Skip to content
Snippets Groups Projects
Commit da0d5ebc authored by ESCOBAR MUNOZ Juan's avatar ESCOBAR MUNOZ Juan
Browse files

Juan 22/04/2022:Juan:ZSOLVER/turb.f90, Bit Reproductible for CRAY -> MNH_BITREP_OMP

parent 8a573328
Branches
Tags
1 merge request!4Jean Wurtz 30/04/2025 : Bugfixes mainly for TEB and for simple precision
Hide whitespace changes
Inline Side-by-side
Showing
with 66 additions and 57 deletions
src/ZSOLVER/turb.f90
+ 66
− 57
View file @ da0d5ebc
...@@ -11,8 +11,7 @@ module mode_turb ...@@ -11,8 +11,7 @@ module mode_turb
use mode_msg use mode_msg
USE MODE_MNH_ZWORK, ONLY: MNH_MEM_GET, MNH_MEM_POSITION_PIN, MNH_MEM_RELEASE USE MODE_MNH_ZWORK, ONLY: MNH_MEM_GET, MNH_MEM_POSITION_PIN, MNH_MEM_RELEASE
#endif #endif
#if defined(MNH_BITREP) || defined(MNH_BITREP_OMP)
#ifdef MNH_BITREP
use modi_bitrep use modi_bitrep
#endif #endif
...@@ -469,6 +468,7 @@ LOGICAL :: GOCEAN !Intermediate variable used to work around a Cray compiler bug ...@@ -469,6 +468,7 @@ LOGICAL :: GOCEAN !Intermediate variable used to work around a Cray compiler bug
! !
REAL, DIMENSION(:,:,:), pointer , contiguous :: ZTEMP_BUD REAL, DIMENSION(:,:,:), pointer , contiguous :: ZTEMP_BUD
! !
LOGICAL :: GTURBLEN_BL89_TURBLEN_RM17_TURBLEN_ADAP_ORMC01
!------------------------------------------------------------------------------------------ !------------------------------------------------------------------------------------------
! !
! IN variables ! IN variables
...@@ -706,13 +706,17 @@ ZRVORD= XRV / XRD ...@@ -706,13 +706,17 @@ ZRVORD= XRV / XRD
! !
GOCEAN = LOCEAN GOCEAN = LOCEAN
! !
GTURBLEN_BL89_TURBLEN_RM17_TURBLEN_ADAP_ORMC01 = &
HTURBLEN=='BL89' .OR. HTURBLEN=='RM17' .OR. HTURBLEN == 'ADAP' .OR. ORMC01
! !
!$acc update device(PTHLT,PRT) !$acc update device(PTHLT,PRT)
!$acc kernels present_cr(ZCOEF_DISS) !$acc kernels present_cr(ZCOEF_DISS,ZTHLM,ZRM)
!Copy data into ZTHLM and ZRM only if needed !Copy data into ZTHLM and ZRM only if needed
IF (HTURBLEN=='BL89' .OR. HTURBLEN=='RM17' .OR. HTURBLEN == 'ADAP' .OR. ORMC01) THEN IF (GTURBLEN_BL89_TURBLEN_RM17_TURBLEN_ADAP_ORMC01) THEN
ZTHLM(:,:,:) = PTHLT(:,:,:) DO CONCURRENT(JI=1:JIU,JJ=1:JJU,JK=1:JKU)
ZRM(:,:,:,:) = PRT(:,:,:,:) ZTHLM(JI,JJ,JK) = PTHLT(JI,JJ,JK)
ZRM(JI,JJ,JK,:) = PRT(JI,JJ,JK,:)
END DO
END IF END IF
! !
