Juan 16/11/2022:MNH/*.f90 , Optimization of NOBITREP part , with mnh_expand

src/MNH/rain_ice_fast_ri.f90

@@ -162,6 +162,7 @@ CALL MNH_MEM_GET( ZLBEXI, SIZE(PRHODREF) )
   zzw(:) = 0.
   GWORK(:) = PRCS(:)>0.0 .AND. PSSI(:)>0.0 .AND. PRIT(:)>XRTMIN(4) .AND. PCIT(:)>0.0
 #if !defined(MNH_BITREP) && !defined(MNH_BITREP_OMP)
+ !$mnh_expand_where(JL=1:JLU)
   WHERE( GWORK(:) )
     ZZW(:) = MIN(1.E8,XLBI*( PRHODREF(:)*PRIT(:)/PCIT(:) )**XLBEXI) ! Lbda_i
     ZZW(:) = MIN( PRCS(:),( PSSI(:) / (PRHODREF(:)*PAI(:)) ) * PCIT(:) * &
@@ -170,6 +171,7 @@ CALL MNH_MEM_GET( ZLBEXI, SIZE(PRHODREF) )
     PRIS(:) = PRIS(:) + ZZW(:)
     PTHS(:) = PTHS(:) + ZZW(:)*(PLSFACT(:)-PLVFACT(:)) ! f(L_f*(RCBERI))
  END WHERE
+ !$mnh_end_expand_where()
 #else
 
 !!$ Le DO concurrent n'est pas bit-reproductible BUG NVHPC 20.7

src/MNH/rain_ice_warm.f90

@@ -291,6 +291,7 @@ zzw(JL) = 0.
        !Evaporation only when there's no cloud (RC must be 0)
        GWORK(:) = PRRT(:)>XRTMIN(3) .AND. PRCT(:)<=XRTMIN(2)
 #if !defined(MNH_BITREP) && !defined(MNH_BITREP_OMP)       
+       !$mnh_expand_where(JL=1:JLU)
        WHERE( GWORK(:) )
           ZZW(:)  = EXP( XALPW - XBETAW/PZT(:) - XGAMW*ALOG(PZT(:) ) ) ! es_w
           PUSW(:) = 1.0 - PRVT(:)*( PPRES(:)-ZZW(:) ) / ( (XMV/XMD) * ZZW(:) )
@@ -303,6 +304,7 @@ zzw(JL) = 0.
           PRVS(:) = PRVS(:) + ZZW(:)
           PTHS(:) = PTHS(:) - ZZW(:)*PLVFACT(:)
        END WHERE
+       !$mnh_end_expand_where()
 #else
        !$acc loop independent
        DO CONCURRENT ( JL=1:JLU )

src/MNH/tke_eps_sources.f90

@@ -423,7 +423,9 @@ CALL MZM_DEVICE(ZKEFF, ZTMP1_DEVICE) !Warning: re-used later
 CALL MZM_DEVICE(PRHODJ,ZTMP2_DEVICE) !Warning: re-used later
 !$acc kernels ! present(ZA)
 #if !defined(MNH_BITREP) && !defined(MNH_BITREP_OMP)
+!$mnh_expand_array(JI=1:JIU,JJ=1:JJU,JK=1:JKU)
 ZA(:,:,:)     = - PTSTEP * XCET * ZTMP1_DEVICE(:,:,:) * ZTMP2_DEVICE(:,:,:) / PDZZ(:,:,:)**2
+!$mnh_end_expand_array()
 #else
 !$acc_nv loop independent collapse(3)
 DO CONCURRENT ( JI=1:JIU,JJ=1:JJU,JK=1:JKU)

