Merge pull request #1082 from su2code/feature_simd_numerics

Fix #1081

Author: talbring

Common/include/basic_types/datatype_structure.hpp

@@ -40,7 +40,7 @@
 #define FORCEINLINE inline
 #endif
 
-#if defined(__GNUC__) || defined(__clang__) || defined(__INTEL_COMPILER)
+#if defined(__GNUC__) || defined(__clang__)
 #define NEVERINLINE inline __attribute__((noinline))
 #else
 #define NEVERINLINE inline

Common/include/containers/C2DContainer.hpp

@@ -519,12 +519,12 @@ class C2DContainer :
   /*!
    * \brief Move ctor, implemented by base class (if fully static works as copy).
    */
-  C2DContainer(C2DContainer&&) noexcept(std::is_nothrow_move_constructible<Scalar>::value) = default;
+  C2DContainer(C2DContainer&&) = default;
 
   /*!
    * \brief Move assign operator, implemented by base class (if fully static works as copy).
    */
-  C2DContainer& operator= (C2DContainer&&) noexcept(std::is_nothrow_move_assignable<Scalar>::value) = default;
+  C2DContainer& operator= (C2DContainer&&) = default;
 
   /*!
    * \overload Set all entries to rhs value (syntax sugar, see "resize").

Common/include/mpi_structure.inl

@@ -32,73 +32,6 @@
 
 #ifdef HAVE_MPI
 
-NEVERINLINE void CBaseMPIWrapper::Error(std::string ErrorMsg, std::string FunctionName){
-
-  /* Set MinRankError to Rank, as the error message is called on this rank. */
-  MinRankError = Rank;
-  int flag = 0;
-
-#if MPI_VERSION >= 3
-  /* Find out whether the error call is collective via MPI_Ibarrier. */
-  Request barrierRequest;
-  MPI_Ibarrier(currentComm, &barrierRequest);
-
-  /* Try to complete the non-blocking barrier call for a second. */
-  double startTime = SU2_MPI::Wtime();
-  while( true ) {
-
-    MPI_Test(&barrierRequest, &flag, MPI_STATUS_IGNORE);
-    if( flag ) break;
-
-    double currentTime = SU2_MPI::Wtime();
-    if(currentTime > startTime + 1.0) break;
-  }
-#else
-  /* MPI_Ibarrier function is not supported. Simply wait for one
-     second to give other ranks the opportunity to reach this point. */
-#ifdef _MSC_VER
-  _sleep(1);
-#else
-  sleep(1);
-#endif
-#endif
-
-  if( flag ) {
-    /* The barrier is completed and hence the error call is collective.
-       Set MinRankError to 0. */
-    MinRankError = 0;
-  }
-  else {
-    /* The error call is not collective and the minimum rank must be
-       determined by one sided communication. Loop over the lower numbered
-       ranks to check if they participate in the error message. */
-    for(int i=0; i<Rank; ++i) {
-
-      int MinRankErrorOther;
-      MPI_Win_lock(MPI_LOCK_SHARED, i, 0, winMinRankError);
-      MPI_Get(&MinRankErrorOther, 1, MPI_INT, i, 0, 1, MPI_INT, winMinRankError);
-      MPI_Win_unlock(i, winMinRankError);
-
-      if(MinRankErrorOther < MinRankError) {
-        MinRankError = MinRankErrorOther;
-        break;
-      }
-    }
-  }
-
-  /* Check if this rank must write the error message and do so. */
-  if (Rank == MinRankError){
-    std::cout << std::endl << std::endl;
-    std::cout << "Error in \"" << FunctionName << "\": " << std::endl;
-    std::cout <<  "-------------------------------------------------------------------------" << std::endl;
-    std::cout << ErrorMsg << std::endl;
-    std::cout <<  "------------------------------ Error Exit -------------------------------" << std::endl;
-    std::cout << std::endl << std::endl;
-  }
-  Abort(currentComm, EXIT_FAILURE);
-}
-
-
 inline int CBaseMPIWrapper::GetRank(){
   return Rank;
 }
@@ -515,20 +448,9 @@ inline void CMediMPIWrapper::Waitany(int nrequests, Request *request,
                                  int *index, Status *status) {
   AMPI_Waitany(nrequests, request, index, status);
 }
-#endif
-#else // HAVE_MPI
+#endif // CODI
 
