Merge pull request #27 from IvanMary69/main

FastC: HPC_layer simplifiee

.gitignore

@@ -2,6 +2,7 @@
 *~
 build
 __pycache__
+*.pyc
 *.egg-info
 core.*
 setup.cfg
@@ -20,3 +21,5 @@ Data*
 ValidData*
 *.cgns
 *.dat
+*.swo
+*.swp

Fast/FastC/FastC/HPC_LAYER/INDICE_RANGE.for

@@ -4,17 +4,13 @@
 #endif
 #endif
            call indice_boucle_ssdom(ndo, extended_range,
-     &                              ibloc , jbloc , kbloc,
      &                              icache, jcache, kcache,
-     &                              param_int(KFLUDOM),
-     &                              topo_s, ithread_sock,thread_pos_tmp,
+     &                              topo_s, thread_pos,
      &                              size_cache,
-     &                              synchro_receive_sock,
      &                              synchro_receive_th  ,
-     &                              synchro_send_sock,
      &                              synchro_send_th  ,
      &                              param_int(NIJK), param_int(IJKV),
-     &                              ind_dm_zone, ind_dm_socket,
+     &                              ind_dm_zone,
      &                              ind_dm_omp, ijkv_sdm,
      &                              ind_sdm , ind_coe,
      &                              ind_grad, ind_rhs, 
@@ -23,7 +19,7 @@
 
 #if CHECK_BLOCK > 0
        if(ithread.eq.param_int( IO_THREAD).and.nitrun.eq.0)then
-          if(ibloc*jbloc*kbloc.le.1) then
+          !if(ibloc*jbloc*kbloc.le.1) then
            write(*,'(a,6i4)')'sdm =',ind_sdm
            write(*,'(a,6i4)')'grad=',ind_grad
            write(*,'(a,6i4)')'coe =',ind_coe
@@ -32,7 +28,7 @@
            write(*,'(a,6i4)')'ssa =',ind_ssa
            write(*,'(a,6i4)')'hrr =',ind_hrr
            write(*,'(a,6i4)')'gcb =',ind_gcb
-          endif
+          !endif
        endif
 #endif
 

Fast/FastC/FastC/HPC_LAYER/LOC_VAR_DECLARATION.for

@@ -2,18 +2,15 @@
 
       character*7 omp_init,omp_wait,omp_go,omp_wait_lu
 
-      INTEGER_E i, icache,jcache,kcache,lmin,
+      INTEGER_E i,l,  icache,jcache,kcache,lmin,
      & size_max, size_loc,thread_parsock,
      & thread_parsock_actif,extended_range,
-     & lok_shap_sock(4),thread_topology(3),
+     & thread_topology(3),
      & lok_shap(4), size_cache(3),
-     & socket_pos(3), synchro_receive_sock(3),synchro_send_sock(3),
-     & synchro_receive_th(3),synchro_send_th(3),ipt_lok_sock,
-     & ithread_sock,ipt_lok,size_max_sock,neq_lok,taille,
-     & ijkv_thread(3),kGbloc,jGbloc,iGbloc,ip,jp,kp,lth,
-     & ibloc,jbloc,kbloc,ijkvloc(3),skip(3),shift(3),test(3),lwait,lgo,
-     & size_thread(3),thread_pos(3),thread_pos_tmp(3),sens(3),
-     & size_target(3), cache(3)
+     & synchro_receive_th(3),synchro_send_th(3),
+     & ipt_lok,size_max_sock,neq_lok,taille,
+     & ijkvloc(3),skip(3),shift(3),test(3),lwait,lgo,
+     & size_thread(3),thread_pos(3),sens(3), size_target(3)
 
       INTEGER_E ind_coe(6),ind_grad(6),ind_sdm(6),ind_rhs(6),ind_mjr(6),
      & ind_ssa(6), ind_hrr(6), ind_gcb(6)