ZTRH(:, :, : ) = XUNDEF ZTRH(:, :, : ) = XUNDEF
...@@ -746,10 +750,12 @@ ELSE ...@@ -746,10 +750,12 @@ ELSE
!PW: "BUG" PGI : results different on CPU and GPU due to the power function !PW: "BUG" PGI : results different on CPU and GPU due to the power function
!See http://www.pgroup.com/userforum/viewtopic.php?t=5364&sid=ec7b762b17fb9bb3332a47f0db57af55 !See http://www.pgroup.com/userforum/viewtopic.php?t=5364&sid=ec7b762b17fb9bb3332a47f0db57af55
!Use of own functions allows bit-reproducible results !Use of own functions allows bit-reproducible results
#ifndef MNH_BITREP #if !defined(MNH_BITREP) && !defined(MNH_BITREP_OMP)
ZEXN(:,:,:) = (PPABST(:,:,:)/XP00) ** (XRD/XCPD) ZEXN(:,:,:) = (PPABST(:,:,:)/XP00) ** (XRD/XCPD)
#else #else
ZEXN(:,:,:) = BR_POW(PPABST(:,:,:)/XP00,XRD/XCPD) DO CONCURRENT(JI=1:JIU,JJ=1:JJU,JK=1:JKU)
ZEXN(JI,JJ,JK) = BR_POW(PPABST(JI,JJ,JK)/XP00,XRD/XCPD)
END DO
#endif #endif
END IF END IF
! !
...@@ -1003,7 +1009,7 @@ IF (ORMC01) THEN ...@@ -1003,7 +1009,7 @@ IF (ORMC01) THEN
#ifdef MNH_OPENACC #ifdef MNH_OPENACC
call Print_msg( NVERB_FATAL, 'GEN', 'TURB', 'OpenACC: ORMC01 not yet implemented' ) call Print_msg( NVERB_FATAL, 'GEN', 'TURB', 'OpenACC: ORMC01 not yet implemented' )
#endif #endif
#ifndef MNH_BITREP #if !defined(MNH_BITREP) && !defined(MNH_BITREP_OMP)
ZUSTAR(:,:) = (PSFU(:,:)**2+PSFV(:,:)**2)**(0.25) ZUSTAR(:,:) = (PSFU(:,:)**2+PSFV(:,:)**2)**(0.25)
#else #else
ZUSTAR(:,:) = BR_POW( BR_P2( PSFU(:,:) ) + BR_P2( PSFV(:,:) ), 0.25 ) ZUSTAR(:,:) = BR_POW( BR_P2( PSFU(:,:) ) + BR_P2( PSFV(:,:) ), 0.25 )
...@@ -1063,17 +1069,15 @@ ENDIF ...@@ -1063,17 +1069,15 @@ ENDIF
! !
!* 4.2 compute the proportionality coefficient between wind and stress !* 4.2 compute the proportionality coefficient between wind and stress
! !
!$acc kernels present_cr(ZCDUEFF,ZTAU22M,ZTAU33M) !$acc kernels ! present_cr(ZCDUEFF,ZUSLOPE,ZVSLOPE,PSFU,PSFV)
#ifndef MNH_BITREP #if !defined(MNH_BITREP) && !defined(MNH_BITREP_OMP)
ZCDUEFF(:,:) =-SQRT ( (PSFU(:,:)**2 + PSFV(:,:)**2) / & ZCDUEFF(:,:) =-SQRT ( (PSFU(:,:)**2 + PSFV(:,:)**2) / &
(XMNH_TINY + ZUSLOPE(:,:)**2 + ZVSLOPE(:,:)**2 ) ) (XMNH_TINY + ZUSLOPE(:,:)**2 + ZVSLOPE(:,:)**2 ) )
#else #else
#ifdef MNH_COMPILER_NVHPC !$acc_nv loop independent collapse(2)
!$acc loop independent collapse(2)
#endif
DO CONCURRENT ( JI=1:JIU,JJ=1:JJU ) DO CONCURRENT ( JI=1:JIU,JJ=1:JJU )
ZCDUEFF(JI,JJ) =-SQRT ( (BR_P2(PSFU(JI,JJ)) + BR_P2(PSFV(JI,JJ))) / & ZCDUEFF(JI,JJ) =-SQRT ( (BR_P2(PSFU(JI,JJ)) + BR_P2(PSFV(JI,JJ))) / &
(XMNH_TINY + BR_P2(ZUSLOPE(JI,JJ)) + BR_P2(ZVSLOPE(JI,JJ)) ) ) (XMNH_TINY + BR_P2(ZUSLOPE(JI,JJ)) + BR_P2(ZVSLOPE(JI,JJ)) ) )
END DO END DO
#endif #endif
!$acc end kernels !$acc end kernels
...@@ -1855,7 +1859,7 @@ CALL MNH_MEM_GET( zdrvsatdt, size( pexn, 1 ), size( pexn, 2 ), size( pexn, 3 ) ) ...@@ -1855,7 +1859,7 @@ CALL MNH_MEM_GET( zdrvsatdt, size( pexn, 1 ), size( pexn, 2 ), size( pexn, 3 ) )
! !
