Fixing issues with growing memory allocation with multiple init calls

lib/gpu/lal_base_atomic.cpp

@@ -25,12 +25,16 @@ BaseAtomicT::BaseAtomic() : _compiled(false), _max_bytes(0)  {
   device=&global_device;
   ans=new Answer<numtyp,acctyp>();
   nbor=new Neighbor();
+  pair_program=NULL;
 }
 
 template <class numtyp, class acctyp>
 BaseAtomicT::~BaseAtomic() {
   delete ans;
   delete nbor;
+  if (pair_program) delete pair_program;
+  k_pair_fast.clear();
+  k_pair.clear();
 }
 
 template <class numtyp, class acctyp>
@@ -109,19 +113,11 @@ void BaseAtomicT::clear_atomic() {
   device->output_times(time_pair,*ans,*nbor,avg_split,_max_bytes+_max_an_bytes,
                        _gpu_overhead,_driver_overhead,_threads_per_atom,screen);
 
-  if (_compiled) {
-    k_pair_fast.clear();
-    k_pair.clear();
-    delete pair_program;
-    _compiled=false;
-  }
-
   time_pair.clear();
   hd_balancer.clear();
 
   nbor->clear();
   ans->clear();
-  device->clear();
 }
 
 // ---------------------------------------------------------------------------
@@ -276,6 +272,7 @@ void BaseAtomicT::compile_kernels(UCL_Device &dev, const void *pair_str,
     return;
 
   std::string s_fast=std::string(kname)+"_fast";
+  if (pair_program) delete pair_program;
   pair_program=new UCL_Program(dev);
   pair_program->load_string(pair_str,device->compile_string().c_str());
   k_pair_fast.set_function(*pair_program,s_fast.c_str());

lib/gpu/lal_base_charge.cpp

@@ -25,12 +25,16 @@ BaseChargeT::BaseCharge() : _compiled(false), _max_bytes(0) {
   device=&global_device;
   ans=new Answer<numtyp,acctyp>();
   nbor=new Neighbor();
+  pair_program=NULL;
 }
 
 template <class numtyp, class acctyp>
 BaseChargeT::~BaseCharge() {
   delete ans;
   delete nbor;
+  if (pair_program) delete pair_program;
+  k_pair_fast.clear();
+  k_pair.clear();
 }
 
 template <class numtyp, class acctyp>
@@ -111,19 +115,11 @@ void BaseChargeT::clear_atomic() {
   device->output_times(time_pair,*ans,*nbor,avg_split,_max_bytes+_max_an_bytes,
                        _gpu_overhead,_driver_overhead,_threads_per_atom,screen);
 
-  if (_compiled) {
-    k_pair_fast.clear();
-    k_pair.clear();
-    delete pair_program;
-    _compiled=false;
-  }
-
   time_pair.clear();
   hd_balancer.clear();
 
   nbor->clear();
   ans->clear();
-  device->clear();
 }
 
 // ---------------------------------------------------------------------------
@@ -291,6 +287,7 @@ void BaseChargeT::compile_kernels(UCL_Device &dev, const void *pair_str,
     return;
 
   std::string s_fast=std::string(kname)+"_fast";
+  if (pair_program) delete pair_program;
   pair_program=new UCL_Program(dev);
   pair_program->load_string(pair_str,device->compile_string().c_str());
   k_pair_fast.set_function(*pair_program,s_fast.c_str());

lib/gpu/lal_base_dipole.cpp

@@ -25,12 +25,16 @@ BaseDipoleT::BaseDipole() : _compiled(false), _max_bytes(0) {
   device=&global_device;
   ans=new Answer<numtyp,acctyp>();
   nbor=new Neighbor();
+  pair_program=NULL;
 }
 
 template <class numtyp, class acctyp>
 BaseDipoleT::~BaseDipole() {
   delete ans;
   delete nbor;
+  if (pair_program) delete pair_program;
+  k_pair_fast.clear();
+  k_pair.clear();
 }
 
 template <class numtyp, class acctyp>
@@ -113,19 +117,11 @@ void BaseDipoleT::clear_atomic() {
   device->output_times(time_pair,*ans,*nbor,avg_split,_max_bytes+_max_an_bytes,
                        _gpu_overhead,_driver_overhead,_threads_per_atom,screen);
 
