Merge pull request #4752 from hnil/linearsolver_timing

-- added more timing to get better coverage of cpr solver

opm/simulators/linalg/FlexibleSolver_impl.hpp

@@ -22,7 +22,7 @@
 #define OPM_FLEXIBLE_SOLVER_IMPL_HEADER_INCLUDED
 
 #include <opm/common/ErrorMacros.hpp>
-
+#include <opm/common/TimingMacros.hpp>
 #include <opm/simulators/linalg/matrixblock.hh>
 #include <opm/simulators/linalg/ilufirstelement.hh>
 #include <opm/simulators/linalg/FlexibleSolver.hpp>

opm/simulators/linalg/ISTLSolverEbos.cpp

@@ -20,7 +20,7 @@
 */
 
 #include <config.h>
-
+#include <opm/common/TimingMacros.hpp>
 #include <opm/simulators/linalg/ISTLSolverEbos.hpp>
 
 #include <dune/istl/schwarz.hh>

opm/simulators/linalg/MILU.cpp

@@ -20,7 +20,7 @@
 
 #include <config.h>
 #include <opm/simulators/linalg/MILU.hpp>
-
+#include <opm/common/TimingMacros.hpp>
 #include <dune/common/version.hh>
 #include <dune/common/fmatrix.hh>
 #include <dune/istl/bcrsmatrix.hh>

opm/simulators/linalg/ParallelOverlappingILU0.hpp

@@ -19,7 +19,7 @@
 */
 #ifndef OPM_PARALLELOVERLAPPINGILU0_HEADER_INCLUDED
 #define OPM_PARALLELOVERLAPPINGILU0_HEADER_INCLUDED
-
+#include <opm/common/TimingMacros.hpp>
 #include <opm/simulators/linalg/MILU.hpp>
 #include <opm/simulators/linalg/PreconditionerWithUpdate.hpp>
 #include <dune/istl/paamg/smoother.hh>

opm/simulators/linalg/ParallelRestrictedAdditiveSchwarz.hpp

@@ -165,6 +165,7 @@ public:
     */
     virtual void pre (Domain& x, Range& b)
     {
+        OPM_TIMEBLOCK(pre);
         communication_.copyOwnerToAll(x,x);     // make dirichlet values consistent
         preconditioner_.pre(x,b);
     }
@@ -182,6 +183,7 @@ public:
     template<bool forward>
     void apply (Domain& v, const Range& d)
     {
+        OPM_TIMEBLOCK(apply);
         // hack us a mutable d to prevent copying.
         Range& md = const_cast<Range&>(d);
         communication_.copyOwnerToAll(md,md);
@@ -198,6 +200,7 @@ public:
     */
     virtual void post (Range& x)
     {
+        OPM_TIMEBLOCK(post);
         preconditioner_.post(x);
     }
 

opm/simulators/linalg/PreconditionerFactory.hpp

@@ -21,7 +21,7 @@
 
 #ifndef OPM_PRECONDITIONERFACTORY_HEADER
 #define OPM_PRECONDITIONERFACTORY_HEADER
-
+#include <opm/common/TimingMacros.hpp>
 #include <opm/simulators/linalg/PreconditionerWithUpdate.hpp>
 
 #include <dune/istl/paamg/aggregates.hh>

opm/simulators/linalg/PreconditionerFactory_impl.hpp

@@ -17,10 +17,11 @@
   You should have received a copy of the GNU General Public License
   along with OPM.  If not, see <http://www.gnu.org/licenses/>.
 */
+#include <opm/common/ErrorMacros.hpp>
+#include <opm/common/TimingMacros.hpp>
 
 #include <opm/simulators/linalg/PreconditionerFactory.hpp>
 
-#include <opm/common/ErrorMacros.hpp>
 
 #include <opm/simulators/linalg/matrixblock.hh>
 #include <opm/simulators/linalg/amgcpr.hh>

opm/simulators/linalg/WellOperators.hpp

@@ -72,12 +72,14 @@ public:
      */
     void apply(const X& x, Y& y) const override
     {
+        OPM_TIMEBLOCK(apply);
         wellMod_.apply(x, y);
     }
 
     //! apply operator to x, scale and add:  \f$ y = y + \alpha A(x) \f$
     virtual void applyscaleadd(field_type alpha, const X& x, Y& y) const override
     {
+        OPM_TIMEBLOCK(applyscaleadd);
         wellMod_.applyScaleAdd(alpha, x, y);
     }
 
