaptrack ./main.GCC_
|
||||
#> heaptrack_gui heaptrack.main.GCC_.<pid>.gz
|
||||
heap: ${PROGRAM}
|
||||
heaptrack ./$^ ${PARAMS} 11
|
||||
heaptrack_gui `ls -1tr heaptrack.$^.* |tail -1` &
|
||||
|
||||
codecheck: $(SOURCES)
|
||||
cppcheck --enable=all --inconclusive --std=c++17 -I${CUDA_INC} --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
|
||||
lscpu
|
||||
inxi -C
|
||||
lstopo
|
||||
|
||||
# Excellent hardware info
|
||||
# hardinfo
|
||||
# Life monitoring of CPU frequency etc.
|
||||
# sudo i7z
|
||||
|
||||
# Memory consumption
|
||||
# vmstat -at -SM 3
|
||||
# xfce4-taskmanager
|
||||
|
||||
|
||||
# https://www.tecmint.com/check-linux-cpu-information/
|
||||
#https://www.tecmint.com/monitor-cpu-and-gpu-temperature-in-ubuntu/
|
||||
|
||||
# Debugging:
|
||||
# https://wiki.archlinux.org/index.php/Debugging
|
||||
151
ex5/ICC_default.mk
Normal file
151
ex5/ICC_default.mk
Normal file
|
|
@ -0,0 +1,151 @@
|
|||
# Basic Defintions for using INTEL compiler suite sequentially
|
||||
# requires setting of COMPILER=ICC_
|
||||
|
||||
# special on my sony [GH]
|
||||
#BINDIR = /opt/save.intel/bin/
|
||||
# very special on my sony [GH]
|
||||
# FIND_LIBS = -L /opt/save.intel/composer_xe_2013.1.117/mkl/lib/intel64/libmkl_intel_lp64.so
|
||||
|
||||
#export KMP_AFFINITY=verbose,compact
|
||||
|
||||
CC = ${BINDIR}icc
|
||||
CXX = ${BINDIR}icpc
|
||||
F77 = ${BINDIR}ifort
|
||||
LINKER = ${CXX}
|
||||
|
||||
WARNINGS = -pedantic -Wall -Weffc++ -Woverloaded-virtual -Wfloat-equal -Wshadow -wd2015,2012
|
||||
#-Winline -Wunreachable-code -Wredundant-decls
|
||||
CXXFLAGS += -std=c++17 -O3 -fma -DNDEBUG ${WARNINGS} -mkl
|
||||
#CXXFLAGS += -std=c++17 -O3 -march=core-avx2 -fma -ftz -fomit-frame-pointer -DNDEBUG ${WARNINGS} -mkl
|
||||
# -fast # fast inludes also -ipo !
|
||||
CXXFLAGS += -fargument-noalias -fargument-noalias-global -ansi-alias
|
||||
CXXFLAGS += -align -qopt-dynamic-align
|
||||
#CXXFLAGS += -xCore-AVX2
|
||||
#CXXFLAGS += -tp=zen
|
||||
# -qopt-subscript-in-range
|
||||
# -vec-threshold0
|
||||
# -xCORE-AVX2
|
||||
# -axcode COMMON-AVX512 -axcode MIC-AVX512 -axcode CORE-AVX512 -axcode CORE-AVX2
|
||||
# -ipo
|
||||
|
||||
# Reports: https://software.intel.com/en-us/articles/getting-the-most-out-of-your-intel-compiler-with-the-new-optimization-reports
|
||||
#CXXFLAGS += -qopt-report=5 -qopt-report-phase=vec,par
|
||||
|
||||
#CXXFLAGS += -qopt-report=5 -qopt-report-phase=cg
|
||||
# Redirect report from *.optrpt to stderr
|
||||
# -qopt-report-file=stderr
|
||||
# Guided paralellization
|
||||
# -guide -parallel
|
||||
# -guide-opts=string -guide-par[=n] -guide-vec[=n]
|
||||
# -auto-p32 -simd
|
||||
|
||||
# interprocedural optimization
|
||||
#CXXFLAGS += -ipo
|
||||
#LINKFLAGS += -ipo
|
||||
|
||||
# annotated Assembler file
|
||||
ANNOTED = -fsource-asm -S
|
||||
|
||||
# OpenMP
|
||||
CXXFLAGS += -qopenmp
|
||||
# -qopt-report-phase=openmp
|
||||
# -diag-enable=sc-full -diag-file=filename -diag-file-append[=filename]
|
||||
LINKFLAGS += -qopenmp
|
||||
|
||||
# use MKL by INTEL
|
||||
# LINKFLAGS += -L${BINDIR}../composer_xe_2013.1.117/mkl/lib/intel64 -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lpthread
|
||||
LINKFLAGS += -O2 -mkl
|
||||
# -ipo
|
||||
|
||||
|
||||
|
||||
default: ${PROGRAM}
|
||||
|
||||
${PROGRAM}: ${OBJECTS}
|
||||
$(LINKER) $^ ${LINKFLAGS} -o $@
|
||||
|
||||
clean:
|
||||
rm -f ${PROGRAM} ${OBJECTS} *.optrpt
|
||||
|
||||
clean_all:: clean
|
||||
@rm -f *_ *~ *.bak *.log *.out *.tar
|
||||
|
||||
run: clean ${PROGRAM}
|
||||
./${PROGRAM}
|
||||
|
||||
# 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 *
|
||||
|
||||
doc:
|
||||
doxygen Doxyfile
|
||||
|
||||
#########################################################################
|
||||
|
||||
.cpp.o:
|
||||
$(CXX) -c $(CXXFLAGS) -o $@ $<
|
||||
|
||||
.c.o:
|
||||
$(CC) -c $(CFLAGS) -o $@ $<
|
||||
|
||||
.f.o:
|
||||
$(F77) -c $(FFLAGS) -o $@ $<
|
||||
|
||||
##################################################################################################
|
||||
# # some tools
|
||||
# # Cache behaviour (CXXFLAGS += -g tracks down to source lines)
|
||||
# cache: ${PROGRAM}
|
||||
# valgrind --tool=callgrind --simulate-cache=yes ./$^
|
||||
# # kcachegrind callgrind.out.<pid> &
|
||||
#
|
||||
# # Check for wrong memory accesses, memory leaks, ...
|
||||
# # use smaller data sets
|
||||
# mem: ${PROGRAM}
|
||||
# valgrind -v --leak-check=yes --tool=memcheck --undef-value-errors=yes --track-origins=yes --log-file=$^.addr.out --show-reachable=yes ./$^
|
||||
#
|
||||
# # Simple run time profiling of your code
|
||||
# # CXXFLAGS += -g -pg
|
||||
# # LINKFLAGS += -pg
|
||||
# prof: ${PROGRAM}
|
||||
# ./$^
|
||||
# gprof -b ./$^ > gp.out
|
||||
# # kprof -f gp.out -p gprof &
|
||||
#
|
||||
|
||||
|
||||
mem: inspector
|
||||
prof: amplifier
|
||||
cache: amplifier
|
||||
|
||||
gap_par_report:
|
||||
${CXX} -c -guide -parallel $(SOURCES) 2> gap.txt
|
||||
|
||||
# GUI for performance report
|
||||
amplifier: ${PROGRAM}
|
||||
echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope
|
||||
echo 0 | sudo tee /proc/sys/kernel/perf_event_paranoid
|
||||
amplxe-gui &
|
||||
|
||||
# GUI for Memory and Thread analyzer (race condition)
|
||||
inspector: ${PROGRAM}
|
||||
echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope
|
||||
#${BINDIR}../inspector_xe_2013/bin64/inspxe-gui &
|
||||
inspxe-gui &
|
||||
|
||||
advisor:
|
||||
echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope
|
||||
# https://software.intel.com/en-us/articles/intel-advisor-2017-update-1-what-s-new
|
||||
export ADVIXE_EXPERIMENTAL=roofline
|
||||
advixe-gui &
|
||||
|
||||
icc-info:
|
||||
icpc -# main.cpp
|
||||
|
||||
|
||||
|
||||
|
||||
181
ex5/ONEAPI_default.mk
Normal file
181
ex5/ONEAPI_default.mk
Normal file
|
|
@ -0,0 +1,181 @@
|
|||
# Basic Defintions for using INTEL compiler suite sequentially
|
||||
# requires setting of COMPILER=ONEAPI_
|
||||
|
||||
# https://software.intel.com/content/www/us/en/develop/tools/oneapi/components/onemkl/link-line-advisor.html
|
||||
# requires
|
||||
# source /opt/intel/oneapi/setvars.sh
|
||||
# on AMD: export MKL_DEBUG_CPU_TYPE=5
|
||||
|
||||
#BINDIR = /opt/intel/oneapi/compiler/latest/linux/bin/
|
||||
#MKL_ROOT = /opt/intel/oneapi/mkl/latest/
|
||||
#export KMP_AFFINITY=verbose,compact
|
||||
|
||||
CC = ${BINDIR}icc
|
||||
CXX = ${BINDIR}dpcpp
|
||||
F77 = ${BINDIR}ifort
|
||||
LINKER = ${CXX}
|
||||
|
||||
## Compiler flags
|
||||
WARNINGS = -Wall -Weffc++ -Woverloaded-virtual -Wfloat-equal -Wshadow -pedantic
|
||||
WARNINGS += -Wpessimizing-move -Wredundant-move
|
||||
#-wd2015,2012,2014 -wn3
|
||||
# -Winline -Wredundant-decls -Wunreachable-code
|
||||
# -qopt-subscript-in-range
|
||||
# -vec-threshold0
|
||||
|
||||
CXXFLAGS += -O3 -std=c++17 -tbb ${WARNINGS}
|
||||
# https://stackoverflow.com/questions/67923287/how-to-resolve-no-member-named-task-in-namespace-tbb-error-when-using-oned
|
||||
# needed on Desktop-PC Haase (not needed on mephisto)
|
||||
CXXFLAGS += -D_GLIBCXX_USE_TBB_PAR_BACKEND=0
|
||||
#CXXFLAGS += -DMKL_ILP64 -I"${MKLROOT}/include"
|
||||
#CXXFLAGS += -DMKL_ILP32 -I"${MKLROOT}/include"
|
||||
LINKFLAGS += -O3 -tbb
|
||||
#LINKFLAGS += -no-prec-div
|
||||
|
||||
# interprocedural optimization
|
||||
CXXFLAGS += -ipo
|
||||
LINKFLAGS += -ipo
|
||||
LINKFLAGS += -flto
|
||||
|
||||
# annotated Assembler file
|
||||
ANNOTED = -fsource-asm -S
|
||||
|
||||
#architecture
|
||||
CPU = -march=core-avx2
|
||||
#CPU += -mtp=zen
|
||||
# -xCORE-AVX2
|
||||
# -axcode COMMON-AVX512 -axcode MIC-AVX512 -axcode CORE-AVX512 -axcode CORE-AVX2
|
||||
CXXFLAGS += ${CPU}
|
||||
LINKFLAGS += ${CPU}
|
||||
|
||||
# use MKL by INTEL
|
||||
# https://software.intel.com/content/www/us/en/develop/tools/oneapi/components/onemkl/link-line-advisor.html
|
||||
# sequential MKL
|
||||
# use the 32 bit interface (LP64) instead of 64 bit interface (ILP64)
|
||||
CXXFLAGS += -qmkl=sequential -UMKL_ILP64
|
||||
LINKFLAGS += -O3 -qmkl=sequential -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lpthread
|
||||
#LINKFLAGS += -O3 -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lpthread
|
||||
|
||||
# shared libs: https://aur.archlinux.org/packages/intel-oneapi-compiler-static
|
||||
# install intel-oneapi-compiler-static
|
||||
# or
|
||||
LINKFLAGS += -shared-intel
|
||||
|
||||
|
||||
OPENMP = -qopenmp
|
||||
CXXFLAGS += ${OPENMP}
|
||||
LINKFLAGS += ${OPENMP}
|
||||
|
||||
|
||||
# profiling tools
|
||||
#CXXFLAGS += -pg
|
||||
#LINKFLAGS += -pg
|
||||
# -vec-report=3
|
||||
# -qopt-report=5 -qopt-report-phase=vec -qopt-report-phase=openmp
|
||||
# -guide -parallel
|
||||
# -guide-opts=string -guide-par[=n] -guide-vec[=n]
|
||||
# -auto-p32 -simd
|
||||
|
||||
# Reports: https://software.intel.com/en-us/articles/getting-the-most-out-of-your-intel-compiler-with-the-new-optimization-reports
|
||||
#CXXFLAGS += -qopt-report=5 -qopt-report-phase=vec,par
|
||||
#CXXFLAGS += -qopt-report=5 -qopt-report-phase=cg
|
||||
# Redirect report from *.optrpt to stderr
|
||||
# -qopt-report-file=stderr
|
||||
# Guided paralellization
|
||||
# -guide -parallel
|
||||
# -guide-opts=string -guide-par[=n] -guide-vec[=n]
|
||||
# -auto-p32 -simd
|
||||
|
||||
## run time checks
|
||||
# https://www.intel.com/content/www/us/en/develop/documentation/fortran-compiler-oneapi-dev-guide-and-reference/top/compiler-reference/compiler-options/offload-openmp-and-parallel-processing-options/par-runtime-control-qpar-runtime-control.html
|
||||
|
||||
|
||||
default: ${PROGRAM}
|
||||
|
||||
${PROGRAM}: ${OBJECTS}
|
||||
$(LINKER) $^ ${LINKFLAGS} -o $@
|
||||
|
||||
clean:
|
||||
rm -f ${PROGRAM} ${OBJECTS} *.optrpt
|
||||
|
||||
clean_all:: clean
|
||||
@rm -f *_ *~ *.bak *.log *.out *.tar
|
||||
|
||||
run: clean ${PROGRAM}
|
||||
./${PROGRAM} ${PARAMS}
|
||||
|
||||
# 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 *
|
||||
|
||||
doc:
|
||||
doxygen Doxyfile
|
||||
|
||||
#########################################################################
|
||||
|
||||
.cpp.o:
|
||||
$(CXX) -c $(CXXFLAGS) -o $@ $<
|
||||
|
||||
.c.o:
|
||||
$(CC) -c $(CFLAGS) -o $@ $<
|
||||
|
||||
.f.o:
|
||||
$(F77) -c $(FFLAGS) -o $@ $<
|
||||
|
||||
##################################################################################################
|
||||
# some tools
|
||||
# Cache behaviour (CXXFLAGS += -g tracks down to source lines)
|