-NEVERINLINE void CBaseMPIWrapper::Error(std::string ErrorMsg, std::string FunctionName){
-  if (Rank == 0){
-    std::cout << std::endl << std::endl;
-    std::cout << "Error in \"" << FunctionName << "\": " << std::endl;
-    std::cout <<  "-------------------------------------------------------------------------" << std::endl;
-    std::cout << ErrorMsg << std::endl;
-    std::cout <<  "------------------------------ Error Exit -------------------------------" << std::endl;
-    std::cout << std::endl << std::endl;
-  }
-  Abort(currentComm, 0);
-}
+#else // HAVE_MPI
 
 inline int CBaseMPIWrapper::GetRank(){
   return Rank;

Common/src/CConfig.cpp

@@ -1814,7 +1814,7 @@ void CConfig::SetConfig_Options() {
   addBoolOption("USE_ACCURATE_FLUX_JACOBIANS", Use_Accurate_Jacobians, false);
   /*!\brief CENTRAL_JACOBIAN_FIX_FACTOR \n DESCRIPTION: Improve the numerical properties (diagonal dominance) of the global Jacobian matrix, 3 to 4 is "optimum" (central schemes) \ingroup Config*/
   addDoubleOption("CENTRAL_JACOBIAN_FIX_FACTOR", Cent_Jac_Fix_Factor, 4.0);
-  
+
   /*!\brief CONV_NUM_METHOD_ADJFLOW
    *  \n DESCRIPTION: Convective numerical method for the adjoint solver.
    *  \n OPTIONS:  See \link Upwind_Map \endlink , \link Centered_Map \endlink. Note: not all methods are guaranteed to be implemented for the adjoint solver. \ingroup Config */
@@ -4659,7 +4659,7 @@ void CConfig::SetPostprocessing(unsigned short val_software, unsigned short val_
        nRough_Wall = nWall;
        Roughness_Height = new su2double [nWall];
        Kind_Wall = new unsigned short [nWall];
-       for (unsigned short iMarker = 0; iMarker < nMarker_HeatFlux; iMarker++) {
+       for (iMarker = 0; iMarker < nMarker_HeatFlux; iMarker++) {
          Roughness_Height[iMarker] = 0.0;
          Kind_Wall[iMarker] = SMOOTH;
        }
@@ -5009,7 +5009,7 @@ void CConfig::SetPostprocessing(unsigned short val_software, unsigned short val_
   if (edgeColorGroupSize==0) edgeColorGroupSize = 1<<30;
 
   /*--- Specifying a deforming surface requires a mesh deformation solver. ---*/
-  if (GetSurface_Movement(DEFORMING)) Deform_Mesh = true; 
+  if (GetSurface_Movement(DEFORMING)) Deform_Mesh = true;
 
