race condition, reduce testcase duration, cleanup

Author: pcarruscag

Common/include/CConfig.hpp

@@ -28,7 +28,7 @@
 
 #pragma once
 
-#include "./parallelization/mpi_structure.hpp"
+#include "parallelization/mpi_structure.hpp"
 
 #include <iostream>
 #include <cstdlib>

Common/include/linear_algebra/CSysMatrix.hpp

@@ -28,9 +28,7 @@
 
 #pragma once
 
-#include "../../include/parallelization/mpi_structure.hpp"
-#include "../../include/parallelization/omp_structure.hpp"
-#include "../../include/parallelization/vectorization.hpp"
+#include "../../include/CConfig.hpp"
 #include "CSysVector.hpp"
 #include "CPastixWrapper.hpp"
 
@@ -75,7 +73,6 @@ struct mkl_jit_wrapper<float> {
 #endif
 #endif
 
-class CConfig;
 class CGeometry;
 
 struct CSysMatrixComms {
@@ -88,7 +85,8 @@ struct CSysMatrixComms {
    * \param[in] commType - Enumerated type for the quantity to be communicated.
    */
   template<class T>
-  static void Initiate(const CSysVector<T>& x, CGeometry *geometry, const CConfig *config, unsigned short commType);
+  static void Initiate(const CSysVector<T>& x, CGeometry *geometry, const CConfig *config,
+                       unsigned short commType = SOLUTION_MATRIX);
 
   /*!
    * \brief Routine to complete the set of non-blocking communications launched by
@@ -99,7 +97,8 @@ struct CSysMatrixComms {
    * \param[in] commType - Enumerated type for the quantity to be unpacked.
    */
   template<class T>
-  static void Complete(CSysVector<T>& x, CGeometry *geometry, const CConfig *config, unsigned short commType);
+  static void Complete(CSysVector<T>& x, CGeometry *geometry, const CConfig *config,
+                       unsigned short commType = SOLUTION_MATRIX);
 };
 
 /*!

Common/src/grid_movement/CVolumetricMovement.cpp

@@ -168,11 +168,11 @@ void CVolumetricMovement::SetVolume_Deformation(CGeometry *geometry, CConfig *co
      so that all nodes have the same solution and r.h.s. entries
      across all partitions. ---*/
 
-    CSysMatrixComms::Initiate(LinSysSol, geometry, config, SOLUTION_MATRIX);
-    CSysMatrixComms::Complete(LinSysSol, geometry, config, SOLUTION_MATRIX);
+    CSysMatrixComms::Initiate(LinSysSol, geometry, config);
+    CSysMatrixComms::Complete(LinSysSol, geometry, config);
 
-    CSysMatrixComms::Initiate(LinSysRes, geometry, config, SOLUTION_MATRIX);
-    CSysMatrixComms::Complete(LinSysRes, geometry, config, SOLUTION_MATRIX);
+    CSysMatrixComms::Initiate(LinSysRes, geometry, config);
+    CSysMatrixComms::Complete(LinSysRes, geometry, config);
 
     /*--- Definition of the preconditioner matrix vector multiplication, and linear solver ---*/
 

Common/src/linear_algebra/CSysMatrix.cpp

@@ -28,8 +28,6 @@
 #include "../../include/linear_algebra/CSysMatrix.inl"
 
 #include "../../include/geometry/CGeometry.hpp"
-#include "../../include/CConfig.hpp"
-#include "../../include/parallelization/omp_structure.hpp"
 #include "../../include/toolboxes/allocation_toolbox.hpp"
 
 #include <cmath>

SU2_CFD/src/integration/CNewtonIntegration.cpp

@@ -222,10 +222,11 @@ void CNewtonIntegration::MultiGrid_Iteration(CGeometry ****geometry_, CSolver **
   bool endStartup = false;
 
   if (startupPeriod) {
-    SU2_OMP_MASTER
-    firstResidual = max(firstResidual, residual);
+    SU2_OMP_MASTER {
+      firstResidual = max(firstResidual, residual);
+      if (startupIters) startupIters -= 1;
+    }
     SU2_OMP_BARRIER
-    if (startupIters) startupIters -= 1;
     endStartup = (startupIters == 0) && (residual - firstResidual < startupResidual);
   }
 
@@ -328,8 +329,8 @@ void CNewtonIntegration::MatrixFreeProduct(const CSysVector<Scalar>& u, CSysVect
     }
   }
 