||||
# https://software.intel.com/content/www/us/en/develop/documentation/vtune-help/top/analyze-performance/microarchitecture-analysis-group/memory-access-analysis.html
|
||||
|
||||
mem: inspector
|
||||
prof: vtune
|
||||
cache: inspector
|
||||
|
||||
gap_par_report:
|
||||
${CXX} -c -guide -parallel $(SOURCES) 2> gap.txt
|
||||
|
||||
# GUI for performance report
|
||||
amplifier: ${PROGRAM}
|
||||
echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope
|
||||
echo 0 | sudo tee /proc/sys/kernel/perf_event_paranoid
|
||||
amplxe-gui &
|
||||
|
||||
# GUI for Memory and Thread analyzer (race condition)
|
||||
inspector: ${PROGRAM}
|
||||
echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope
|
||||
# inspxe-gui &
|
||||
vtune-gui ./${PROGRAM} &
|
||||
|
||||
advisor:
|
||||
# source /opt/intel/oneapi/advisor/2021.2.0/advixe-vars.sh
|
||||
# /opt/intel/oneapi/advisor/latest/bin64/advixe-gui &
|
||||
advisor --collect=survey ./${PROGRAM} ${PARAMS}
|
||||
# advisor --collect=roofline ./${PROGRAM} ${PARAMS}
|
||||
advisor --report=survey --project-dir=./ src:r=./ --format=csv --report-output=./out/survey.csv
|
||||
# advisor-gui
|
||||
|
||||
vtune:
|
||||
echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope
|
||||
# https://software.intel.com/en-us/articles/intel-advisor-2017-update-1-what-s-new
|
||||
export ADVIXE_EXPERIMENTAL=roofline
|
||||
vtune -collect hotspots ./${PROGRAM} ${PARAMS}
|
||||
vtune -report hotspots -r r000hs > vtune.out
|
||||
# vtune-gui ./${PROGRAM} &
|
||||
|
||||
icc-info:
|
||||
icpc -# main.cpp
|
||||
|
||||
# MKL on AMD
|
||||
# https://www.computerbase.de/2019-11/mkl-workaround-erhoeht-leistung-auf-amd-ryzen/
|
||||
#
|
||||
# https://sites.google.com/a/uci.edu/mingru-yang/programming/mkl-has-bad-performance-on-an-amd-cpu
|
||||
# export MKL_DEBUG_CPU_TYPE=5
|
||||
# export MKL_NUM_THRAEDS=1
|
||||
# export MKL_DYNAMIC=false
|
||||
# on Intel compiler
|
||||
# http://publicclu2.blogspot.com/2013/05/intel-complier-suite-reference-card.html
|
||||
96
ex5/PGI_default.mk
Normal file
96
ex5/PGI_default.mk
Normal file
|
|
@ -0,0 +1,96 @@
|
|||
# Basic Defintions for using PGI-compiler suite sequentially
|
||||
# requires setting of COMPILER=PGI_
|
||||
# OPTIRUN = optirun
|
||||
# on mephisto:
|
||||
#CXXFLAGS += -I/share/apps/atlas/include
|
||||
#LINKFLAGS += -L/share/apps/atlas/lib
|
||||
#LINKFLAGS += -lcblas -latlas
|
||||
|
||||
LINKFLAGS += -lblas
|
||||
|
||||
CC = pgcc
|
||||
CXX = pgc++
|
||||
F77 = pgfortran
|
||||
LINKER = ${CXX}
|
||||
|
||||
|
||||
WARNINGS = -Minform=warn
|
||||
|
||||
#PGI_PROFILING = -Minfo=loop,vect,opt,intensity,mp,accel
|
||||
PGI_PROFILING = -Minfo=ccff,accel,ipa,loop,lre,mp,opt,par,unified,vect,intensity
|
||||
|
||||
# -Minfo
|
||||
# -Mprof=lines
|
||||
|
||||
CXXFLAGS += -std=c++14 -O3 -fast -DNDEBUG ${PGI_PROFILING} ${WARNINGS}
|
||||
CXXFLAGS += -Mvect -Mcache_align -Msafeptr -Mprefetch -Mlre -Mdepchk
|
||||
#-Msmart
|
||||
|
||||
LINKFLAGS += ${PGI_PROFILING}
|
||||
#-lcblas
|
||||
# OpenMP
|
||||
CXXFLAGS += -mp=align,bind,numa -Mneginfo=mp
|
||||
LINKFLAGS += -mp=allcores,bind,numa
|
||||
|
||||
default: ${PROGRAM}
|
||||
|
||||
${PROGRAM}: ${OBJECTS}
|
||||
$(LINKER) $^ ${LINKFLAGS} -o $@
|
||||
|
||||
clean:
|
||||
@rm -f ${PROGRAM} ${OBJECTS}
|
||||
|
||||
clean_all:: clean
|
||||
@rm -f *_ *~ *.bak *.log *.out *.tar
|
||||
|
||||
run: clean ${PROGRAM}
|
||||
./${PROGRAM}
|
||||
|
||||
# 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 *
|
||||
|
||||
doc:
|
||||
doxygen Doxyfile
|
||||
|
||||
#########################################################################
|
||||
|
||||
.cpp.o:
|
||||
$(CXX) -c $(CXXFLAGS) -o $@ $<
|
||||
|
||||
.c.o:
|
||||
$(CC) -c $(CFLAGS) -o $@ $<
|
||||
|
||||
.f.o:
|
||||
$(F77) -c $(FFLAGS) -o $@ $<
|
||||
|
||||
##################################################################################################
|
||||
# # some tools
|
||||
# # Simple run time profiling of your code
|
||||
# # CXXFLAGS += -g
|
||||
# # LINKFLAGS +=
|
||||
|
||||
|
||||
# Profiling options PGI, see: pgprof -h
|
||||
PROF_FILE = jac.pgprof
|
||||
# CPU_PROF = -allcache
|
||||
CPU_PROF = --cpu-profiling on --analysis-metrics
|
||||
# GPU_PROF = -cuda=gmem,branch,cc13 -cudainit
|
||||
#GPU_PROF = -cuda=branch:cc20
|
||||
#
|
||||
|
||||
cache: prof
|
||||
|
||||
prof: ${PROGRAM}
|
||||
# ./$^
|
||||
# $(CUDA_HOME)/bin/nvvp &
|
||||
# more /opt/pgi/linux86-64/16.10/bin/pgcollectrc
|
||||
${OPTIRUN} ${BINDIR}pgprof ${CPU_PROF} -o $(PROF_FILE) ./$^
|
||||
${OPTIRUN} ${BINDIR}pgprof -i $(PROF_FILE) 2> prof.out
|
||||
|
||||
|
||||
30
ex5/ex5_1/Makefile
Normal file
30
ex5/ex5_1/Makefile
Normal file
|
|
@ -0,0 +1,30 @@
|
|||
#
|
||||
# use GNU-Compiler tools
|
||||
COMPILER=GCC_
|
||||
# alternatively from the shell
|
||||
# export COMPILER=GCC_
|
||||
# or, alternatively from the shell
|
||||
# make COMPILER=GCC_
|
||||
|
||||
# use Intel compilers
|
||||
#COMPILER=ICC_
|
||||
|
||||
# use PGI compilers
|
||||
# COMPILER=PGI_
|
||||
|
||||
|
||||
SOURCES = main.cpp mylib.cpp
|
||||
OBJECTS = $(SOURCES:.cpp=.o)
|
||||
|
||||
PROGRAM = main.${COMPILER}
|
||||
|
||||
# uncomment the next to lines for debugging and detailed performance analysis
|
||||
CXXFLAGS += -g
|
||||
LINKFLAGS += -g
|
||||
# do not use -pg with PGI compilers
|
||||
|
||||
ifndef COMPILER
|
||||
COMPILER=GCC_
|
||||
endif
|
||||
|
||||
include ../${COMPILER}default.mk
|
||||
99
ex5/ex5_1/check_env.h
Normal file
99
ex5/ex5_1/check_env.h
Normal file
|
|
@ -0,0 +1,99 @@
|
|||
#pragma once
|
||||
|
||||
#include <iostream>
|
||||
#ifdef _OPENMP
|
||||
#include <omp.h>
|
||||
#endif
|
||||
#include <unordered_map>
|
||||
|
||||
//#####################################
|
||||
// G.Haase
|
||||
// See https://sourceforge.net/p/predef/wiki/Compilers/
|
||||
// http://www.cplusplus.com/doc/tutorial/preprocessor/
|
||||
// also: export OMP_DISPLAY_ENV=VERBOSE
|
||||
//#####################################
|
||||
/** Checks for compilers, its versions, threads etc.
|
||||
*
|
||||
@param[in] argc number of command line arguemnts
|
||||
@param[in] argv command line arguments as array of C-strings
|
||||
*/
|
||||
template <class T>
|
||||
void check_env(T argc, char const *argv[])
|
||||
{
|
||||
std::cout << "\n#######################################################################\n";
|
||||
std::cout << "Code :";
|
||||
for (T k = 0; k < argc; ++k) std::cout << " " << argv[k];
|
||||
std::cout << std::endl;
|
||||
|
||||
// compiler: https://sourceforge.net/p/predef/wiki/Compilers/
|
||||
std::cout << "Compiler: ";
|
||||
#if defined __INTEL_COMPILER
|
||||
#pragma message(" ########## INTEL ###############")
|
||||
std::cout << "INTEL " << __INTEL_COMPILER;
|
||||
// Ignore warnings for #pragma acc unrecognice
|
||||
#pragma warning disable 161
|
||||
// Ignore warnings for #pragma omp unrecognice
|
||||
#pragma warning disable 3180
|
||||
|
||||
#elif defined __PGI
|
||||
#pragma message(" ########## PGI ###############")
|
||||
std::cout << "PGI " << __PGIC__ << "." << __PGIC_MINOR__ << "." << __PGIC_PATCHLEVEL__;
|
||||
#elif defined __clang__
|
||||
#pragma message(" ########## CLANG ###############")
|
||||
std::cout << "CLANG " << __clang_major__ << "." << __clang_minor__ << "."; // << __clang_patchlevel__;
|
||||
#elif defined __GNUC__
|
||||
#pragma message(" ########## Gnu ###############")
|
||||
std::cout << "Gnu " << __GNUC__ << "." << __GNUC_MINOR__ << "." << __GNUC_PATCHLEVEL__;
|
||||
#else
|
||||
#pragma message(" ########## unknown Compiler ###############")
|
||||
std::cout << "unknown";
|
||||
#endif
|
||||
std::cout << " C++ standard: " << __cplusplus << std::endl;
|
||||
|
||||
// Parallel environments
|
||||
std::cout << "Parallel: ";
|
||||
#if defined MPI_VERSION
|
||||
#pragma message(" ########## MPI ###############")
|
||||
#ifdef OPEN_MPI
|
||||
std::cout << "OpenMPI ";
|
||||
#else
|
||||
std::cout << "MPI ";
|
||||
#endif
|
||||
std::cout << MPI_VERSION << "." << MPI_SUBVERSION << " ";
|
||||
#endif
|
||||
|
||||
#ifdef _OPENMP
|
||||
//https://www.openmp.org/specifications/
|
||||
//https://stackoverflow.com/questions/1304363/how-to-check-the-version-of-openmp-on-linux
|
||||
std::unordered_map<unsigned, std::string> const map{
|
||||
{200505, "2.5"}, {200805, "3.0"}, {201107, "3.1"}, {201307, "4.0"}, {201511, "4.5"}, {201611, "5.0"}, {201811, "5.0"}};
|
||||
#pragma message(" ########## OPENMP ###############")
|
||||
//std::cout << _OPENMP;
|
||||
std::cout << "OpenMP ";
|
||||
try {
|
||||
std::cout << map.at(_OPENMP);
|
||||
}
|
||||
catch (...) {
|
||||
std::cout << _OPENMP;
|
||||
}
|
||||
#pragma omp parallel
|
||||
{
|
||||
#pragma omp master
|
||||
{
|
||||
const int nn = omp_get_num_threads(); // OpenMP
|
||||
std::cout << " ---> " << nn << " Threads ";
|
||||
}
|
||||
#pragma omp barrier
|
||||
}
|
||||
|
||||
#endif
|
||||
#ifdef _OPENACC
|
||||
#pragma message(" ########## OPENACC ###############")
|
||||
std::cout << "OpenACC ";
|
||||
#endif
|
||||
std::cout << std::endl;
|
||||
std::cout << "Date : " << __DATE__ << " " << __TIME__;
|
||||
std::cout << "\n#######################################################################\n";
|
||||
}
|
||||
// HG
|
||||
|
||||
142
ex5/ex5_1/main.cpp
Normal file
142
ex5/ex5_1/main.cpp
Normal file
|
|
@ -0,0 +1,142 @@
|
|||
#include "check_env.h"
|
||||
#include "mylib.h"
|
||||
#include <cstdlib> // atoi()
|
||||
#include <cstring> // strncmp()
|
||||
#include <ctime>
|
||||
#include <iostream>
|
||||
#include <omp.h> // OpenMP
|
||||
#include <sstream>
|
||||
#include <string>
|
||||
using namespace std;
|
||||
|
||||
int main(int argc, char const *argv[])
|
||||
{
|
||||
omp_set_schedule(omp_sched_static, 2000000);
|
||||
//omp_set_schedule(omp_sched_dynamic, 1000000);
|
||||
//omp_set_schedule(omp_sched_guided, 1000000);
|
||||
//omp_set_schedule(omp_sched_auto, 1); // chunk size does not matter for auto
|
||||
|
||||
|
||||
// Speedup for different number of cores (incl. hyperthreading)
|
||||
omp_set_num_threads(8);
|
||||
|
||||
// Print number of available processors
|
||||
cout << "Number of available processors: " << omp_get_num_procs() << endl;
|
||||
|
||||
// Currently executing parallel code? -> no
|
||||
cout << "Currently in parallel? " << omp_in_parallel() << endl;
|
||||
|
||||
|
||||
int const NLOOPS = 10; // chose a value such that the benchmark runs at least 10 sec.
