Fortran并行计算中OpenMP，MPI及Coarrays之单节点多线程应用比较

chiangtp

[Fortran] 纯文本查看 复制代码

!---------------------------------------
! Intel Core2 Quad @2.83GHz, Windows 10
! Compiler default optimizations
!
! Intel         9 (sec)
!
! GNU          14
! Absooft      21
! G95          21
! Lahey        24
!
! NAG         121
! PGI         127
!
! Silverfrost 331
! Compaq      546
!---------------------------------------
program no_parallel
  implicit none

  integer, parameter :: rk = SELECTED_REAL_KIND(P=15)
  real(kind=rk) :: pi

  integer :: i, n_limit, c1, c2, c_rate, c_max

  !-----------------

  n_limit = 2000000000

  call system_clock(c1, c_rate, c_max)

  pi = 0.0_rk
  do i = 1, n_limit
    pi = pi + (-1)**(i+1) / real( 2*i-1, kind=rk )
  end do
  pi = pi * 4.0_rk

  call system_clock(c2, c_rate)
  IF( c2 < c1 ) c2 = c2 + c_max

  write(*,*) pi, real(c2-c1)/real(c_rate)

end program no_parallel                                 

Intel的效率確實沒得比, GNU/Absoft/G95/Lahey持平, NAG/PGI令人跌破眼鏡, 還是我的測試有問題?