-  CSysMatrixComms::Initiate(v, geometry, config, SOLUTION_MATRIX);
-  CSysMatrixComms::Complete(v, geometry, config, SOLUTION_MATRIX);
+  CSysMatrixComms::Initiate(v, geometry, config);
+  CSysMatrixComms::Complete(v, geometry, config);
 }
 
 void CNewtonIntegration::Preconditioner(const CSysVector<Scalar>& u, CSysVector<Scalar>& v) const {
@@ -350,7 +351,7 @@ void CNewtonIntegration::Preconditioner(const CSysVector<Scalar>& u, CSysVector<
         v(iPoint,iVar) = SU2_TYPE::GetValue(delta) * u(iPoint,iVar);
     }
 
-    CSysMatrixComms::Initiate(v, geometry, config, SOLUTION_MATRIX);
-    CSysMatrixComms::Complete(v, geometry, config, SOLUTION_MATRIX);
+    CSysMatrixComms::Initiate(v, geometry, config);
+    CSysMatrixComms::Complete(v, geometry, config);
   }
 }

SU2_CFD/src/solvers/CFEASolver.cpp

@@ -2544,8 +2544,8 @@ void CFEASolver::GeneralizedAlpha_UpdateLoads(const CGeometry *geometry, const C
 void CFEASolver::Solve_System(CGeometry *geometry, CConfig *config) {
 
   /*--- Enforce solution at some halo points possibly not covered by essential BC markers. ---*/
-  CSysMatrixComms::Initiate(LinSysSol, geometry, config, SOLUTION_MATRIX);
-  CSysMatrixComms::Complete(LinSysSol, geometry, config, SOLUTION_MATRIX);
+  CSysMatrixComms::Initiate(LinSysSol, geometry, config);
+  CSysMatrixComms::Complete(LinSysSol, geometry, config);
 
   for (auto iPoint : ExtraVerticesToEliminate) {
     Jacobian.EnforceSolutionAtNode(iPoint, LinSysSol.GetBlock(iPoint), LinSysRes);

TestCases/hybrid_regression.py

@@ -182,8 +182,8 @@ def main():
     turb_oneram6_nk           = TestCase('turb_oneram6_nk')
     turb_oneram6_nk.cfg_dir   = "rans/oneram6"
     turb_oneram6_nk.cfg_file  = "turb_ONERAM6_nk.cfg"
-    turb_oneram6_nk.test_iter = 100
-    turb_oneram6_nk.test_vals = [-7.015278, -6.587369, -10.394193, 0.271661, 0.019845, 4, -0.626403, 2.8101e+02]
+    turb_oneram6_nk.test_iter = 20
+    turb_oneram6_nk.test_vals = [-4.893470, -4.511977, -11.437109, 0.221926, 0.045766, 2, -0.893674, 31.384]
     test_list.append(turb_oneram6_nk)
 
     # NACA0012 (SA, FUN3D finest grid results: CL=1.0983, CD=0.01242)

TestCases/rans/oneram6/turb_ONERAM6_nk.cfg

@@ -28,8 +28,8 @@ LINEAR_SOLVER_ERROR= 0.25
 % The tolerance for NK iterations is initially relaxed by factor "ft", and reaches
 % LINEAR_SOLVER_ERROR after "rf" residual reduction (additional to "r0").
 % The Jacobian-free products are based on finite differences with step "e".
-NEWTON_KRYLOV_IPARAM= (0, 3, 2) % n0, np, ft
-NEWTON_KRYLOV_DPARAM= (-1.0, 0.1, -6.0, 1e-5) % r0, tp, rf, e
+NEWTON_KRYLOV_IPARAM= (10, 3, 2) % n0, np, ft
+NEWTON_KRYLOV_DPARAM= (1.0, 0.1, -6.0, 1e-5) % r0, tp, rf, e
 
 CFL_ADAPT= YES % it's needed
 CFL_NUMBER= 10

config_template.cfg

@@ -1093,6 +1093,8 @@ CENTRAL_JACOBIAN_FIX_FACTOR= 4.0
 TIME_DISCRE_FLOW= EULER_IMPLICIT
 %
 % Use a Newton-Krylov method on the flow equations, see TestCases/rans/oneram6/turb_ONERAM6_nk.cfg
+% For multizone discrete adjoint it will use FGMRES on inner iterations with restart frequency
+% equal to "QUASI_NEWTON_NUM_SAMPLES".
 NEWTON_KRYLOV= NO
 
 % ------------------- FEM FLOW NUMERICAL METHOD DEFINITION --------------------%