|
||||
unsigned int N = 500000001;
|
||||
|
||||
|
||||
//##########################################################################
|
||||
// Read Parameter from command line (C++ style)
|
||||
cout << "Checking command line parameters for: -n <number> " << endl;
|
||||
for (int i = 1; i < argc; i++)
|
||||
{
|
||||
cout << " arg[" << i << "] = " << argv[i] << endl;
|
||||
string ss(argv[i]);
|
||||
if ("-n"==ss && i + 1 < argc) // found "-n" followed by another parameter
|
||||
{
|
||||
N = static_cast<unsigned int>(atoi(argv[i + 1]));
|
||||
}
|
||||
else
|
||||
{
|
||||
cout << "Corect call: " << argv[0] << " -n <number>\n";
|
||||
}
|
||||
}
|
||||
|
||||
cout << "\nN = " << N << endl;
|
||||
|
||||
check_env(argc, argv);
|
||||
//########################################################################
|
||||
int nthreads; // OpenMP
|
||||
#pragma omp parallel default(none) shared(cout,nthreads)
|
||||
{
|
||||
stringstream inparallel;
|
||||
inparallel << "Currently in parallel? " << omp_in_parallel() << endl;
|
||||
|
||||
|
||||
int const th_id = omp_get_thread_num(); // OpenMP
|
||||
int const nthrds = omp_get_num_threads(); // OpenMP
|
||||
stringstream ss;
|
||||
ss << "C++: Hello World from thread " << th_id << " / " << nthrds << endl;
|
||||
#pragma omp critical
|
||||
{
|
||||
cout << ss.str(); // output to a shared ressource
|
||||
cout << inparallel.str() << endl;
|
||||
}
|
||||
#pragma omp master
|
||||
nthreads = nthrds; // transfer nn to to master thread
|
||||
}
|
||||
cout << " " << nthreads << " threads have been started." << endl;
|
||||
|
||||
//##########################################################################
|
||||
// Memory allocation
|
||||
cout << "Memory allocation\n";
|
||||
|
||||
vector<double> x(N), y(N);
|
||||
|
||||
cout.precision(2);
|
||||
cout << 2.0 * N *sizeof(x[0]) / 1024 / 1024 / 1024 << " GByte Memory allocated\n";
|
||||
cout.precision(6);
|
||||
|
||||
//##########################################################################
|
||||
// Data initialization
|
||||
// Special: x_i = i+1; y_i = 1/x_i ==> <x,y> == N
|
||||
for (unsigned int i = 0; i < N; ++i)
|
||||
{
|
||||
x[i] = i + 1;
|
||||
y[i] = 1.0 / x[i];
|
||||
}
|
||||
|
||||
//##########################################################################
|
||||
cout << "\nStart Benchmarking\n";
|
||||
|
||||
// Do calculation
|
||||
double tstart = omp_get_wtime(); // OpenMP
|
||||
|
||||
double sk(0.0);
|
||||
for (int i = 0; i < NLOOPS; ++i)
|
||||
{
|
||||
//sk = scalar(x, y);
|
||||
sk = scalar_parallel(x, y);
|
||||
//sk = scalar_trans(x, y);
|
||||
//sk = norm(x);
|
||||
}
|
||||
|
||||
double t1 = omp_get_wtime() - tstart; // OpenMP
|
||||
|
||||
t1 /= NLOOPS; // divide by number of function calls
|
||||
|
||||
//##########################################################################
|
||||
// Check the correct result
|
||||
cout << "\n <x,y> = " << sk << endl;
|
||||
if (static_cast<unsigned int>(sk) != N)
|
||||
{
|
||||
cout << " !! W R O N G result !!\n";
|
||||
}
|
||||
cout << endl;
|
||||
|
||||
//##########################################################################
|
||||
// Timings and Performance
|
||||
cout << endl;
|
||||
cout.precision(2);
|
||||
cout << "Total benchmarking time: " << t1*NLOOPS << endl;
|
||||
cout << "Timing in sec. : " << t1 << endl;
|
||||
cout << "GFLOPS : " << 2.0 * N / t1 / 1024 / 1024 / 1024 << endl;
|
||||
cout << "GiByte/s : " << 2.0 * N / t1 / 1024 / 1024 / 1024 * sizeof(x[0]) << endl;
|
||||
|
||||
//#########################################################################
|
||||
|
||||
cout << "\n Try the reduction with an STL-vektor \n";
|
||||
|
||||
auto vr = reduction_vec_append(5);
|
||||
cout << "done\n";
|
||||
cout << vr << endl;
|
||||
|
||||
|
||||
return 0;
|
||||
} // memory for x and y will be deallocated their destructors
|
||||
137
ex5/ex5_1/mylib.cpp
Normal file
137
ex5/ex5_1/mylib.cpp
Normal file
|
|
@ -0,0 +1,137 @@
|
|||
#include "mylib.h"
|
||||
#include <cassert> // assert()
|
||||
#include <cmath>
|
||||
#include <iostream>
|
||||
#include <functional> // multiplies<>{}
|
||||
#include <list>
|
||||
#include <numeric> // iota()
|
||||
#ifdef _OPENMP
|
||||
#include <omp.h>
|
||||
#endif
|
||||
#include <vector>
|
||||
using namespace std;
|
||||
|
||||
|
||||
double scalar_parallel(vector<double> const &x, vector<double> const &y)
|
||||
{
|
||||
assert(x.size() == y.size()); // switch off via compile flag: -DNDEBUG
|
||||
size_t const N = x.size();
|
||||
double sum = 0.0;
|
||||
#pragma omp parallel default(none) shared(x,y,N, cout) reduction(+:sum)
|
||||
{
|
||||
const size_t nthreads = omp_get_num_threads();
|
||||
const size_t threadnum = omp_get_thread_num();
|
||||
const size_t chunksize = N/nthreads;
|
||||
|
||||
|
||||
size_t start = threadnum*chunksize;
|
||||
size_t end = start + chunksize;
|
||||
if (threadnum == nthreads - 1)
|
||||
end = N;
|
||||
|
||||
|
||||
for (size_t i = start; i < end; ++i)
|
||||
{
|
||||
sum += x[i] * y[i];
|
||||
}
|
||||
|
||||
}
|
||||
return sum;
|
||||
}
|
||||
|
||||
|
||||
vector<int> reduction_vec_append(int n)
|
||||
{
|
||||
vector<int> vec(n);
|
||||
#pragma omp parallel default(none) shared(cout) reduction(VecAppend:vec)
|
||||
{
|
||||
#pragma omp barrier
|
||||
#pragma omp critical
|
||||
cout << omp_get_thread_num() << " : " << vec.size() << endl;
|
||||
#pragma omp barrier
|
||||
iota( vec.begin(),vec.end(), omp_get_thread_num() );
|
||||
#pragma omp barrier
|
||||
|
||||
}
|
||||
return vec;
|
||||
}
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
double scalar(vector<double> const &x, vector<double> const &y)
|
||||
{
|
||||
assert(x.size() == y.size()); // switch off via compile flag: -DNDEBUG
|
||||
size_t const N = x.size();
|
||||
double sum = 0.0;
|
||||
#pragma omp parallel for default(none) shared(x,y,N) reduction(+:sum) schedule(runtime) // added schedule(runtime)
|
||||
for (size_t i = 0; i < N; ++i)
|
||||
{
|
||||
sum += x[i] * y[i];
|
||||
//sum += exp(x[i])*log(y[i]);
|
||||
}
|
||||
return sum;
|
||||
}
|
||||
|
||||
|
||||
double norm(vector<double> const &x)
|
||||
{
|
||||
size_t const N = x.size();
|
||||
double sum = 0.0;
|
||||
#pragma omp parallel for default(none) shared(x,N) reduction(+:sum) schedule(runtime) // added schedule(runtime)
|
||||
for (size_t i = 0; i < N; ++i)
|
||||
{
|
||||
sum += x[i]*x[i];
|
||||
}
|
||||
return sum;
|
||||
}
|
||||
|
||||
|
||||
vector<int> reduction_vec(int n)
|
||||
{
|
||||
vector<int> vec(n);
|
||||
#pragma omp parallel default(none) shared(cout) reduction(VecAdd:vec)
|
||||
{
|
||||
#pragma omp barrier
|
||||
#pragma omp critical
|
||||
cout << omp_get_thread_num() << " : " << vec.size() << endl;
|
||||
#pragma omp barrier
|
||||
iota( vec.begin(),vec.end(), omp_get_thread_num() );
|
||||
#pragma omp barrier
|
||||
|
||||
}
|
||||
return vec;
|
||||
}
|
||||
|
||||
|
||||
double scalar_trans(vector<double> const &x, vector<double> const &y)
|
||||
{
|
||||
assert(x.size() == y.size()); // switch off via compile flag: -DNDEBUG
|
||||
vector<double> z(x.size());
|
||||
//list<double> z(x.size()); // parallel for-loop on iterators not possible (missing 'operator-')
|
||||
// c++-20 CLANG_, ONEAPI_:condition of OpenMP for loop must be a relational comparison
|
||||
|
||||
transform(cbegin(x),cend(x),cbegin(y),begin(z),std::multiplies<>{});
|
||||
|
||||
double sum = 0.0;
|
||||
#pragma omp parallel for default(none) shared(z) reduction(+:sum) schedule(runtime) // added schedule(runtime)
|
||||
for (auto pi = cbegin(z); pi!=cend(z); ++pi)
|
||||
{
|
||||
sum += *pi;
|
||||
}
|
||||
//for (auto val: z)
|
||||
//{
|
||||
//sum += val;
|
||||
//}
|
||||
return sum;
|
||||
}
|
||||
|
||||
|
||||
|
||||
88
ex5/ex5_1/mylib.h
Normal file
88
ex5/ex5_1/mylib.h
Normal file
|
|
@ -0,0 +1,88 @@
|
|||
#pragma once
|
||||
#include <cassert>
|
||||
#include <iomanip> // setw()
|
||||
#include <iostream>
|
||||
#include <omp.h>
|
||||
#include <vector>
|
||||
|
||||
/** Inner product
|
||||
@param[in] x vector
|
||||
@param[in] y vector
|
||||
@return resulting Euclidian inner product <x,y>
|
||||
*/
|
||||
double scalar_parallel(std::vector<double> const &x, std::vector<double> const &y);
|
||||
double scalar(std::vector<double> const &x, std::vector<double> const &y);
|
||||
double scalar_trans(std::vector<double> const &x, std::vector<double> const &y);
|
||||
|
||||
|
||||
|
||||
// Declare additional reduction operation in OpenMP for STL-vector
|
||||
#pragma omp declare reduction(VecAppend : std::vector<int> : omp_out.insert(omp_out.end(), omp_in.begin(), omp_in.end())) \
|
||||
initializer (omp_priv=omp_orig)
|
||||
|
||||
std::vector<int> reduction_vec_append(int n);
|
||||
|
||||
|
||||
/** l2-norm
|
||||
@param[in] x vector
|
||||
@return resulting Euclidian norm
|
||||
*/
|
||||
double norm(std::vector<double> const &x);
|
||||
|
||||
/** Vector @p b adds its elements to vector @p a .
