Changeset 1616 for LMDZ5/trunk/libf/dyn3d/leapfrog.F

Timestamp:

Feb 17, 2012, 12:59:00 PM (12 years ago)

Author:

Ehouarn Millour

Message:

Some cleanup around what is done during the integration step of dynamical tendencies and namely removed computation of (unused) finvmaold, thereby saving us the expense of a call to the (costly) filter at every dynamical time step.
Checked (on Vargas, in seq, omp, mpi and mixed mode) that this doesn't change the GCM results, as expected.
EM

File:

• LMDZ5/trunk/libf/dyn3d/leapfrog.F (modified) (4 diffs)

LMDZ5/trunk/libf/dyn3d/leapfrog.F

@@ -117 +117 @@
 
       REAL  SSUM
-      REAL time_0 , finvmaold(ip1jmp1,llm)
+      REAL time_0
+!     REAL finvmaold(ip1jmp1,llm)
 
 cym      LOGICAL  lafin
@@ -257 +258 @@
 
 c   ...    P.Le Van .26/04/94  ....
-
-      CALL SCOPY   ( ijp1llm,   masse, 1, finvmaold,     1 )
-      CALL filtreg ( finvmaold ,jjp1, llm, -2,2, .TRUE., 1 )
+! Ehouarn: finvmaold is actually not used
+!      CALL SCOPY   ( ijp1llm,   masse, 1, finvmaold,     1 )
+!      CALL filtreg ( finvmaold ,jjp1, llm, -2,2, .TRUE., 1 )
 
    2  CONTINUE ! Matsuno backward or leapfrog step begins here
@@ -318 +319 @@
 
       IF( forward. OR . leapf )  THEN
-
+! Ehouarn: NB: at this point p with ps are not synchronized
+!              (whereas mass and ps are...)
          CALL caladvtrac(q,pbaru,pbarv,
      *        p, masse, dq,  teta,
@@ -343 +345 @@
 
        CALL integrd ( 2,vcovm1,ucovm1,tetam1,psm1,massem1 ,
-     $         dv,du,dteta,dq,dp,vcov,ucov,teta,q,ps,masse,phis ,
-     $              finvmaold                                    )
+     $         dv,du,dteta,dq,dp,vcov,ucov,teta,q,ps,masse,phis )
+!     $              finvmaold                                    )