!* 1.1 Lv/Cph at t !* 1.1 Lv/Cph at t
! !
!$acc kernels present(PLOCPEXN,ZDRVSATDT) !$acc kernels ! present(ZRVSAT,ZDRVSATDT) ! present(PLOCPEXN) ! present ZDRVSATDT)
PLOCPEXN(:,:,:) = ( PLTT + (XCPV-PC) * (PT(:,:,:)-XTT) ) / PCP(:,:,:) PLOCPEXN(:,:,:) = ( PLTT + (XCPV-PC) * (PT(:,:,:)-XTT) ) / PCP(:,:,:)
! !
!* 1.2 Saturation vapor pressure at t !* 1.2 Saturation vapor pressure at t
...@@ -1863,13 +1867,15 @@ CALL MNH_MEM_GET( zdrvsatdt, size( pexn, 1 ), size( pexn, 2 ), size( pexn, 3 ) ) ...@@ -1863,13 +1867,15 @@ CALL MNH_MEM_GET( zdrvsatdt, size( pexn, 1 ), size( pexn, 2 ), size( pexn, 3 ) )
!PW: "BUG" PGI : results different on CPU and GPU due to the EXP and LOG functions !PW: "BUG" PGI : results different on CPU and GPU due to the EXP and LOG functions
!See http://www.pgroup.com/userforum/viewtopic.php?t=5364&sid=ec7b762b17fb9bb3332a47f0db57af55 !See http://www.pgroup.com/userforum/viewtopic.php?t=5364&sid=ec7b762b17fb9bb3332a47f0db57af55
!Use of own functions allows bit-reproducible results !Use of own functions allows bit-reproducible results
#ifndef MNH_BITREP #if !defined(MNH_BITREP) && !defined(MNH_BITREP_OMP)
ZRVSAT(:,:,:) = EXP( PALP - PBETA/PT(:,:,:) - PGAM*ALOG( PT(:,:,:) ) ) ZRVSAT(:,:,:) = EXP( PALP - PBETA/PT(:,:,:) - PGAM*ALOG( PT(:,:,:) ) )
#else #else
DO CONCURRENT ( JI=1:JIU,JJ=1:JJU,JK=1:JKU) DO CONCURRENT ( JI=1:JIU,JJ=1:JJU,JK=1:JKU)
ZRVSAT(JI,JJ,JK) = BR_EXP( PALP - PBETA/PT(JI,JJ,JK) - PGAM*BR_LOG( PT(JI,JJ,JK) ) ) ZRVSAT(JI,JJ,JK) = BR_EXP( PALP - PBETA/PT(JI,JJ,JK) - PGAM*BR_LOG( PT(JI,JJ,JK) ) )
END DO END DO
#endif #endif
!$acc end kernels
!$acc kernels present(ZRVSAT,ZDRVSATDT)
! !
!* 1.3 saturation mixing ratio at t !* 1.3 saturation mixing ratio at t
! !
...@@ -1884,24 +1890,36 @@ CALL MNH_MEM_GET( zdrvsatdt, size( pexn, 1 ), size( pexn, 2 ), size( pexn, 3 ) ) ...@@ -1884,24 +1890,36 @@ CALL MNH_MEM_GET( zdrvsatdt, size( pexn, 1 ), size( pexn, 2 ), size( pexn, 3 ) )
! !
PAMOIST(:,:,:)= 0.5 / ( 1.0 + ZDRVSATDT(:,:,:) * PLOCPEXN(:,:,:) ) PAMOIST(:,:,:)= 0.5 / ( 1.0 + ZDRVSATDT(:,:,:) * PLOCPEXN(:,:,:) )
! !