|
||||
@param[in] a vector
|
||||
@param[in] b vector
|
||||
@return a+=b componentwise
|
||||
*/
|
||||
template<class T>
|
||||
std::vector<T> &operator+=(std::vector<T> &a, std::vector<T> const &b)
|
||||
{
|
||||
assert(a.size()==b.size());
|
||||
for (size_t k = 0; k < a.size(); ++k) {
|
||||
a[k] += b[k];
|
||||
}
|
||||
return a;
|
||||
}
|
||||
|
||||
// Declare the reduction operation in OpenMP for an STL-vector
|
||||
// omp_out += omp_in requires operator+=(vector<int> &, vector<int> const &) from above
|
||||
// ------------------------------------------------------------
|
||||
// https://scc.ustc.edu.cn/zlsc/tc4600/intel/2016.0.109/compiler_c/common/core/GUID-7312910C-D175-4544-99C5-29C12D980744.htm
|
||||
// https://gist.github.com/eruffaldi/7180bdec4c8c9a11f019dd0ba9a2d68c
|
||||
// https://stackoverflow.com/questions/29633531/user-defined-reduction-on-vector-of-varying-size
|
||||
// see also p.74ff in https://www.fz-juelich.de/ias/jsc/EN/AboutUs/Staff/Hagemeier_A/docs-parallel-programming/OpenMP-Slides.pdf
|
||||
#pragma omp declare reduction(VecAdd : std::vector<int> : omp_out += omp_in) \
|
||||
initializer (omp_priv=omp_orig)
|
||||
|
||||
// Templates are n o t possible, i.e. the reduction has to be declared fore a specified type.
|
||||
//template <class T>
|
||||
//#pragma omp declare reduction(VecAdd : std::vector<T> : omp_out += omp_in) initializer (omp_priv(omp_orig))
|
||||
// MS: template nach #pragma !?
|
||||
|
||||
// ------------------------------------------------------------
|
||||
|
||||
/** Test for vector reduction.
|
||||
*
|
||||
* The thread-private vectors of size @p n are initialized via @f$v_k^{tID}=tID+k@f$.
|
||||
* Afterwards these vectors are accumulated, i.e.,
|
||||
* @f$v_k= \sum_{tID=0}^{numThreads} v_k^{tID}@f$.
|
||||
*
|
||||
* @param[in] n size of global/private vector
|
||||
* @return resulting global vector.
|
||||
*/
|
||||
std::vector<int> reduction_vec(int n);
|
||||
|
||||
|
||||
|
||||
/** Output of a vector.
|
||||
@param[in,out] s output stream
|
||||
@param[in] x vector
|
||||
@return modified output stream
|
||||
*/
|
||||
template <class T>
|
||||
std::ostream &operator<<(std::ostream &s, std::vector<T> const &x)
|
||||
{
|
||||
for (auto const &v : x) s << std::setw(4) << v << " ";
|
||||
return s;
|
||||
}
|
||||
|
||||
70
ex5/ex5_1/timing.h
Normal file
70
ex5/ex5_1/timing.h
Normal file
|
|
@ -0,0 +1,70 @@
|
|||
#pragma once
|
||||
#include <chrono> // timing
|
||||
#include <stack>
|
||||
|
||||
using Clock = std::chrono::system_clock; //!< The wall clock timer chosen
|
||||
//using Clock = std::chrono::high_resolution_clock;
|
||||
using TPoint= std::chrono::time_point<Clock>;
|
||||
|
||||
// [Galowicz, C++17 STL Cookbook, p. 29]
|
||||
inline
|
||||
std::stack<TPoint> MyStopWatch; //!< starting time of stopwatch
|
||||
|
||||
/** Starts stopwatch timer.
|
||||
* Use as @code tic(); myfunction(...) ; double tsec = toc(); @endcode
|
||||
*
|
||||
* The timining is allowed to be nested and the recent time is stored on top of the stack.
|
||||
*
|
||||
* @return recent time
|
||||
* @see toc
|
||||
*/
|
||||
inline auto tic()
|
||||
{
|
||||
MyStopWatch.push(Clock::now());
|
||||
return MyStopWatch.top();
|
||||
}
|
||||
|
||||
/** Returns the elapsed time from stopwatch.
|
||||
*
|
||||
* The time from top of the stack is used
|
||||
* if time point @p t_b is not passed as input parameter.
|
||||
* Use as @code tic(); myfunction(...) ; double tsec = toc(); @endcode
|
||||
* or as @code auto t_b = tic(); myfunction(...) ; double tsec = toc(t_b); @endcode
|
||||
* The last option is to be used in the case of
|
||||
* non-nested but overlapping time measurements.
|
||||
*
|
||||
* @param[in] t_b start time of some stop watch
|
||||
* @return elapsed time in seconds.
|
||||
*
|
||||
*/
|
||||
inline double toc(TPoint const &t_b = MyStopWatch.top())
|
||||
{
|
||||
// https://en.cppreference.com/w/cpp/chrono/treat_as_floating_point
|
||||
using Unit = std::chrono::seconds;
|
||||
using FpSeconds = std::chrono::duration<double, Unit::period>;
|
||||
auto t_e = Clock::now();
|
||||
MyStopWatch.pop();
|
||||
return FpSeconds(t_e-t_b).count();
|
||||
}
|
||||
|
||||
#include <iostream>
|
||||
#include <string>
|
||||
/** Executes function @p f and measures/prints elapsed wall clock time in seconds
|
||||
*
|
||||
* Call as
|
||||
* @code measure("Time for (b = b + 1)", [&]() {
|
||||
thrust::transform(b.begin(), b.end(), b.begin(), increment());
|
||||
}); @endcode
|
||||
*
|
||||
* @param[in] label additional string to be printed with the measurement.
|
||||
* @param[in] f function to execute.
|
||||
* @author Therese Bösmüller, 2025
|
||||
*
|
||||
*/
|
||||
auto measure = [](const std::string& label, auto&& f) {
|
||||
auto start = std::chrono::high_resolution_clock::now();
|
||||
f();
|
||||
auto stop = std::chrono::high_resolution_clock::now();
|
||||
auto duration = std::chrono::duration_cast<std::chrono::microseconds>(stop - start).count();
|
||||
std::cout << label << ": " << duration << " microseconds" << std::endl;
|
||||
}; // ';' is needed for a visible documentation of this lambda-function
|
||||
31
ex5/ex5_2/Makefile
Normal file
31
ex5/ex5_2/Makefile
Normal file
|
|
@ -0,0 +1,31 @@
|
|||
#
|
||||
# use GNU-Compiler tools
|
||||
COMPILER=GCC_
|
||||
# alternatively from the shell
|
||||
# export COMPILER=GCC_
|
||||
# or, alternatively from the shell
|
||||
# make COMPILER=GCC_
|
||||
|
||||
# use Intel compilers
|
||||
#COMPILER=ICC_
|
||||
|
||||
# use PGI compilers
|
||||
# COMPILER=PGI_
|
||||
|
||||
|
||||
SOURCES = main.cpp mylib.cpp
|
||||
OBJECTS = $(SOURCES:.cpp=.o)
|
||||
|
||||
PROGRAM = main.${COMPILER}
|
||||
|
||||
# uncomment the next to lines for debugging and detailed performance analysis
|
||||
CXXFLAGS += -g
|
||||
LINKFLAGS += -g
|
||||
# do not use -pg with PGI compilers
|
||||
|
||||
ifndef COMPILER
|
||||
COMPILER=GCC_
|
||||
endif
|
||||
|
||||
|
||||
include ../${COMPILER}default.mk
|
||||
501
ex5/ex5_2/data_1.txt
Normal file
501
ex5/ex5_2/data_1.txt
Normal file
|
|
@ -0,0 +1,501 @@
|
|||
141
|
||||
261
|
||||
87
|
||||
430
|
||||
258
|
||||
298
|
||||
425
|
||||
120
|
||||
496
|
||||
707
|
||||
244
|
||||
786
|
||||
75
|
||||
394
|
||||
4
|
||||
221
|
||||
2
|
||||
190
|
||||
143
|
||||
269
|
||||
175
|
||||
139
|
||||
599
|
||||
902
|
||||
940
|
||||
222
|
||||
483
|
||||
377
|
||||
524
|
||||
265
|
||||
69
|
||||
437
|
||||
174
|
||||
27
|
||||
955
|
||||
431
|
||||
962
|
||||
763
|
||||
8
|
||||
681
|
||||
706
|
||||
646
|
||||
553
|
||||
219
|
||||
773
|
||||
229
|
||||
371
|
||||
891
|
||||
857
|
||||
403
|
||||
319
|
||||
609
|
||||
911
|
||||
910
|
||||
592
|
||||
333
|
||||
854
|
||||
443
|
||||
905
|
||||
34
|
||||
533
|
||||
717
|
||||
180
|
||||
337
|
||||
188
|
||||
322
|
||||
404
|
||||
549
|
||||
49
|
||||
553
|
||||
275
|
||||
242
|
||||
244
|
||||
155
|
||||
957
|
||||
936
|
||||
819
|
||||
729
|
||||
176
|
||||
361
|
||||
189
|
||||
2
|
||||
317
|
||||
700
|
||||
626
|
||||
544
|
||||
440
|
||||
288
|
||||
502
|
||||
762
|
||||
763
|
||||
577
|
||||
748
|
||||
646
|
||||
124
|
||||
505
|
||||
348
|
||||
93
|
||||
148
|
||||
199
|
||||
673
|
||||
432
|
||||
695
|
||||
257
|
||||
10
|
||||
533
|
||||
280
|
||||
947
|
||||
907
|
||||
393
|
||||
25
|
||||
672
|
||||
838
|
||||
972
|
||||
57
|
||||
451
|
||||
583
|
||||
687
|
||||
720
|
||||
651
|
||||
727
|
||||
374
|
||||
582
|
||||
117
|
||||
58
|
||||
980
|
||||
285
|
||||
595
|
||||
963
|
||||
186
|
||||
194
|
||||
342
|
||||
933
|
||||
391
|
||||
274
|
||||
152
|
||||
398
|
||||
375
|
||||
132
|
||||
436
|
||||
92
|
||||
615
|
||||
11
|
||||
574
|
||||
790
|
||||
236
|
||||
449
|
||||
570
|
||||
62
|
||||
497
|
||||
643
|
||||
222
|
||||
838
|
||||
972
|
||||
847
|
||||
506
|
||||
279
|
||||
747
|
||||
237
|
||||
958
|
||||
621
|
||||
601
|
||||
173
|
||||
91
|
||||
256
|
||||
859
|
||||
912
|
||||
700
|
||||
726
|
||||
230
|
||||
577
|
||||
811
|
||||
404
|
||||
989
|
||||
90
|
||||
321
|
||||
512
|
||||
61
|
||||
726
|
||||
557
|
||||
530
|
||||
830
|
||||
859
|
||||
790
|
||||
318
|
||||
453
|
||||
753
|
||||
110
|
||||
110
|
||||
270
|
||||
525
|
||||
973
|
||||
711
|
||||
312
|
||||
292
|
||||
851
|
||||
912
|
||||
640
|
||||
256
|
||||
89
|
||||
839
|
||||
585
|
||||
949
|
||||
62
|
||||
585
|
||||
286
|
||||
828
|
||||
191
|
||||
443
|
||||
394
|
||||
827
|
||||
677
|
||||
208
|
||||
319
|
||||
134
|
||||
672
|
||||
571
|
||||
170
|
||||
148
|
||||
477
|
||||
909
|
||||
553
|
||||
33
|
||||
54
|
||||
806
|
||||
452
|
||||
383
|
||||
790
|
||||
365
|
||||
533
|
||||
712
|
||||
872
|
||||
329
|
||||
651
|
||||
975
|
||||
76
|
||||
588
|
||||
414
|
||||
310
|
||||
264
|
||||
759
|
||||
996
|
||||
187
|
||||
782
|
||||
196
|
||||
993
|
||||
803
|
||||
425
|
||||
729
|
||||
499
|
||||
809
|
||||
357
|
||||
74
|
||||
591
|
||||
911
|
||||
194
|
||||
433
|
||||
750
|
||||
40
|
||||
947
|
||||
764
|
||||
559
|
||||
184
|
||||
498
|
||||
518
|
||||
995
|
||||
855
|
||||
963
|
||||
679
|
||||
404
|
||||
935
|
||||
480
|
||||
232
|
||||
397
|
||||
706
|
||||
559
|
||||
757
|
||||
996
|
||||
963
|
||||
536
|
||||
964
|
||||
116
|
||||
52
|
||||
305
|
||||
581
|
||||
531
|
||||
902
|
||||
541
|
||||
432
|
||||
543
|
||||
713
|
||||
17
|
||||
801
|
||||
143
|
||||
479
|
||||
257
|
||||
370
|
||||
662
|
||||
170
|
||||
279
|
||||
199
|
||||
196
|
||||
327
|
||||
881
|
||||
472
|
||||
404
|
||||
180
|
||||
969
|
||||
408
|
||||
845
|
||||
616
|
||||
377
|
||||
878
|
||||
785
|
||||
465
|
||||
814
|
||||
899
|
||||
430
|
||||
335
|
||||
597
|
||||
902
|
||||
703
|
||||
378
|
||||
735
|
||||
955
|
||||
543
|
||||
541
|
||||
312
|
