fix walltime output nompi + omp, some simd optimization

Author: pcarruscag

Common/include/linear_algebra/CSysMatrix.hpp

@@ -29,6 +29,7 @@
 #pragma once
 
 #include "../../include/mpi_structure.hpp"
+#include "../../include/omp_structure.hpp"
 #include "CSysVector.hpp"
 #include "CPastixWrapper.hpp"
 
@@ -169,58 +170,60 @@ class CSysMatrix {
    * \param[in] vector
    * \param[out] product
    */
-  inline void MatrixVectorProduct(const ScalarType *matrix, const ScalarType *vector, ScalarType *product) const;
+  void MatrixVectorProduct(const ScalarType *matrix, const ScalarType *vector, ScalarType *product) const;
 
   /*!
    * \brief Calculates the matrix-vector product: product += matrix*vector
    * \param[in] matrix
    * \param[in] vector
    * \param[in,out] product
    */
-  inline void MatrixVectorProductAdd(const ScalarType *matrix, const ScalarType *vector, ScalarType *product) const;
+  void MatrixVectorProductAdd(const ScalarType *matrix, const ScalarType *vector, ScalarType *product) const;
 
   /*!
    * \brief Calculates the matrix-vector product: product -= matrix*vector
    * \param[in] matrix
    * \param[in] vector
    * \param[in,out] product
    */
-  inline void MatrixVectorProductSub(const ScalarType *matrix, const ScalarType *vector, ScalarType *product) const;
+  void MatrixVectorProductSub(const ScalarType *matrix, const ScalarType *vector, ScalarType *product) const;
 
   /*!
    * \brief Calculates the matrix-vector product: product += matrix^T * vector
    * \param[in] matrix
    * \param[in] vector
    * \param[in,out] product
    */
-  inline void MatrixVectorProductTransp(const ScalarType *matrix, const ScalarType *vector, ScalarType *product) const;
+  void MatrixVectorProductTransp(const ScalarType *matrix, const ScalarType *vector, ScalarType *product) const;
 
   /*!
    * \brief Calculates the matrix-matrix product
    */
-  inline void MatrixMatrixProduct(const ScalarType *matrix_a, const ScalarType *matrix_b, ScalarType *product) const;
+  void MatrixMatrixProduct(const ScalarType *matrix_a, const ScalarType *matrix_b, ScalarType *product) const;
 
   /*!
    * \brief Subtract b from a and store the result in c.
    */
-  inline void VectorSubtraction(const ScalarType *a, const ScalarType *b, ScalarType *c) const {
+  FORCEINLINE void VectorSubtraction(const ScalarType *a, const ScalarType *b, ScalarType *c) const {
     for(unsigned long iVar = 0; iVar < nVar; iVar++)
       c[iVar] = a[iVar] - b[iVar];
   }
 
   /*!
    * \brief Subtract b from a and store the result in c.
    */
-  inline void MatrixSubtraction(const ScalarType *a, const ScalarType *b, ScalarType *c) const {
+  FORCEINLINE void MatrixSubtraction(const ScalarType *a, const ScalarType *b, ScalarType *c) const {
+    SU2_OMP_SIMD
     for(unsigned long iVar = 0; iVar < nVar*nEqn; iVar++)
       c[iVar] = a[iVar] - b[iVar];
   }
 
   /*!
    * \brief Copy matrix src into dst, transpose if required.
    */
-  inline void MatrixCopy(const ScalarType *src, ScalarType *dst, bool transposed = false) const {
+  FORCEINLINE void MatrixCopy(const ScalarType *src, ScalarType *dst, bool transposed = false) const {
     if (!transposed) {
+      SU2_OMP_SIMD
       for(auto iVar = 0ul; iVar < nVar*nEqn; ++iVar)
         dst[iVar] = src[iVar];
     }
@@ -446,6 +449,7 @@ class CSysMatrix {
 
     auto mat_ij = GetBlock(block_i, block_j);
     if (!mat_ij) return;
+    SU2_OMP_SIMD
     for (auto iVar = 0ul; iVar < nVar*nEqn; ++iVar) {
       mat_ij[iVar] = (Overwrite? ScalarType(0) : mat_ij[iVar]) + PassiveAssign(alpha * val_block[iVar]);
     }
@@ -657,6 +661,7 @@ class CSysMatrix {
     unsigned long iVar, index = dia_ptr[block_i]*nVar*nVar;
 
     /*--- Clear entire block before setting its diagonal. ---*/
+    SU2_OMP_SIMD
     for (iVar = 0; iVar < nVar*nVar; iVar++)
       matrix[index+iVar] = 0.0;
 