@@ -92,10 +94,12 @@ public:
     }
     void addWellPressureEquations(PressureMatrix& jacobian, const X& weights,const bool use_well_weights) const override
     {
+        OPM_TIMEBLOCK(addWellPressureEquations);
         wellMod_.addWellPressureEquations(jacobian, weights, use_well_weights);
     }
     void addWellPressureEquationsStruct(PressureMatrix& jacobian) const override
     {
+        OPM_TIMEBLOCK(addWellPressureEquationsStruct);
         wellMod_.addWellPressureEquationsStruct(jacobian);
     }
     int getNumberOfExtraEquations() const override
@@ -149,6 +153,7 @@ public:
 
   virtual void apply( const X& x, Y& y ) const override
   {
+    OPM_TIMEBLOCK(apply);
     A_.mv( x, y );
 
     // add well model modification to y
@@ -163,6 +168,7 @@ public:
   // y += \alpha * A * x
   virtual void applyscaleadd (field_type alpha, const X& x, Y& y) const override
   {
+    OPM_TIMEBLOCK(applyscaleadd);
     A_.usmv(alpha,x,y);
 
     // add scaled well model modification to y
@@ -178,10 +184,12 @@ public:
 
     void addWellPressureEquations(PressureMatrix& jacobian, const X& weights,const bool use_well_weights) const
     {
+        OPM_TIMEBLOCK(addWellPressureEquations);
         wellOper_.addWellPressureEquations(jacobian, weights, use_well_weights);
     }
     void addWellPressureEquationsStruct(PressureMatrix& jacobian) const
     {
+        OPM_TIMEBLOCK(addWellPressureEquations);
         wellOper_.addWellPressureEquationsStruct(jacobian);
     }
     int getNumberOfExtraEquations() const
@@ -235,6 +243,7 @@ public:
 
     virtual void apply( const X& x, Y& y ) const override
     {
+        OPM_TIMEBLOCK(apply);
         for (auto row = A_.begin(); row.index() < interiorSize_; ++row)
         {
             y[row.index()]=0;
@@ -252,6 +261,7 @@ public:
     // y += \alpha * A * x
     virtual void applyscaleadd (field_type alpha, const X& x, Y& y) const override
     {
+        OPM_TIMEBLOCK(applyscaleadd);
         for (auto row = A_.begin(); row.index() < interiorSize_; ++row)
         {
             auto endc = (*row).end();
@@ -268,10 +278,12 @@ public:
 
     void addWellPressureEquations(PressureMatrix& jacobian, const X& weights,const bool use_well_weights) const
     {
+        OPM_TIMEBLOCK(addWellPressureEquations);
         wellOper_.addWellPressureEquations(jacobian, weights, use_well_weights);
     }
     void addWellPressureEquationsStruct(PressureMatrix& jacobian) const
     {
+        OPM_TIMEBLOCK(addWellPressureEquationsStruct);
         wellOper_.addWellPressureEquationsStruct(jacobian);
     }
     int getNumberOfExtraEquations() const

opm/simulators/linalg/amgcpr.hh

@@ -7,7 +7,7 @@
 // dune-istl release 2.6.0. Modifications have been kept as minimal as possible.
 
 #include <opm/simulators/linalg/PreconditionerWithUpdate.hpp>
-
+#include <opm/common/TimingMacros.hpp>
 #include <dune/common/exceptions.hh>
 #include <dune/common/version.hh>
 #include <dune/istl/paamg/amg.hh>
@@ -42,6 +42,7 @@ namespace Dune
   redistributeMatrixAmg(M& mat, M& matRedist, PI& info, PI& infoRedist,
                         Dune::RedistributeInformation<PI>& redistInfo)
   {
+    OPM_TIMEBLOCK(redistributeMatrixAmg);
     info.buildGlobalLookup(mat.N());
     redistributeMatrixEntries(mat, matRedist, info, infoRedist, redistInfo);
     info.freeGlobalLookup();
@@ -192,6 +193,7 @@ namespace Dune
        */
       void recalculateHierarchy()
       {
+        OPM_TIMEBLOCK(recalculateHierarch);
         auto copyFlags = NegateSet<typename PI::OwnerSet>();
         const auto& matrices =  matrices_->matrices();
         const auto& aggregatesMapHierarchy = matrices_->aggregatesMaps();
@@ -260,6 +262,7 @@ namespace Dune
       void createHierarchies(C& criterion, Operator& matrix,
                              const PI& pinfo)
       {
+        //OPM_TIMEBLOCK(createHierarchies);
         // create shared_ptr with empty deleter
         std::shared_ptr< Operator > op( &matrix, []( Operator* ) {});
         std::shared_ptr< PI > pifo( const_cast< PI* > (&pinfo), []( PI * ) {});
@@ -466,6 +469,7 @@ namespace Dune
     template<class M, class X, class S, class PI, class A>
     void AMGCPR<M,X,S,PI,A>::update()
     {
+      OPM_TIMEBLOCK(update);
       Timer watch;
       smoothers_.reset(new Hierarchy<Smoother,A>);
       solver_.reset();
@@ -488,6 +492,7 @@ namespace Dune
     void AMGCPR<M,X,S,PI,A>::createHierarchies(C& criterion, std::shared_ptr< Operator > matrix,
                                                std::shared_ptr< PI > pinfo )
     {
+      OPM_TIMEBLOCK(createHierarchies);
       Timer watch;
       matrices_.reset(new OperatorHierarchy(matrix, pinfo));
 
