GPU Package: Switching to parallel GPU initialization / JIT compilation.

lib/gpu/lal_beck_ext.cpp

@@ -76,7 +76,7 @@ int beck_gpu_init(const int ntypes, double **cutsq, double **aa,
                         special_lj, inum, nall, max_nbors, maxspecial,
                         cell_size, gpu_split, screen);
 
-    BLMF.device->gpu_barrier();
+    BLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_born_coul_long_cs_ext.cpp

@@ -84,7 +84,7 @@ int bornclcs_gpu_init(const int ntypes, double **cutsq, double **host_rhoinv,
                             gpu_split, screen, host_cut_ljsq, host_cut_coulsq,
                             host_special_coul, qqrd2e, g_ewald);
 
-    BCLCSMF.device->gpu_barrier();
+    BCLCSMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_born_coul_long_ext.cpp

@@ -84,7 +84,7 @@ int borncl_gpu_init(const int ntypes, double **cutsq, double **host_rhoinv,
                             gpu_split, screen, host_cut_ljsq, host_cut_coulsq,
                             host_special_coul, qqrd2e, g_ewald);
 
-    BORNCLMF.device->gpu_barrier();
+    BORNCLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_born_coul_wolf_cs_ext.cpp

@@ -86,7 +86,7 @@ int borncwcs_gpu_init(const int ntypes, double **cutsq, double **host_rhoinv,
                             host_cut_coulsq, host_special_coul, qqrd2e,
                             alf, e_shift, f_shift);
 
-    BornCWCST.device->gpu_barrier();
+    BornCWCST.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_born_coul_wolf_ext.cpp

@@ -86,7 +86,7 @@ int borncw_gpu_init(const int ntypes, double **cutsq, double **host_rhoinv,
                             host_cut_coulsq, host_special_coul, qqrd2e,
                             alf, e_shift, f_shift);
 
-    BORNCWMF.device->gpu_barrier();
+    BORNCWMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_born_ext.cpp

@@ -80,7 +80,7 @@ int born_gpu_init(const int ntypes, double **cutsq, double **host_rhoinv,
                           offset, special_lj, inum, nall, max_nbors,
                           maxspecial, cell_size, gpu_split, screen);
 
-    BORNMF.device->gpu_barrier();
+    BORNMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }
@@ -114,7 +114,7 @@ void born_gpu_reinit(const int ntypes, double **host_rhoinv,
       BORNMF.reinit(ntypes, host_rhoinv, host_born1, host_born2,
                     host_born3, host_a, host_c, host_d, offset);
 
-    BORNMF.device->gpu_barrier();
+    BORNMF.device->serialize_init();
   }
 }
 

lib/gpu/lal_buck_coul_ext.cpp

@@ -83,7 +83,7 @@ int buckc_gpu_init(const int ntypes, double **cutsq, double **host_rhoinv,
                        host_cut_ljsq, host_cut_coulsq,
                        host_special_coul, qqrd2e);
 
-    BUCKCMF.device->gpu_barrier();
+    BUCKCMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_buck_coul_long_ext.cpp

@@ -82,7 +82,7 @@ int buckcl_gpu_init(const int ntypes, double **cutsq, double **host_rhoinv,
                         maxspecial, cell_size, gpu_split, screen, host_cut_ljsq,
                         host_cut_coulsq, host_special_coul, qqrd2e, g_ewald);
 
-    BUCKCLMF.device->gpu_barrier();
+    BUCKCLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_buck_ext.cpp

@@ -77,7 +77,7 @@ int buck_gpu_init(const int ntypes, double **cutsq, double **host_rhoinv,
                        host_a, host_c, offset, special_lj, inum, nall, max_nbors,
                        maxspecial, cell_size, gpu_split, screen);
 
