How to limit FillPatch to the first subcycle in AmrLevelAdv advance()?

[Afraid31]

Hi everyone,

I am studying the AmrLevelAdv example in AMReX and looking at the advance() function. In particular, I want to understand how to control ghost cell filling during AMR subcycling.

Currently, the code looks like this:

for (int k = 0; k < NUM_STATE_TYPE; k++) {
    state[k].allocOldData();
    state[k].swapTimeLevels(dt);
}

MultiFab& S_new = get_new_data(Phi_Type);

// State with ghost cells
MultiFab Sborder(grids, dmap, NUM_STATE, NUM_GROW);
FillPatch(*this, Sborder, NUM_GROW, time, Phi_Type, 0, NUM_STATE);

My goals are:

1. I want the FillPatch to happen only during the first subcycling of a recursive advance. For the remaining subcycles, it should not be performed.
2. The ghost cells should be filled only with the latest coarse grid data, without any time interpolation.

My question: How can I modify this code so that FillPatch is limited to the first subcycle and only uses the current coarse grid data?

[WeiqunZhang]

Something like

diff --git a/Tests/Amr/Advection_AmrLevel/Source/AmrLevelAdv.cpp b/Tests/Amr/Advection_AmrLevel/Source/AmrLevelAdv.cpp
index 8ff8bad0b6..62933465cc 100644
--- a/Tests/Amr/Advection_AmrLevel/Source/AmrLevelAdv.cpp
+++ b/Tests/Amr/Advection_AmrLevel/Source/AmrLevelAdv.cpp
@@ -227,7 +227,7 @@ Real
 AmrLevelAdv::advance (Real time,
                       Real dt,
                       int  iteration,
-                      int  /*ncycle*/)
+                      int  ncycle)
 {
     MultiFab& S_mm = get_new_data(Phi_Type);
     Real maxval = S_mm.max(0);
@@ -277,10 +277,16 @@ AmrLevelAdv::advance (Real time,
         }
     }
 
-    // State with ghost cells
-    MultiFab Sborder(grids, dmap, NUM_STATE, NUM_GROW);
-    // We use FillPatcher to do fillpatch here if we can
-    FillPatcherFill(Sborder, 0, NUM_STATE, NUM_GROW, time, Phi_Type, 0);
+    static Vector<std::unique_ptr<MultiFab>> Sborder_all(parent->maxLevel()+1);
+    if (iteration == 1) {
+       Sborder_all[level].reset(new MultiFab Sborder(grids, dmap, NUM_STATE, NUM_GROW));
+       if (level == 0) {
+           FillPatchSingleLevel(...);
+       } else {
+           FillPatchTwoLevels(...);
+       }
+    }
+    MultiFab& Sborder = *Sborder_all[level];
 
     // MF to hold the mac velocity
     MultiFab Umac[BL_SPACEDIM];
@@ -396,6 +402,10 @@ AmrLevelAdv::advance (Real time,
     }
 #endif
 
+    if (iteration == ncycle) {
+        Sborder_all[level].reset();
+    }
+
     return dt;
 }

Note that you are using an old version of amrex (or maybe you are using amrex-tutorials?). My diff is against the latest amrex.

[Afraid31]

Thank you very much for your reply and for pointing out the difference with the latest AMReX version!

I was thinking about a simpler approach and would like to check if it makes sense. Could I just do something like:

if ((iteration == 1) && (level > 0))
    FillPatcherFill(Sborder, 0, NUM_STATE, NUM_GROW,
                    getLevel(level - 1).state[Phi_Type].curTime(),
                    phi_Type, 0);
else{......}

My idea is that this would let the fine grid ghost cells use only the latest coarse data, and that this filling would only happen on the first subcycling iteration.

Would this be a valid way to ensure the ghost cells are initialized correctly, or could it break consistency later in the subcycling?