Fast/FastC/FastC/HPC_LAYER/SYNCHRO_GO.for

@@ -1,14 +1,9 @@
               call synchro_omp_scater(param_int, ithread,
-     &                          lth, sens,lgo,lwait,Nbre_socket,
-     &                          Nbre_thread_actif,thread_parsock,
-     &                          lok_shap_sock, lok_shap,neq_lok,
-     &                          socket , socket_topology, socket_pos,
-     &                          ithread, thread_topology,thread_pos_tmp,
-     &                          synchro_receive_sock, synchro_send_sock,
-     &                          synchro_receive_th  , synchro_send_th,
-     &                          ibloc , jbloc , kbloc , ijkv_thread,
+     &                          Nbre_thread_actif,
+     &                          lok_shap, neq_lok,
+     &                          ithread, thread_topology, thread_pos,
+     &                          synchro_receive_th, synchro_send_th,
      &                          icache, jcache, kcache, ijkv_sdm,
      &                          size_cache,
      &                          ind_dm_omp,
-     &                          lok(1),lok(ipt_lok_sock),
-     &                          lok(ipt_lok), omp_go )
+     &                          lok(ipt_lok),  omp_go )

Fast/FastC/FastC/HPC_LAYER/SYNCHRO_WAIT.for

@@ -1,14 +1,9 @@
               call synchro_omp_scater(param_int, ithread,
-     &                          lth, sens,lgo,lwait,Nbre_socket,
-     &                          Nbre_thread_actif,thread_parsock,
-     &                          lok_shap_sock, lok_shap,neq_lok,
-     &                          socket , socket_topology, socket_pos,
-     &                          ithread, thread_topology,thread_pos_tmp,
-     &                          synchro_receive_sock, synchro_send_sock,
-     &                          synchro_receive_th  , synchro_send_th,
-     &                          ibloc , jbloc , kbloc , ijkv_thread,
+     &                          Nbre_thread_actif,
+     &                          lok_shap, neq_lok,
+     &                          ithread, thread_topology, thread_pos,
+     &                          synchro_receive_th, synchro_send_th,
      &                          icache, jcache, kcache, ijkv_sdm,
      &                          size_cache,
      &                          ind_dm_omp,
-     &                          lok(1),lok(ipt_lok_sock),
      &                          lok(ipt_lok), omp_wait )

Fast/FastC/FastC/HPC_LAYER/WORK_DISTRIBUTION_BEGIN.for

@@ -11,9 +11,51 @@
       omp_wait_lu   = 'wait_lu'
       lerr          =.false.
 
-      thread_parsock =  Nbre_thread_actif/Nbre_socket
-#include   "../FastC/FastC/HPC_LAYER/topo_cache.for"
-#include   "../FastC/FastC/HPC_LAYER/loopBloc_begin.for"
-#include   "../FastC/FastC/HPC_LAYER/loop_scater.for"
+      thread_topology(1) =topo_s(1)
+      thread_topology(2) =topo_s(2)
+      thread_topology(3) =topo_s(3)
+
+      !cible taille cache bloc applique a la souszone omp
+      size_cache(1)       = param_int(CACHEBLCKI)
+      size_cache(2)       = param_int(CACHEBLCKJ)
+      size_cache(3)       = param_int(CACHEBLCKK)
+
+      do i=1,3
+        if(thread_topology(i).ne.1) then
+          l            =(ind_dm_omp(2*i)-ind_dm_omp(2*i-1)+1)/2
+          size_cache(i)=min( l, size_cache(i) )
+          size_cache(i)=max( 1, size_cache(i) )
+        endif
+      enddo
+
+      thread_pos(3) = 1 + (ithread-1)/(topo_s(1)*topo_s(2))
+      l             = ithread -(thread_pos(3)-1)*topo_s(1)*topo_s(2)
+      thread_pos(2) = 1 + (l-1)/topo_s(1)
+      thread_pos(1) = l - (thread_pos(2)-1)*topo_s(1)
+      !on determine synchro_thrread + nbre sous-domaine )cache !bloc)
+
+#if defined(__INTEL_COMPILER)
+#if not defined(__INTEL_LLVM_COMPILER)
+!DIR$ ATTRIBUTES FORCEINLINE :: crsdm_scater
+#endif
+#endif
+      !in : ind_dm_zone: taille zone, ind_dm_omp: sousdomaine omp
+      !out: loop cache bloc (ijkv_sdm), synchro thread
+       call crsdm_scater( ndo, topo_s, size_cache,
+     &                    synchro_receive_th, synchro_send_th,
+     &                    ind_dm_zone   , ind_dm_omp, ijkv_sdm )
+
+        if(ithread.eq.param_int( IO_THREAD).and.nitrun.eq.0)then
+           write(*,'(a,3i4)')'thread_pos =',thread_pos
+           !write(*,'(a,3i4)')'IJKV socket=',ijkv_thread
+           write(*,'(a,3i4)')'IJKV thread=',ijkv_sdm
+           write(*,'(a,9i4)')'topo thread',thread_topology,ind_dm_omp
+           write(*,'(a,9i4)')'synchro_receive_th',synchro_receive_th
+           write(*,'(a,9i4)')'synchro_send_th',synchro_send_th
+       endif
+
+ccc#include   "../FastC/FastC/HPC_LAYER/topo_cache.for"
+ccc#include   "../FastC/FastC/HPC_LAYER/loopBloc_begin.for"
+ccc#include   "../FastC/FastC/HPC_LAYER/loop_scater.for"
 #include   "../FastC/FastC/HPC_LAYER/verif_loksize.for"
 