 }
 
@@ -5042,7 +5042,7 @@ void CConfig::SetMarkers(unsigned short val_software) {
   nMarker_CfgFile = nMarker_Euler + nMarker_FarField + nMarker_SymWall +
   nMarker_PerBound + nMarker_NearFieldBound + nMarker_Fluid_InterfaceBound +
   nMarker_CHTInterface + nMarker_Inlet + nMarker_Riemann + nMarker_Smoluchowski_Maxwell +
-  nMarker_Giles + nMarker_Outlet + nMarker_Isothermal + 
+  nMarker_Giles + nMarker_Outlet + nMarker_Isothermal +
   nMarker_HeatFlux +
   nMarker_EngineInflow + nMarker_EngineExhaust + nMarker_Internal +
   nMarker_Supersonic_Inlet + nMarker_Supersonic_Outlet + nMarker_Displacement + nMarker_Load +
@@ -5317,7 +5317,7 @@ void CConfig::SetMarkers(unsigned short val_software) {
     Marker_CfgFile_KindBC[iMarker_CfgFile] = ISOTHERMAL;
     iMarker_CfgFile++;
   }
- 
+
   for (iMarker_Smoluchowski_Maxwell = 0; iMarker_Smoluchowski_Maxwell < nMarker_Smoluchowski_Maxwell; iMarker_Smoluchowski_Maxwell++) {
     Marker_CfgFile_TagBound[iMarker_CfgFile] = Marker_Smoluchowski_Maxwell[iMarker_Smoluchowski_Maxwell];
     Marker_CfgFile_KindBC[iMarker_CfgFile] = SMOLUCHOWSKI_MAXWELL;
@@ -5581,14 +5581,14 @@ void CConfig::SetOutput(unsigned short val_software, unsigned short val_izone) {
         break;
       case NEMO_EULER:
         if (Kind_Regime == COMPRESSIBLE) cout << "Compressible two-temperature thermochemical non-equilibrium Euler equations." << endl;
-        if(Kind_FluidModel == USER_DEFINED_NONEQ){ 
+        if(Kind_FluidModel == USER_DEFINED_NONEQ){
           if ((GasModel != "N2") && (GasModel != "AIR-5"))
           SU2_MPI::Error("The GAS_MODEL given as input is not valid. Choose one of the options: N2, AIR-5.", CURRENT_FUNCTION);
         }
         break;
-      case NEMO_NAVIER_STOKES: 
+      case NEMO_NAVIER_STOKES:
         if (Kind_Regime == COMPRESSIBLE) cout << "Compressible two-temperature thermochemical non-equilibrium Navier-Stokes equations." << endl;
-        if(Kind_FluidModel == USER_DEFINED_NONEQ){  
+        if(Kind_FluidModel == USER_DEFINED_NONEQ){
           if ((GasModel != "N2") && (GasModel != "AIR-5"))
           SU2_MPI::Error("The GAS_MODEL given as input is not valid. Choose one of the options: N2, AIR-5.", CURRENT_FUNCTION);
         }
@@ -6384,7 +6384,7 @@ void CConfig::SetOutput(unsigned short val_software, unsigned short val_izone) {
 
     if ((Kind_Solver == EULER) || (Kind_Solver == NAVIER_STOKES) || (Kind_Solver == RANS) ||
         (Kind_Solver == INC_EULER) || (Kind_Solver == INC_NAVIER_STOKES) || (Kind_Solver == INC_RANS) ||
-        (Kind_Solver == NEMO_EULER) || (Kind_Solver == NEMO_NAVIER_STOKES) || 
+        (Kind_Solver == NEMO_EULER) || (Kind_Solver == NEMO_NAVIER_STOKES) ||
         (Kind_Solver == DISC_ADJ_INC_EULER) || (Kind_Solver == DISC_ADJ_INC_NAVIER_STOKES) || (Kind_Solver == DISC_ADJ_INC_RANS) ||
         (Kind_Solver == DISC_ADJ_EULER) || (Kind_Solver == DISC_ADJ_NAVIER_STOKES) || (Kind_Solver == DISC_ADJ_RANS) ||
         (Kind_Solver == DISC_ADJ_FEM_EULER) || (Kind_Solver == DISC_ADJ_FEM_NS) || (Kind_Solver == DISC_ADJ_FEM_RANS)) {
@@ -9790,7 +9790,7 @@ void CConfig::SetMultizone(CConfig *driver_config, CConfig **config_container){
     switch (config_container[iZone]->GetKind_Solver()) {
     case EULER: case NAVIER_STOKES: case RANS:
     case INC_EULER: case INC_NAVIER_STOKES: case INC_RANS:
-    case NEMO_EULER: case NEMO_NAVIER_STOKES:     
+    case NEMO_EULER: case NEMO_NAVIER_STOKES:
       fluid_zone = true;
       break;
     case FEM_ELASTICITY:

