Merge Pull Request trilinos#6623 from trilinos/Trilinos/csiefer-8683cc9

Automatically Merged using Trilinos Pull Request AutoTester PR Title: MueLu: Adding timer granularity to Kokkos-ized prolongator smoothing PR Author: csiefer2
tcad-charon · Jan 21, 2020 · 6c249fc · 6c249fc
2 parents 2a8abf5 + 8683cc9
commit 6c249fc
Show file tree

Hide file tree

Showing 3 changed files with 15 additions and 7 deletions.
diff --git a/packages/muelu/src/Transfers/Smoothed-Aggregation/MueLu_SaPFactory_kokkos_def.hpp b/packages/muelu/src/Transfers/Smoothed-Aggregation/MueLu_SaPFactory_kokkos_def.hpp
@@ -173,13 +173,19 @@ namespace MueLu {
 
       {
         SubFactoryMonitor m2(*this, "Fused (I-omega*D^{-1} A)*Ptent", coarseLevel);
-        RCP<Vector> invDiag = Utilities_kokkos::GetMatrixDiagonalInverse(*A);
-
+        RCP<Vector> invDiag;
+        {
+          SubFactoryMonitor m3(*this, "Diagonal Extraction", coarseLevel);
+          invDiag = Utilities_kokkos::GetMatrixDiagonalInverse(*A);
+        }
         SC omega = dampingFactor / lambdaMax;
         TEUCHOS_TEST_FOR_EXCEPTION(!std::isfinite(Teuchos::ScalarTraits<SC>::magnitude(omega)), Exceptions::RuntimeError, "Prolongator damping factor needs to be finite.");
 
-        // finalP = Ptent + (I - \omega D^{-1}A) Ptent
-        finalP = Xpetra::IteratorOps<Scalar, LocalOrdinal, GlobalOrdinal, Node>::Jacobi(omega, *invDiag, *A, *Ptent, finalP, GetOStream(Statistics2), std::string("MueLu::SaP-") + toString(coarseLevel.GetLevelID()), APparams);
+        {
+          SubFactoryMonitor m3(*this, "Xpetra::IteratorOps::Jacobi", coarseLevel);
+          // finalP = Ptent + (I - \omega D^{-1}A) Ptent
+          finalP = Xpetra::IteratorOps<Scalar, LocalOrdinal, GlobalOrdinal, Node>::Jacobi(omega, *invDiag, *A, *Ptent, finalP, GetOStream(Statistics2), std::string("MueLu::SaP-") + toString(coarseLevel.GetLevelID()), APparams);
+        }
       }
 
     } else {

diff --git a/packages/muelu/src/Utils/MueLu_UtilitiesBase_decl.hpp b/packages/muelu/src/Utils/MueLu_UtilitiesBase_decl.hpp
@@ -164,6 +164,7 @@ namespace MueLu {
     NOTE -- it's assumed that A has been fillComplete'd.
     */
     static RCP<Vector> GetMatrixDiagonalInverse(const Matrix& A, Magnitude tol = Teuchos::ScalarTraits<Scalar>::eps()*100) {
+      Teuchos::TimeMonitor MM = *Teuchos::TimeMonitor::getNewTimer("UtilitiesBase::GetMatrixDiagonalInverse");
 
       RCP<const Matrix> rcpA = Teuchos::rcpFromRef(A);
 

diff --git a/packages/muelu/src/Utils/MueLu_Utilities_kokkos_def.hpp b/packages/muelu/src/Utils/MueLu_Utilities_kokkos_def.hpp
@@ -138,6 +138,7 @@ namespace MueLu {
   Teuchos::RCP<Xpetra::Vector<Scalar,LocalOrdinal,GlobalOrdinal,Node> >
   Utilities_kokkos<Scalar, LocalOrdinal, GlobalOrdinal, Node>::
   GetMatrixDiagonalInverse(const Matrix& A, Magnitude tol) {
+    Teuchos::TimeMonitor MM = *Teuchos::TimeMonitor::getNewTimer("Utilities_kokkos::GetMatrixDiagonalInverse");
     // Some useful type definitions
     using local_matrix_type = typename Matrix::local_matrix_type;
     using local_graph_type  = typename local_matrix_type::staticcrsgraph_type;
@@ -154,14 +155,13 @@ namespace MueLu {
 
     // Get/Create distributed objects
     RCP<const Map> rowMap = A.getRowMap();
-    RCP<Vector> diag      = VectorFactory::Build(rowMap);
+    RCP<Vector> diag      = VectorFactory::Build(rowMap,false);
 
     // Now generate local objects
     local_matrix_type localMatrix = A.getLocalMatrix();
-    local_graph_type  localGraph  = localMatrix.graph;
     auto diagVals = diag->template getLocalView<memory_space>();
 
-    ordinal_type numRows = localGraph.numRows();
+    ordinal_type numRows = localMatrix.graph.numRows();
 
     // Note: 2019-11-21, LBV
     // This could be implemented with a TeamPolicy over the rows
@@ -180,6 +180,7 @@ namespace MueLu {
                                } else {
                                  diagVals(rowIdx, 0) = KAT::zero();
                                }
+                               break;
                              }
                            }