-    BUCKMF.device->gpu_barrier();
+    BUCKMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }
@@ -110,7 +110,7 @@ void buck_gpu_reinit(const int ntypes, double **cutsq, double **host_rhoinv,
       BUCKMF.reinit(ntypes, cutsq, host_rhoinv, host_buck1, host_buck2,
                     host_a, host_c, offset);
 
-    BUCKMF.device->gpu_barrier();
+    BUCKMF.device->serialize_init();
   }
 }
 

lib/gpu/lal_charmm_ext.cpp

@@ -88,7 +88,7 @@ int crm_gpu_init(const int ntypes, double cut_bothsq, double **host_lj1,
                           qqrd2e, cut_lj_innersq, cut_coul_innersq, denom_lj,
                           denom_coul, epsilon, sigma, mix_arithmetic);
 
-    CRMMF.device->gpu_barrier();
+    CRMMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_charmm_long_ext.cpp

@@ -86,7 +86,7 @@ int crml_gpu_init(const int ntypes, double cut_bothsq, double **host_lj1,
                           qqrd2e, g_ewald,  cut_lj_innersq, denom_lj, epsilon,
                           sigma, mix_arithmetic);
 
-    CRMLMF.device->gpu_barrier();
+    CRMLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_colloid_ext.cpp

@@ -83,7 +83,7 @@ int colloid_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
                           inum, nall, max_nbors, maxspecial,
                           cell_size, gpu_split, screen);
 
-    COLLMF.device->gpu_barrier();
+    COLLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_coul_debye_ext.cpp

@@ -74,7 +74,7 @@ int cdebye_gpu_init(const int ntypes, double **host_scale, double **cutsq,
       init_ok=CDEMF.init(ntypes, host_scale, cutsq, host_special_coul, inum, nall, max_nbors,
                          maxspecial, cell_size, gpu_split, screen, qqrd2e, kappa);
 
-    CDEMF.device->gpu_barrier();
+    CDEMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }
@@ -103,7 +103,7 @@ void cdebye_gpu_reinit(const int ntypes, double **host_scale) {
     if (gpu_rank==i && world_me!=0)
       CDEMF.reinit(ntypes, host_scale);
 
-    CDEMF.device->gpu_barrier();
+    CDEMF.device->serialize_init();
   }
 }
 

lib/gpu/lal_coul_dsf_ext.cpp

@@ -77,7 +77,7 @@ int cdsf_gpu_init(const int ntypes, const int inum, const int nall,
                         gpu_split, screen, host_cut_coulsq, host_special_coul,
                         qqrd2e, e_shift, f_shift, alpha);
 
-    CDMF.device->gpu_barrier();
+    CDMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_coul_ext.cpp

@@ -74,7 +74,7 @@ int coul_gpu_init(const int ntypes, double **host_scale,
       init_ok=COULMF.init(ntypes, host_scale, cutsq, special_coul, inum, nall, max_nbors,
                           maxspecial, cell_size, gpu_split, screen, qqrd2e);
 
-    COULMF.device->gpu_barrier();
+    COULMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }
@@ -103,7 +103,7 @@ void coul_gpu_reinit(const int ntypes, double **host_scale) {
     if (gpu_rank==i && world_me!=0)
       COULMF.reinit(ntypes, host_scale);
 
-    COULMF.device->gpu_barrier();
+    COULMF.device->serialize_init();
   }
 }
 

lib/gpu/lal_coul_long_cs_ext.cpp

@@ -76,7 +76,7 @@ int clcs_gpu_init(const int ntypes, double **host_scale,
                         cell_size, gpu_split, screen, host_cut_coulsq,
                         host_special_coul, qqrd2e, g_ewald);
 
-    CLCSMF.device->gpu_barrier();
+    CLCSMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }
@@ -105,7 +105,7 @@ void clcs_gpu_reinit(const int ntypes, double **host_scale) {
     if (gpu_rank==i && world_me!=0)
       CLCSMF.reinit(ntypes, host_scale);
 