!$acc end kernels
!$acc kernels
!* 1.6 compute Atheta !* 1.6 compute Atheta
! #if !defined(MNH_BITREP) && !defined(MNH_BITREP_OMP)
PATHETA(:,:,:)= PAMOIST(:,:,:) * PEXN(:,:,:) * & PATHETA(:,:,:)= PAMOIST(:,:,:) * PEXN(:,:,:) * &
( ( ZRVSAT(:,:,:) - PRT(:,:,:,1) ) * PLOCPEXN(:,:,:) / & ( ( ZRVSAT(:,:,:) - PRT(:,:,:,1) ) * PLOCPEXN(:,:,:) / &
( 1. + ZDRVSATDT(:,:,:) * PLOCPEXN(:,:,:) ) * & ( 1. + ZDRVSATDT(:,:,:) * PLOCPEXN(:,:,:) ) * &
( & ( &
ZRVSAT(:,:,:) * (1. + ZRVSAT(:,:,:)/ZEPS) & ZRVSAT(:,:,:) * (1. + ZRVSAT(:,:,:)/ZEPS) &
#ifndef MNH_BITREP
* ( -2.*PBETA/PT(:,:,:) + PGAM ) / PT(:,:,:)**2 & * ( -2.*PBETA/PT(:,:,:) + PGAM ) / PT(:,:,:)**2 &
#else
* ( -2.*PBETA/PT(:,:,:) + PGAM ) / BR_P2(PT(:,:,:)) &
#endif
+ZDRVSATDT(:,:,:) * (1. + 2. * ZRVSAT(:,:,:)/ZEPS) & +ZDRVSATDT(:,:,:) * (1. + 2. * ZRVSAT(:,:,:)/ZEPS) &
* ( PBETA/PT(:,:,:) - PGAM ) / PT(:,:,:) & * ( PBETA/PT(:,:,:) - PGAM ) / PT(:,:,:) &
) & ) &
- ZDRVSATDT(:,:,:) & - ZDRVSATDT(:,:,:) &
) )
! #else
DO CONCURRENT( JI=1:JIU,JJ=1:JJU,JK=1:JKU)
PATHETA(JI,JJ,JK)= PAMOIST(JI,JJ,JK) * PEXN(JI,JJ,JK) * &
( ( ZRVSAT(JI,JJ,JK) - PRT(JI,JJ,JK,1) ) * PLOCPEXN(JI,JJ,JK) / &
( 1. + ZDRVSATDT(JI,JJ,JK) * PLOCPEXN(JI,JJ,JK) ) * &
( &
ZRVSAT(JI,JJ,JK) * (1. + ZRVSAT(JI,JJ,JK)/ZEPS) &
* ( -2.*PBETA/PT(JI,JJ,JK) + PGAM ) / BR_P2(PT(JI,JJ,JK)) &
+ZDRVSATDT(JI,JJ,JK) * (1. + 2. * ZRVSAT(JI,JJ,JK)/ZEPS) &
* ( PBETA/PT(JI,JJ,JK) - PGAM ) / PT(JI,JJ,JK) &
) &
- ZDRVSATDT(JI,JJ,JK) &
)
ENDDO
#endif
!* 1.7 Lv/Cph/Exner at t-1 !* 1.7 Lv/Cph/Exner at t-1
! !
PLOCPEXN(:,:,:) = PLOCPEXN(:,:,:) / PEXN(:,:,:) PLOCPEXN(:,:,:) = PLOCPEXN(:,:,:) / PEXN(:,:,:)
...@@ -2114,6 +2132,8 @@ END SUBROUTINE CLOUD_MODIF_LM ...@@ -2114,6 +2132,8 @@ END SUBROUTINE CLOUD_MODIF_LM
! !