||||
72
|
||||
182
|
||||
93
|
||||
464
|
||||
10
|
||||
916
|
||||
643
|
||||
2
|
||||
31
|
||||
209
|
||||
455
|
||||
128
|
||||
9
|
||||
728
|
||||
355
|
||||
781
|
||||
437
|
||||
437
|
||||
50
|
||||
50
|
||||
92
|
||||
595
|
||||
242
|
||||
842
|
||||
858
|
||||
964
|
||||
489
|
||||
221
|
||||
227
|
||||
537
|
||||
763
|
||||
348
|
||||
462
|
||||
640
|
||||
918
|
||||
162
|
||||
716
|
||||
578
|
||||
434
|
||||
885
|
||||
394
|
||||
179
|
||||
634
|
||||
625
|
||||
328
|
||||
803
|
||||
1000
|
||||
981
|
||||
128
|
||||
233
|
||||
24
|
||||
608
|
||||
111
|
||||
408
|
||||
885
|
||||
549
|
||||
370
|
||||
209
|
||||
441
|
||||
957
|
||||
125
|
||||
471
|
||||
857
|
||||
44
|
||||
692
|
||||
979
|
||||
284
|
||||
134
|
||||
686
|
||||
910
|
||||
611
|
||||
900
|
||||
194
|
||||
755
|
||||
347
|
||||
419
|
||||
156
|
||||
820
|
||||
625
|
||||
739
|
||||
806
|
||||
68
|
||||
951
|
||||
498
|
||||
756
|
||||
743
|
||||
832
|
||||
157
|
||||
458
|
||||
619
|
||||
933
|
||||
836
|
||||
896
|
||||
583
|
||||
583
|
||||
855
|
||||
35
|
||||
886
|
||||
408
|
||||
37
|
||||
747
|
||||
155
|
||||
144
|
||||
606
|
||||
255
|
||||
325
|
||||
402
|
||||
407
|
||||
387
|
||||
610
|
||||
167
|
||||
189
|
||||
95
|
||||
324
|
||||
770
|
||||
235
|
||||
741
|
||||
693
|
||||
825
|
||||
828
|
||||
294
|
||||
310
|
||||
524
|
||||
326
|
||||
832
|
||||
811
|
||||
557
|
||||
263
|
||||
681
|
||||
234
|
||||
457
|
||||
385
|
||||
539
|
||||
992
|
||||
756
|
||||
981
|
||||
235
|
||||
529
|
||||
52
|
||||
757
|
||||
602
|
||||
858
|
||||
989
|
||||
930
|
||||
410
|
||||
1
|
||||
541
|
||||
208
|
||||
220
|
||||
326
|
||||
96
|
||||
748
|
||||
749
|
||||
544
|
||||
339
|
||||
833
|
||||
553
|
||||
958
|
||||
893
|
||||
357
|
||||
547
|
||||
347
|
||||
623
|
||||
797
|
||||
746
|
||||
126
|
||||
823
|
||||
26
|
||||
415
|
||||
732
|
||||
782
|
||||
368
|
||||
|
||||
130
ex5/ex5_2/main.cpp
Normal file
130
ex5/ex5_2/main.cpp
Normal file
|
|
@ -0,0 +1,130 @@
|
|||
#include "mylib.h"
|
||||
#include <fstream>
|
||||
#include <iostream>
|
||||
#include <omp.h>
|
||||
#include <vector>
|
||||
using namespace std;
|
||||
|
||||
|
||||
int main()
|
||||
{
|
||||
// read vector from file
|
||||
vector<size_t> data_vector = {};
|
||||
|
||||
ifstream input_stream("data_1.txt");
|
||||
|
||||
size_t line;
|
||||
while(input_stream >> line)
|
||||
{
|
||||
data_vector.push_back(line);
|
||||
}
|
||||
data_vector.shrink_to_fit();
|
||||
|
||||
|
||||
|
||||
|
||||
// specify loops
|
||||
size_t NLOOPS = 10000;
|
||||
|
||||
|
||||
|
||||
// ############# Parallelization with openMP #############
|
||||
// calculate arithmetic mean, geometric mean and harmonic mean
|
||||
double am_omp, gm_omp, hm_omp;
|
||||
|
||||
double tstart = omp_get_wtime();
|
||||
|
||||
for (size_t i = 0; i < NLOOPS; ++i)
|
||||
means_omp(data_vector, am_omp, gm_omp, hm_omp);
|
||||
|
||||
double t_means_omp = (omp_get_wtime() - tstart)/NLOOPS;
|
||||
|
||||
// calculate minimum and maximum
|
||||
size_t min, max;
|
||||
|
||||
tstart = omp_get_wtime();
|
||||
|
||||
for (size_t i = 0; i < NLOOPS; ++i)
|
||||
minmax_omp(data_vector, min, max);
|
||||
|
||||
double t_minmax_omp = (omp_get_wtime() - tstart)/NLOOPS;
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
// ############# Parallelization with C++ algorithms #############
|
||||
// calculate arithmetic mean, geometric mean and harmonic mean
|
||||
double am_cpp, gm_cpp, hm_cpp;
|
||||
|
||||
tstart = omp_get_wtime();
|
||||
|
||||
for (size_t i = 0; i < NLOOPS; ++i)
|
||||
means_cpp(data_vector, am_cpp, gm_cpp, hm_cpp);
|
||||
|
||||
double t_means_cpp = (omp_get_wtime() - tstart)/NLOOPS;
|
||||
|
||||
// calculate minimum and maximum
|
||||
size_t min_cpp, max_cpp;
|
||||
|
||||
tstart = omp_get_wtime();
|
||||
|
||||
for (size_t i = 0; i < NLOOPS; ++i)
|
||||
minmax_cpp(data_vector, min_cpp, max_cpp);
|
||||
|
||||
double t_minmax_cpp = (omp_get_wtime() - tstart)/NLOOPS;
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
// print results
|
||||
cout << "####### OpenMP #######" << endl;
|
||||
cout << "minimum: " << min << endl;
|
||||
cout << "maximum: " << max << endl;
|
||||
cout << "duration: " << t_minmax_omp << endl << endl;
|
||||
|
||||
cout << "arithmetic mean: " << am_omp << endl;
|
||||
cout << "geometric mean: " << gm_omp << endl;
|
||||
cout << "harmonic mean: " << hm_omp << endl;
|
||||
cout << "duration: " << t_means_omp << endl << endl;
|
||||
|
||||
|
||||
cout << "####### C++ #######" << endl;
|
||||
cout << "minimum: " << min_cpp << endl;
|
||||
cout << "maximum: " << max_cpp << endl;
|
||||
cout << "duration: " << t_minmax_cpp << endl << endl;
|
||||
|
||||
cout << "arithmetic mean: " << am_cpp << endl;
|
||||
cout << "geometric mean: " << gm_cpp << endl;
|
||||
cout << "harmonic mean: " << hm_cpp << endl;
|
||||
cout << "duration: " << t_means_cpp << endl << endl;
|
||||
|
||||
|
||||
|
||||
// ####### OpenMP #######
|
||||
// minimum: 1
|
||||
// maximum: 1000
|
||||
// duration: 3.52086e-06
|
||||
|
||||
// arithmetic mean: 498.184
|
||||
// geometric mean: 364.412
|
||||
// harmonic mean: 95.6857
|
||||
// duration: 5.90171e-06
|
||||
|
||||
// ####### C++ #######
|
||||
// minimum: 1
|
||||
// maximum: 1000
|
||||
// duration: 1.76816e-05
|
||||
|
||||
// arithmetic mean: 498.184
|
||||
// geometric mean: 364.412
|
||||
// harmonic mean: 95.6857
|
||||
// duration: 2.35728e-05
|
||||
|
||||
// --> the openMP variant is faster in both cases
|
||||
|
||||
|
||||
return 0;
|
||||
}
|
||||
103
ex5/ex5_2/mylib.cpp
Normal file
103
ex5/ex5_2/mylib.cpp
Normal file
|
|
@ -0,0 +1,103 @@
|
|||
#include "mylib.h"
|
||||
#include <algorithm>
|
||||
#include <cmath>
|
||||
#include <execution>
|
||||
#include <iostream>
|
||||
#include <numeric>
|
||||
#include <omp.h>
|
||||
|
||||
using namespace std;
|
||||
|
||||
|
||||
void means_omp(const std::vector<size_t> numbers, double &am, double &gm, double &hm)
|
||||
{
|
||||
size_t const n = numbers.size();
|
||||
|
||||
am = 0.;
|
||||
gm = 0.;
|
||||
hm = 0.;
|
||||
|
||||
#pragma omp parallel for shared(numbers, n, cout) reduction(+:am, gm, hm)
|
||||
for (size_t i = 0; i < n; ++i)
|
||||
{
|
||||
am += numbers[i];
|
||||
gm += log(numbers[i]);
|
||||
hm += 1.0/numbers[i];
|
||||
|
||||
// #pragma omp critical
|
||||
// {
|
||||
// cout << "Thread number " << omp_get_thread_num() << " processes value " << numbers[i] << endl;
|
||||
// }
|
||||
}
|
||||
|
||||
am /= n;
|
||||
gm = exp(gm/n);
|
||||
hm = n/hm;
|
||||
}
|
||||
|
||||
|
||||
void minmax_omp(const std::vector<size_t> numbers, size_t &global_min, size_t &global_max)
|
||||
{
|
||||
size_t const n = numbers.size();
|
||||
|
||||
global_min = -1; // gives the maximum size_t value
|
||||
global_max = 0;
|
||||
|
||||
#pragma omp parallel shared(numbers, n, global_min, global_max)
|
||||
{
|
||||
const size_t nthreads = omp_get_num_threads();
|
||||
const size_t threadnum = omp_get_thread_num();
|
||||
const size_t chunksize = n/nthreads;
|
||||
|
||||
|
||||
size_t start = threadnum*chunksize;
|
||||
size_t end = start + chunksize;
|
||||
if (threadnum == nthreads - 1)
|
||||
end = n;
|
||||
|
||||
|
||||
size_t local_min = -1;
|
||||
size_t local_max = 0;
|
||||
for (size_t i = start; i < end ; ++i)
|
||||
{
|
||||
if (numbers[i] < local_min)
|
||||
local_min = numbers[i];
|
||||
|
||||
if (numbers[i] > local_max)
|
||||
local_max = numbers[i];
|
||||
}
|
||||
|
||||
#pragma omp critical
|
||||
{
|
||||
if (local_min < global_min)
|
||||
global_min = local_min;
|
||||
|
||||
if (local_max > global_max)
|
||||
global_max = local_max;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
void means_cpp(const std::vector<size_t> numbers, double &am, double &gm, double &hm)
|
||||
{
|
||||
size_t const n = numbers.size();
|
||||
|
||||
am = reduce(std::execution::par, numbers.begin(), numbers.end());
|
||||
gm = transform_reduce(std::execution::par, numbers.begin(), numbers.end(), 0.0, plus{}, [] (size_t x) -> double { return log(x); } );
|
||||
hm = transform_reduce(std::execution::par, numbers.begin(), numbers.end(), 0.0, plus{}, [] (size_t x) -> double { return 1.0/x; });
|
||||
|
||||
am /= n;
|
||||
gm = exp(gm/n);
|
||||
hm = n/hm;
|
||||
}
|
||||
|
||||
|
||||
void minmax_cpp(const std::vector<size_t> numbers, size_t &global_min, size_t &global_max)
|
||||
{
|
||||
auto min_it = min_element(std::execution::par, numbers.begin(), numbers.end());
|
||||
auto max_it = max_element(std::execution::par, numbers.begin(), numbers.end());
|
||||
|
||||
global_min = *min_it;
|
||||
global_max = *max_it;
|
||||
}
|
||||
42
ex5/ex5_2/mylib.h
Normal file
42
ex5/ex5_2/mylib.h
Normal file
|
|
@ -0,0 +1,42 @@
|
|||
#include <vector>
|
||||
|
||||
/**
|
||||
This function calculates arithmetic mean, geometric mean and harmonic mean of an integer vector.
|
||||
Uses openMP parallelization.
|
||||
@param[in] numbers vector containing integers
|
||||
@param[out] am arithmetic mean
|
||||
@param[out] gm geometric mean
|
||||
@param[out] hm harmonic mean
|
||||
*/
|
||||
void means_omp(const std::vector<size_t> numbers, double &am, double &gm, double &hm);
|
||||
|
||||
|
||||
/**
|
||||
This function calculates the minimum and maximum of a vector.
|
||||
Uses openMP parallelization.
|
||||
@param[in] numbers vector containing integers
|
||||
@param[out] global_min minimum
|
||||
@param[out] global_max maximum
|
||||
*/
|
||||
void minmax_omp(const std::vector<size_t> numbers, size_t &global_min, size_t &global_max);
|
||||
|
||||
|
||||
/**
|
||||
This function calculates arithmetic mean, geometric mean and harmonic mean of an integer vector.
|
||||
Uses C++ parallelization.