Common/src/mpi_structure.cpp

@@ -6,7 +6,7 @@
  *
  * SU2 Project Website: https://su2code.github.io
  *
- * The SU2 Project is maintained by the SU2 Foundation 
+ * The SU2 Project is maintained by the SU2 Foundation
  * (http://su2foundation.org)
  *
  * Copyright 2012-2020, SU2 Contributors (cf. AUTHORS.md)
@@ -35,6 +35,85 @@ CBaseMPIWrapper::Comm CBaseMPIWrapper::currentComm = MPI_COMM_WORLD;
 int  CBaseMPIWrapper::MinRankError;
 bool CBaseMPIWrapper::winMinRankErrorInUse = false;
 CBaseMPIWrapper::Win CBaseMPIWrapper::winMinRankError;
+
+void CBaseMPIWrapper::Error(std::string ErrorMsg, std::string FunctionName){
+
+  /* Set MinRankError to Rank, as the error message is called on this rank. */
+  MinRankError = Rank;
+  int flag = 0;
+
+#if MPI_VERSION >= 3
+  /* Find out whether the error call is collective via MPI_Ibarrier. */
+  Request barrierRequest;
+  MPI_Ibarrier(currentComm, &barrierRequest);
+
+  /* Try to complete the non-blocking barrier call for a second. */
+  double startTime = SU2_MPI::Wtime();
+  while( true ) {
+
+    MPI_Test(&barrierRequest, &flag, MPI_STATUS_IGNORE);
+    if( flag ) break;
+
+    double currentTime = SU2_MPI::Wtime();
+    if(currentTime > startTime + 1.0) break;
+  }
+#else
+  /* MPI_Ibarrier function is not supported. Simply wait for one
+     second to give other ranks the opportunity to reach this point. */
+#ifdef _MSC_VER
+  _sleep(1);
+#else
+  sleep(1);
+#endif
+#endif
+
+  if( flag ) {
+    /* The barrier is completed and hence the error call is collective.
+       Set MinRankError to 0. */
+    MinRankError = 0;
+  }
+  else {
+    /* The error call is not collective and the minimum rank must be
+       determined by one sided communication. Loop over the lower numbered
+       ranks to check if they participate in the error message. */
+    for(int i=0; i<Rank; ++i) {
+
+      int MinRankErrorOther;
+      MPI_Win_lock(MPI_LOCK_SHARED, i, 0, winMinRankError);
+      MPI_Get(&MinRankErrorOther, 1, MPI_INT, i, 0, 1, MPI_INT, winMinRankError);
+      MPI_Win_unlock(i, winMinRankError);
+
+      if(MinRankErrorOther < MinRankError) {
+        MinRankError = MinRankErrorOther;
+        break;
+      }
+    }
+  }
+
+  /* Check if this rank must write the error message and do so. */
+  if (Rank == MinRankError){
+    std::cout << std::endl << std::endl;
+    std::cout << "Error in \"" << FunctionName << "\": " << std::endl;
+    std::cout <<  "-------------------------------------------------------------------------" << std::endl;
+    std::cout << ErrorMsg << std::endl;
+    std::cout <<  "------------------------------ Error Exit -------------------------------" << std::endl;
+    std::cout << std::endl << std::endl;
+  }
+  Abort(currentComm, EXIT_FAILURE);
+}
+#else // HAVE_MPI
+
+void CBaseMPIWrapper::Error(std::string ErrorMsg, std::string FunctionName){
+  if (Rank == 0){
+    std::cout << std::endl << std::endl;
+    std::cout << "Error in \"" << FunctionName << "\": " << std::endl;
+    std::cout <<  "-------------------------------------------------------------------------" << std::endl;
+    std::cout << ErrorMsg << std::endl;
+    std::cout <<  "------------------------------ Error Exit -------------------------------" << std::endl;
+    std::cout << std::endl << std::endl;
+  }
+  Abort(currentComm, 0);
+}
 #endif
 
 #ifdef HAVE_MPI
@@ -43,4 +122,4 @@ MediTypes* mediTypes;
 #include <medi/medi.cpp>
 #endif // defined CODI_REVERSE_TYPE || defined CODI_FORWARD_TYPE
 