Fast/FastC/FastC/HPC_LAYER/WORK_DISTRIBUTION_END.for

@@ -1,18 +1,18 @@
-      enddo
-      enddo
-      enddo  !boucle patern bloc
+c      enddo
+c      enddo
+c      enddo  !boucle patern bloc
 
 
- 9999 continue
+c 9999 continue
 
-#if CHECK_SPLIT > 0
-#include "../FastC/FastC/HPC_LAYER/check_split1.for"
-#endif
+c#if CHECK_SPLIT > 0
+c#include "../FastC/FastC/HPC_LAYER/check_split1.for"
+c#endif
 
-      enddo
-      enddo
-      enddo  !boucle bloc_thread
+c      enddo
+c      enddo
+c      enddo  !boucle bloc_thread
 
-#if CHECK_SPLIT > 0
-#include "../FastC/FastC/HPC_LAYER/check_split2.for"
-#endif
+c#if CHECK_SPLIT > 0
+c#include "../FastC/FastC/HPC_LAYER/check_split2.for"
+c#endif

Fast/FastC/FastC/HPC_LAYER/crsdm_scater.for

@@ -3,9 +3,9 @@ c     $Date: 2011-12-07 15:30:38 +0100 (mer 07 déc 2011) $
 c     $Revision: 56 $
 c     $Author: MarcTerracol $
 c***********************************************************************
-      subroutine crsdm_scater(ndo,ith,jth,kth, topo_th,
+      subroutine crsdm_scater(ndo, topo_th,
      &                 size_cache, synchro_receive, synchro_send,
-     &                 ind_dm_glob, ind_dm, ijkv_sdm )
+     &                 ind_dm_zone, ind_dm, ijkv_sdm )
 c***********************************************************************
 c_P                          O N E R A
 c     ACT
@@ -20,9 +20,9 @@ c=======================================================================
 
       include "omp_lib.h"
 
-      INTEGER_E ndo, ith,jth,kth, topo_th(3),
+      INTEGER_E ndo, topo_th(3),
      & size_cache(3), synchro_receive(3),synchro_send(3),
-     & ind_dm_glob(6), ind_dm(6), ijkv_sdm(3)
+     & ind_dm_zone(6), ind_dm(6), ijkv_sdm(3)
 
 C Var loc
       INTEGER_E l,i,j,k,ivloc,iverbs
@@ -34,9 +34,9 @@ C Var loc
 
          synchro_receive(i) = 0
          synchro_send   (i) = 0
-         if( ind_dm_glob(2*i  ).ne. ind_dm(2*i).and.topo_th(i).ne.1) 
+         if( ind_dm_zone(2*i  ).ne. ind_dm(2*i).and.topo_th(i).ne.1) 
      &          synchro_receive(i)= 1
-         if( ind_dm_glob(2*i-1).ne. ind_dm(2*i-1).and.topo_th(i).ne.1) 
+         if( ind_dm_zone(2*i-1).ne. ind_dm(2*i-1).and.topo_th(i).ne.1) 
      &       synchro_send(i)   = 1
 
          ivloc       = ind_dm(2*i)- ind_dm(2*i-1) + 1

Fast/FastC/FastC/HPC_LAYER/distributeThreads.cpp

@@ -204,8 +204,8 @@ void K_FASTC::distributeThreads_c( E_Int**& param_int, E_Float**& param_real, E_
    FldArrayI newtab_nozone(mxzone);   E_Int* ipt_nozone_new  = newtab_nozone.begin();
    FldArrayI newtab_nosszone(mxzone); E_Int* ipt_nosszone_new= newtab_nosszone.begin();
 
