[WIP] Add the Paraview parallel output format (.pvtu with one piece per rank)

.github/workflows/regression.yml

@@ -235,7 +235,7 @@ jobs:
         uses: docker://ghcr.io/su2code/su2/test-su2:261011-1107
         with:
           # -t <Tutorials-branch> -c <Testcases-branch>
-          args: -b ${{github.ref}} -t develop -c develop -s ${{matrix.testscript}}
+          args: -b ${{github.ref}} -t develop -c paraviewParallelOutput -s ${{matrix.testscript}}
       - name: Cleanup
         uses: docker://ghcr.io/su2code/su2/test-su2:261011-1107
         with:
@@ -284,7 +284,7 @@ jobs:
         uses: docker://ghcr.io/su2code/su2/test-su2:261011-1107
         with:
           # -t <Tutorials-branch> -c <Testcases-branch>
-          args: -b ${{github.ref}} -t develop -c develop -s ${{matrix.testscript}} -a "--tapetests"
+          args: -b ${{github.ref}} -t develop -c paraviewParallelOutput -s ${{matrix.testscript}} -a "--tapetests"
       - name: Cleanup
         uses: docker://ghcr.io/su2code/su2/test-su2:261011-1107
         with:
@@ -334,7 +334,7 @@ jobs:
           PMIX_MCA_gds: hash
         with:
           # -t <Tutorials-branch> -c <Testcases-branch>
-          args: -b ${{github.ref}} -t develop -c develop -s ${{matrix.testscript}} -a "--tsan"
+          args: -b ${{github.ref}} -t develop -c paraviewParallelOutput -s ${{matrix.testscript}} -a "--tsan"
       - name: Cleanup
         uses: docker://ghcr.io/su2code/su2/test-su2-tsan:261011-1107
         with:
@@ -379,7 +379,7 @@ jobs:
         uses: docker://ghcr.io/su2code/su2/test-su2-asan:261011-1107
         with:
           # -t <Tutorials-branch> -c <Testcases-branch>
-          args: -b ${{github.ref}} -t develop -c develop -s ${{matrix.testscript}} -a "--asan"
+          args: -b ${{github.ref}} -t develop -c paraviewParallelOutput -s ${{matrix.testscript}} -a "--asan"
       - name: Cleanup
         uses: docker://ghcr.io/su2code/su2/test-su2-asan:261011-1107
         with:

Common/include/option_structure.hpp

@@ -2408,6 +2408,8 @@ enum class OUTPUT_TYPE {
   PARAVIEW_XML,            /*!< \brief Paraview XML with binary data format */
   SURFACE_PARAVIEW_XML,    /*!< \brief Surface Paraview XML with binary data format */
   PARAVIEW_MULTIBLOCK,     /*!< \brief Paraview XML Multiblock */
+  PARAVIEW_PARALLEL,       /*!< \brief Paraview parallel XML (.pvtu), one piece per rank. */
+  SURFACE_PARAVIEW_PARALLEL, /*!< \brief Surface Paraview parallel XML (.pvtu), one piece per rank. */
   CGNS,                    /*!< \brief CGNS format. */
   SURFACE_CGNS,            /*!< \brief CGNS format. */
   STL_ASCII,               /*!< \brief STL ASCII format for surface solution output. */
@@ -2427,6 +2429,8 @@ static const MapType<std::string, OUTPUT_TYPE> Output_Map = {
   MakePair("PARAVIEW", OUTPUT_TYPE::PARAVIEW_XML)
   MakePair("SURFACE_PARAVIEW", OUTPUT_TYPE::SURFACE_PARAVIEW_XML)
   MakePair("PARAVIEW_MULTIBLOCK", OUTPUT_TYPE::PARAVIEW_MULTIBLOCK)
+  MakePair("PARAVIEW_PARALLEL", OUTPUT_TYPE::PARAVIEW_PARALLEL)
+  MakePair("SURFACE_PARAVIEW_PARALLEL", OUTPUT_TYPE::SURFACE_PARAVIEW_PARALLEL)
   MakePair("MESH", OUTPUT_TYPE::MESH)
   MakePair("MESH_BINARY", OUTPUT_TYPE::MESH_BINARY)
   MakePair("RESTART_ASCII", OUTPUT_TYPE::RESTART_ASCII)
@@ -2449,6 +2453,8 @@ inline bool isParaview(OUTPUT_TYPE format) {
     case OUTPUT_TYPE::PARAVIEW_XML:
     case OUTPUT_TYPE::SURFACE_PARAVIEW_XML:
     case OUTPUT_TYPE::PARAVIEW_MULTIBLOCK:
+    case OUTPUT_TYPE::PARAVIEW_PARALLEL:
+    case OUTPUT_TYPE::SURFACE_PARAVIEW_PARALLEL:
       return true;
     default:
       return false;