-  if (_compiled) {
-    k_pair_fast.clear();
-    k_pair.clear();
-    delete pair_program;
-    _compiled=false;
-  }
-
   time_pair.clear();
   hd_balancer.clear();
 
   nbor->clear();
   ans->clear();
-  device->clear();
 }
 
 // ---------------------------------------------------------------------------
@@ -299,6 +295,7 @@ void BaseDipoleT::compile_kernels(UCL_Device &dev, const void *pair_str,
     return;
 
   std::string s_fast=std::string(kname)+"_fast";
+  if (pair_program) delete pair_program;
   pair_program=new UCL_Program(dev);
   pair_program->load_string(pair_str,device->compile_string().c_str());
   k_pair_fast.set_function(*pair_program,s_fast.c_str());

lib/gpu/lal_base_dpd.cpp

@@ -25,12 +25,16 @@ BaseDPDT::BaseDPD() : _compiled(false), _max_bytes(0) {
   device=&global_device;
   ans=new Answer<numtyp,acctyp>();
   nbor=new Neighbor();
+  pair_program=NULL;
 }
 
 template <class numtyp, class acctyp>
 BaseDPDT::~BaseDPD() {
   delete ans;
   delete nbor;
+  if (pair_program) delete pair_program;
+  k_pair_fast.clear();
+  k_pair.clear();
 }
 
 template <class numtyp, class acctyp>
@@ -112,19 +116,11 @@ void BaseDPDT::clear_atomic() {
   device->output_times(time_pair,*ans,*nbor,avg_split,_max_bytes+_max_an_bytes,
                        _gpu_overhead,_driver_overhead,_threads_per_atom,screen);
 
-  if (_compiled) {
-    k_pair_fast.clear();
-    k_pair.clear();
-    delete pair_program;
-    _compiled=false;
-  }
-
   time_pair.clear();
   hd_balancer.clear();
 
   nbor->clear();
   ans->clear();
-  device->clear();
 }
 
 // ---------------------------------------------------------------------------
@@ -297,6 +293,7 @@ void BaseDPDT::compile_kernels(UCL_Device &dev, const void *pair_str,
     return;
 
   std::string s_fast=std::string(kname)+"_fast";
+  if (pair_program) delete pair_program;
   pair_program=new UCL_Program(dev);
   pair_program->load_string(pair_str,device->compile_string().c_str());
   k_pair_fast.set_function(*pair_program,s_fast.c_str());

lib/gpu/lal_base_ellipsoid.cpp

@@ -33,12 +33,25 @@ BaseEllipsoidT::BaseEllipsoid() : _compiled(false), _max_bytes(0) {
   device=&global_device;
   ans=new Answer<numtyp,acctyp>();
   nbor=new Neighbor();
+  nbor_program=NULL;
+  ellipsoid_program=NULL;
+  lj_program=NULL;
 }
 
 template <class numtyp, class acctyp>
 BaseEllipsoidT::~BaseEllipsoid() {
   delete ans;
   delete nbor;
+  k_nbor_fast.clear();
+  k_nbor.clear();
+  k_ellipsoid.clear();
+  k_ellipsoid_sphere.clear();
+  k_sphere_ellipsoid.clear();
+  k_lj_fast.clear();
+  k_lj.clear();
+  if (nbor_program) delete nbor_program;
+  if (ellipsoid_program) delete ellipsoid_program;
+  if (lj_program) delete lj_program;
 }
 
 template <class numtyp, class acctyp>
@@ -146,20 +159,6 @@ void BaseEllipsoidT::clear_base() {
   output_times();
   host_olist.clear();
 
