cam6_4_131: Performance improvements for CSLAM (from John Dennis, CISL, and Lauritzen)

cime_config/config_pes.xml

@@ -572,43 +572,6 @@
       </pes>
     </mach>
   </grid>
-  <grid name="a%T31">
-    <mach name="hera|sierra" >
-      <pes pesize="any" compset="any">
-	<comment>none</comment>
-	<ntasks>
-	  <ntasks_atm>64</ntasks_atm>
-	  <ntasks_lnd>64</ntasks_lnd>
-	  <ntasks_rof>64</ntasks_rof>
-	  <ntasks_ice>64</ntasks_ice>
-	  <ntasks_ocn>64</ntasks_ocn>
-	  <ntasks_glc>64</ntasks_glc>
-	  <ntasks_wav>64</ntasks_wav>
-	  <ntasks_cpl>64</ntasks_cpl>
-	</ntasks>
-	<nthrds>
-	  <nthrds_atm>1</nthrds_atm>
-	  <nthrds_lnd>1</nthrds_lnd>
-	  <nthrds_rof>1</nthrds_rof>
-	  <nthrds_ice>1</nthrds_ice>
-	  <nthrds_ocn>1</nthrds_ocn>
-	  <nthrds_glc>1</nthrds_glc>
-	  <nthrds_wav>1</nthrds_wav>
-	  <nthrds_cpl>1</nthrds_cpl>
-	</nthrds>
-	<rootpe>
-	  <rootpe_atm>0</rootpe_atm>
-	  <rootpe_lnd>0</rootpe_lnd>
-	  <rootpe_rof>0</rootpe_rof>
-	  <rootpe_ice>0</rootpe_ice>
-	  <rootpe_ocn>0</rootpe_ocn>
-	  <rootpe_glc>0</rootpe_glc>
-	  <rootpe_wav>0</rootpe_wav>
-	  <rootpe_cpl>0</rootpe_cpl>
-	</rootpe>
-      </pes>
-    </mach>
-  </grid>
   <grid name="a%1.9x2.5">
     <mach name="derecho">
       <pes pesize='any' compset='any'>