Common/include/parallelization/mpi_structure.cpp

@@ -187,7 +187,7 @@ void CBaseMPIWrapper<ScalarType>::CopyData(const void* sendbuf, void* recvbuf, i
 }
 
 template class CBaseMPIWrapper<su2double>;
-#if defined CODI_REVERSE_TYPE
+#if defined(CODI_REVERSE_TYPE) || defined(CODI_FORWARD_TYPE)
 template class CBaseMPIWrapper<passivedouble>;
 #endif
 #if defined(USE_MIXED_PRECISION) && !defined(USE_SINGLE_PRECISION)

Common/include/parallelization/mpi_structure.hpp

@@ -628,7 +628,7 @@ struct SelectMPIWrapper {
 };
 
 /*--- In AD we specialize for the passive wrapper. ---*/
-#if defined CODI_REVERSE_TYPE
+#if defined(CODI_REVERSE_TYPE) || defined(CODI_FORWARD_TYPE)
 template <>
 struct SelectMPIWrapper<passivedouble> {
 #if defined HAVE_MPI

SU2_CFD/include/output/filewriter/CFileWriter.hpp

@@ -41,6 +41,10 @@ using namespace std;
 
 class CFileWriter{
 protected:
+  static passivedouble WallTime();
+  passivedouble fileStartTime = 0.0; //!< Open-to-close timer, independent of per-write scratch timers.
+  void UpdateBandwidth(passivedouble elapsed);
+
 
   /*!
    * \brief The MPI rank

SU2_CFD/include/output/filewriter/CParallelDataSorter.hpp

@@ -122,6 +122,9 @@ class CParallelDataSorter{
   void PrepareSendBuffers(std::vector<unsigned long>& globalID);
 
 public:
+  /*! Reject unsupported counts before narrowing IDs or multiplying int MPI counts. */
+  static void CheckIntCount(unsigned long count, unsigned long entries = 1);
+
 
   /*!
    * \brief Constructor

SU2_CFD/include/output/filewriter/CParaviewFieldLayout.hpp

@@ -0,0 +1,29 @@
+/* Shared field selection for legacy VTK, VTU and PVTU output. LGPL-2.1-or-later. */
+#pragma once
+#include <algorithm>
+#include <string>
+#include <vector>
+
+struct CParaviewField {
+  std::size_t first;
+  unsigned short components;
+  std::string name;
+};
+
+inline std::vector<CParaviewField> GetParaviewFields(const std::vector<std::string>& fieldNames,
+                                                  unsigned short nDim) {
+  auto names = fieldNames;
+  for (auto& name : names) name.erase(std::remove(name.begin(), name.end(), '"'), name.end());
+  std::vector<CParaviewField> fields;
+  for (std::size_t i = nDim; i < names.size();) {
+    const auto& name = names[i];
+    const bool startsVector = name.size() > 2 && name.compare(name.size() - 2, 2, "_x") == 0;
+    const auto base = startsVector ? name.substr(0, name.size() - 2) : name;
+    // A component selected on its own remains a scalar; never read missing or unrelated fields.
+    const bool complete = (nDim == 2 || nDim == 3) && startsVector && i + nDim <= names.size() && names[i + 1] == base + "_y" &&
+                          (nDim == 2 || names[i + 2] == base + "_z");
+    fields.push_back({i, static_cast<unsigned short>(complete ? 3 : 1), complete ? base : name});
+    i += complete ? nDim : 1;
+  }
+  return fields;
+}