-    CLCSMF.device->gpu_barrier();
+    CLCSMF.device->serialize_init();
   }
 }
 

lib/gpu/lal_coul_long_ext.cpp

@@ -76,7 +76,7 @@ int cl_gpu_init(const int ntypes, double **host_scale,
                         cell_size, gpu_split, screen, host_cut_coulsq,
                         host_special_coul, qqrd2e, g_ewald);
 
-    CLMF.device->gpu_barrier();
+    CLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }
@@ -105,7 +105,7 @@ void cl_gpu_reinit(const int ntypes, double **host_scale) {
     if (gpu_rank==i && world_me!=0)
       CLMF.reinit(ntypes, host_scale);
 
-    CLMF.device->gpu_barrier();
+    CLMF.device->serialize_init();
   }
 }
 

lib/gpu/lal_device.cpp

@@ -328,7 +328,7 @@ int DeviceT::init_device(MPI_Comm world, MPI_Comm replica, const int ngpu,
   for (int i=0; i<_procs_per_gpu; i++) {
     if (_gpu_rank==i)
       flag=compile_kernels();
-    gpu_barrier();
+    serialize_init();
   }
 
   // check if double precision support is available

lib/gpu/lal_device.h

@@ -217,6 +217,12 @@ class Device {
   inline int gpu_rank() const { return _gpu_rank; }
   /// MPI Barrier for gpu
   inline void gpu_barrier() { MPI_Barrier(_comm_gpu); }
+  /// Serialize GPU initialization and JIT for unsafe platforms
+  inline void serialize_init() {
+    #ifdef LAL_SERIALIZE_INIT
+    gpu_barrier();
+    #endif
+  }
   /// Return the 'mode' for acceleration: GPU_FORCE, GPU_NEIGH or GPU_HYB_NEIGH
   inline int gpu_mode() const { return _gpu_mode; }
   /// Index of first device used by a node

lib/gpu/lal_dipole_lj_ext.cpp

@@ -80,7 +80,7 @@ int dpl_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
                          cell_size, gpu_split, screen, host_cut_ljsq,
                          host_cut_coulsq, host_special_coul, qqrd2e);
 
-    DPLMF.device->gpu_barrier();
+    DPLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_dipole_lj_sf_ext.cpp

@@ -80,7 +80,7 @@ int dplsf_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
                            cell_size, gpu_split, screen, host_cut_ljsq,
                            host_cut_coulsq, host_special_coul, qqrd2e);
 
-    DPLSFMF.device->gpu_barrier();
+    DPLSFMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_dipole_long_lj_ext.cpp

@@ -81,7 +81,7 @@ int dplj_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
                          cell_size, gpu_split, screen, host_cut_ljsq,
                          host_cut_coulsq, host_special_coul, qqrd2e, g_ewald);
 
-    DPLJMF.device->gpu_barrier();
+    DPLJMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_dpd_ext.cpp

@@ -76,7 +76,7 @@ int dpd_gpu_init(const int ntypes, double **cutsq, double **host_a0,
                          host_cut, special_lj, false, inum, nall, max_nbors,
                          maxspecial, cell_size, gpu_split, screen);
 
-    DPDMF.device->gpu_barrier();
+    DPDMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_dpd_tstat_ext.cpp

@@ -76,7 +76,7 @@ int dpd_tstat_gpu_init(const int ntypes, double **cutsq, double **host_a0,
                          host_cut, special_lj, true, inum, nall, 300,
                          maxspecial, cell_size, gpu_split, screen);
 
-    DPDTMF.device->gpu_barrier();
+    DPDTMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_eam_alloy_ext.cpp

@@ -90,7 +90,7 @@ int eam_alloy_gpu_init(const int ntypes, double host_cutforcesq,
                          nz2r, nfrho, nr, nlocal, nall, max_nbors, maxspecial,
                          cell_size, gpu_split, screen);
 
-    EAMALMF.device->gpu_barrier();
+    EAMALMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_eam_ext.cpp