|
||||
@param[in] numbers vector containing integers
|
||||
@param[out] am arithmetic mean
|
||||
@param[out] gm geometric mean
|
||||
@param[out] hm harmonic mean
|
||||
*/
|
||||
void means_cpp(const std::vector<size_t> numbers, double &am, double &gm, double &hm);
|
||||
|
||||
|
||||
/**
|
||||
This function calculates the minimum and maximum of a vector.
|
||||
Uses C++ parallelization.
|
||||
@param[in] numbers vector containing integers
|
||||
@param[out] global_min minimum
|
||||
@param[out] global_max maximum
|
||||
*/
|
||||
void minmax_cpp(const std::vector<size_t> numbers, size_t &global_min, size_t &global_max);
|
||||
30
ex5/ex5_3/Makefile
Normal file
30
ex5/ex5_3/Makefile
Normal file
|
|
@ -0,0 +1,30 @@
|
|||
#
|
||||
# use GNU-Compiler tools
|
||||
COMPILER=GCC_
|
||||
# alternatively from the shell
|
||||
# export COMPILER=GCC_
|
||||
# or, alternatively from the shell
|
||||
# make COMPILER=GCC_
|
||||
|
||||
# use Intel compilers
|
||||
#COMPILER=ICC_
|
||||
|
||||
# use PGI compilers
|
||||
# COMPILER=PGI_
|
||||
|
||||
|
||||
SOURCES = main.cpp goldbach.cpp
|
||||
OBJECTS = $(SOURCES:.cpp=.o)
|
||||
|
||||
PROGRAM = main.${COMPILER}
|
||||
|
||||
# uncomment the next to lines for debugging and detailed performance analysis
|
||||
CXXFLAGS += -g
|
||||
LINKFLAGS += -g
|
||||
# do not use -pg with PGI compilers
|
||||
|
||||
ifndef COMPILER
|
||||
COMPILER=GCC_
|
||||
endif
|
||||
|
||||
include ../${COMPILER}default.mk
|
||||
46
ex5/ex5_3/goldbach.cpp
Normal file
46
ex5/ex5_3/goldbach.cpp
Normal file
|
|
@ -0,0 +1,46 @@
|
|||
#include "goldbach.h"
|
||||
#include <iostream>
|
||||
#include <iterator>
|
||||
#include <omp.h>
|
||||
|
||||
size_t single_goldbach(size_t k)
|
||||
{
|
||||
const std::vector<size_t> relevant_primes = get_primes(k);
|
||||
size_t m = relevant_primes.size();
|
||||
|
||||
size_t counter = 0;
|
||||
|
||||
#pragma omp parallel for shared(relevant_primes, m, k) reduction(+:counter)
|
||||
for(size_t i = 0; i < m; ++i)
|
||||
{
|
||||
for(size_t j = i; j < m; ++j)
|
||||
{
|
||||
if(relevant_primes[i] + relevant_primes[j] == k)
|
||||
++counter;
|
||||
}
|
||||
}
|
||||
|
||||
return counter;
|
||||
}
|
||||
|
||||
|
||||
std::vector<size_t> count_goldbach(size_t n)
|
||||
{
|
||||
const std::vector<size_t> relevant_primes = get_primes(n);
|
||||
size_t m = relevant_primes.size();
|
||||
|
||||
std::vector<size_t> counter_vector(n + 1, 0);
|
||||
|
||||
#pragma omp parallel for shared(relevant_primes, m, n) reduction(VecAdd:counter_vector)
|
||||
for(size_t i = 0; i < m; ++i)
|
||||
{
|
||||
for(size_t j = i; j < m; ++j)
|
||||
{
|
||||
size_t sum = relevant_primes[i] + relevant_primes[j];
|
||||
if(sum <= n)
|
||||
++counter_vector[relevant_primes[i] + relevant_primes[j]];
|
||||
}
|
||||
}
|
||||
|
||||
return counter_vector;
|
||||
}
|
||||
45
ex5/ex5_3/goldbach.h
Normal file
45
ex5/ex5_3/goldbach.h
Normal file
|
|
@ -0,0 +1,45 @@
|
|||
#pragma once
|
||||
#include "mayer_primes.h"
|
||||
#include <cassert>
|
||||
#include <vector>
|
||||
|
||||
|
||||
/**
|
||||
This function returns the number of possible decompositions of an integer into a sum of two prime numbers.
|
||||
@param[in] k first integer
|
||||
@param[out] count number of decompositions
|
||||
*/
|
||||
size_t single_goldbach(size_t k);
|
||||
|
||||
|
||||
/**
|
||||
This function returns the number of possible decompositions into a sum of two prime numbers of all even integers in the interval [4,n].
|
||||
@param[in] n upper integer bound
|
||||
@param[out] count_vector vector containing the number of decompositions for a natural number the corresponding index
|
||||
*/
|
||||
std::vector<size_t> count_goldbach(size_t n);
|
||||
|
||||
|
||||
/** Vector @p b adds its elements to vector @p a .
|
||||
@param[in] a vector
|
||||
@param[in] b vector
|
||||
@return a+=b componentwise
|
||||
*/
|
||||
template<class T>
|
||||
std::vector<T> &operator+=(std::vector<T> &a, std::vector<T> const &b)
|
||||
{
|
||||
assert(a.size()==b.size());
|
||||
for (size_t k = 0; k < a.size(); ++k) {
|
||||
a[k] += b[k];
|
||||
}
|
||||
return a;
|
||||
}
|
||||
|
||||
// Declare the reduction operation in OpenMP for an STL-vector
|
||||
// omp_out += omp_in requires operator+=(vector<int> &, vector<int> const &) from above
|
||||
// ------------------------------------------------------------
|
||||
// https://scc.ustc.edu.cn/zlsc/tc4600/intel/2016.0.109/compiler_c/common/core/GUID-7312910C-D175-4544-99C5-29C12D980744.htm
|
||||
// https://gist.github.com/eruffaldi/7180bdec4c8c9a11f019dd0ba9a2d68c
|
||||
// https://stackoverflow.com/questions/29633531/user-defined-reduction-on-vector-of-varying-size
|
||||
// see also p.74ff in https://www.fz-juelich.de/ias/jsc/EN/AboutUs/Staff/Hagemeier_A/docs-parallel-programming/OpenMP-Slides.pdf
|
||||
#pragma omp declare reduction(VecAdd : std::vector<size_t> : omp_out += omp_in) initializer (omp_priv=omp_orig)
|
||||
45
ex5/ex5_3/main.cpp
Normal file
45
ex5/ex5_3/main.cpp
Normal file
|
|
@ -0,0 +1,45 @@
|
|||
#include "goldbach.h"
|
||||
#include <algorithm>
|
||||
#include <iostream>
|
||||
#include <omp.h>
|
||||
using namespace std;
|
||||
|
||||
|
||||
int main()
|
||||
{
|
||||
cout << "Check: 694 has "<< single_goldbach(694) << " decompositions." << endl << "----------------------------------------" << endl;
|
||||
|
||||
for(size_t n : {10000, 100000, 400000, 1000000, 2000000})
|
||||
{
|
||||
double t_start = omp_get_wtime();
|
||||
|
||||
auto goldbach_vector = count_goldbach(n);
|
||||
|
||||
auto max_it = max_element(goldbach_vector.begin(), goldbach_vector.end());
|
||||
size_t max_number = distance(goldbach_vector.begin(), max_it);
|
||||
|
||||
double t_end = omp_get_wtime() - t_start;
|
||||
|
||||
cout << "The number " << max_number << " has " << *max_it << " decompositions. Duration: " << t_end << endl;
|
||||
}
|
||||
|
||||
/*
|
||||
###### WITHOUT PARALLELIZATION ######
|
||||
The number 9240 has 329 decompositions. Duration: 0.00307696
|
||||
The number 99330 has 2168 decompositions. Duration: 0.189839
|
||||
The number 390390 has 7094 decompositions. Duration: 1.3042
|
||||
The number 990990 has 15594 decompositions. Duration: 5.45034
|
||||
The number 1981980 has 27988 decompositions. Duration: 47.1807
|
||||
|
||||
|
||||
###### WITH PARALLELIZATION ######
|
||||
The number 9240 has 329 decompositions. Duration: 0.000734854
|
||||
The number 99330 has 2168 decompositions. Duration: 0.0251322
|
||||
The number 390390 has 7094 decompositions. Duration: 0.487375
|
||||
The number 990990 has 15594 decompositions. Duration: 6.16972
|
||||
The number 1981980 has 27988 decompositions. Duration: 31.5699
|
||||
*/
|
||||
|
||||
|
||||
return 0;
|
||||
}
|
||||
73
ex5/ex5_3/mayer_primes.h
Normal file
73
ex5/ex5_3/mayer_primes.h
Normal file
|
|
@ -0,0 +1,73 @@
|
|||
#pragma once
|
||||
|
||||
#include <cstring> //memset
|
||||
#include <vector>
|
||||
//using namespace std;
|
||||
|
||||
/** \brief Determines all prime numbers in interval [2, @p max].
|
||||
*
|
||||
* The sieve of Eratosthenes is used.
|
||||
*
|
||||
* The implementation originates from <a href="http://code.activestate.com/recipes/576559-fast-prime-generator/">Florian Mayer</a>.
|
||||
*
|
||||
* \param[in] max end of interval for the prime number search.
|
||||
* \return vector of prime numbers @f$2,3,5, ..., p<=max @f$.
|
||||
*
|
||||
* \copyright
|
||||
* Copyright (c) 2008 Florian Mayer (adapted by Gundolf Haase 2018)
|
||||
*
|
||||
* Permission is hereby granted, free of charge, to any person obtaining a copy
|
||||
* of this software and associated documentation files (the "Software"), to deal
|
||||
* in the Software without restriction, including without limitation the rights
|
||||
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
||||
* copies of the Software, and to permit persons to whom the Software is
|
||||
* furnished to do so, subject to the following conditions:
|
||||
*
|
||||
* The above copyright notice and this permission notice shall be included in
|
||||
* all copies or substantial portions of the Software.
|
||||
*
|
||||
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED,
|
||||
* INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
||||
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
||||
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
||||
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
||||
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
|
||||
*
|
||||
*/
|
||||
template <class T>
|
||||
std::vector<T> get_primes(T max)
|
||||
{
|
||||
std::vector<T> primes;
|
||||
char *sieve;
|
||||
sieve = new char[max / 8 + 1];
|
||||
// Fill sieve with 1
|
||||
memset(sieve, 0xFF, (max / 8 + 1) * sizeof(char));
|
||||
for (T x = 2; x <= max; x++)
|
||||
{
|
||||
if (sieve[x / 8] & (0x01 << (x % 8))) {
|
||||
primes.push_back(x);
|
||||
// Is prime. Mark multiplicates.
|
||||
for (T j = 2 * x; j <= max; j += x)
|
||||
{
|
||||
sieve[j / 8] &= ~(0x01 << (j % 8));
|
||||
}
|
||||
}
|
||||
}
|
||||
delete[] sieve;
|
||||
return primes;
|
||||
}
|
||||
|
||||
//---------------------------------------------------------------
|
||||
//int main() // by Florian Mayer
|
||||
//{g++ -O3 -std=c++14 -fopenmp main.cpp && ./a.out
|
||||
// vector<unsigned long> primes;
|
||||
// primes = get_primes(10000000);
|
||||
// // return 0;
|
||||
// // Print out result.
|
||||
// vector<unsigned long>::iterator it;
|
||||
// for(it=primes.begin(); it < primes.end(); it++)
|
||||
// cout << *it << " ";
|
||||
//
|
||||
// cout << endl;
|
||||
// return 0;
|
||||
//}
|
||||
30
ex5/ex5_4/Makefile
Normal file
30
ex5/ex5_4/Makefile
Normal file
|
|
@ -0,0 +1,30 @@
|
|||
#
|
||||
# use GNU-Compiler tools
|
||||
COMPILER=GCC_
|
||||
# alternatively from the shell
|
||||
# export COMPILER=GCC_
|
||||
# or, alternatively from the shell
|
||||
# make COMPILER=GCC_
|
||||
|
||||
# use Intel compilers
|
||||
#COMPILER=ICC_
|
||||
|
||||
# use PGI compilers
|
||||
# COMPILER=PGI_
|
||||
|
||||
|
||||
SOURCES = main.cpp benchmarks.cpp benchmark_tests.cpp
|
||||
OBJECTS = $(SOURCES:.cpp=.o)
|
||||
|
||||
PROGRAM = main.${COMPILER}
|
||||
|
||||
# uncomment the next to lines for debugging and detailed performance analysis
|
||||
CXXFLAGS += -g
|
||||
LINKFLAGS += -g
|
||||
# do not use -pg with PGI compilers
|
||||
|
||||
ifndef COMPILER
|
||||
COMPILER=GCC_
|
||||
endif
|
||||
|
||||
include ../${COMPILER}default.mk
|
||||
375
ex5/ex5_4/benchmark_tests.cpp
Normal file
375
ex5/ex5_4/benchmark_tests.cpp
Normal file
|
|
@ -0,0 +1,375 @@
|
|||
#include "benchmark_tests.h"
|
||||
#include "benchmarks.h"
|
||||
#include <chrono>
|
||||
#include <iostream>
|
||||
#include <math.h>
|
||||
using namespace std::chrono;
|
||||
|
||||
vector<double> test_A(const size_t &NLOOPS, const size_t &N)
|
||||
{
|
||||
cout << "#################### (A) ####################" << endl;
|
||||
cout << "\nLOOPS = " << NLOOPS << endl;
|
||||
cout << "\nN = " << N << endl;
|
||||
|
||||
|
||||
// Memory allocation
|
||||
cout << "Memory allocation\n";
|
||||
|
||||
vector<double> x(N), y(N);
|
||||
|
||||
cout.precision(2);
|
||||
cout << 2.0*N *sizeof(x[0]) / 1024 / 1024 / 1024 << " GByte Memory allocated\n";
|
||||
cout.precision(6);
|
||||
|
||||
|
||||
// Data initialization
|
||||
// Special: x_i = i+1; y_i = 1/x_i ==> <x,y> == N
|
||||
|
||||
for (size_t i = 0; i < N; ++i)
|
||||
{
|
||||
x[i] = i % 219 + 1;
|
||||
y[i] = 1.0/x[i];
|
||||
}
|
||||
|
||||
|
||||
cout << "\nStart Benchmarking scalar\n";
|
||||
|
||||
auto t1 = system_clock::now(); // start timer
|
||||
// Do calculation
|
||||
double check(0.0),ss(0.0);
|
||||
for (size_t i = 0; i < NLOOPS; ++i)
|
||||
{
|
||||
check = scalar_parallel(x, y);
|
||||
ss += check; // prevents the optimizer from removing unused calculation results.