-   FldArrayF    tab_HPC_CUPS(mxzone); E_Float* ipt_HPC_CUPS    = tab_HPC_CUPS.begin();
-   FldArrayF newtab_HPC_CUPS(mxzone); E_Float* ipt_HPC_CUPS_new= newtab_HPC_CUPS.begin();
+   FldArrayF    tab_hpccups(mxzone); E_Float* ipt_hpccups    = tab_hpccups.begin();
+   FldArrayF newtab_hpccups(mxzone); E_Float* ipt_hpccups_new= newtab_hpccups.begin();
 
    E_Int c = 0;
    E_Int ndimt=0;
@@ -228,8 +228,8 @@ void K_FASTC::distributeThreads_c( E_Int**& param_int, E_Float**& param_real, E_
             E_Int* nozone   = ipt_nozone   +   c;
             E_Int* nosszone = ipt_nosszone +   c;
 
-            ipt_HPC_CUPS[c] = param_real[nd][HPC_CUPS];
-            if (ipt_HPC_CUPS[c] > CupsMax) CupsMax =  ipt_HPC_CUPS[c];
+            ipt_hpccups[c] = param_real[nd][HPC_CUPS];
+            if (ipt_hpccups[c] > CupsMax) CupsMax =  ipt_hpccups[c];
 
 
             ijk_start[0]= ipt_ind_dm[nd][0+shift];
@@ -274,31 +274,28 @@ void K_FASTC::distributeThreads_c( E_Int**& param_int, E_Float**& param_real, E_
        {  
             E_Int* ndimdx = ipt_ndimdx  + c1;
 
-            E_Float* hpc_cups = ipt_HPC_CUPS+   c1;
+            E_Float* hpc_cups = ipt_hpccups+   c1;
 
             //if(ndimdx[0] > ndimdx_max) { ndimdx_max= ndimdx[0]; c_tg = c1;}
             if( float(ndimdx[0])/hpc_cups[0] > ndimdx_max) { ndimdx_max= float(ndimdx[0])/hpc_cups[0]; c_tg = c1;}
        }// loop recherche plus grosse zone
-
-
+
        E_Int* nijk     = ipt_nijk     + 3*c_tg;
        E_Int* ijk_start= ipt_nijk     + 3*c_tg +3*mxzone;
        E_Int* ndimdx   = ipt_ndimdx   +   c_tg;
 
        E_Int* nozone   = ipt_nozone   +   c_tg;
        E_Int* nosszone = ipt_nosszone +   c_tg;
 
-       E_Float* hpc_cups = ipt_HPC_CUPS+   c_tg;
-
-       //printf("zone %d %d %d %d %d \n", c, ndimdx_max, c_tg, nozone[0],  nosszone[0]);
+       E_Float* hpc_cups = ipt_hpccups+   c_tg;
 
        E_Int* nijk_new     = ipt_nijk_new     + 3*c;
        E_Int* ijk_startnew = ipt_nijk_new     + 3*c +3*mxzone;
        E_Int* ndimdx_new   = ipt_ndimdx_new   +   c;
        E_Int* nozone_new   = ipt_nozone_new   +   c;
        E_Int* nosszone_new = ipt_nosszone_new +   c;
 
-       E_Float* hpc_cups_new = ipt_HPC_CUPS_new +   c;
+       E_Float* hpc_cups_new = ipt_hpccups_new +   c;
 
        nijk_new[0]     = nijk[0];
        nijk_new[1]     = nijk[1];
@@ -337,7 +334,7 @@ void K_FASTC::distributeThreads_c( E_Int**& param_int, E_Float**& param_real, E_
         E_Int* nozone_new   = ipt_nozone_new   +   c;
         //E_Int  No_zone      = nozone_new[0];
 