END SUBROUTINE TURB END SUBROUTINE TURB
!################### !###################
SUBROUTINE DELT(KKA,KKU,KKL,KKB, KKE,KKTB, KKTE,ORMC01,HTURBDIM,PDXX, PDYY,PZZ,PDIRCOSZW,PLM, ODZ) SUBROUTINE DELT(KKA,KKU,KKL,KKB, KKE,KKTB, KKTE,ORMC01,HTURBDIM,PDXX, PDYY,PZZ,PDIRCOSZW,PLM, ODZ)
!################### !###################
...@@ -2173,6 +2193,7 @@ LOGICAL, INTENT(IN) :: ODZ ...@@ -2173,6 +2193,7 @@ LOGICAL, INTENT(IN) :: ODZ
! !
integer :: ji, jj, jk integer :: ji, jj, jk
integer :: JIU,JJU,JKU
LOGICAL :: GOCEAN !Intermediate variable used to work around a Cray compiler bug (CCE 13.0.0) LOGICAL :: GOCEAN !Intermediate variable used to work around a Cray compiler bug (CCE 13.0.0)
REAL :: ZALPHA ! proportionnality constant between Dz/2 and REAL :: ZALPHA ! proportionnality constant between Dz/2 and
! ! BL89 mixing length near the surface ! ! BL89 mixing length near the surface
...@@ -2181,12 +2202,10 @@ REAL :: ZD ! distance to the surface ...@@ -2181,12 +2202,10 @@ REAL :: ZD ! distance to the surface
real, dimension(:,:,:), pointer , contiguous :: ztmp1_device, ztmp2_device real, dimension(:,:,:), pointer , contiguous :: ztmp1_device, ztmp2_device
#endif #endif
! !
integer :: jiu, jju, jku
!------------------------------------------------------------------------------- !-------------------------------------------------------------------------------
JIU=SIZE(pdxx,1)
jiu = size( pdxx, 1 ) JJU=SIZE(pdxx,2)
jju = size( pdxx, 2 ) JKU=SIZE(pdxx,3)
jku = size( pdxx, 3 )
if ( mppdb_initialized ) then if ( mppdb_initialized ) then
!Check all in arrays !Check all in arrays
...@@ -2197,8 +2216,6 @@ if ( mppdb_initialized ) then ...@@ -2197,8 +2216,6 @@ if ( mppdb_initialized ) then
end if end if
#ifdef MNH_OPENACC #ifdef MNH_OPENACC
!!$allocate( ztmp1_device( JIU,JJU,JKU ) )
!!$allocate( ztmp2_device( JIU,JJU,JKU ) )
!Pin positions in the pools of MNH memory !Pin positions in the pools of MNH memory
CALL MNH_MEM_POSITION_PIN() CALL MNH_MEM_POSITION_PIN()
CALL MNH_MEM_GET( ztmp1_device, JIU,JJU,JKU ) CALL MNH_MEM_GET( ztmp1_device, JIU,JJU,JKU )
...@@ -2231,7 +2248,7 @@ IF (ODZ) THEN ...@@ -2231,7 +2248,7 @@ IF (ODZ) THEN
#endif #endif
ELSE ELSE
#ifndef MNH_OPENACC #ifndef MNH_OPENACC
#ifndef MNH_BITREP #if !defined(MNH_BITREP) && !defined(MNH_BITREP_OMP)
PLM(:,:,:) = (PLM(:,:,:)*MXF(PDXX(:,:,:))*MYF(PDYY(:,:,:)) ) ** (1./3.) PLM(:,:,:) = (PLM(:,:,:)*MXF(PDXX(:,:,:))*MYF(PDYY(:,:,:)) ) ** (1./3.)
#else #else
PLM(:,:,:) = BR_POW( PLM(:, :, : ) * MXF( PDXX(:, :, : ) ) * MYF( PDYY(:, :, : ) ), 1. / 3. ) PLM(:,:,:) = BR_POW( PLM(:, :, : ) * MXF( PDXX(:, :, : ) ) * MYF( PDYY(:, :, : ) ), 1. / 3. )
...@@ -2240,10 +2257,12 @@ IF (ODZ) THEN ...@@ -2240,10 +2257,12 @@ IF (ODZ) THEN
CALL MXF_DEVICE( PDXX, ZTMP1_DEVICE ) CALL MXF_DEVICE( PDXX, ZTMP1_DEVICE )
CALL MYF_DEVICE( PDYY, ZTMP2_DEVICE ) CALL MYF_DEVICE( PDYY, ZTMP2_DEVICE )
!$acc kernels !$acc kernels
#ifndef MNH_BITREP #if !defined(MNH_BITREP) && !defined(MNH_BITREP_OMP)
PLM(:,:,:) = ( PLM(:,:,:) * ZTMP1_DEVICE(:,:,:) * ZTMP2_DEVICE(:,:,:) ) ** (1./3.) PLM(:,:,:) = ( PLM(:,:,:) * ZTMP1_DEVICE(:,:,:) * ZTMP2_DEVICE(:,:,:) ) ** (1./3.)