|
||||
}
|
||||
|
||||
auto t2 = system_clock::now(); // stop timer
|
||||
auto duration = duration_cast<microseconds>(t2 - t1); // duration in microseconds
|
||||
double t_diff = static_cast<double>(duration.count()) / 1e6; // overall duration in seconds
|
||||
t_diff = t_diff/NLOOPS; // duration per loop seconds
|
||||
|
||||
|
||||
|
||||
// Check the correct result
|
||||
cout << "\n <x,y> = " << check << endl;
|
||||
if (static_cast<unsigned int>(check) != N)
|
||||
cout << " !! W R O N G result !!\n";
|
||||
cout << endl;
|
||||
|
||||
|
||||
// Timings and Performance
|
||||
cout << endl;
|
||||
cout.precision(2);
|
||||
|
||||
|
||||
double Gflops = 2.0*N / t_diff / 1024 / 1024 / 1024;
|
||||
double MemBandwidth = 2.0*N / t_diff / 1024 / 1024 / 1024 * sizeof(x[0]);
|
||||
|
||||
cout << "Total duration : " << t_diff*NLOOPS << endl;
|
||||
cout << "Timing in sec. : " << t_diff << endl;
|
||||
cout << "GFLOPS : " << Gflops << endl;
|
||||
cout << "GiByte/s : " << MemBandwidth << endl;
|
||||
|
||||
|
||||
return vector<double>{t_diff, Gflops, MemBandwidth};
|
||||
}
|
||||
|
||||
vector<double> test_A_sum(const size_t &NLOOPS, const size_t &N)
|
||||
{
|
||||
cout << "#################### (A) sum ####################" << endl;
|
||||
cout << "\nLOOPS = " << NLOOPS << endl;
|
||||
cout << "\nN = " << N << endl;
|
||||
|
||||
|
||||
// Memory allocation
|
||||
cout << "Memory allocation\n";
|
||||
|
||||
vector<double> x(N);
|
||||
|
||||
cout.precision(2);
|
||||
cout << 1.0*N *sizeof(x[0]) / 1024 / 1024 / 1024 << " GByte Memory allocated\n";
|
||||
cout.precision(6);
|
||||
|
||||
|
||||
// Data initialization
|
||||
|
||||
for (size_t i = 0; i < N; ++i)
|
||||
{
|
||||
x[i] = 1;
|
||||
}
|
||||
|
||||
|
||||
cout << "\nStart Benchmarking sum\n";
|
||||
|
||||
auto t1 = system_clock::now(); // start timer
|
||||
// Do calculation
|
||||
double check(0.0),ss(0.0);
|
||||
for (size_t i = 0; i < NLOOPS; ++i)
|
||||
{
|
||||
check = sum(x);
|
||||
ss += check; // prevents the optimizer from removing unused calculation results.
|
||||
}
|
||||
|
||||
auto t2 = system_clock::now(); // stop timer
|
||||
auto duration = duration_cast<microseconds>(t2 - t1); // duration in microseconds
|
||||
double t_diff = static_cast<double>(duration.count()) / 1e6; // overall duration in seconds
|
||||
t_diff = t_diff/NLOOPS; // duration per loop seconds
|
||||
|
||||
|
||||
|
||||
// Check the correct result
|
||||
cout << "\n <x,y> = " << check << endl;
|
||||
if (static_cast<unsigned int>(check) != N)
|
||||
cout << " !! W R O N G result !!\n";
|
||||
cout << endl;
|
||||
|
||||
|
||||
// Timings and Performance
|
||||
cout << endl;
|
||||
cout.precision(2);
|
||||
|
||||
|
||||
double Gflops = 1.0*N / t_diff / 1024 / 1024 / 1024;
|
||||
double MemBandwidth = 1.0*N / t_diff / 1024 / 1024 / 1024 * sizeof(x[0]);
|
||||
|
||||
cout << "Total duration : " << t_diff*NLOOPS << endl;
|
||||
cout << "Timing in sec. : " << t_diff << endl;
|
||||
cout << "GFLOPS : " << Gflops << endl;
|
||||
cout << "GiByte/s : " << MemBandwidth << endl;
|
||||
|
||||
|
||||
return vector<double>{t_diff, Gflops, MemBandwidth};
|
||||
}
|
||||
|
||||
|
||||
vector<double> test_B(const size_t &NLOOPS, const size_t &N, const size_t &M)
|
||||
{
|
||||
cout << "#################### (B) ####################" << endl;
|
||||
|
||||
cout << "\nLOOPS = " << NLOOPS << endl;
|
||||
cout << "\nN = " << N << endl;
|
||||
cout << "\nM = " << M << endl;
|
||||
|
||||
// Memory allocation
|
||||
cout << "Memory allocation\n";
|
||||
|
||||
vector<double> A(M*N);
|
||||
vector<double> x(N);
|
||||
|
||||
cout.precision(2);
|
||||
cout << (1.0*M*N + N) * sizeof(x[0]) / 1024 / 1024 / 1024 << " GByte Memory allocated\n";
|
||||
cout.precision(6);
|
||||
|
||||
// Data initialization
|
||||
|
||||
for (size_t i = 0; i < M; ++i)
|
||||
for (size_t j = 0; j < N; ++j)
|
||||
A[N*i + j] = (i + j) % 219 + 1;
|
||||
|
||||
|
||||
for (size_t j = 0; j < N; ++j)
|
||||
{
|
||||
x[j] = 1.0/A[N*17 + j];
|
||||
}
|
||||
|
||||
cout << "\nStart Benchmarking MatVec\n";
|
||||
|
||||
auto t1 = system_clock::now(); // start timer
|
||||
// Do calculation
|
||||
vector<double> b(M);
|
||||
|
||||
for (size_t i = 0; i < NLOOPS; ++i)
|
||||
{
|
||||
b = MatVec_parallel(A, x);
|
||||
}
|
||||
|
||||
auto t2 = system_clock::now(); // stop timer
|
||||
auto duration = duration_cast<microseconds>(t2 - t1); // duration in microseconds
|
||||
double t_diff = static_cast<double>(duration.count()) / 1e6; // overall duration in seconds
|
||||
t_diff = t_diff/NLOOPS; // duration per loop seconds
|
||||
|
||||
|
||||
// Check the correct result
|
||||
cout << "\n <A[17,*],x> = " << b[17] << endl;
|
||||
if (static_cast<size_t>(b[17]) != N)
|
||||
{
|
||||
cout << " !! W R O N G result !!\n";
|
||||
}
|
||||
cout << endl;
|
||||
|
||||
|
||||
// Timings and Performance
|
||||
cout << endl;
|
||||
cout.precision(2);
|
||||
|
||||
double Gflops = (2.0*N*M) / t_diff / 1024 / 1024 / 1024;
|
||||
double MemBandwidth = (2.0*N*M + M)/ t_diff / 1024 / 1024 / 1024 * sizeof(x[0]);
|
||||
|
||||
cout << "Total duration : " << t_diff*NLOOPS << endl;
|
||||
cout << "Timing in sec. : " << t_diff << endl;
|
||||
cout << "GFLOPS : " << Gflops << endl;
|
||||
cout << "GiByte/s : " << MemBandwidth << endl;
|
||||
|
||||
|
||||
|
||||
return vector<double>{t_diff, Gflops, MemBandwidth};
|
||||
}
|
||||
|
||||
|
||||
vector<double> test_C(const size_t &NLOOPS, const size_t &L, const size_t &M, const size_t &N)
|
||||
{
|
||||
cout << "#################### (C) ####################" << endl;
|
||||
cout << "\nLOOPS = " << NLOOPS << endl;
|
||||
cout << "\nL = " << L << endl;
|
||||
cout << "\nM = " << M << endl;
|
||||
cout << "\nN = " << N << endl;
|
||||
|
||||
|
||||
// Memory allocation
|
||||
cout << "Memory allocation\n";
|
||||
|
||||
vector<double> A(M*L);
|
||||
vector<double> B(L*N);
|
||||
|
||||
cout.precision(2);
|
||||
cout << (1.0*M*L + L*N) *sizeof(A[0]) / 1024 / 1024 / 1024 << " GByte Memory allocated\n";
|
||||
cout.precision(6);
|
||||
|
||||
|
||||
// Data initialization
|
||||
|
||||
for (size_t i = 0; i < M; ++i)
|
||||
for (size_t k = 0; k < L; ++k)
|
||||
A[L*i + k] = (i + k) % 219 + 1;
|
||||
|
||||
for (size_t k = 0; k < L; ++k)
|
||||
for (size_t j = 0; j < N; ++j)
|
||||
B[N*k + j] = 1.0/A[L*17 + k];
|
||||
|
||||
|
||||
cout << "\nStart Benchmarking MatMat\n";
|
||||
|
||||
auto t1 = system_clock::now(); // start timer
|
||||
// Do calculation
|
||||
vector<double> C(M*N);
|
||||
double check;
|
||||
double check_sum = 0;
|
||||
|
||||
for (size_t i = 0; i < NLOOPS; ++i)
|
||||
{
|
||||
C = MatMat_parallel(A, B, L);
|
||||
|
||||
check = C[N*17];
|
||||
check_sum += check; // prevents the optimizer from removing unused calculation results.
|
||||
}
|
||||
cout << check_sum;
|
||||
|
||||
auto t2 = system_clock::now(); // stop timer
|
||||
auto duration = duration_cast<microseconds>(t2 - t1); // duration in microseconds
|
||||
double t_diff = static_cast<double>(duration.count()) / 1e6; // overall duration in seconds
|
||||
t_diff = t_diff/NLOOPS; // duration per loop seconds
|
||||
|
||||
|
||||
// Check the correct result
|
||||
cout << "\n C[17,0] = " << check << endl;
|
||||
if (static_cast<unsigned int>(check) != L)
|
||||
{
|
||||
cout << " !! W R O N G result !!, should be " << L <<"\n";
|
||||
}
|
||||
cout << endl;
|
||||
|
||||
// Timings and Performance
|
||||
cout << endl;
|
||||
cout.precision(2);
|
||||
|
||||
|
||||
double Gflops = (2.0*L*N*M) / t_diff / 1024 / 1024 / 1024;
|
||||
double MemBandwidth = (2.0*L*N*M + M*N)/ t_diff / 1024 / 1024 / 1024 * sizeof(A[0]);
|
||||
|
||||
cout << "Total duration : " << t_diff*NLOOPS << endl;
|
||||
cout << "Timing in sec. : " << t_diff << endl;
|
||||
cout << "GFLOPS : " << Gflops << endl;
|
||||
cout << "GiByte/s : " << MemBandwidth << endl;
|
||||
|
||||
|
||||
|
||||
return vector<double>{t_diff, Gflops, MemBandwidth};
|
||||
}
|
||||
|
||||
|
||||
vector<double> test_D(const size_t &NLOOPS, const size_t &N, const size_t &p)
|
||||
{
|
||||
cout << "#################### (D) ####################" << endl;
|
||||
cout << "\nLOOPS = " << NLOOPS << endl;
|
||||
cout << "\nN = " << N << endl;
|
||||
cout << "\np = " << p << endl;
|
||||
|
||||
// Memory allocation
|
||||
cout << "Memory allocation\n";
|
||||
|
||||
vector<double> a(p + 1, 0);
|
||||
vector<double> x(N);
|
||||
|
||||
cout.precision(2);
|
||||
cout << (1.0*(p + 1) + N) *sizeof(x[0]) / 1024 / 1024 / 1024 << " GByte Memory allocated\n";
|
||||
cout.precision(6);
|
||||
|
||||
// Data initialization
|
||||
|
||||
for (size_t j = 0; j < N; ++j)
|
||||
x[j] = 1.0*j;
|
||||
|
||||
for (size_t k = 0; k < p + 1; ++k)
|
||||
a[k] = pow(-1.0, k); // poly(x) = 1 - x + x^2 - x^3 + x^4 - ...