@@ -90,7 +90,7 @@ int eam_gpu_init(const int ntypes, double host_cutforcesq,
                          nz2r, nfrho, nr, nlocal, nall, max_nbors, maxspecial,
                          cell_size, gpu_split, screen);
 
-    EAMMF.device->gpu_barrier();
+    EAMMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_eam_fs_ext.cpp

@@ -90,7 +90,7 @@ int eam_fs_gpu_init(const int ntypes, double host_cutforcesq,
                          nz2r, nfrho, nr, nlocal, nall, max_nbors, maxspecial,
                          cell_size, gpu_split, screen);
 
-    EAMFSMF.device->gpu_barrier();
+    EAMFSMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_gauss_ext.cpp

@@ -76,7 +76,7 @@ int gauss_gpu_init(const int ntypes, double **cutsq, double **host_a,
                         offset, special_lj, inum, nall, max_nbors, maxspecial,
                         cell_size, gpu_split, screen);
 
-    GLMF.device->gpu_barrier();
+    GLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }
@@ -106,7 +106,7 @@ void gauss_gpu_reinit(const int ntypes, double **cutsq, double **host_a,
     if (gpu_rank==i && world_me!=0)
       GLMF.reinit(ntypes, cutsq, host_a, host_b, offset);
 
-    GLMF.device->gpu_barrier();
+    GLMF.device->serialize_init();
   }
 }
 

lib/gpu/lal_gayberne_ext.cpp

@@ -83,7 +83,7 @@ int gb_gpu_init(const int ntypes, const double gamma,
                         host_lj3, host_lj4, offset, special_lj,  inum, nall,
                         max_nbors, maxspecial, cell_size, gpu_split,  screen);
 
-    GBMF.device->gpu_barrier();
+    GBMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_lj96_ext.cpp

@@ -76,7 +76,7 @@ int lj96_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
                           offset, special_lj, inum,  nall, max_nbors, maxspecial,
                           cell_size, gpu_split, screen);
 
-    LJ96MF.device->gpu_barrier();
+    LJ96MF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_lj_class2_long_ext.cpp

@@ -81,7 +81,7 @@ int c2cl_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
                           cell_size, gpu_split, screen, host_cut_ljsq,
                           host_cut_coulsq, host_special_coul, qqrd2e, g_ewald);
 
-    C2CLMF.device->gpu_barrier();
+    C2CLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_lj_coul_debye_ext.cpp

@@ -81,7 +81,7 @@ int ljcd_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
                           cell_size, gpu_split, screen, host_cut_ljsq,
                           host_cut_coulsq, host_special_coul, qqrd2e, kappa);
 
-    LJCDMF.device->gpu_barrier();
+    LJCDMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_lj_coul_ext.cpp

@@ -80,7 +80,7 @@ int ljc_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
                          cell_size, gpu_split, screen, host_cut_ljsq,
                          host_cut_coulsq, host_special_coul, qqrd2e);
 
-    LJCMF.device->gpu_barrier();
+    LJCMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_lj_coul_long_ext.cpp

@@ -81,7 +81,7 @@ int ljcl_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
                           cell_size, gpu_split, screen, host_cut_ljsq,
                           host_cut_coulsq, host_special_coul, qqrd2e, g_ewald);
 
-    LJCLMF.device->gpu_barrier();
+    LJCLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }
@@ -112,7 +112,7 @@ void ljcl_gpu_reinit(const int ntypes, double **cutsq, double **host_lj1,
     if (gpu_rank==i && world_me!=0)
       LJCLMF.reinit(ntypes, cutsq, host_lj1, host_lj2, host_lj3, host_lj4,
                     offset, host_cut_ljsq);
-    LJCLMF.device->gpu_barrier();
+    LJCLMF.device->serialize_init();
   }
 }
 

lib/gpu/lal_lj_coul_msm_ext.cpp

@@ -83,7 +83,7 @@ int ljcm_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
                           cell_size, gpu_split, screen, host_cut_ljsq,
                           host_cut_coulsq, host_special_coul, order, qqrd2e);
 
