Параллельные и распределенные вычисления

Семинар 2

OpenMP

Пономаренко Роман
@rerand0m
rerandom@ispras.ru

OpenMP

Спецификация, описывающая программный интерфейс параллельного программирования.

  • Первая спецификация (для Fortran) была опубликована в октябре 1997г.

  • В настоящий момент актуальной является версия 5.2. (с ноября 2021)

Ссылка-источник для любопытных

Шпаргалка по OpenMP v4.5 для всех

Hello, world!


						Num of processors = 10
						Consecutive part, myid = 0, num of threads  = 1
						Parallel part, myid = 0, num of threads  = 2
						Parallel part, myid = 1, num of threads  = 2

Как компилировать и запускать?



$ g++ -fopenmp main.cpp
$ ./a.out
  Num of processors = 10
  Consecutive part, myid = 0, num of threads  = 1
  Parallel part, myid = 0, num of threads  = 2
  Parallel part, myid = 1, num of threads  = 2
								
# sudo apt install libomp-dev
# sudo pacman -Sy openmp
$ clang++ -fopenmp=libomp main.cpp
$ ./a.out
  Num of processors = 10
  Consecutive part, myid = 0, num of threads  = 1
  Parallel part, myid = 0, num of threads  = 2
  Parallel part, myid = 1, num of threads  = 2

Как компилировать и запускать? (mac os)

$ brew install llvm libomp
...
If you need to have llvm first in your PATH, run:
  echo 'export PATH="/opt/homebrew/opt/llvm/bin:$PATH"' >> ~/.zshrc

For compilers to find llvm you may need to set:
  export LDFLAGS="-L/opt/homebrew/opt/llvm/lib"
  export CPPFLAGS="-I/opt/homebrew/opt/llvm/include"
...
For compilers to find libomp you may need to set:
  export LDFLAGS="-L/opt/homebrew/opt/libomp/lib"
  export CPPFLAGS="-I/opt/homebrew/opt/libomp/include"
$ /opt/homebrew/opt/llvm/bin/clang++ -fopenmp=libomp \
	-I/opt/homebrew/opt/libomp/include \
	-L/opt/homebrew/opt/libomp/lib \
	main.cpp
$
$ ./a.out
  Num of processors = 10
  Consecutive part, myid = 0, num of threads  = 1
  Parallel part, myid = 0, num of threads  = 2
  Parallel part, myid = 1, num of threads  = 2

Hello, world!


						Num of processors = 10
						Consecutive part, myid = 0, num of threads  = 1
						Parallel part, myid = 0, num of threads  = 2
						Parallel part, myid = 1, num of threads  = 2

Атрибуты доступа к переменным

  • shared(list)
  • private(list)
  • default(list) (deprecated in 5.2)
  • firstprivate(list)
  • lastprivate(list)

world! Hello,

Num of processors = 10
Consecutive part, myid = 0, num of threads  = 1
Parallel part, myid = 0, num of threads  = Parallel part, myid = 810, num of threads  = Parallel part, myid = 10
Parallel part, myid = 4, num of threads  = 10
Parallel part, myid = Parallel part, myid = 5, num of threads  = 10

3, num of threads  = 10
Parallel part, myid = 9, num of threads  = 10
1, num of threads  = 10
Parallel part, myid = 6, num of threads  = 10
Parallel part, myid = 7, num of threads  = 10
Parallel part, myid = 2, num of threads  = 10

Critical

Num of processors = 10
Consecutive part, myid = 0, num of threads  = 1
Parallel part, myid = 0, num of threads  = 10
Parallel part, myid = 8, num of threads  = 10
Parallel part, myid = 4, num of threads  = 10
Parallel part, myid = 7, num of threads  = 10
Parallel part, myid = 9, num of threads  = 10
Parallel part, myid = 2, num of threads  = 10
Parallel part, myid = 6, num of threads  = 10
Parallel part, myid = 3, num of threads  = 10
Parallel part, myid = 1, num of threads  = 10
Parallel part, myid = 5, num of threads  = 10

Barrier

Parallel part 1, myid = 0
Parallel part 1, myid = 4
Parallel part 1, myid = 2
Parallel part 1, myid = 3
Parallel part 1, myid = 1
Parallel part 2, myid = 4
Parallel part 2, myid = 3
Parallel part 2, myid = 0
Parallel part 2, myid = 2
Parallel part 2, myid = 1

Atomic

0) cntr = 1
8) cntr = 3
9) cntr = 3
7) cntr = 4
6) cntr = 6
4) cntr = 7
3) cntr = 9
2) cntr = 9
1) cntr = 10
5) cntr = 10
10

Sections

Running in thread 0
Message from thread 2
Report from thread 1

Tasks

Tasks / single

Tasks / single

Почему программа выводит 80?

For

Подсчёт времени работы

double omp_get_wtime(void)

Задание

Описать класс с перегруженными конструкторами и деструктором, и попытаться передать объект такого класса при помощи атрибутов shared, private.

Оценить результат.