doc/ChangeLog

@@ -1,3 +1,86 @@
+===============================================================
+
+Tag name: cam6_4_131
+Originator(s): johnmauff, pel, cacraig, nusbaume
+Date: Nov 26, 2025
+One-line Summary: Performance improvements for CSLAM
+Github PR URL: https://github.com/ESCOMP/CAM/pull/1365
+
+Purpose of changes (include the issue number and title text for each relevant GitHub issue):
+   Excessive data movement in extend_panel_interpolate (CSLAM): https://github.com/ESCOMP/CAM/issues/1360
+
+   The subroutine extend_panel_interpolate is written such that the compiler will generate more data movement than is necessary.
+   This excessive data movement intensifies a computational load imbalance in the CSLAM advection. While it is impossible to eliminate
+   the load imbalance that is caused by the special treatment of panels at the corners of the cubed sphere, it is possible to reduce
+   the cost of this subroutine by changing the way that the subroutine is written.
+
+Describe any changes made to build system: N/A
+
+Describe any changes made to the namelist: N/A
+
+List any changes to the defaults for the boundary datasets: N/A
+
+Describe any substantial timing or memory changes: N/A
+
+Code reviewed by: nusbaume
+
+List all files eliminated: N/A
+
+List all files added and what they do: N/A
+
+List all existing files that have been modified, and describe the changes:
+
+M       cime_config/config_pes.xml
+          - remove dead config code originally used by Eulerian dycore
+
+M       src/dynamics/se/dycore/fvm_consistent_se_cslam.F90
+M       src/dynamics/se/dycore/fvm_reconstruction_mod.F90
+          - mods for SE dycore as described above
+
+If there were any failures reported from running test_driver.sh on any test
+platform, and checkin with these failures has been OK'd by the gatekeeper,
+then copy the lines from the td.*.status files for the failed tests to the
+appropriate machine below.  All failed tests must be justified.
+
+derecho/intel/aux_cam:
+  SMS_D_Ln9_P1536x1.ne0CONUSne30x8_ne0CONUSne30x8_mt12.FCHIST.derecho_intel.cam-outfrq9s (Overall: FAIL) details:
+      - intermittent failure in CTSM code (lnd_set_decomp_and_domain.F90)
+
+  ERC_D_Ln9.ne30pg2_ne30pg2_mt232.QPC7.derecho_intel.cam-outfrq9s (Overall: DIFF) details:
+  ERC_D_Ln9.ne30pg3_ne30pg3_mt232.F1850C_LTso.derecho_intel.cam-outfrq9s (Overall: DIFF) details:
+  ERI_D_Ln18.ne16pg3_ne16pg3_mt232.FHIST_C4.derecho_intel.cam-outfrq3s_eri (Overall: DIFF) details:
+  ERI_D_Ln18.ne30pg3_ne30pg3_mt232.FHISTC_LTso.derecho_intel.cam-outfrq3s_eri (Overall: DIFF) details:
+  ERP_D_Ln9.ne30pg3_ne30pg3_mt232.F1850C_MTso.derecho_intel.cam-outfrq9s (Overall: DIFF) details:
+  ERP_Ld3.ne16pg3_ne16pg3_mg17.FHISTC_WAt1ma.derecho_intel.cam-reduced_hist1d (Overall: DIFF) details:
+  ERP_Ld3.ne30pg3_ne30pg3_mt232.FHISTC_MTt4s.derecho_intel.cam-outfrq1d_aoa (Overall: DIFF) details:
+  ERP_Ln9.ne30pg3_ne30pg3_mg17.FCnudged.derecho_intel.cam-outfrq9s (Overall: DIFF) details:
+  ERP_Ln9.ne30pg3_ne30pg3_mg17.FHISTC_WAma.derecho_intel.cam-outfrq9s (Overall: DIFF) details:
+  ERR_Ln9.ne16pg3_ne16pg3_mt232.FHISTC_LTso.derecho_intel.cam-outfrq9s_bwic (Overall: DIFF) details:
+  ERS_Ln9.ne30pg3_ne30pg3_mg17.FHISTC_WXma.derecho_intel.cam-outfrq9s_ctem (Overall: DIFF) details:
+  SMS_C2_D_Ln9.ne16pg3_ne16pg3_mg17.FHISTC_WXma.derecho_intel.cam-outfrq9s (Overall: DIFF) details:
+  SMS_D_Ln9.ne30pg3_ne30pg3_mt232.FHISTC_MTso.derecho_intel.cam-outfrq9s (Overall: DIFF) details:
+  SMS_D_Ln9_P1280x1.ne30pg3_ne30pg3_mt232.FHISTC_MTt1s.derecho_intel.cam-outfrq9s_Leung_dust (Overall: DIFF) details:
+  SMS_Ld1.ne30pg3_ne30pg3_mg17.FC2010climo.derecho_intel.cam-outfrq1d (Overall: DIFF) details:
+  SMS_Ln9.ne30pg3_ne30pg3_mg17.FW2000climo.derecho_intel.cam-outfrq9s_rrtmgp (Overall: DIFF) details:
+      - answer differences for CSLAM runs
+
+derecho/nvhpc/aux_cam:
+  ERS_Ln9.ne30pg3_ne30pg3_mt232.FHISTC_LTso.derecho_nvhpc.cam-outfrq9s_gpu_default (Overall: FAIL) details:
+     - timing issue - Jian has determined this is due to changes on derecho with the the last upgrade.  He has
+                      reported the issue to CISL.  Note cam6_4_128 is the last CAM tag with baselines to use for comparison
+                      but answer changes are expected starting with cam6_4_130
+
+izumi/nag/aux_cam:
+  ERC_D_Ln27.ne3pg3_ne3pg3_mt232.FKESSLER.izumi_nag.cam-outfrq9s (Overall: DIFF) details:
+  ERC_D_Ln9.ne3pg3_ne3pg3_mt232.FHISTC_LTso.izumi_nag.cam-cosp_rad_diags (Overall: DIFF) details:
+  SMS_D_Ln3.ne5pg3_ne5pg3_mg37.QPX2000.izumi_nag.cam-outfrq3s (Overall: DIFF) details:
+      - answer differences for CSLAM runs
+
+
+izumi/gnu/aux_cam: all BFB
+
+Summarize any changes to answers:
+	Answer changing, bug not climate changing as reported by pel
 
