📄 tstep.f90
字号:
#include <misc.h>#include <params.h>! Note that this routine has 2 complete blocks of code for PVP vs. non-PVP.! This is due to the fact that spectral coefficients are stored consecutively! along diagonals of M-N wavenumber space when the target architecture is ! PVP (optimal for vectorization), and along total wavenumber N otherwise! (optimal for message-passing).#if ( defined PVP )subroutine tstep(n ,zdt ,ztdtsq )!----------------------------------------------------------------------- ! ! Purpose: ! ! Method: ! Solution of the vertically coupled system of equations arising! from the semi-impicit equations for each spectral element along! the n(th) diagonal. (Note, n is distinct from the two dimensional! wavenumber which is also often denoted n.) The inverse matrix depends! only on two dimensional wavenumber and the reference atmosphere.! It is precomputed and stored for use during the forecast. The routine! overwrites the d,T and lnps coefficients with the new values.! ! Author: ! Original version: CCM1! Standardized: J. Rosinski, June 1992! Reviewed: B. Boville, D. Williamson, August 1992! Reviewed: B. Boville, D. Williamson, April 1996!!-----------------------------------------------------------------------!! $Id: tstep.F90,v 1.2 2000/12/20 18:02:03 rosinski Exp $! $Author: rosinski $!!----------------------------------------------------------------------- use pmgrid use pspect use comspe!----------------------------------------------------------------------- implicit none!------------------------------Commons----------------------------------#include <comhyb.h>!-----------------------------------------------------------------------use commap!------------------------------Arguments--------------------------------!! Input arguments! integer, intent(in) :: n ! index of spectral diagonal being calculated! this call (not two dimensional wavenumber) real(r8), intent(in) :: zdt ! timestep, dt (seconds) real(r8), intent(in) :: ztdtsq(2*pnmax) ! dt*(n(n+1)/a^2 where n is 2-d wavenumber!!---------------------------Local workspace-----------------------------! real(r8) z(2*pmmax,plev) ! workspace for computation of spectral array d real(r8) hhref ! href/2 (reference hydrostatic matrix / 2) real(r8) hbps ! bps/2 (ref. coeff. for lnps term in div. eq. / 2) real(r8) ztemp ! temporary workspace integer ne ! index into ztdtsq integer m ! diagonal element (index) of complex array integer k,kk ! level indices integer irh ! index into levels of spectral arrays integer irhr,irhi ! index into real, imaginary coefficients integer isp ! index into spectral arrays!!-----------------------------------------------------------------------!! Complete rhs of helmholtz eq.! Set offsets for beginning of diagonal being calculated this call! isp = nco2(n) - 2 ne = 2*(n-1) do k=1,plev!! Coefficients for diagonal terms! hhref = 0.5*href(k,k) hbps = 0.5*bps(k)!! Loop along current diagonal (in spectral space)! Add lnps and diagonal (vertical space) T terms to d(t-1)! do m=1,2*nm(n) d(isp+m,k) = d(isp+m,k) + & ztdtsq(ne+m)*(hhref*t(isp+m,k) + hbps*alps(isp+m)) end do if (k.lt.plev) then do kk=k+1,plev!! Add off-diagonal (vertical space) T terms to d(t-1)! hhref = 0.5*href(kk,k) do m=1,2*nm(n) d(isp+m,k) = d(isp+m,k) + ztdtsq(ne+m)*hhref*t(isp+m,kk) end do end do end if end do ! k=1,plev (calculation level)!! Solution of helmholtz equation! First: initialize temporary space for solution! do k=1,plev do m=1,2*pmmax z(m,k) = 0. end do end do do k=1,plev!! Initialize offset for diagonals (inner loop over levels)! Start inner loop over levels (for matrix multiply)! irhr = nco2(n) - 3 irhi = irhr + 1 do kk=1,plev!! Multiply right hand side by inverse matrix!!DIR$ IVDEP do m=1,nm(n) z(2*m-1,k) = z(2*m-1,k) + bm1(kk,k,m+n-1)*d(irhr+2*m,kk) z(2*m ,k) = z(2*m ,k) + bm1(kk,k,m+n-1)*d(irhi+2*m,kk) end do end do ! inner loop over levels end do ! outer loop over levels!! Move solution for divergence to d! irh = nco2(n) - 2 do k=1,plev do m=1,2*nm(n) d(irh+m,k) = z(m,k) end do end do!! Complete ln(pstar) and T forecasts! Add semi-implicit part to surface pressure (vector multiply)! do k=1,plev ztemp = zdt*hypd(k)/hypi(plevp) do m=1,2*nm(n) alps(isp+m) = alps(isp+m) - ztemp*d(isp+m,k) end do end do!! Add semi-implicit part to temperature (matrix multiply)! do k=1,plev do kk=1,plev ztemp = zdt*tau(kk,k) do m=1,2*nm(n) t(isp+m,k) = t(isp+m,k) - ztemp*d(isp+m,kk) end do end do end do! return#else subroutine tstep(m ,zdt ,ztdtsq )!