src/MNH/turb_hor_uv.f90

@@ -465,13 +465,10 @@ ZFLX(:,:,IKB) = - XCMFS * ZTMP2_DEVICE(:,:,1) * ( ZTMP5_DEVICE(:,:,1) + ZTMP6_DE
 #endif
 ! 
 ! extrapolates this flux under the ground with the surface flux
-#ifdef MNH_COMPILER_CCE
-!$acc kernels present(ZFLX)
-#else
-!$acc kernels 
-#endif
+!$acc parallel present_cr(ZFLX)
 #ifndef MNH_BITREP
-ZFLX(:,:,IKB-1) =                                                           &
+!$mnh_expand_array(JI=1:JIU,JJ=1:JJU)
+ ZFLX(:,:,IKB-1) =                                                           &
    PTAU11M(:,:) * PCOSSLOPE(:,:) * PSINSLOPE(:,:) * PDIRCOSZW(:,:)**2         &
   +PTAU12M(:,:) * (PCOSSLOPE(:,:)**2 - PSINSLOPE(:,:)**2) *                   &
                   PDIRCOSZW(:,:)**2                                           &
@@ -481,25 +478,23 @@ ZFLX(:,:,IKB-1) =                                                           &
     2. * PUSLOPEM(:,:) * PCOSSLOPE(:,:) * PSINSLOPE(:,:) *                    &
           PDIRCOSZW(:,:) * ZDIRSINZW(:,:)                                     &
     +PVSLOPEM(:,:) * (PCOSSLOPE(:,:)**2 - PSINSLOPE(:,:)**2) * ZDIRSINZW(:,:) )
+!$mnh_end_expand_array()
 #else
 !PW: BUG: commented 'acc loop independent collapse(2)' to workaround compiler bug (NVHPC 21.1)
-#ifdef MNH_COMPILER_NVHPC
-!acc loop independent collapse(2)
-#endif
-DO CONCURRENT ( JI=1:JIU,JJ=1:JJU )
-ZFLX(JI,JJ,IKB-1) =                                                           &
-   PTAU11M(JI,JJ) * PCOSSLOPE(JI,JJ) * PSINSLOPE(JI,JJ) * BR_P2(PDIRCOSZW(JI,JJ))     &
-  +PTAU12M(JI,JJ) * (BR_P2(PCOSSLOPE(JI,JJ)) - BR_P2(PSINSLOPE(JI,JJ))) *           &
-                  BR_P2(PDIRCOSZW(JI,JJ))                                       &
-  -PTAU22M(JI,JJ) * PCOSSLOPE(JI,JJ) * PSINSLOPE(JI,JJ)                             &
-  +PTAU33M(JI,JJ) * PCOSSLOPE(JI,JJ) * PSINSLOPE(JI,JJ) * BR_P2(ZDIRSINZW(JI,JJ))     &
-  -PCDUEFF(JI,JJ) * (                                                           &
-    2. * PUSLOPEM(JI,JJ) * PCOSSLOPE(JI,JJ) * PSINSLOPE(JI,JJ) *                    &
-          PDIRCOSZW(JI,JJ) * ZDIRSINZW(JI,JJ)                                     &
-          +PVSLOPEM(JI,JJ) * (BR_P2(PCOSSLOPE(JI,JJ)) - BR_P2(PSINSLOPE(JI,JJ))) * ZDIRSINZW(JI,JJ) )
-END DO ! CONCURRENT
+!$mnh_expand_array(JI=1:JIU,JJ=1:JJU)
+ ZFLX(:,:,IKB-1) =                                                           &
+   PTAU11M(:,:) * PCOSSLOPE(:,:) * PSINSLOPE(:,:) * BR_P2(PDIRCOSZW(:,:))     &
+  +PTAU12M(:,:) * (BR_P2(PCOSSLOPE(:,:)) - BR_P2(PSINSLOPE(:,:))) *           &
+                  BR_P2(PDIRCOSZW(:,:))                                       &
+  -PTAU22M(:,:) * PCOSSLOPE(:,:) * PSINSLOPE(:,:)                             &
+  +PTAU33M(:,:) * PCOSSLOPE(:,:) * PSINSLOPE(:,:) * BR_P2(ZDIRSINZW(:,:))     &
+  -PCDUEFF(:,:) * (                                                           &
+    2. * PUSLOPEM(:,:) * PCOSSLOPE(:,:) * PSINSLOPE(:,:) *                    &
+          PDIRCOSZW(:,:) * ZDIRSINZW(:,:)                                     &
+          +PVSLOPEM(:,:) * (BR_P2(PCOSSLOPE(:,:)) - BR_P2(PSINSLOPE(:,:))) * ZDIRSINZW(:,:) )
+!$mnh_end_expand_array()
 #endif
-!$acc end kernels
+!$acc end parallel
 !  
 #ifndef MNH_OPENACC
 ZFLX(:,:,IKB-1:IKB-1) = 2. * MXM( MYM( ZFLX(:,:,IKB-1:IKB-1) ) )  &