SU2_CFD/include/output/filewriter/CParaviewPVTUFileWriter.hpp

@@ -0,0 +1,105 @@
+/*!
+ * \file CParaviewPVTUFileWriter.hpp
+ * \brief Headers of the Paraview parallel XML file writer (.pvtu and one .vtu piece per rank).
+ * \author A. Rausa
+ * \version 8.5.0 "Harrier"
+ *
+ * SU2 Project Website: https://su2code.github.io
+ *
+ * The SU2 Project is maintained by the SU2 Foundation
+ * (http://su2foundation.org)
+ *
+ * Copyright 2012-2026, SU2 Contributors (cf. AUTHORS.md)
+ *
+ * SU2 is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * SU2 is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with SU2. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#pragma once
+
+#include <cstdint>
+
+#include "CFileWriter.hpp"
+
+/*!
+ * \class CParaviewPVTUFileWriter
+ * \brief Writes the data as a Paraview parallel unstructured grid: a .pvtu file that lists one .vtu
+ *        piece per rank. Each rank writes its own piece, so no rank gathers data of other ranks and
+ *        ParaView can read, filter and render the pieces in parallel (unlike a single .vtu file,
+ *        which one process has to load as a whole).
+ */
+class CParaviewPVTUFileWriter final : public CFileWriter {
+ private:
+  /*!
+   * \brief Boolean storing whether we are on a big or little endian machine.
+   */
+  bool bigEndian;
+
+  /*!
+   * \brief True to write the coordinates and fields in double precision instead of single.
+   */
+  bool doublePrecision;
+
+  /*!
+   * \brief Points of the piece of this rank: those used by its elements, wherever they belong.
+   */
+  vector<passivedouble> pieceData;    /*!< \brief Fields of the points of the piece, field by field. */
+  vector<unsigned long> usedPoints;   /*!< \brief Global indices of the points of the piece, sorted. */
+
+  /*!
+   * \brief Collect the points used by the elements of this rank and get their data from the ranks that own
+   *        them, so that the piece of this rank is self-contained.
+   * \returns The number of points of the piece.
+   */
+  unsigned long CollectExternalPoints();
+
+  /*!
+   * \brief Write the piece of this rank.
+   * \param[in] val_filename - Name of the piece file.
+   * \param[in] nPointsPiece - Number of points of the piece.
+   * \param[in] connInt64 - Write the connectivity and offsets as Int64 instead of Int32.
+   */
+  bool WritePiece(const string& val_filename, unsigned long nPointsPiece, bool connInt64);
+
+  /*!
+   * \brief Write the .pvtu file that lists the pieces, only the master node writes it.
+   * \param[in] val_filename - Name of the .pvtu file.
+   * \param[in] pieceNames - Names of the pieces, relative to the .pvtu file.
+   * \param[in] connInt64 - The connectivity and offsets of the pieces are Int64 instead of Int32.
+   */
+  bool WriteIndexFile(const string& val_filename, const vector<string>& pieceNames, bool connInt64);
+
+ public:
+  /*!
+   * \brief File extension
+   */
+  const static string fileExt;
+
+  /*!
+   * \brief Construct a file writer using field names and the data sorter.
+   * \param[in] valDataSorter - The parallel sorted data to write.
+   * \param[in] valDoublePrecision - Write the coordinates and fields in double precision.
+   */
+  CParaviewPVTUFileWriter(CParallelDataSorter* valDataSorter, bool valDoublePrecision = false);
+
+  /*!
+   * \brief Destructor
+   */
+  ~CParaviewPVTUFileWriter() override;
+
+  /*!
+   * \brief Write sorted data to the .pvtu file and its pieces.
+   * \param[in] val_filename - The name of the file.
+   */
+  void WriteData(string val_filename) override;
+};

SU2_CFD/src/meson.build

@@ -51,6 +51,7 @@ su2_cfd_src += files(['output/COutputFactory.cpp',
                       'output/filewriter/CSU2BinaryFileWriter.cpp',
                       'output/filewriter/CParaviewXMLFileWriter.cpp',
                       'output/filewriter/CParaviewVTMFileWriter.cpp',
+                      'output/filewriter/CParaviewPVTUFileWriter.cpp',
                       'output/filewriter/CSU2MeshFileWriter.cpp',
                       'output/filewriter/CSU2MeshBinaryFileWriter.cpp',
                       'output/filewriter/CCGNSFileWriter.cpp',