-----------------------------------------------------------------------!! Solution of the vertically coupled system of equations arising! from the semi-impicit equations for each spectral element along! two dimensional wavenumber n. The inverse matrix depends! only on two dimensional wavenumber and the reference atmosphere.! It is precomputed and stored for use during the forecast. The routine! overwrites the d,T and lnps coefficients with the new values.!!---------------------------Code history--------------------------------!! Original version: CCM1! Standardized: J. Rosinski, June 1992! Reviewed: B. Boville, D. Williamson, August 1992! Reviewed: B. Boville, D. Williamson, April 1996!!-----------------------------------------------------------------------!! $Id: tstep.F90,v 1.2 2000/12/20 18:02:03 rosinski Exp $! $Author: rosinski $!!----------------------------------------------------------------------- use precision use pmgrid use pspect use comspe use commap implicit none!-----------------------------------------------------------------------#include <comhyb.h>!-----------------------------------------------------------------------!! Input arguments! integer, intent(in) :: m ! Fourier wavenumber real(r8), intent(in) :: zdt ! timestep, dt (seconds) real(r8), intent(in) :: ztdtsq(pnmax) ! dt*(n(n+1)/a^2 where n is 2-d wavenumber!!---------------------------Local workspace-----------------------------! real(r8) z(2*pnmax,plev) ! workspace for computation of spectral array d real(r8) hhref ! href/2 (reference hydrostatic matrix / 2) real(r8) hbps ! bps/2 (ref. coeff. for lnps term in div. eq. / 2) real(r8) ztemp ! temporary workspace integer n,j ! 2-d wavenumber index integer k,kk ! level indices integer mr,mc ! real and imaginary spectral indices integer ir,ii ! real and imaginary spectral indices!!-----------------------------------------------------------------------!! Complete rhs of helmholtz eq.! mr = nstart(m) mc = 2*mr do k=1,plev!! Coefficients for diagonal terms! hhref = 0.5*href(k,k) hbps = 0.5*bps(k)!! Loop along total wavenumber index (in spectral space)! Add lnps and diagonal (vertical space) T terms to d(t-1)! do n=1,nlen(m) ir = mc + 2*n - 1 ii = ir + 1 d(ir,k) = d(ir,k) + ztdtsq(n+m-1)*(hhref*t(ir,k) + hbps*alps(ir)) d(ii,k) = d(ii,k) + ztdtsq(n+m-1)*(hhref*t(ii,k) + hbps*alps(ii)) end do if (k.lt.plev) then do kk=k+1,plev!! Add off-diagonal (vertical space) T terms to d(t-1)! hhref = 0.5*href(kk,k) do n=1,nlen(m) ir = mc + 2*n - 1 ii = ir + 1 d(ir,k) = d(ir,k) + ztdtsq(n+m-1)*hhref*t(ir,kk) d(ii,k) = d(ii,k) + ztdtsq(n+m-1)*hhref*t(ii,kk) end do end do end if end do ! k=1,plev (calculation level)!! Solution of helmholtz equation! First: initialize temporary space for solution! do k=1,plev do j=1,2*pnmax z(j,k) = 0. end do end do do k=1,plev do kk=1,plev!! Multiply right hand side by inverse matrix!!DIR$ IVDEP do n=1,nlen(m) ir = mc + 2*n - 1 ii = ir + 1 z(2*n-1,k) = z(2*n-1,k) + bm1(kk,k,m+n-1)*d(ir,kk) z(2*n ,k) = z(2*n ,k) + bm1(kk,k,m+n-1)*d(ii,kk) end do end do ! inner loop over levels end do ! outer loop over levels!! Move solution for divergence to d! do k=1,plev do n=1,nlen(m) ir = mc + 2*n - 1 ii = ir + 1 d(ir,k) = z(2*n-1,k) d(ii,k) = z(2*n ,k) end do end do!! Complete ln(pstar) and T forecasts! Add semi-implicit part to surface pressure (vector multiply)! do k=1,plev ztemp = zdt*hypd(k)/hypi(plevp) do n=1,nlen(m) ir = mc + 2*n - 1 ii = ir + 1 alps(ir) = alps(ir) - ztemp*d(ir,k) alps(ii) = alps(ii) - ztemp*d(ii,k) end do end do!! Add semi-implicit part to temperature (matrix multiply)! do k=1,plev do kk=1,plev ztemp = zdt*tau(kk,k) do n=1,nlen(m) ir = mc + 2*n - 1 ii = ir + 1 t(ir,k) = t(ir,k) - ztemp*d(ir,kk) t(ii,k) = t(ii,k) - ztemp*d(ii,kk) end do end do end do! return#endif end subroutine tstep
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -