Merge pull request #92 from IvanMary69/dev

Fast: detection interp O4 pour prevoir 2 passes transfert

Author: IvanMary69

Fast/Fast/Fast/PyTree.py

@@ -214,7 +214,7 @@ def warmup(t, tc=None, graph=None, infos_ale=None, Adjoint=False, tmy=None, list
     # Compactage arbre transfert et mise a jour FastC.HOOK
     #
     if tc is not None:
-        X.miseAPlatDonorTree__(zones, tc, graph=graph, list_graph=list_graph)
+        X.miseAPlatDonorTree__( t, tc, graph=graph, list_graph=list_graph)
 
         FastC.HOOK['param_int_tc'] = Internal.getNodeFromName1( tc, 'Parameter_int' )[1]
         param_real_tc = Internal.getNodeFromName1( tc, 'Parameter_real')

Fast/FastC/FastC/Com/flux_conservatif_5eq_Recep.cpp

@@ -3,7 +3,8 @@
 
          E_Int pt_bcs   = param_int[ NoR ][PT_BC];
          E_Int nb_bc    = param_int[ NoR ][pt_bcs];
-         E_Int idir     = param_int[ NoR ][pt_bcs + 1 + nobcR +BC_IDIR];
+         E_Int pt_bc    = param_int[ NoR ][pt_bcs + 1 + nobcR];
+         E_Int idir     = param_int[ NoR ][pt_bc +BC_IDIR];
          E_Int adrFlu   = param_int[ NoR ][pt_bcs + 1 + nobcR + nb_bc] +3;
          E_Float* fluxR = param_real[ NoR ] + adrFlu;
          E_Float* ratio = fluxR-3;
@@ -16,7 +17,7 @@
 
          E_Int shift   = nvars_loc * (nbRcvPts + shift_fluR);
          shift_fluR   += sizefluR;
-         //printf("No raccord Recep %d , nbRcvPts: %d ,  sizefluR: %d, shift_fluR: %d \n", nobcR, nbRcvPts, sizefluR, shift_fluR);
+         //printf("NoR: %d , No raccord Recep %d , sizefluD: %d ,  sizefluR: %d, shift_fluR: %d idir: %d ratio: %d %d %d \n", NoR, nobcR, sizefluD, sizefluR, shift_fluR, idir, ratio_i,ratio_j,ratio_k);
          #pragma omp for nowait
          for (E_Int l = 0; l < sizefluR*nvars_loc; l++)
            {

Fast/FastC/FastC/PyTree.py

@@ -1179,7 +1179,7 @@ def _buildOwnData(t, Padding):
     size_omp = MX_OMP_SIZE_INT
 
     nitCyclLBM = 2**(maxlevel-1)
-    dtdim = 12 + nitCyclLBM
+    dtdim = 13 + nitCyclLBM
     #print('ncycl_LBM', nitCyclLBM, maxlevel)
 
     datap = numpy.zeros((dtdim+ size_omp), Internal.E_NpyInt)
@@ -1195,13 +1195,14 @@ def _buildOwnData(t, Padding):
     datap[9] = maxlevel
     datap[10]= LBMCycleIteration # No it du cycle integration temporelle LBM
     datap[11]= dtdim             # shift pour acceder au info OMP
+    datap[12]= 1                 # nbre de pass pour transfert
 
     # level max pour chaque it du cycle LBM
     for it in range(nitCyclLBM):
         for level in range(maxlevel,0,-1):
             it_tg = 2**(level-1)
             if it%it_tg == 0:
-                datap[12+it] = level
+                datap[13+it] = level
                 #print('Nblevel',datap[12+it],'itCycl=',it )
                 break
 