|
||||
|
||||
|
||||
|
||||
cout << "\nStart Benchmarking poly\n";
|
||||
|
||||
auto t1 = system_clock::now(); // start timer
|
||||
// Do calculation
|
||||
vector<double> y(N);
|
||||
double check;
|
||||
double check_sum;
|
||||
|
||||
for (size_t i = 0; i < NLOOPS; ++i)
|
||||
{
|
||||
y = poly_parallel(a, x);
|
||||
check = y[0];
|
||||
|
||||
check_sum += check; // prevents the optimizer from removing unused calculation results.
|
||||
}
|
||||
|
||||
auto t2 = system_clock::now(); // stop timer
|
||||
auto duration = duration_cast<microseconds>(t2 - t1); // duration in microseconds
|
||||
double t_diff = static_cast<double>(duration.count()) / 1e6; // overall duration in seconds
|
||||
t_diff = t_diff/NLOOPS; // duration per loop seconds
|
||||
|
||||
|
||||
|
||||
// Check the correct result
|
||||
cout << "\n poly(" << x[0] << ") = " << check << endl;
|
||||
if (abs(check - 1.0) > 1.0/1e6)
|
||||
{
|
||||
cout << " !! W R O N G result !!\n";
|
||||
}
|
||||
cout << endl;
|
||||
|
||||
|
||||
// Timings and Performance
|
||||
cout << endl;
|
||||
cout.precision(2);
|
||||
|
||||
|
||||
double Gflops = (N*(p + 1)*3.0) / t_diff / 1024 / 1024 / 1024;
|
||||
double MemBandwidth = (N*(2.0 + 3.0*(p + 1)))/ t_diff / 1024 / 1024 / 1024 * sizeof(x[0]);
|
||||
|
||||
cout << "Total duration : " << t_diff*NLOOPS << endl;
|
||||
cout << "Timing in sec. : " << t_diff << endl;
|
||||
cout << "GFLOPS : " << Gflops << endl;
|
||||
cout << "GiByte/s : " << MemBandwidth << endl;
|
||||
|
||||
|
||||
|
||||
return vector<double>{t_diff, Gflops, MemBandwidth};
|
||||
}
|
||||
13
ex5/ex5_4/benchmark_tests.h
Normal file
13
ex5/ex5_4/benchmark_tests.h
Normal file
|
|
@ -0,0 +1,13 @@
|
|||
#pragma once
|
||||
#include <vector>
|
||||
using namespace std;
|
||||
|
||||
vector<double> test_A(const size_t &NLOOPS, const size_t &N);
|
||||
|
||||
vector<double> test_A_sum(const size_t &NLOOPS, const size_t &N);
|
||||
|
||||
vector<double> test_B(const size_t &NLOOPS, const size_t &N, const size_t &M);
|
||||
|
||||
vector<double> test_C(const size_t &NLOOPS, const size_t &L, const size_t &M, const size_t &N);
|
||||
|
||||
vector<double> test_D(const size_t &NLOOPS, const size_t &N, const size_t &p);
|
||||
141
ex5/ex5_4/benchmarks.cpp
Normal file
141
ex5/ex5_4/benchmarks.cpp
Normal file
|
|
@ -0,0 +1,141 @@
|
|||
#include "benchmarks.h"
|
||||
#include <cassert> // assert()
|
||||
#include <cmath>
|
||||
#include <iostream>
|
||||
#include <vector>
|
||||
#include <omp.h>
|
||||
|
||||
// (A) Inner product of two vectors (from skalar_stl)
|
||||
double scalar_parallel(vector<double> const &x, vector<double> const &y)
|
||||
{
|
||||
assert(x.size() == y.size());
|
||||
size_t const N = x.size();
|
||||
double sum = 0.0;
|
||||
//#pragma omp parallel for default(none) shared(x, y, N) reduction(+:sum) schedule(runtime)
|
||||
#pragma omp parallel for shared(x, y, N) reduction(+:sum)
|
||||
for (size_t i = 0; i < N; ++i)
|
||||
{
|
||||
sum += x[i] * y[i];
|
||||
}
|
||||
return sum;
|
||||
}
|
||||
|
||||
// (A) Vector entry sum
|
||||
double sum(vector<double> const &x)
|
||||
{
|
||||
double sum = 0.0;
|
||||
#pragma omp parallel for shared(x) reduction(+:sum)
|
||||
for (size_t i = 0; i < x.size(); ++i)
|
||||
{
|
||||
sum += x[i];
|
||||
}
|
||||
return sum;
|
||||
}
|
||||
|
||||
|
||||
// (B) Matrix-vector product (from intro_vector_densematrix)
|
||||
vector<double> MatVec_parallel(vector<double> const &A, vector<double> const &x)
|
||||
{
|
||||
size_t const nelem = A.size();
|
||||
size_t const N = x.size();
|
||||
assert(nelem % N == 0); // make sure multiplication is possible
|
||||
size_t const M = nelem/N;
|
||||
|
||||
vector<double> b(M);
|
||||
|
||||
#pragma omp parallel for shared(A, x, N, M, b)
|
||||
for (size_t i = 0; i < M; ++i)
|
||||
{
|
||||
double tmp = 0.0;
|
||||
for (size_t j = 0; j < N; ++j)
|
||||
tmp += A[N*i + j] * x[j];
|
||||
b[i] = tmp;
|
||||
}
|
||||
|
||||
return b;
|
||||
}
|
||||
|
||||
|
||||
// (C) Matrix-matrix product
|
||||
vector<double> MatMat_parallel(vector<double> const &A, vector<double> const &B, size_t const &L)
|
||||
{
|
||||
size_t const nelem_A = A.size();
|
||||
size_t const nelem_B = B.size();
|
||||
|
||||
assert(nelem_A % L == 0 && nelem_B % L == 0);
|
||||
|
||||
size_t const M = nelem_A/L;
|
||||
size_t const N = nelem_B/L;
|
||||
|
||||
|
||||
vector<double> C(M*N);
|
||||
|
||||
|
||||
#pragma omp parallel for shared(A, B, M, N, L, C)
|
||||
for (size_t i = 0; i < M; ++i)
|
||||
{
|
||||
for (size_t k = 0; k < L; ++k)
|
||||
{
|
||||
for (size_t j = 0; j < N; ++j)
|
||||
{
|
||||
C[N*i + j] += A[L*i + k]*B[N*k + j];
|
||||
}
|
||||
|
||||
}
|
||||
}
|
||||
|
||||
return C;
|
||||
}
|
||||
|
||||
|
||||
// (D) Evaluation of a polynomial function
|
||||
vector<double> poly_parallel(vector<double> const &a, vector<double> const &x)
|
||||
{
|
||||
size_t const N = x.size();
|
||||
size_t const p = a.size() - 1;
|
||||
vector<double> y(N, 0);
|
||||
|
||||
#pragma omp parallel for shared(a, x, N, p, y)
|
||||
for (size_t i = 0; i < N; ++i)
|
||||
{
|
||||
double x_temp = x[i];
|
||||
double y_temp = 0;
|
||||
for (size_t k = 0; k < p + 1; ++k)
|
||||
{
|
||||
y_temp += x_temp*y_temp + a[p - k];
|
||||
}
|
||||
y[i] = y_temp;
|
||||
}
|
||||
|
||||
return y;
|
||||
}
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
55
ex5/ex5_4/benchmarks.h
Normal file
55
ex5/ex5_4/benchmarks.h
Normal file
|
|
@ -0,0 +1,55 @@
|
|||
#pragma once
|
||||
#include <vector>
|
||||
using namespace std;
|
||||
|
||||
/** (A) Inner product of two vectors (from skalar_stl)
|
||||
@param[in] x vector
|
||||
@param[in] y vector
|
||||
@return resulting Euclidian inner product <x,y>
|
||||
*/
|
||||
double scalar_parallel(vector<double> const &x, vector<double> const &y);
|
||||
|
||||
|
||||
/** (A) Sum entries of vector
|
||||
@param[in] x vector
|
||||
@return sum
|
||||
*/
|
||||
double sum(vector<double> const &x);
|
||||
|
||||
|
||||
/** (B) Matrix-vector product (from intro_vector_densematrix)
|
||||
* @param[in] A dense matrix (1D access)
|
||||
* @param[in] u vector
|
||||
*
|
||||
* @return resulting vector
|
||||
*/
|
||||
vector<double> MatVec_parallel(vector<double> const &A, vector<double> const &x);
|
||||
|
||||
|
||||
/** (C) Matrix-matrix product
|
||||
* @param[in] A MxL dense matrix (1D access)
|
||||
* @param[in] B LxN dense matrix (1D access)
|
||||
* @param[in] shared_dim shared dimension L
|
||||
*
|
||||
* @return resulting MxN matrix
|
||||
*/
|
||||
vector<double> MatMat_parallel(vector<double> const &A, vector<double> const &B, size_t const &shared_dim);
|
||||
|
||||
|
||||
/** (D) Evaluation of a polynomial function using Horner's scheme
|
||||
* @param[in] a coefficient vector
|
||||
* @param[in] x vector with input values
|
||||
*
|
||||
* @return vector with output values
|
||||
*/
|
||||
vector<double> poly_parallel(vector<double> const &a, vector<double> const &x);
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
84
ex5/ex5_4/main.cpp
Normal file
84
ex5/ex5_4/main.cpp
Normal file
|
|
@ -0,0 +1,84 @@
|
|||
#include "benchmark_tests.h"
|
||||
#include <iostream>
|
||||
#include <cmath>
|
||||
|
||||
int main()
|
||||
{
|
||||
vector<vector<double>> results_scalar;
|
||||
results_scalar.push_back(test_A(2000000, pow(10,3)));
|
||||
results_scalar.push_back(test_A(1000000, pow(10,4)));
|
||||
results_scalar.push_back(test_A(100000, pow(10,5)));
|
||||
results_scalar.push_back(test_A(10000, pow(10,6)));
|
||||
results_scalar.push_back(test_A(750, pow(10,7)));
|
||||
results_scalar.push_back(test_A(125, pow(10,8)));
|
||||
|
||||
|
||||
vector<vector<double>> results_sum;
|
||||
results_sum.push_back(test_A_sum(3000000, pow(10,3)));
|
||||
results_sum.push_back(test_A_sum(2000000, pow(10,4)));
|
||||
results_sum.push_back(test_A_sum(1000000, pow(10,5)));
|
||||
results_sum.push_back(test_A_sum(50000, pow(10,6)));
|
||||
results_sum.push_back(test_A_sum(2000, pow(10,7)));
|
||||
results_sum.push_back(test_A_sum(250, pow(10,8)));
|
||||
|
||||
|
||||
test_B(100, 20000, 10000);
|
||||
|
||||
test_C(25, 500, 1000, 1500);
|
||||
|
||||
test_D(100, 100, 1000000);
|
||||
|
||||
|
||||
|
||||
cout << endl << "###### Scalar ######" << endl;
|
||||
cout << "Timing\tGFLOPS\tGiByte/s" << endl;
|
||||
cout << "------------------------------" << endl;
|
||||
for (size_t i = 0; i < results_scalar.size(); ++i)
|
||||
cout << results_scalar[i][0] << "\t" << results_scalar[i][1] << "\t" << results_scalar[i][2] << endl;
|
||||
|
||||
cout << endl << "###### Sum ######" << endl;
|
||||
cout << "Timing\tGFLOPS\tGiByte/s" << endl;
|
||||
cout << "------------------------------" << endl;
|
||||
for (size_t i = 0; i < results_sum.size(); ++i)
|
||||
cout << results_sum[i][0] << "\t" << results_sum[i][1] << "\t" << results_sum[i][2] << endl;
|
||||
|
||||
|
||||
|
||||
|
||||
// ###### Scalar ######
|
||||
// Timing GFLOPS GiByte/s
|
||||
// ------------------------------
|
||||
// 3.4e-06 0.54 4.3
|
||||
// 4.6e-06 4 32
|
||||
// 1.6e-05 12 95
|
||||
// 0.0011 1.7 13
|
||||
// 0.0097 1.9 15
|
||||
// 0.075 2.5 20
|
||||
|
||||
|
||||
// ###### Sum ######
|
||||
// Timing GFLOPS GiByte/s
|
||||
// ------------------------------
|
||||
// 5.5e-06 0.17 1.3
|
||||
// 5.4e-06 1.7 14
|
||||
// 1.5e-05 6.1 49
|
||||
// 0.00013 7.2 57
|
||||
// 0.0033 2.8 23
|
||||
// 0.032 2.9 23
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
// ######### NOT PARALLEL (from exercise sheet 2) #########
|
||||
// Timing GFLOPS GiByte/s
|
||||
// ----------------------------------
|
||||
// (A) 0.038 2.5 20
|
||||
// (B) 0.13 2.9 23
|
||||
// (C) 0.44 3.2 25
|
||||
// (D) 0.19 1.5 12
|
||||
|
||||
|
||||
|
||||
return 0;
|
||||
}
|
||||
Loading…
Add table
Add a link
Reference in a new issue