-        E_Float* hpc_cups = ipt_HPC_CUPS_new +   c;
+        E_Float* hpc_cups = ipt_hpccups_new +   c;
 
         ind_dm[0]= ijk_start[0];
         ind_dm[2]= ijk_start[1];
@@ -353,8 +350,6 @@ void K_FASTC::distributeThreads_c( E_Int**& param_int, E_Float**& param_real, E_
 	E_Int size_k = 0;
         if(nijk[2] != 1) size_k = (nijk[2]+1)*nijk[0]*nijk[1];
 
-        //poids = cupsmoy/ipt_HPC_CUPS[c];
-        //poids = cupsmoy/ipt_HPC_CUPS[No_zone];
         poids = cupsmoy/hpc_cups[0];
         //if(nstep==1) printf("poids %f %f %f %d %d \n", poids, cupsmoy,ipt_HPC_CUPS[No_zone], c, No_zone);
 
@@ -411,7 +406,7 @@ void K_FASTC::distributeThreads_c( E_Int**& param_int, E_Float**& param_real, E_
         E_Int  No_sszone    = nosszone_new[0];
         //E_Int* ndimdx       = ipt_ndimdx_new   +   c;
 
-        E_Float* hpc_cups = ipt_HPC_CUPS_new +   c;
+        E_Float* hpc_cups = ipt_hpccups_new +   c;
 
         E_Int* ipt_nidom_loc = ipt_ind_dm[No_zone] + param_int[No_zone][ MXSSDOM_LU ]*6*nssiter + nssiter;     //nidom_loc(nssiter)
         E_Int nb_subzone     = ipt_nidom_loc [nstep-1];   
@@ -433,8 +428,8 @@ void K_FASTC::distributeThreads_c( E_Int**& param_int, E_Float**& param_real, E_
         if(nstep==1) { for (E_Int socket = 0; socket < NBR_SOCKET; socket++) { numa_socket[socket] = 0;} }
 
         poids = cupsmoy/hpc_cups[0];
-        //poids = cupsmoy/ipt_HPC_CUPS[c];
-        //poids = cupsmoy/ipt_HPC_CUPS[No_zone];
+        //poids = cupsmoy/ipt_hpccups[c];
+        //poids = cupsmoy/ipt_hpccups[No_zone];
         //printf("Nozone %d %d %f %d \n",c, No_zone, hpc_cups[0], nstep); 
 
         E_Int size_c = nijk[0]*nijk[1]*nijk[2]*poids;
@@ -639,9 +634,9 @@ void K_FASTC::distributeThreads_c( E_Int**& param_int, E_Float**& param_real, E_
                   }
                 */
 
-                //poids = cupsmoy/ipt_HPC_CUPS[No_zone];
+                //poids = cupsmoy/ipt_hpccups[No_zone];
                 poids = cupsmoy/hpc_cups[0];
-                //poids = cupsmoy/ipt_HPC_CUPS[c];
+                //poids = cupsmoy/ipt_hpccups[c];
 
                 E_Int res = (cc*dim_i[0]*dim_j[0]*dim_k[0] + ci*dim_i[0]*dim_j[0]*dim_k[0] + cj*dim_j[0]*dim_i[0]*dim_k[0] + ck*dim_k[0]*dim_i[0]*dim_j[0] )*poids - cells_tg_loc;
                 E_Float sign = 1.;
@@ -894,9 +889,9 @@ void K_FASTC::distributeThreads_c( E_Int**& param_int, E_Float**& param_real, E_
               { list_affected[th]=-1; 
                 ipt_omp[PtTask + 2 + th] = -2;} //Thread inactif
 
-            //poids = cupsmoy/ipt_HPC_CUPS[No_zone];
+            //poids = cupsmoy/ipt_hpccups[No_zone];
             poids = cupsmoy/hpc_cups[0];
-            //poids = cupsmoy/ipt_HPC_CUPS[c];
+            //poids = cupsmoy/ipt_hpccups[c];
             //Recherche thread disponible
             E_Int th = 0;
             for (E_Int k = 0; k < topo_lu[2]; k++){