-  if (_compiled) {
-    k_nbor_fast.clear();
-    k_nbor.clear();
-    k_ellipsoid.clear();
-    k_ellipsoid_sphere.clear();
-    k_sphere_ellipsoid.clear();
-    k_lj_fast.clear();
-    k_lj.clear();
-    delete nbor_program;
-    delete ellipsoid_program;
-    delete lj_program;
-    _compiled=false;
-  }
-
   time_nbor1.clear();
   time_ellipsoid.clear();
   time_nbor2.clear();
@@ -171,7 +170,6 @@ void BaseEllipsoidT::clear_base() {
 
   nbor->clear();
   ans->clear();
-  device->clear();
 }
 
 template <class numtyp, class acctyp>
@@ -437,6 +435,7 @@ int** BaseEllipsoidT::compute(const int ago, const int inum_full, const int nall
   ans->copy_answers(eflag,vflag,eatom,vatom);
   device->add_ans_object(ans);
   hd_balancer.stop_timer();
+
   return nbor->host_jlist.begin()-host_start;
 }
 
@@ -462,18 +461,21 @@ void BaseEllipsoidT::compile_kernels(UCL_Device &dev,
 
   std::string flags=device->compile_string();
 
+  if (nbor_program) delete nbor_program;
   nbor_program=new UCL_Program(dev);
   nbor_program->load_string(ellipsoid_nbor,flags.c_str());
   k_nbor_fast.set_function(*nbor_program,"kernel_nbor_fast");
   k_nbor.set_function(*nbor_program,"kernel_nbor");
   neigh_tex.get_texture(*nbor_program,"pos_tex");
 
+  if (ellipsoid_program) delete ellipsoid_program;
   ellipsoid_program=new UCL_Program(dev);
   ellipsoid_program->load_string(ellipsoid_string,flags.c_str());
   k_ellipsoid.set_function(*ellipsoid_program,kname);
   pos_tex.get_texture(*ellipsoid_program,"pos_tex");
   quat_tex.get_texture(*ellipsoid_program,"quat_tex");
 
+  if (lj_program) delete lj_program;
   lj_program=new UCL_Program(dev);
   lj_program->load_string(lj_string,flags.c_str());
   k_sphere_ellipsoid.set_function(*lj_program,s_sphere_ellipsoid.c_str());

lib/gpu/lal_base_three.cpp

@@ -27,6 +27,7 @@ BaseThreeT::BaseThree() : _compiled(false), _max_bytes(0) {
   #ifdef THREE_CONCURRENT
   ans2=new Answer<numtyp,acctyp>();
   #endif
+  pair_program=NULL;
 }
 
 template <class numtyp, class acctyp>
@@ -36,6 +37,12 @@ BaseThreeT::~BaseThree() {
   #ifdef THREE_CONCURRENT
   delete ans2;
   #endif
+  if (pair_program) delete pair_program;
+  k_three_center.clear();
+  k_three_end.clear();
+  k_three_end_vatom.clear();
+  k_pair.clear();
+  k_short_nbor.clear();
 }
 
 template <class numtyp, class acctyp>
@@ -139,16 +146,6 @@ void BaseThreeT::clear_atomic() {
   device->output_times(time_pair,*ans,*nbor,avg_split,_max_bytes+_max_an_bytes,
                        _gpu_overhead,_driver_overhead,_threads_per_atom,screen);
 
-  if (_compiled) {
-    k_three_center.clear();
-    k_three_end.clear();
-    k_three_end_vatom.clear();
-    k_pair.clear();
-    k_short_nbor.clear();
-    delete pair_program;
-    _compiled=false;
-  }
-
   time_pair.clear();
   hd_balancer.clear();
 
@@ -161,7 +158,6 @@ void BaseThreeT::clear_atomic() {
   // ucl_device will clean up the command queue in its destructor
 //  ucl_device->pop_command_queue();
   #endif
-  device->clear();
 }
 
 // ---------------------------------------------------------------------------
@@ -378,7 +374,7 @@ void BaseThreeT::compile_kernels(UCL_Device &dev, const void *pair_str,
     return;
 
   std::string vatom_name=std::string(three_end)+"_vatom";
-
+  if (pair_program) delete pair_program;
   pair_program=new UCL_Program(dev);
   pair_program->load_string(pair_str,device->compile_string().c_str());
   k_three_center.set_function(*pair_program,three_center);