 ===============================================================
 

src/dynamics/se/dycore/fvm_consistent_se_cslam.F90

@@ -1,3 +1,4 @@
+#define FVM_TIMERS .FALSE.
 module fvm_consistent_se_cslam
   use shr_kind_mod,           only: r8=>shr_kind_r8
   use dimensions_mod,         only: nc, nhe, nlev, ntrac, np, nhr, nhc, ngpc, ns, nht
@@ -107,7 +108,7 @@ subroutine run_consistent_se_cslam(elem,fvm,hybrid,dt_fvm,tl,nets,nete,hvcoord,&
     endif
 
     kblk = kmax-kmin+1
-    !call t_startf('fvm:before_Qnhc')
+    if(FVM_TIMERS) call t_startf('fvm:before_Qnhc')
     do ie=nets,nete
        do k=kmin,kmax
           elem(ie)%sub_elem_mass_flux(:,:,:,k) = dt_fvm*elem(ie)%sub_elem_mass_flux(:,:,:,k)*fvm(ie)%dp_ref_inverse(k)
@@ -120,11 +121,11 @@ subroutine run_consistent_se_cslam(elem,fvm,hybrid,dt_fvm,tl,nets,nete,hvcoord,&
           call ghostpack(ghostbufQnhc,fvm(ie)%c(1-nhc:nc+nhc,1-nhc:nc+nhc,kmin:kmax,q),kblk,kptr,ie)
        enddo
     end do
-    !call t_stopf('fvm:before_Qnhc')
-    !call t_startf('fvm:ghost_exchange:Qnhc')
+    if(FVM_TIMERS) call t_stopf('fvm:before_Qnhc')
+    if(FVM_TIMERS) call t_startf('fvm:ghost_exchange:Qnhc')
     call ghost_exchange(hybridnew,ghostbufQnhc,location='ghostbufQnhc')
-    !call t_stopf('fvm:ghost_exchange:Qnhc')
-    !call t_startf('fvm:orthogonal_swept_areas')
+    if(FVM_TIMERS) call t_stopf('fvm:ghost_exchange:Qnhc')
+    if(FVM_TIMERS) call t_startf('fvm:orthogonal_swept_areas')
     do ie=nets,nete
       do k=kmin,kmax
         fvm(ie)%se_flux    (1:nc,1:nc,:,k) = elem(ie)%sub_elem_mass_flux(:,:,:,k)
@@ -152,14 +153,14 @@ subroutine run_consistent_se_cslam(elem,fvm,hybrid,dt_fvm,tl,nets,nete,hvcoord,&
       end do
     enddo
 