SU2_CFD/src/output/COutput.cpp

@@ -42,6 +42,7 @@
 #include "../../include/output/filewriter/CSTLFileWriter.hpp"
 #include "../../include/output/filewriter/CParaviewBinaryFileWriter.hpp"
 #include "../../include/output/filewriter/CParaviewXMLFileWriter.hpp"
+#include "../../include/output/filewriter/CParaviewPVTUFileWriter.hpp"
 #include "../../include/output/filewriter/CParaviewVTMFileWriter.hpp"
 #include "../../include/output/filewriter/CTecplotFileWriter.hpp"
 #include "../../include/output/filewriter/CTecplotBinaryFileWriter.hpp"
@@ -560,6 +561,46 @@ void COutput::WriteToFile(CConfig *config, CGeometry *geometry, OUTPUT_TYPE form
 
       break;
 
+    case OUTPUT_TYPE::PARAVIEW_PARALLEL:
+
+      extension = CParaviewPVTUFileWriter::fileExt;
+
+      if (fileName.empty())
+        fileName = config->GetFilename(volumeFilename, "", curTimeIter);
+
+      if (!config->GetWrt_Volume_Overwrite())
+        filename_iter = config->GetFilename_Iter(fileName, curInnerIter, curOuterIter);
+
+      /*--- Load the connectivity without sorting it: each rank then writes the elements of its own
+       partition, which are next to each other, so its piece needs few points of other ranks. ---*/
+
+      volumeDataSorter->SortConnectivity(config, geometry, false);
+
+      LogOutputFiles("Paraview parallel");
+      fileWriter = new CParaviewPVTUFileWriter(volumeDataSorter, config->GetVolume_Output_Double_Precision());
+
+      break;
+
+    case OUTPUT_TYPE::SURFACE_PARAVIEW_PARALLEL:
+
+      extension = CParaviewPVTUFileWriter::fileExt;
+
+      if (fileName.empty())
+        fileName = config->GetFilename(surfaceFilename, "", curTimeIter);
+
+      if (!config->GetWrt_Surface_Overwrite())
+        filename_iter = config->GetFilename_Iter(fileName, curInnerIter, curOuterIter);
+
+      /*--- Load and sort the output data and connectivity. ---*/
+
+      surfaceDataSorter->SortConnectivity(config, geometry);
+      surfaceDataSorter->SortOutputData();
+
+      LogOutputFiles("Paraview parallel surface");
+      fileWriter = new CParaviewPVTUFileWriter(surfaceDataSorter, config->GetVolume_Output_Double_Precision());
+
+      break;
+
     case OUTPUT_TYPE::PARAVIEW_LEGACY_BINARY:
 
       extension = CParaviewBinaryFileWriter::fileExt;

SU2_CFD/src/output/filewriter/CCGNSFileWriter.cpp

@@ -46,7 +46,7 @@ void CCGNSFileWriter::WriteData(string val_filename) {
   val_filename.append(fileExt);
 
   /*--- Set a timer for the file writing. ---*/
-  startTime = SU2_MPI::Wtime();
+  startTime = WallTime();
 
   /*--- Names of all the nodes, before the file is opened: a duplicate name would stop the run with the file open. ---*/
   PrepareNames();
@@ -75,10 +75,10 @@ void CCGNSFileWriter::WriteData(string val_filename) {
 #endif
 
   /*--- Compute and store the write time and the bandwidth. ---*/
-  stopTime = SU2_MPI::Wtime();
+  stopTime = WallTime();
   usedTime = stopTime - startTime;
   fileSize = DetermineFilesize(val_filename);
-  bandwidth = fileSize / (1.0e6) / usedTime;
+  UpdateBandwidth(SU2_TYPE::GetValue(usedTime));
 
 #endif
 }

SU2_CFD/src/output/filewriter/CFEMDataSorter.cpp

@@ -66,6 +66,7 @@ CFEMDataSorter::CFEMDataSorter(CConfig *config, CGeometry *geometry, const vecto
 
   /*--- Create a linear partition --- */
 
+  CheckIntCount(nGlobalPointBeforeSort);
   linearPartitioner.Initialize(nGlobalPointBeforeSort, 0);
 