@@ -504,6 +509,7 @@ namespace Dune
     template<class M, class X, class S, class PI, class A>
     void AMGCPR<M,X,S,PI,A>::setupCoarseSolver()
     {
+      OPM_TIMEBLOCK(setupCoarseSolver);
       // test whether we should solve on the coarse level. That is the case if we
       // have that level and if there was a redistribution on this level then our
       // communicator has to be valid (size()>0) as the smoother might try to communicate
@@ -607,6 +613,7 @@ namespace Dune
     template<class M, class X, class S, class PI, class A>
     void AMGCPR<M,X,S,PI,A>::pre(Domain& x, Range& b)
     {
+      OPM_TIMEBLOCK(pre);
       // Detect Matrix rows where all offdiagonal entries are
       // zero and set x such that  A_dd*x_d=b_d
       // Thus users can be more careless when setting up their linear
@@ -700,6 +707,7 @@ namespace Dune
     template<class M, class X, class S, class PI, class A>
     void AMGCPR<M,X,S,PI,A>::apply(Domain& v, const Range& d)
     {
+      OPM_TIMEBLOCK(apply);
       LevelContext levelContext;
 
       if(additive) {
@@ -729,6 +737,7 @@ namespace Dune
     template<class M, class X, class S, class PI, class A>
     void AMGCPR<M,X,S,PI,A>::initIteratorsWithFineLevel(LevelContext& levelContext)
     {
+      OPM_TIMEBLOCK(initIteratorsWithFineLevel);
       levelContext.smoother = smoothers_->finest();
       levelContext.matrix = matrices_->matrices().finest();
       levelContext.pinfo = matrices_->parallelInformation().finest();
@@ -744,7 +753,7 @@ namespace Dune
     bool AMGCPR<M,X,S,PI,A>
     ::moveToCoarseLevel(LevelContext& levelContext)
     {
-
+      OPM_TIMEBLOCK(moveToCoarseLevel);
       bool processNextLevel=true;
 
       if(levelContext.redist->isSetup()) {
@@ -793,6 +802,7 @@ namespace Dune
     void AMGCPR<M,X,S,PI,A>
     ::moveToFineLevel(LevelContext& levelContext, bool processNextLevel)
     {
+      OPM_TIMEBLOCK(moveToFineLevel);
       if(processNextLevel) {
         if(levelContext.matrix != matrices_->matrices().coarsest() || matrices_->levels()<matrices_->maxlevels()) {
           // previous level is not the globally coarsest one
@@ -841,6 +851,7 @@ namespace Dune
 
     template<class M, class X, class S, class PI, class A>
     void AMGCPR<M,X,S,PI,A>::mgc(LevelContext& levelContext){
+      //OPM_TIMEBLOCK(mgc);
       if(levelContext.matrix == matrices_->matrices().coarsest() && levels()==maxlevels()) {
         // Solve directly
         InverseOperatorResult res;
@@ -895,7 +906,7 @@ namespace Dune
 
     template<class M, class X, class S, class PI, class A>
     void AMGCPR<M,X,S,PI,A>::additiveMgc(){
-
+      OPM_TIMEBLOCK(additiveMgc);
       // restrict residual to all levels
       typename ParallelInformationHierarchy::Iterator pinfo=matrices_->parallelInformation().finest();
       typename Hierarchy<Range,A>::Iterator rhs=rhs_->finest();
@@ -946,6 +957,7 @@ namespace Dune
     template<class M, class X, class S, class PI, class A>
     void AMGCPR<M,X,S,PI,A>::post([[maybe_unused]] Domain& x)
     {
+      OPM_TIMEBLOCK(post);
       // Postprocess all smoothers
       typedef typename Hierarchy<Smoother,A>::Iterator Iterator;
       typedef typename Hierarchy<Domain,A>::Iterator DIterator;