-    !call t_stopf('fvm:orthogonal_swept_areas')
+    if(FVM_TIMERS) call t_stopf('fvm:orthogonal_swept_areas')
     do ie=nets,nete
        ! Intel compiler version 2023.0.0 on derecho had significant slowdown on subroutine interface without
        ! these pointers.
       fcube => fvm(ie)%c(:,:,:,:)
       spherecentroid => fvm(ie)%spherecentroid(:,1-nhe:nc+nhe,1-nhe:nc+nhe)
       do k=kmin,kmax
-         !call t_startf('FVM:tracers_reconstruct')
+         if(FVM_TIMERS) call t_startf('FVM:tracers_reconstruct')
          call reconstruction(fcube,nlev,k,&
              ctracer(:,:,:,:),irecons_tracer,llimiter,ntrac,&
              nc,nhe,nhr,nhc,nht,ns,nhr+(nhe-1),&
@@ -170,10 +171,10 @@ subroutine run_consistent_se_cslam(elem,fvm,hybrid,dt_fvm,tl,nets,nete,hvcoord,&
              fvm(ie)%rot_matrix,fvm(ie)%centroid_stretch,&
              fvm(ie)%vertex_recons_weights,fvm(ie)%vtx_cart,&
              irecons_tracer_lev(k))
-         !call t_stopf('FVM:tracers_reconstruct')
-         !call t_startf('fvm:swept_flux')
+         if(FVM_TIMERS) call t_stopf('FVM:tracers_reconstruct')
+         if(FVM_TIMERS) call t_startf('fvm:swept_flux')
          call swept_flux(elem(ie),fvm(ie),k,ctracer,irecons_tracer_lev(k),gsweights,gspts)
-         !call t_stopf('fvm:swept_flux')
+         if(FVM_TIMERS) call t_stopf('fvm:swept_flux')
       end do
     end do
     !
@@ -193,7 +194,7 @@ subroutine run_consistent_se_cslam(elem,fvm,hybrid,dt_fvm,tl,nets,nete,hvcoord,&
     !
     !
     if (large_Courant_incr) then
-      !call t_startf('fvm:fill_halo_fvm:large_Courant')
+      if(FVM_TIMERS) call t_startf('fvm:fill_halo_fvm:large_Courant')
       !if (kmin_jet<kmin.or.kmax_jet>kmax) then
       !  call endrun('ERROR: kmax_jet must be .le. kmax passed to run_consistent_se_cslam')
       !end if      
@@ -203,19 +204,19 @@ subroutine run_consistent_se_cslam(elem,fvm,hybrid,dt_fvm,tl,nets,nete,hvcoord,&
       kmax_jet_local = min(kmax_jet,kmax)
       klev = kmax_jet-kmin_jet+1
       call fill_halo_fvm(ghostbufQ1,elem,fvm,hybridnew,nets,nete,1,kmin_jet_local,kmax_jet_local,klev,active=ActiveJetThread)
-      !call t_stopf('fvm:fill_halo_fvm:large_Courant')
-      !call t_startf('fvm:large_Courant_number_increment')
+      if(FVM_TIMERS) call t_stopf('fvm:fill_halo_fvm:large_Courant')
+      if(FVM_TIMERS) call t_startf('fvm:large_Courant_number_increment')
       if(ActiveJetThread) then 
         do k=kmin_jet_local,kmax_jet_local !1,nlev
           do ie=nets,nete
             call large_courant_number_increment(fvm(ie),k)
           end do
         end do
       endif
-      !call t_stopf('fvm:large_Courant_number_increment')
+      if(FVM_TIMERS) call t_stopf('fvm:large_Courant_number_increment')
     end if
 
-    !call t_startf('fvm:end_of_reconstruct_subroutine')
+    if(FVM_TIMERS) call t_startf('fvm:end_of_reconstruct_subroutine')
     do k=kmin,kmax
       !
       ! convert to mixing ratio
@@ -251,7 +252,7 @@ subroutine run_consistent_se_cslam(elem,fvm,hybrid,dt_fvm,tl,nets,nete,hvcoord,&
         elem(ie)%sub_elem_mass_flux(:,:,:,k)=0
       end do
     end do
-    !call t_stopf('fvm:end_of_reconstruct_subroutine')
+    if(FVM_TIMERS) call t_stopf('fvm:end_of_reconstruct_subroutine')
     !$OMP END PARALLEL 
     call omp_set_nested(.false.)
   end subroutine run_consistent_se_cslam
@@ -281,7 +282,7 @@ subroutine swept_flux(elem,fvm,ilev,ctracer,irecons_tracer_actual,gsweights,gspt
     REAL(KIND=r8), dimension(2,8) :: x_start, dgam_vec
     REAL(KIND=r8) :: gamma_max, displ_first_guess
 
-    REAL(KIND=r8) :: flux,flux_tracer(ntrac)
+    REAL(KIND=r8) :: flux,flux_tracer(ntrac),w
 