#else #else
PLM(:,:,:) = BR_POW( PLM(:,:,:) * ZTMP1_DEVICE(:,:,:) * ZTMP2_DEVICE(:,:,:), 1./3. ) DO CONCURRENT ( JI=1:JIU,JJ=1:JJU,JK=1:JKU)
PLM(JI,JJ,JK) = BR_POW( PLM(JI,JJ,JK) * ZTMP1_DEVICE(JI,JJ,JK) * ZTMP2_DEVICE(JI,JJ,JK), 1./3. )
ENDDO
#endif #endif
!$acc end kernels !$acc end kernels
#endif #endif
...@@ -2263,7 +2282,7 @@ ELSE ...@@ -2263,7 +2282,7 @@ ELSE
#endif #endif
ELSE ELSE
#ifndef MNH_OPENACC #ifndef MNH_OPENACC
#ifndef MNH_BITREP #if !defined(MNH_BITREP) && !defined(MNH_BITREP_OMP)
PLM(:,:,:) = (MXF(PDXX(:,:,:))*MYF(PDYY(:,:,:)) ) ** (1./2.) PLM(:,:,:) = (MXF(PDXX(:,:,:))*MYF(PDYY(:,:,:)) ) ** (1./2.)
#else #else
PLM(:,:,:) = BR_POW(MXF(PDXX(:,:,:))*MYF(PDYY(:,:,:)), 1. / 2. ) PLM(:,:,:) = BR_POW(MXF(PDXX(:,:,:))*MYF(PDYY(:,:,:)), 1. / 2. )
...@@ -2272,10 +2291,12 @@ ELSE ...@@ -2272,10 +2291,12 @@ ELSE
CALL MXF_DEVICE( PDXX, ZTMP1_DEVICE ) CALL MXF_DEVICE( PDXX, ZTMP1_DEVICE )
CALL MYF_DEVICE( PDYY, ZTMP2_DEVICE ) CALL MYF_DEVICE( PDYY, ZTMP2_DEVICE )
!$acc kernels !$acc kernels
#ifndef MNH_BITREP #if !defined(MNH_BITREP) && !defined(MNH_BITREP_OMP)
PLM(:,:,:) = ( ZTMP1_DEVICE * ZTMP2_DEVICE ) ** (1./2.) PLM(:,:,:) = ( ZTMP1_DEVICE * ZTMP2_DEVICE ) ** (1./2.)
#else #else
PLM(:,:,:) = BR_POW( ZTMP1_DEVICE(:,:,:) * ZTMP2_DEVICE(:,:,:), 1. / 2. ) DO CONCURRENT( JI=1:JIU, JJ=1:JJU, JK=1:JKU )
PLM(JI,JJ,JK) = BR_POW( ZTMP1_DEVICE(JI,JJ,JK) * ZTMP2_DEVICE(JI,JJ,JK), 1. / 2. )
END DO
#endif #endif
!$acc end kernels !$acc end kernels
#endif #endif
...@@ -2287,9 +2308,8 @@ END IF ...@@ -2287,9 +2308,8 @@ END IF
! mixing length limited by the distance normal to the surface ! mixing length limited by the distance normal to the surface
! (with the same factor as for BL89) ! (with the same factor as for BL89)
! !