opm/simulators/linalg/bda/BdaBridge.cpp

@@ -18,7 +18,7 @@
 */
 
 #include <config.h>
-
+#include <opm/common/TimingMacros.hpp>
 #include "dune/istl/bcrsmatrix.hh"
 #include <opm/simulators/linalg/matrixblock.hh>
 

opm/simulators/linalg/bda/MultisegmentWellContribution.cpp

@@ -19,7 +19,7 @@
 
 
 #include <config.h> // CMake
-
+#include <opm/common/TimingMacros.hpp>
 #if HAVE_UMFPACK
 #include <dune/istl/umfpack.hh>
 #endif // HAVE_UMFPACK
@@ -69,6 +69,7 @@ MultisegmentWellContribution::~MultisegmentWellContribution()
 // y -= (C^T * (D^-1 * (B * x)))
 void MultisegmentWellContribution::apply(double *h_x, double *h_y)
 {
+    OPM_TIMEBLOCK(apply);
     // reset z1 and z2
     std::fill(z1.begin(), z1.end(), 0.0);
     std::fill(z2.begin(), z2.end(), 0.0);

opm/simulators/linalg/bda/opencl/CPR.cpp

@@ -18,7 +18,7 @@
 */
 
 #include <config.h>
-
+#include <opm/common/TimingMacros.hpp>
 #include <opm/common/OpmLog/OpmLog.hpp>
 #include <opm/common/ErrorMacros.hpp>
 #include <dune/common/timer.hh>

opm/simulators/linalg/bda/opencl/Preconditioner.cpp

@@ -19,7 +19,7 @@
 
 #include <config.h>
 #include <memory>
-
+#include <opm/common/TimingMacros.hpp>
 #include <opm/common/ErrorMacros.hpp>
 
 #include <opm/simulators/linalg/bda/opencl/BILU0.hpp>

opm/simulators/wells/MSWellHelpers.cpp

@@ -24,6 +24,7 @@
 #include <opm/simulators/wells/MSWellHelpers.hpp>
 
 #include <opm/common/ErrorMacros.hpp>
+#include <opm/common/TimingMacros.hpp>
 #include <opm/common/OpmLog/OpmLog.hpp>
 
 #include <opm/input/eclipse/Schedule/MSW/SICD.hpp>

opm/simulators/wells/MultisegmentWellEquations.cpp

@@ -20,11 +20,11 @@
 */
 
 #include <config.h>
-#include <opm/simulators/wells/MultisegmentWellEquations.hpp>
+#include <opm/common/TimingMacros.hpp>
 
+#include <opm/simulators/wells/MultisegmentWellEquations.hpp>
 #include <dune/istl/umfpack.hh>
 
-#include <opm/common/ErrorMacros.hpp>
 
 #include <opm/input/eclipse/Schedule/MSW/WellSegments.hpp>
 

opm/simulators/wells/StandardWellEquations.cpp

@@ -20,9 +20,11 @@
 */
 
 #include <config.h>
+#include <opm/common/Exceptions.hpp>
+#include <opm/common/TimingMacros.hpp>
 #include <opm/simulators/wells/StandardWellEquations.hpp>
 
-#include <opm/common/Exceptions.hpp>
+
 
 #include <opm/simulators/linalg/bda/WellContributions.hpp>
 #include <opm/simulators/linalg/istlsparsematrixadapter.hh>

opm/simulators/wells/StandardWellEquations.hpp

@@ -25,7 +25,7 @@
 
 #include <opm/simulators/utils/ParallelCommunication.hpp>
 #include <opm/simulators/wells/WellHelpers.hpp>
-
+#include <opm/common/TimingMacros.hpp>
 #include <dune/common/dynmatrix.hh>
 #include <dune/common/dynvector.hh>
 #include <dune/istl/bcrsmatrix.hh>

opm/simulators/wells/WellHelpers.cpp

@@ -20,7 +20,7 @@
 */
 
 #include <config.h>
-
+#include <opm/common/TimingMacros.hpp>
 #include <opm/simulators/wells/WellHelpers.hpp>
 
 #include <opm/common/OpmLog/OpmLog.hpp>