     REAL(KIND=r8), dimension(num_area) :: dp_area
 
@@ -306,7 +307,6 @@ subroutine swept_flux(elem,fvm,ilev,ctracer,irecons_tracer_actual,gsweights,gspt
     !
     ! prepare for air/tracer update
     !
-!    dp = fvm%dp_fvm(1-nhe:nc+nhe,1-nhe:nc+nhe,ilev)
     dp = fvm%dp_fvm(1-nhc:nc+nhc,1-nhc:nc+nhc,ilev)
     fvm%dp_fvm(1:nc,1:nc,ilev) = fvm%dp_fvm(1:nc,1:nc,ilev)*fvm%area_sphere
     do itr=1,ntrac
@@ -538,14 +538,14 @@ subroutine swept_flux(elem,fvm,ilev,ctracer,irecons_tracer_actual,gsweights,gspt
             !
             ! iterate to get flux area
             !
-            !call t_startf('fvm:swept_area:get_gamma')
+            if(FVM_TIMERS) call t_startf('fvm:swept_area:get_gamma')
             do iarea=1,num_area
               dp_area(iarea) = dp(idx(1,iarea,i,j,iside),idx(2,iarea,i,j,iside))
             end do
             call get_flux_segments_area_iterate(x,x_static,dx_static,dx,x_start,dgam_vec,num_seg,num_seg_static,&
                  num_seg_max,num_area,dp_area,flowcase,gamma,mass_flux_se(i,j,iside),0.0_r8,gamma_max,          &
                  gsweights,gspts,ilev)
-            !call t_stopf('fvm:swept_area:get_gamma')
+            if(FVM_TIMERS) call t_stopf('fvm:swept_area:get_gamma')
             !
             ! pack segments for high-order weights computation
             !
@@ -560,27 +560,28 @@ subroutine swept_flux(elem,fvm,ilev,ctracer,irecons_tracer_actual,gsweights,gspt
             !
             ! compute higher-order weights
             !
-            !call t_startf('fvm:swept_area:get_high_order_w')
+            if(FVM_TIMERS) call t_startf('fvm:swept_area:get_high_order_w')
             call get_high_order_weights_over_areas(x,dx,num_seg,num_seg_max,num_area,weights,ngpc,&
                  gsweights, gspts,irecons_tracer)
-            !call t_stopf('fvm:swept_area:get_high_order_w')
+            if(FVM_TIMERS) call t_stopf('fvm:swept_area:get_high_order_w')
             !
             !**************************************************
             !
             ! remap air and tracers
             !
             !**************************************************
             !
-            !call t_startf('fvm:swept_area:remap')
+            if(FVM_TIMERS) call t_startf('fvm:swept_area:remap')
             flux=0.0_r8; flux_tracer=0.0_r8
             do iarea=1,num_area
               if (num_seg(iarea)>0) then
                 ii=idx(1,iarea,i,j,iside); jj=idx(2,iarea,i,j,iside)
                 flux=flux+weights(1,iarea)*dp(ii,jj)
-                do itr=1,ntrac
-                  do iw=1,irecons_tracer_actual
-                    flux_tracer(itr) = flux_tracer(itr)+weights(iw,iarea)*ctracer(iw,ii,jj,itr)
-                  end do
+                do iw=1,irecons_tracer_actual
+                   w = weights(iw,iarea)
+                   do itr=1,ntrac
+                      flux_tracer(itr) = flux_tracer(itr)+w*ctracer(iw,ii,jj,itr)
+                   end do
                 end do
               end if
             end do
@@ -614,7 +615,7 @@ subroutine swept_flux(elem,fvm,ilev,ctracer,irecons_tracer_actual,gsweights,gspt
               fvm%dp_fvm(i-1,j,ilev        ) = fvm%dp_fvm(i-1,j,ilev        )+flux
               fvm%     c(i-1,j,ilev,1:ntrac) = fvm%     c(i-1,j,ilev,1:ntrac)+flux_tracer(1:ntrac)
             end if
-            !call t_stopf('fvm:swept_area:remap')
+            if(FVM_TIMERS) call t_stopf('fvm:swept_area:remap')
           end if
         end do
       end do