IF (.NOT. ORMC01) THEN IF (.NOT. ORMC01) THEN
#ifndef MNH_BITREP #if !defined(MNH_BITREP) && !defined(MNH_BITREP_OMP)
ZALPHA = 0.5**(-1.5) ZALPHA = 0.5**(-1.5)
#else #else
ZALPHA = BR_POW( 0.5, -1.5 ) ZALPHA = BR_POW( 0.5, -1.5 )
...@@ -2339,7 +2359,8 @@ if ( mppdb_initialized ) then ...@@ -2339,7 +2359,8 @@ if ( mppdb_initialized ) then
end if end if
#ifdef MNH_OPENACC #ifdef MNH_OPENACC
CALL MNH_MEM_RELEASE() !Release all memory allocated with MNH_MEM_GET calls since last call to MNH_MEM_POSITION_PIN
CALL MNH_MEM_RELEASE()
#endif #endif
END SUBROUTINE DELT END SUBROUTINE DELT
...@@ -2512,7 +2533,7 @@ IF ( HTURBDIM /= '1DIM' ) THEN ! 3D turbulence scheme ...@@ -2512,7 +2533,7 @@ IF ( HTURBDIM /= '1DIM' ) THEN ! 3D turbulence scheme
!PW: "BUG" PGI : results different on CPU and GPU due to the power function !PW: "BUG" PGI : results different on CPU and GPU due to the power function
!See http://www.pgroup.com/userforum/viewtopic.php?t=5364&sid=ec7b762b17fb9bb3332a47f0db57af55 !See http://www.pgroup.com/userforum/viewtopic.php?t=5364&sid=ec7b762b17fb9bb3332a47f0db57af55
!Use of own functions allows bit-reproducible results !Use of own functions allows bit-reproducible results
#ifndef MNH_BITREP #if !defined(MNH_BITREP) && !defined(MNH_BITREP_OMP)
! !
#ifndef MNH_OPENACC #ifndef MNH_OPENACC
PLM(:,:,:) = (PLM(:,:,:)*MXF(PDXX(:,:,:))*MYF(PDYY(:,:,:)) ) ** (1./3.) PLM(:,:,:) = (PLM(:,:,:)*MXF(PDXX(:,:,:))*MYF(PDYY(:,:,:)) ) ** (1./3.)
...@@ -2542,7 +2563,9 @@ IF ( HTURBDIM /= '1DIM' ) THEN ! 3D turbulence scheme ...@@ -2542,7 +2563,9 @@ IF ( HTURBDIM /= '1DIM' ) THEN ! 3D turbulence scheme
call Mppdb_check( plm, "Dear mid1:plm" ) call Mppdb_check( plm, "Dear mid1:plm" )
end if end if
!$acc kernels !$acc kernels
PLM(:,:,:) = BR_POW( PLM(:,:,:)*ZTMP1_DEVICE *ZTMP2_DEVICE , 1./3. ) DO CONCURRENT ( JI=1:JIU,JJ=1:JJU,JK=1:JKU)
PLM(JI,JJ,JK) = BR_POW( PLM(JI,JJ,JK)*ZTMP1_DEVICE(JI,JJ,JK) *ZTMP2_DEVICE(JI,JJ,JK) , 1./3. )
ENDDO
!$acc end kernels !$acc end kernels
if ( mppdb_initialized ) then if ( mppdb_initialized ) then
call Mppdb_check( plm, "Dear mid2:plm" ) call Mppdb_check( plm, "Dear mid2:plm" )
...@@ -2564,12 +2587,7 @@ CALL EMOIST(KRR,KRRI,PTHLT,PRT,PLOCPEXNM,PAMOIST,PSRCT,ZEMOIST) ...@@ -2564,12 +2587,7 @@ CALL EMOIST(KRR,KRRI,PTHLT,PRT,PLOCPEXNM,PAMOIST,PSRCT,ZEMOIST)