   /*--- Prepare the send buffers ---*/
@@ -150,6 +151,7 @@ void CFEMDataSorter::SortVolumetricConnectivity(CConfig *config, CGeometry *geom
 
   /* Allocate the memory to store the connectivity if the size is
      larger than zero. */
+  CheckIntCount(nSubElem_Local, NODES_PER_ELEMENT);
   int *Conn_SubElem = nullptr;
   if(nSubElem_Local > 0) Conn_SubElem = new int[nSubElem_Local*NODES_PER_ELEMENT];
 

SU2_CFD/src/output/filewriter/CFVMDataSorter.cpp

@@ -98,6 +98,7 @@ void CFVMDataSorter::SetHaloPoints(CGeometry *geometry, CConfig *config){
 }
 
 void CFVMDataSorter::SortConnectivity(CConfig *config, CGeometry *geometry, bool val_sort) {
+  CheckIntCount(geometry->GetnElem());
 
   /*--- Sort connectivity for each type of element (excluding halos). Note
    In these routines, we sort the connectivity into a linear partitioning
@@ -242,7 +243,9 @@ void CFVMDataSorter::SortVolumetricConnectivity(CConfig *config,
     if ((ii != rank) && (nElem_Send[ii+1] > 0)) nSends++;
     if ((ii != rank) && (nElem_Cum[ii+1] > 0)) nRecvs++;
 
+    CheckIntCount(static_cast<unsigned long>(nElem_Send[ii]) + nElem_Send[ii+1], NODES_PER_ELEMENT);
     nElem_Send[ii+1] += nElem_Send[ii];
+    CheckIntCount(static_cast<unsigned long>(nElem_Cum[ii]) + nElem_Cum[ii+1], NODES_PER_ELEMENT);
     nElem_Cum[ii+1] += nElem_Cum[ii];
   }
 

SU2_CFD/src/output/filewriter/CParallelDataSorter.cpp

@@ -28,6 +28,12 @@
 #include "../../../include/output/filewriter/CParallelDataSorter.hpp"
 #include <cassert>
 #include <numeric>
+#include <limits>
+
+void CParallelDataSorter::CheckIntCount(unsigned long count, unsigned long entries) {
+  if (entries && count > static_cast<unsigned long>(std::numeric_limits<int>::max()) / entries)
+    SU2_MPI::Error("Output exceeds the supported 32-bit node ID or MPI count range", CURRENT_FUNCTION);
+}
 
 CParallelDataSorter::CParallelDataSorter(CConfig *config, const vector<string> &valFieldNames) :
   rank(SU2_MPI::GetRank()),
@@ -240,6 +246,8 @@ void CParallelDataSorter::SortOutputData() {
 }
 
 void CParallelDataSorter::PrepareSendBuffers(std::vector<unsigned long>& globalID){
+  CheckIntCount(nGlobalPointBeforeSort);
+  CheckIntCount(nLocalPointsBeforeSort);
 
   unsigned long iPoint;
   unsigned short iProcessor;
@@ -281,7 +289,9 @@ void CParallelDataSorter::PrepareSendBuffers(std::vector<unsigned long>& globalI
     if ((ii != rank) && (nPoint_Send[ii+1] > 0)) nSends++;
     if ((ii != rank) && (nPoint_Recv[ii+1] > 0)) nRecvs++;
 
+    CheckIntCount(static_cast<unsigned long>(nPoint_Send[ii]) + nPoint_Send[ii+1], 1);
     nPoint_Send[ii+1] += nPoint_Send[ii];
+    CheckIntCount(static_cast<unsigned long>(nPoint_Recv[ii]) + nPoint_Recv[ii+1], 1);
     nPoint_Recv[ii+1] += nPoint_Recv[ii];
   }
 
@@ -378,6 +388,7 @@ void CParallelDataSorter::SetTotalElements(){
 
   nElemGlobal = std::accumulate(nElemPerTypeGlobal.begin(), nElemPerTypeGlobal.end(), 0ul);
   nElem  = std::accumulate(nElemPerType.begin(), nElemPerType.end(), 0ul);
+  CheckIntCount(nElemGlobal);
 
   nConn = 0;
   nConnGlobal   = 0;