Merge pull request #1149 from su2code/more_vectorization

Support for UQ and NICF with vectorized (SIMD) centered schemes

Author: pcarruscag

Common/include/grid_movement/CFreeFormDefBox.hpp

@@ -756,13 +756,4 @@ class CFreeFormDefBox : public CGridMovement {
    */
   inline unsigned short GetLevel() const { return Level; }
 
-  /*!
-   * \brief Compute the determinant of a 3 by 3 matrix.
-   * \param[in] val_matrix 3 by 3 matrix.
-   * \result Determinant of the matrix
-   */
-  inline su2double Determinant_3x3(su2double A00, su2double A01, su2double A02, su2double A10, su2double A11, su2double A12, su2double A20, su2double A21, su2double A22) {
-    return A00*(A11*A22-A12*A21) - A01*(A10*A22-A12*A20) + A02*(A10*A21-A11*A20);
-  }
-
 };

Common/include/grid_movement/CVolumetricMovement.hpp

@@ -361,26 +361,6 @@ class CVolumetricMovement : public CGridMovement {
    */
   void UpdateGridCoord_Derivatives(CGeometry *geometry, CConfig *config);
 
-  /*!
-   * \brief Compute the determinant of a 3 by 3 matrix.
-   *        3 by 3 matrix elements
-   * \param[in] A00
-   * \param[in] A01
-   * \param[in] A02
-   * \param[in] A10
-   * \param[in] A11
-   * \param[in] A12
-   * \param[in] A20
-   * \param[in] A21
-   * \param[in] A22
-   * \result Determinant of the matrix
-   */
-  inline su2double Determinant_3x3(su2double A00, su2double A01, su2double A02, su2double A10, su2double A11, su2double A12, su2double A20, su2double A21, su2double A22) {
-    return A00*(A11*A22-A12*A21) - A01*(A10*A22-A12*A20) + A02*(A10*A21-A11*A20);
-  }
-
-
-
   /*!
    * \brief Store the number of iterations when moving the mesh.
    * \param[in] val_nIterMesh - Number of iterations.

Common/include/linear_algebra/CSysVector.hpp

@@ -67,8 +67,6 @@ class CSysVector : public VecExpr::CVecExpr<CSysVector<ScalarType>, ScalarType>
   unsigned long nElm = 0;          /*!< \brief Total number of elements (or number elements on this processor). */
   unsigned long nElmDomain = 0;    /*!< \brief Total number of elements without Ghost cells. */
   unsigned long nVar = 0;          /*!< \brief Number of elements in a block. */
-  mutable ScalarType dotRes = 0.0; /*!< \brief Result of dot product. to perform a reduction with OpenMP the
-                                               variable needs to be declared outside the parallel region. */
 
   /*!
    * \brief Generic initialization from a scalar or array.
@@ -101,7 +99,7 @@ class CSysVector : public VecExpr::CVecExpr<CSysVector<ScalarType>, ScalarType>
   /*!
    * \brief Default constructor of the class.
    */
-  CSysVector() {}
+  CSysVector() = default;
 
   /*!
    * \brief Destructor
@@ -291,8 +289,10 @@ class CSysVector : public VecExpr::CVecExpr<CSysVector<ScalarType>, ScalarType>
    */
   template <class T>
   ScalarType dot(const VecExpr::CVecExpr<T, ScalarType>& expr) const {
+    static ScalarType dotRes;
     /*--- All threads get the same "view" of the vectors and shared variable. ---*/
     SU2_OMP_BARRIER
+    SU2_OMP_MASTER
     dotRes = 0.0;
     SU2_OMP_BARRIER