!$acc kernels present(ZWORK2D,PLM) !$acc kernels present(ZWORK2D,PLM)
IF (KRR>0) THEN IF (KRR>0) THEN
!$acc_nv loop independent collapse(3) private(ZVAR) !$acc_nv loop independent collapse(3) private(ZVAR)
!dir$ concurrent DO CONCURRENT( JI=1:JIU, JJ=1:JJU, JK = KKTB+1:KKTE-1)
DO JK = KKTB+1,KKTE-1
!dir$ concurrent
DO JJ=1,SIZE(PLM,2)
!dir$ concurrent
DO JI=1,SIZE(PLM,1)
ZDTHLDZ(JI,JJ,JK)= 0.5*((PTHLT(JI,JJ,JK+KKL)-PTHLT(JI,JJ,JK ))/PDZZ(JI,JJ,JK+KKL)+ & ZDTHLDZ(JI,JJ,JK)= 0.5*((PTHLT(JI,JJ,JK+KKL)-PTHLT(JI,JJ,JK ))/PDZZ(JI,JJ,JK+KKL)+ &
(PTHLT(JI,JJ,JK )-PTHLT(JI,JJ,JK-KKL))/PDZZ(JI,JJ,JK )) (PTHLT(JI,JJ,JK )-PTHLT(JI,JJ,JK-KKL))/PDZZ(JI,JJ,JK ))
ZDRTDZ(JI,JJ,JK) = 0.5*((PRT(JI,JJ,JK+KKL,1)-PRT(JI,JJ,JK ,1))/PDZZ(JI,JJ,JK+KKL)+ & ZDRTDZ(JI,JJ,JK) = 0.5*((PRT(JI,JJ,JK+KKL,1)-PRT(JI,JJ,JK ,1))/PDZZ(JI,JJ,JK+KKL)+ &
...@@ -2585,17 +2603,10 @@ IF (KRR>0) THEN ...@@ -2585,17 +2603,10 @@ IF (KRR>0) THEN
PLM(JI,JJ,JK)=MAX(XMNH_EPSILON,MIN(PLM(JI,JJ,JK), & PLM(JI,JJ,JK)=MAX(XMNH_EPSILON,MIN(PLM(JI,JJ,JK), &
0.76* SQRT(PTKET(JI,JJ,JK)/ZVAR))) 0.76* SQRT(PTKET(JI,JJ,JK)/ZVAR)))
END IF END IF
END DO
END DO
END DO END DO
ELSE! For dry atmos or unsalted ocean runs ELSE! For dry atmos or unsalted ocean runs
!$acc_nv loop independent collapse(3) private(ZVAR) !$acc_nv loop independent collapse(3) private(ZVAR)
!dir$ concurrent DO CONCURRENT( JI=1:JIU, JJ=1:JJU, JK = KKTB+1:KKTE-1)
DO JK = KKTB+1,KKTE-1
!dir$ concurrent
DO JJ=1,SIZE(PLM,2)
!dir$ concurrent
DO JI=1,SIZE(PLM,1)
ZDTHLDZ(JI,JJ,JK)= 0.5*((PTHLT(JI,JJ,JK+KKL)-PTHLT(JI,JJ,JK ))/PDZZ(JI,JJ,JK+KKL)+ & ZDTHLDZ(JI,JJ,JK)= 0.5*((PTHLT(JI,JJ,JK+KKL)-PTHLT(JI,JJ,JK ))/PDZZ(JI,JJ,JK+KKL)+ &
(PTHLT(JI,JJ,JK )-PTHLT(JI,JJ,JK-KKL))/PDZZ(JI,JJ,JK )) (PTHLT(JI,JJ,JK )-PTHLT(JI,JJ,JK-KKL))/PDZZ(JI,JJ,JK ))
IF (GOCEAN) THEN IF (GOCEAN) THEN
...@@ -2608,8 +2619,6 @@ ELSE! For dry atmos or unsalted ocean runs ...@@ -2608,8 +2619,6 @@ ELSE! For dry atmos or unsalted ocean runs
PLM(JI,JJ,JK)=MAX(XMNH_EPSILON,MIN(PLM(JI,JJ,JK), & PLM(JI,JJ,JK)=MAX(XMNH_EPSILON,MIN(PLM(JI,JJ,JK), &
0.76* SQRT(PTKET(JI,JJ,JK)/ZVAR))) 0.76* SQRT(PTKET(JI,JJ,JK)/ZVAR)))
END IF END IF
END DO
END DO
END DO END DO
END IF END IF
! special case near the surface ! special case near the surface
......
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment

Documentation | Mentions légales | CGU | Accessibilité : non conforme