Common/include/linear_algebra/CSysMatrix.inl

@@ -47,18 +47,15 @@ FORCEINLINE void CSysMatrix<ScalarType>::SetBlock_ILUMatrix(unsigned long block_
 
   auto ilu_ij = GetBlock_ILUMatrix(block_i, block_j);
   if (!ilu_ij) return;
-  for (auto iVar = 0ul; iVar < nVar*nVar; ++iVar)
-    ilu_ij[iVar] = val_block[iVar];
+  MatrixCopy(val_block, ilu_ij, false);
 }
 
 template<class ScalarType>
 FORCEINLINE void CSysMatrix<ScalarType>::SetBlockTransposed_ILUMatrix(unsigned long block_i, unsigned long block_j, ScalarType *val_block) {
 
   auto ilu_ij = GetBlock_ILUMatrix(block_i, block_j);
   if (!ilu_ij) return;
-  for (auto iVar = 0ul; iVar < nVar; iVar++)
-    for (auto jVar = 0ul; jVar < nVar; jVar++)
-      ilu_ij[iVar*nVar+jVar] = val_block[jVar*nVar+iVar];
+  MatrixCopy(val_block, ilu_ij, true);
 }
 
 template<class T, bool alpha, bool beta, bool transp>

Common/include/mpi_structure.inl

@@ -512,7 +512,11 @@ inline void CMediMPIWrapper::Waitany(int nrequests, Request *request,
 }
 #endif
 #else // HAVE_MPI
+#ifdef _OPENMP
+#include <omp.h>
+#else
 #include <ctime>
+#endif
 
 inline void CBaseMPIWrapper::Error(std::string ErrorMsg, std::string FunctionName){
   if (Rank == 0){
@@ -688,6 +692,10 @@ inline void CBaseMPIWrapper::CopyData(void *sendbuf, void *recvbuf, int size, Da
 }
 
 inline passivedouble CBaseMPIWrapper::Wtime(void) {
+#ifdef _OPENMP
+  return omp_get_wtime();
+#else
   return passivedouble(clock()) / CLOCKS_PER_SEC;
+#endif
 }
 #endif

Common/include/omp_structure.hpp

@@ -110,6 +110,11 @@ inline void omp_destroy_lock(omp_lock_t*){}
 #define SU2_OMP_FOR_DYN(CHUNK) SU2_OMP(for schedule(dynamic,CHUNK))
 #define SU2_OMP_FOR_STAT(CHUNK) SU2_OMP(for schedule(static,CHUNK))
 
+/*--- Disable some unsupported features on MSVC. ---*/
+#if defined(_MSC_VER)
+#undef SU2_OMP_SIMD
+#define SU2_OMP_SIMD
+#endif
 
 /*--- Convenience functions (e.g. to compute chunk sizes). ---*/
 

SU2_CFD/include/output/CMultizoneOutput.hpp

@@ -37,8 +37,6 @@
 #endif
 #include <fstream>
 #include <cmath>
-#include <time.h>
-#include <fstream>
 
 #include "COutput.hpp"
 #include "../../../Common/include/CConfig.hpp"

SU2_CFD/include/output/COutput.hpp

@@ -30,7 +30,6 @@
 #include <fstream>
 #include <cmath>
 #include <map>
-#include <time.h>
 #include <sstream>
 #include <iomanip>
 #include <limits>

SU2_CFD/include/output/COutputLegacy.hpp

@@ -38,8 +38,6 @@
 #endif
 #include <fstream>
 #include <cmath>
-#include <time.h>
-#include <fstream>
 #include <vector>
 
 #include "../../../Common/include/option_structure.hpp"

SU2_CFD/include/output/filewriter/CFileWriter.hpp

@@ -33,7 +33,6 @@
 #include <string>
 #include <cstring>
 #include <fstream>
-#include <time.h>
 
 #include "../../output/filewriter/CParallelDataSorter.hpp"
 
@@ -83,7 +82,7 @@ class CFileWriter{
    * \brief The parallel data sorter
    */
   CParallelDataSorter* dataSorter;
-  
+
 #ifdef HAVE_MPI
   /*!
    * \brief The displacement that every process has in the current file view
@@ -143,9 +142,9 @@ class CFileWriter{
    * \return The time used to write to file.
    */
   su2double Get_UsedTime() const {return usedTime;}
-  
+
 protected:
-  
+
   /*!
    * \brief Collectively write a binary data array distributed over all processors to file using MPI I/O.
    * \param[in] data - Pointer to the data to write.
@@ -165,27 +164,27 @@ class CFileWriter{
    * \return Boolean indicating whether the writing was successful.
    */
   bool WriteMPIBinaryData(const void *data, unsigned long sizeInBytes, unsigned short processor);
-  
+
   /*!
    * \brief Write a string to a currently opened file using MPI I/O. Note: routine must be called collectively,
    *  although only one processor writes the string.
    * \param[in] str - The string to write to file.
    * \param[in] processor - Rank of the processor that should the string.
-   * \return 
+   * \return
    */
   bool WriteMPIString(const std::string& str, unsigned short processor);
-  
+
   /*!
    * \brief Open a file to write using MPI I/O. Already existing file is deleted.
    * \return Boolean indicating whether the opening was successful.
    */
   bool OpenMPIFile();
-  
+
   /*!
-   * \brief Close a file using MPI I/O. 
+   * \brief Close a file using MPI I/O.
    * \return Boolean indicating whether the closing was successful.
    */
   bool CloseMPIFile();
-  
+
 };
 

SU2_CFD/src/solvers/CFEASolver.cpp

@@ -1959,7 +1959,7 @@ void CFEASolver::Postprocessing(CGeometry *geometry, CSolver **solver_container,
 
     SU2_OMP_PARALLEL
     {
-#if !defined(CODI_REVERSE_TYPE) && !defined(USE_MIXED_PRECISION)
+#if !(defined(CODI_REVERSE_TYPE) || defined(USE_MIXED_PRECISION)) || defined(CODI_FORWARD_TYPE)
     Jacobian.ComputeResidual(LinSysSol, LinSysRes, LinSysAux);
 #else
     sol.PassiveCopy(LinSysSol);