-    LJCMLMF.device->gpu_barrier();
+    LJCMLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_lj_cubic_ext.cpp

@@ -80,7 +80,7 @@ int ljcb_gpu_init(const int ntypes, double **cutsq, double **cut_inner_sq,
                               special_lj, inum, nall, max_nbors, maxspecial,
                               cell_size, gpu_split, screen);
 
-    LJCubicLMF.device->gpu_barrier();
+    LJCubicLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_lj_dsf_ext.cpp

@@ -84,7 +84,7 @@ int ljd_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
                          host_cut_coulsq, host_special_coul, qqrd2e, e_shift,
                          f_shift, alpha);
 
-    LJDMF.device->gpu_barrier();
+    LJDMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_lj_expand_coul_long_ext.cpp

@@ -81,7 +81,7 @@ int ljecl_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
                           cell_size, gpu_split, screen, host_cut_ljsq,
                           host_cut_coulsq, host_special_coul, qqrd2e, g_ewald);
 
-    LJECLMF.device->gpu_barrier();
+    LJECLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }
@@ -112,7 +112,7 @@ void ljecl_gpu_reinit(const int ntypes, double **cutsq, double **host_lj1,
     if (gpu_rank==i && world_me!=0)
       LJECLMF.reinit(ntypes, cutsq, host_lj1, host_lj2, host_lj3, host_lj4,
                     offset, shift, host_cut_ljsq);
-    LJECLMF.device->gpu_barrier();
+    LJECLMF.device->serialize_init();
   }
 }
 

lib/gpu/lal_lj_expand_ext.cpp

@@ -108,7 +108,7 @@ void lje_gpu_reinit(const int ntypes, double **cutsq, double **host_lj1,
     if (gpu_rank==i && world_me!=0)
       LJEMF.reinit(ntypes, cutsq, host_lj1, host_lj2, host_lj3, host_lj4,
                    offset, shift);
-    LJEMF.device->gpu_barrier();
+    LJEMF.device->serialize_init();
   }
 }
 

lib/gpu/lal_lj_ext.cpp

@@ -76,7 +76,7 @@ int ljl_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
                          offset, special_lj, inum, nall, max_nbors, maxspecial,
                          cell_size, gpu_split, screen);
 
-    LJLMF.device->gpu_barrier();
+    LJLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }
@@ -105,7 +105,7 @@ void ljl_gpu_reinit(const int ntypes, double **cutsq, double **host_lj1,
   for (int i=0; i<procs_per_gpu; i++) {
     if (gpu_rank==i && world_me!=0)
       LJLMF.reinit(ntypes, cutsq, host_lj1, host_lj2, host_lj3, host_lj4, offset);
-    LJLMF.device->gpu_barrier();
+    LJLMF.device->serialize_init();
   }
 }
 

lib/gpu/lal_lj_gromacs_ext.cpp

@@ -81,7 +81,7 @@ int ljgrm_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
                            gpu_split, screen, host_ljsw1, host_ljsw2, host_ljsw3,
                            host_ljsw4, host_ljsw5, cut_inner, cut_inner_sq);
 
-    LJGRMMF.device->gpu_barrier();
+    LJGRMMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_lj_smooth_ext.cpp

@@ -80,7 +80,7 @@ int ljsmt_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
                          cell_size, gpu_split, screen, host_ljsw0, host_ljsw1, host_ljsw2, host_ljsw3,
                          host_ljsw4, cut_inner, cut_inner_sq);
 