-#endif// HAVE_MPI
+#endif // HAVE_MPI

SU2_CFD/include/numerics/CNumerics.hpp

@@ -227,7 +227,7 @@ class CNumerics {
 
   /* Supporting data structures for the eigenspace perturbation for UQ methodology */
   su2double **A_ij, **newA_ij, **Eig_Vec, **New_Eig_Vec, **Corners;
-  su2double *Eig_Val, *Barycentric_Coord, *New_Coord;  
+  su2double *Eig_Val, *Barycentric_Coord, *New_Coord;
 
 public:
   /*!
@@ -777,7 +777,7 @@ class CNumerics {
   void GetInviscidProjFlux(const su2double *val_density, const su2double *val_velocity,
                            const su2double *val_pressure, const su2double *val_enthalpy,
                            const su2double *val_normal, su2double *val_Proj_Flux) const;
-  
+
   /*!
    * \brief Compute the projected inviscid flux vector for incompresible simulations
    * \param[in] val_density - Pointer to the density.
@@ -1168,7 +1168,7 @@ class CNumerics {
    * \return A lightweight const-view (read-only) of the residual/flux and Jacobians.
    */
   inline virtual ResidualType<> ComputeVibRelaxation(const CConfig* config) { return ResidualType<>(nullptr,nullptr,nullptr); }
-  
+
   /*!
    * \brief Calculation of the chemistry source term
    * \param[in] config - Definition of the particular problem.
@@ -1381,15 +1381,15 @@ class CNumerics {
    */
   static void tql2(su2double **V, su2double *d, su2double *e, unsigned short n);
 
-  virtual inline void SetdPdU(su2double *val_dPdU_i, su2double *val_dPdU_j)       { } 
-        
-  virtual inline void SetdTdU(su2double *val_dTdU_i, su2double *val_dTdU_j)       { }   
-  
-  virtual inline void SetdTvedU(su2double *val_dTvedU_i, su2double *val_dTvedU_j) { }  
-  
-  virtual inline void SetEve(su2double *val_Eve_i, su2double *val_Eve_j)          { }  
-  
-  virtual inline void SetCvve(su2double *val_Cvve_i, su2double *val_Cvve_j)       { }  
+  virtual inline void SetdPdU(su2double *val_dPdU_i, su2double *val_dPdU_j)       { }
+
+  virtual inline void SetdTdU(su2double *val_dTdU_i, su2double *val_dTdU_j)       { }
+
+  virtual inline void SetdTvedU(su2double *val_dTvedU_i, su2double *val_dTvedU_j) { }
+
+  virtual inline void SetEve(su2double *val_Eve_i, su2double *val_Eve_j)          { }
+
+  virtual inline void SetCvve(su2double *val_Cvve_i, su2double *val_Cvve_j)       { }
 
 };
 

SU2_CFD/include/numerics_simd/flow/convection/centered.hpp

@@ -175,8 +175,7 @@ class CCenteredBase : public Base {
 
     /*--- Stop preaccumulation. ---*/
 
-    AD::SetPreaccOut(flux, nVar, Double::Size);
-    AD::EndPreacc();
+    stopPreacc(flux);
 
     /*--- Update the vector and system matrix. ---*/
 

SU2_CFD/include/numerics_simd/flow/convection/roe.hpp

@@ -204,8 +204,7 @@ class CRoeBase : public Base {
 
     /*--- Stop preaccumulation. ---*/
 
-    AD::SetPreaccOut(flux, nVar, Double::Size);
-    AD::EndPreacc();
+    stopPreacc(flux);
 
     /*--- Update the vector and system matrix. ---*/
 

SU2_CFD/include/numerics_simd/util.hpp

@@ -145,6 +145,15 @@ FORCEINLINE MatrixDbl<nRows,nCols> gatherVariables(Int iPoint, const Container&
   return x;
 }
 
+/*!
+ * \brief Stop the AD preaccumulation.
+ */
+template<size_t nVar>
+FORCEINLINE void stopPreacc(VectorDbl<nVar>& x) {
+  AD::SetPreaccOut(x, nVar, Double::Size);
+  AD::EndPreacc();
+}
+
 /*!
  * \brief Distance vector, from point i to point j.
  */

externals/cgns/meson.build

@@ -1,4 +1,4 @@
-if build_machine.system() == 'windows'
+if build_machine.system() == 'windows' or meson.get_compiler('cpp').get_id() == 'intel'
   cgns_default_warnings = []
 else
   cgns_default_warnings = ['-Wno-unused-result']