@@ -2656,10 +2657,10 @@ def switchPointersLBM__(zones, dtloc):
         level = param_int[VSHARE.LEVEL]
         maxlevel    = dtloc[ 9]
         it_cycl_lbm = dtloc[10]
-        level_tg = dtloc[12 +it_cycl_lbm]
+        level_tg = dtloc[13 +it_cycl_lbm]
         max_it        = 2**( maxlevel-1)
         level_next_it =  maxlevel;
-        if it_cycl_lbm != max_it -1 : level_next_it = dtloc[12 +it_cycl_lbm +1]
+        if it_cycl_lbm != max_it -1 : level_next_it = dtloc[13 +it_cycl_lbm +1]
 
         #if level==1 or (level >=2 and level <= level_next_it) :
         if level <= level_next_it :

Fast/FastS/FastS/Compute/gsdr3.cpp

@@ -363,7 +363,7 @@ E_Int nitcfg_stk = nitcfg;
 if(lexit_lu ==0 && layer_mode>=1)
 {   
   //remplissage ghost transfert
-  for (E_Int ipass = 0; ipass < 1; ipass++)
+  for (E_Int ipass = 0; ipass < iptdtloc[12]; ipass++)
   {
   #include "FastS/Compute/transfert_multiblock.cpp"
   }

Fast/FastS/FastS/Compute/lhs.cpp

@@ -152,6 +152,8 @@ if( kimpli == 1  && param_int[0][LU_MATCH]==1 && param_int_tc != NULL)
 
                     if(param_int[nd][LU_MATCH]==1 || param_int[nd][NB_RELAX]>1) iptdrodm_out = ipt_ssortmp_shift;
 
+                    //printf("lu %d %d %d %d %d %d \n",ipt_shift_lu[0], ipt_shift_lu[1], ipt_shift_lu[2], ipt_shift_lu[3], ipt_shift_lu[4], ipt_shift_lu[5]);
+                    
 		    invlu_(nd                     , nitcfg                  , nitrun                ,
 			   param_int[nd]          , param_real[nd]          ,
 			   ipt_shift_lu           , ipt_ind_dm_thread       , mjrnewton             ,

Fast/FastS/FastS/Mpi.py

@@ -892,7 +892,11 @@ def warmup(t, tc, graph=None, infos_ale=None, Adjoint=False, tmy=None, list_grap
             # print(h0, -1, nbpts_linelets)
             _createTBLESA2(t, tc, h0=h0, hn=-1, nbpts_linelets=nbpts_linelets)
 
-        X.miseAPlatDonorTree__(zones, tc, graph=graph,list_graph=list_graph, nbpts_linelets=nbpts_linelets)
+        X.miseAPlatDonorTree__(t, tc, graph=graph,list_graph=list_graph, nbpts_linelets=nbpts_linelets)
+
+        npass_transfer = dtloc[12]
+        npass_transfer = Cmpi.allreduce(npass_transfer)
+        if npass_transfer > Cmpi.size: dtloc[12]=2
 
         FastC.HOOK['param_int_tc'] = Internal.getNodeFromName1( tc, 'Parameter_int')[1]
         param_real_tc               = Internal.getNodeFromName1( tc, 'Parameter_real')

Fast/FastS/FastS/PyTree.py

@@ -495,7 +495,7 @@ def warmup(t, tc, graph=None, infos_ale=None, Adjoint=False, tmy=None, list_grap
             _createTBLESA(tc, h0=h0, hn=-1, nbpts_linelets=nbpts_linelets)
             _createTBLESA2(t, tc, h0=h0, hn=-1, nbpts_linelets=nbpts_linelets)
 
-        X.miseAPlatDonorTree__(zones, tc, graph=graph, list_graph=list_graph, nbpts_linelets=nbpts_linelets)
+        X.miseAPlatDonorTree__(t, tc, graph=graph, list_graph=list_graph, nbpts_linelets=nbpts_linelets)
 
         FastC.HOOK['param_int_tc'] = Internal.getNodeFromName1( tc, 'Parameter_int' )[1]
         param_real_tc = Internal.getNodeFromName1(tc, 'Parameter_real')