Implement deleting Vppph

atrip.org

@@ -2955,6 +2955,7 @@ namespace atrip {
         GROUP_AND_SORT,
       };
 
+      ADD_ATTRIBUTE(bool, deleteVppph, false)
       ADD_ATTRIBUTE(bool, rankRoundRobin, false)
       ADD_ATTRIBUTE(bool, chrono, false)
       ADD_ATTRIBUTE(bool, barrier, false)
@@ -3063,14 +3064,6 @@ Atrip::Output Atrip::run(Atrip::Input<F> const& in) {
     MPI_Comm_size(child_comm, &child_size);
   }
 
-
-  // BUILD SLICES PARAMETRIZED BY NV ==================================={{{1
-  WITH_CHRONO("nv-slices",
-    LOG(0,"Atrip") << "BUILD NV-SLICES\n";
-    TAPHH<F> taphh(*in.Tpphh, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
-    HHHA<F>  hhha(*in.Vhhhp, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
-  )
-
   // BUILD SLICES PARAMETRIZED BY NV x NV =============================={{{1
   WITH_CHRONO("nv-nv-slices",
     LOG(0,"Atrip") << "BUILD NV x NV-SLICES\n";
@@ -3079,6 +3072,18 @@ Atrip::Output Atrip::run(Atrip::Input<F> const& in) {
     TABHH<F> tabhh(*in.Tpphh, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
   )
 
+  // delete the Vppph so that we don't have a HWM situation for the NV slices
+  if (in.deleteVppph) {
+    delete in.Vppph;
+  }
+
+  // BUILD SLICES PARAMETRIZED BY NV ==================================={{{1
+  WITH_CHRONO("nv-slices",
+    LOG(0,"Atrip") << "BUILD NV-SLICES\n";
+    TAPHH<F> taphh(*in.Tpphh, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
+    HHHA<F>  hhha(*in.Vhhhp, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
+  )
+
   // all tensors
   std::vector< SliceUnion<F>* > unions = {&taphh, &hhha, &abph, &abhh, &tabhh};
 
@@ -3251,12 +3256,12 @@ Atrip::Output Atrip::run(Atrip::Input<F> const& in) {
 
   const double doublesFlops
     = double(No)
-    * double(No)
+    ,* double(No)
-    * double(No)
+    ,* double(No)
-    * (double(No) + double(Nv))
+    ,* (double(No) + double(Nv))
-    * 2.0
+    ,* 2.0
-    * (traits::isComplex<F>() ? 2.0 : 1.0)
+    ,* (traits::isComplex<F>() ? 2.0 : 1.0)
-    * 6.0
+    ,* 6.0
     / 1e9
     ;
 

bench/test_main.cxx

@@ -17,13 +17,15 @@ int main(int argc, char** argv) {
   MPI_Init(&argc, &argv);
 
   int no(10), nv(10), itMod(-1), percentageMod(10);
-  bool nochrono(false), barrier(false), rankRoundRobin(false);
+  bool nochrono(false), barrier(false), rankRoundRobin(false),
+       keepVppph(false);
   std::string tuplesDistributionString = "naive";
 
   CLI::App app{"Main bench for atrip"};
   app.add_option("--no", no, "Occupied orbitals");
   app.add_option("--nv", nv, "Virtual orbitals");
   app.add_option("--mod", itMod, "Iteration modifier");
+  app.add_flag("--keep-vppph", keepVppph, "Do not delete Vppph");
   app.add_flag("--nochrono", nochrono, "Do not print chrono");
   app.add_flag("--rank-round-robin", rankRoundRobin, "Do rank round robin");
   app.add_flag("--barrier", barrier, "Use the first barrier");
@@ -38,7 +40,8 @@ int main(int argc, char** argv) {
   constexpr double elem_to_gb = 8.0 / 1024.0 / 1024.0 / 1024.0;
 
   // USER PRINTING TEST BEGIN
-  const double doublesFlops = no * no * no
+  const double doublesFlops
+       = no * no * no
        * (no + nv)
        * 2.0
        * 6.0
@@ -110,7 +113,8 @@ int main(int argc, char** argv) {
   Vppph.fill_random(0, 1);
 
   atrip::Atrip::init();
-  const auto in = atrip::Atrip::Input<double>()
+  const auto in
+       = atrip::Atrip::Input<double>()
        // Tensors
        .with_epsilon_i(&ei)
        .with_epsilon_a(&ea)
@@ -120,6 +124,7 @@ int main(int argc, char** argv) {
        .with_Vijka(&Vhhhp)
        .with_Vabci(&Vppph)
        // some options
+       .with_deleteVppph(!keepVppph)
        .with_barrier(barrier)
        .with_chrono(!nochrono)
        .with_rankRoundRobin(rankRoundRobin)

include/atrip/Atrip.hpp

@@ -60,6 +60,7 @@ namespace atrip {
         GROUP_AND_SORT,
       };
 
+      ADD_ATTRIBUTE(bool, deleteVppph, false)
       ADD_ATTRIBUTE(bool, rankRoundRobin, false)
       ADD_ATTRIBUTE(bool, chrono, false)
       ADD_ATTRIBUTE(bool, barrier, false)

src/atrip/Atrip.cxx

@@ -96,14 +96,6 @@ Atrip::Output Atrip::run(Atrip::Input<F> const& in) {
     MPI_Comm_size(child_comm, &child_size);
   }
 
-
-  // BUILD SLICES PARAMETRIZED BY NV ==================================={{{1
-  WITH_CHRONO("nv-slices",
-    LOG(0,"Atrip") << "BUILD NV-SLICES\n";
-    TAPHH<F> taphh(*in.Tpphh, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
-    HHHA<F>  hhha(*in.Vhhhp, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
-  )
-
   // BUILD SLICES PARAMETRIZED BY NV x NV =============================={{{1
   WITH_CHRONO("nv-nv-slices",
     LOG(0,"Atrip") << "BUILD NV x NV-SLICES\n";
@@ -112,6 +104,18 @@ Atrip::Output Atrip::run(Atrip::Input<F> const& in) {
     TABHH<F> tabhh(*in.Tpphh, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
   )
 
+  // delete the Vppph so that we don't have a HWM situation for the NV slices
+  if (in.deleteVppph) {
+    delete in.Vppph;
+  }
+
+  // BUILD SLICES PARAMETRIZED BY NV ==================================={{{1
+  WITH_CHRONO("nv-slices",
+    LOG(0,"Atrip") << "BUILD NV-SLICES\n";
+    TAPHH<F> taphh(*in.Tpphh, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
+    HHHA<F>  hhha(*in.Vhhhp, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
+  )
+
   // all tensors
   std::vector< SliceUnion<F>* > unions = {&taphh, &hhha, &abph, &abhh, &tabhh};