-    LJSMTMF.device->gpu_barrier();
+    LJSMTMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }
@@ -110,7 +110,7 @@ void ljsmt_gpu_reinit(const int ntypes, double **cutsq, double **host_lj1,
   for (int i=0; i<procs_per_gpu; i++) {
     if (gpu_rank==i && world_me!=0)
       LJSMTMF.reinit(ntypes, cutsq, host_lj1, host_lj2, host_lj3, host_lj4, offset, host_ljsw0, host_ljsw1, host_ljsw2, host_ljsw3, host_ljsw4, cut_inner, cut_inner_sq);
-    LJSMTMF.device->gpu_barrier();
+    LJSMTMF.device->serialize_init();
   }
 }
 

lib/gpu/lal_lj_spica_ext.cpp

@@ -77,7 +77,7 @@ int spica_gpu_init(const int ntypes, double **cutsq, int **cg_types,
                          host_lj4, offset, special_lj, inum, nall, max_nbors,
                          maxspecial, cell_size, gpu_split, screen);
 
-    CMMMF.device->gpu_barrier();
+    CMMMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_lj_spica_long_ext.cpp

@@ -81,7 +81,7 @@ int spical_gpu_init(const int ntypes, double **cutsq, int **cg_type,
                           maxspecial, cell_size, gpu_split, screen,
                           host_cut_ljsq, host_cut_coulsq, host_special_coul,
                           qqrd2e, g_ewald);
-    CMMLMF.device->gpu_barrier();
+    CMMLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_lj_tip4p_long_ext.cpp

@@ -89,7 +89,7 @@ int ljtip4p_long_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
           host_special_coul, qqrd2e,
           g_ewald, map_size, max_same);
 
-    LJTIP4PLMF.device->gpu_barrier();
+    LJTIP4PLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_mie_ext.cpp

@@ -80,7 +80,7 @@ int mie_gpu_init(const int ntypes, double **cutsq, double **host_mie1,
                         offset, special_lj, inum, nall, max_nbors, maxspecial,
                         cell_size, gpu_split, screen);
 
-    MLMF.device->gpu_barrier();
+    MLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_morse_ext.cpp

@@ -77,7 +77,7 @@ int mor_gpu_init(const int ntypes, double **cutsq,
                          offset, special_lj, inum, nall, max_nbors, maxspecial,
                          cell_size, gpu_split, screen);
 
-    MORMF.device->gpu_barrier();
+    MORMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_pppm_ext.cpp

@@ -81,7 +81,7 @@ grdtyp * pppm_gpu_init(memtyp &pppm, const int nlocal, const int nall,
                            vd_brick,slab_volfactor,nx_pppm,ny_pppm,nz_pppm,
                            split,success);
 
-    pppm.device->gpu_barrier();
+    pppm.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_re_squared_ext.cpp

@@ -80,7 +80,7 @@ int re_gpu_init(const int ntypes, double **shape, double **well, double **cutsq,
                         host_lj4, offset, special_lj,  inum, nall,
                         max_nbors, maxspecial, cell_size, gpu_split, screen);
 
-    REMF.device->gpu_barrier();
+    REMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_soft_ext.cpp

@@ -76,7 +76,7 @@ int soft_gpu_init(const int ntypes, double **cutsq, double **host_prefactor,
                         special_lj, inum, nall, max_nbors, maxspecial,
                         cell_size, gpu_split, screen);
 
-    SLMF.device->gpu_barrier();
+    SLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }
@@ -106,7 +106,7 @@ void soft_gpu_reinit(const int ntypes, double **cutsq, double **host_prefactor,
     if (gpu_rank==i && world_me!=0)
       SLMF.reinit(ntypes, cutsq, host_prefactor, host_cut);
 
-    SLMF.device->gpu_barrier();
+    SLMF.device->serialize_init();
   }
 }
 

lib/gpu/lal_sw_ext.cpp

@@ -84,7 +84,7 @@ int sw_gpu_init(const int ntypes, const int inum, const int nall,
                         sigma_gamma, c1, c2, c3, c4, c5, c6, lambda_epsilon,
                         costheta, map, e2param);
 
-    SWMF.device->gpu_barrier();
+    SWMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_table_ext.cpp

