Fixing some bugs when using multiple GPU styles hybrid.

lib/gpu/atomic_gpu_memory.cpp

@@ -112,8 +112,7 @@ int * AtomicGPUMemoryT::reset_nbors(const int nall, const int inum, int *ilist,
   success=true;
 
   nbor_time_avail=true;
-
-  int mn=nbor->max_nbor_loop(inum,numj);
+  int mn=nbor->max_nbor_loop(inum,numj,ilist);
   resize_atom(inum,nall,success);
   resize_local(inum,mn,success);
   if (!success)
@@ -170,6 +169,7 @@ void AtomicGPUMemoryT::compute(const int f_ago, const int inum_full,
                                bool &success) {
   acc_timers();
   if (inum_full==0) {
+    host_start=0;
     zero_timers();
     return;
   }
@@ -208,6 +208,7 @@ int * AtomicGPUMemoryT::compute(const int ago, const int inum_full,
                                 const double cpu_time, bool &success) {
   acc_timers();
   if (inum_full==0) {
+    host_start=0;
     zero_timers();
     return NULL;
   }

lib/gpu/charge_gpu_memory.cpp

@@ -114,7 +114,7 @@ int * ChargeGPUMemoryT::reset_nbors(const int nall, const int inum, int *ilist,
 
   nbor_time_avail=true;
 
-  int mn=nbor->max_nbor_loop(inum,numj);
+  int mn=nbor->max_nbor_loop(inum,numj,ilist);
   resize_atom(inum,nall,success);
   resize_local(inum,mn,success);
   if (!success)
@@ -171,6 +171,7 @@ void ChargeGPUMemoryT::compute(const int f_ago, const int inum_full,
                                bool &success, double *host_q) {
   acc_timers();
   if (inum_full==0) {
+    host_start=0;
     zero_timers();
     return;
   }
@@ -213,6 +214,7 @@ int * ChargeGPUMemoryT::compute(const int ago, const int inum_full,
                                 double *host_q) {
   acc_timers();
   if (inum_full==0) {
+    host_start=0;
     zero_timers();
     return NULL;
   }

lib/gpu/gb_gpu.cpp

@@ -163,7 +163,7 @@ void _gb_gpu_reset_nbors(gbmtyp &gbm, const int nall,
     
   gbm.nbor_time_avail=true;
 
-  int mn=gbm.nbor->max_nbor_loop(inum,numj);
+  int mn=gbm.nbor->max_nbor_loop(inum,numj,ilist);
   gbm.resize_atom(inum,nall,success);
   gbm.resize_local(inum,0,mn,osize,success);
   if (!success)
@@ -336,6 +336,7 @@ inline int * _gb_gpu_compute_n(gbmtyp &gbm, const int ago,
 			       double **host_quat) {
   gbm.acc_timers();
   if (inum_full==0) {
+    host_start=0;
     gbm.zero_timers();
     return NULL;
   }
@@ -394,6 +395,7 @@ inline int * _gb_gpu_compute(gbmtyp &gbm, const int f_ago, const int inum_full,
                              bool &success, double **host_quat) {
   gbm.acc_timers();
   if (inum_full==0) {
+    host_start=0;
     gbm.zero_timers();
     return NULL;
   }

lib/gpu/pair_gpu_nbor.h

@@ -126,10 +126,10 @@ class PairGPUNbor {
   inline int max_nbors() const { return _max_nbors; }
 
   /// Loop through neighbor count array and return maximum nbors for a particle
-  inline int max_nbor_loop(const int inum, int *numj) const {
+  inline int max_nbor_loop(const int inum, int *numj, int *ilist) const {
     int mn=0;
     for (int i=0; i<inum; i++)
-      mn=std::max(mn,numj[i]);
+      mn=std::max(mn,numj[ilist[i]]);
     return mn;
   }
 

src/GPU/pair_lj_cut_coul_cut_gpu.cpp

@@ -92,7 +92,7 @@ void PairLJCutCoulCutGPU::compute(int eflag, int vflag)
   int inum, host_start;
   
   bool success = true;
-  
+
   if (gpu_mode == GPU_NEIGH) {
     inum = atom->nlocal;
     gpulist = ljc_gpu_compute_n(neighbor->ago, inum, nall, atom->x,