@@ -76,7 +76,7 @@ int table_gpu_init(const int ntypes, double **cutsq, double ***table_coeffs,
                       special_lj, inum, nall, max_nbors, maxspecial, cell_size,
                       gpu_split, screen, tabstyle, ntables, tablength);
 
-    TBMF.device->gpu_barrier();
+    TBMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_tersoff_ext.cpp

@@ -91,7 +91,7 @@ int tersoff_gpu_init(const int ntypes, const int inum, const int nall, const int
                         ts_c1, ts_c2, ts_c3, ts_c4, ts_c, ts_d, ts_h,
                         ts_gamma, ts_beta, ts_powern, ts_cutsq);
 
-    TSMF.device->gpu_barrier();
+    TSMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_tersoff_mod_ext.cpp

@@ -91,7 +91,7 @@ int tersoff_mod_gpu_init(const int ntypes, const int inum, const int nall,
                         ts_c3, ts_c4, ts_c5, ts_h, ts_beta, ts_powern,
                         ts_powern_del, ts_ca1, ts_cutsq);
 
-    TSMMF.device->gpu_barrier();
+    TSMMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_tersoff_zbl_ext.cpp

@@ -102,7 +102,7 @@ int tersoff_zbl_gpu_init(const int ntypes, const int inum, const int nall,
                         ts_ZBLcut, ts_ZBLexpscale, global_e, global_a_0,
                         global_epsilon_0, ts_cutsq);
 
-    TSZMF.device->gpu_barrier();
+    TSZMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_ufm_ext.cpp

@@ -78,7 +78,7 @@ int ufml_gpu_init(const int ntypes, double **cutsq, double **host_uf1,
                          offset, special_lj, inum, nall, max_nbors, maxspecial,
                          cell_size, gpu_split, screen);
 
-    UFMLMF.device->gpu_barrier();
+    UFMLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }
@@ -106,7 +106,7 @@ void ufml_gpu_reinit(const int ntypes, double **cutsq, double **host_uf1,
   for (int i=0; i<procs_per_gpu; i++) {
     if (gpu_rank==i && world_me!=0)
       UFMLMF.reinit(ntypes, cutsq, host_uf1, host_uf2, host_uf3, offset);
-    UFMLMF.device->gpu_barrier();
+    UFMLMF.device->serialize_init();
   }
 }
 

lib/gpu/lal_vashishta_ext.cpp

@@ -89,7 +89,7 @@ int vashishta_gpu_init(const int ntypes, const int inum, const int nall, const i
                         lam4inv, zizj, mbigd, dvrc, big6w, heta, bigh, bigw,
                         c0, costheta, bigb, big2b, bigc);
 
-    VashishtaMF.device->gpu_barrier();
+    VashishtaMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_yukawa_colloid_ext.cpp

@@ -76,7 +76,7 @@ int ykcolloid_gpu_init(const int ntypes, double **cutsq, double **host_a,
                             inum, nall, max_nbors, maxspecial, cell_size, gpu_split,
                             screen, kappa);
 
-    YKCOLLMF.device->gpu_barrier();
+    YKCOLLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_yukawa_ext.cpp

@@ -76,7 +76,7 @@ int yukawa_gpu_init(const int ntypes, double **cutsq, double kappa,
                       inum, nall, max_nbors, maxspecial, cell_size,
                       gpu_split, screen);
 
-    YKMF.device->gpu_barrier();
+    YKMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }

lib/gpu/lal_zbl_ext.cpp

@@ -79,7 +79,7 @@ int zbl_gpu_init(const int ntypes, double **cutsq, double **host_sw1,
                          cut_globalsq, cut_innersq, cut_inner,
                          inum, nall, max_nbors, maxspecial, cell_size, gpu_split, screen);
 
-    ZBLMF.device->gpu_barrier();
+    ZBLMF.device->serialize_init();
     if (message)
       fprintf(screen,"Done.\n");
   }