From 45c315466e4952a73c0c2228bea1cae074ae167c Mon Sep 17 00:00:00 2001
From: Anna Wellmann <a.wellmann@tu-bs.de>
Date: Tue, 31 May 2022 11:44:42 +0000
Subject: [PATCH] Add doxygen documentation to IndexRearrangementForStreams
---
.../IndexRearrangementForStreams.cpp | 21 +--------
.../IndexRearrangementForStreams.h | 47 +++++++++++++++++--
.../CumulantK17CompChimStreamDevice.cu | 4 +-
.../VirtualFluids_GPU/Parameter/Parameter.h | 4 +-
4 files changed, 50 insertions(+), 26 deletions(-)
diff --git a/src/gpu/VirtualFluids_GPU/DataStructureInitializer/GridReaderGenerator/IndexRearrangementForStreams.cpp b/src/gpu/VirtualFluids_GPU/DataStructureInitializer/GridReaderGenerator/IndexRearrangementForStreams.cpp
index 7eb844583..e830b21c1 100644
--- a/src/gpu/VirtualFluids_GPU/DataStructureInitializer/GridReaderGenerator/IndexRearrangementForStreams.cpp
+++ b/src/gpu/VirtualFluids_GPU/DataStructureInitializer/GridReaderGenerator/IndexRearrangementForStreams.cpp
@@ -468,11 +468,6 @@ void IndexRearrangementForStreams::reorderRecvIndicesForCommAfterFtoC(
void IndexRearrangementForStreams::splitFineToCoarseIntoBorderAndBulk(const uint &level)
{
- // For communication hiding, the interpolation cells from the fine to the coarse grid need to be split into two
- // groups:
- // - cells which are at the border between two gpus --> "border"
- // - the other cells which are not directly related to the communication betweeen the two gpus --> "bulk"
-
this->getGridInterfaceIndicesBorderBulkFC(level);
para->getParD(level)->intFCBorder.kFC = para->getParH(level)->intFCBorder.kFC;
@@ -487,10 +482,6 @@ void IndexRearrangementForStreams::splitFineToCoarseIntoBorderAndBulk(const uint
void IndexRearrangementForStreams::getGridInterfaceIndicesBorderBulkFC(int level)
{
- // This function reorders the arrays of FCC/FCF indices and return pointers and sizes of the new subarrays:
- // The coarse cells for interpolation from fine to coarse (iCellFCC) are divided into two subgroups: border and
- // bulk. The fine cells (iCellFCF) are reordered accordingly.
-
// create some local variables for better readability
uint *iCellFccAll = para->getParH(level)->intFC.ICellFCC;
uint *iCellFcfAll = para->getParH(level)->intFC.ICellFCF;
@@ -520,6 +511,7 @@ void IndexRearrangementForStreams::getGridInterfaceIndicesBorderBulkFC(int level
para->getParH(level)->intFCBulk.ICellFCF = iCellFcfAll + para->getParH(level)->intFCBorder.kFC;
// copy the created vectors to the memory addresses of the old arrays
+ // this is inefficient :(
for (uint i = 0; i < (uint)iCellFccBorderVector.size(); i++) {
iCellFccAll[i] = iCellFccBorderVector[i];
iCellFcfAll[i] = iCellFcfBorderVector[i];
@@ -532,11 +524,6 @@ void IndexRearrangementForStreams::getGridInterfaceIndicesBorderBulkFC(int level
void IndexRearrangementForStreams::splitCoarseToFineIntoBorderAndBulk(const uint &level)
{
- // For communication hiding, the interpolation cells from the coarse to the fine grid need to be split into two
- // groups:
- // - cells which are at the border between two gpus --> "border"
- // - the other cells which are not directly related to the communication betweeen the two gpus --> "bulk"
-
this->getGridInterfaceIndicesBorderBulkCF(level);
para->getParD(level)->intCFBorder.kCF = para->getParH(level)->intCFBorder.kCF;
@@ -554,11 +541,6 @@ void IndexRearrangementForStreams::splitCoarseToFineIntoBorderAndBulk(const uint
void IndexRearrangementForStreams::getGridInterfaceIndicesBorderBulkCF(int level)
{
- // This function reorders the arrays of CFC/CFF indices and sets the pointers and sizes of the new subarrays:
- // The coarse cells for interpolation from coarse to fine (iCellCFC) are divided into two subgroups: border and
- // bulk. The fine cells (iCellCFF) are reordered accordingly. The offset cells (xOffCF, yOffCF, zOffCF) must be
- // reordered in the same way.
-
// create some local variables for better readability
uint *iCellCfcAll = para->getParH(level)->intCF.ICellCFC;
uint *iCellCffAll = para->getParH(level)->intCF.ICellCFF;
@@ -621,6 +603,7 @@ void IndexRearrangementForStreams::getGridInterfaceIndicesBorderBulkCF(int level
para->getParH(level)->offCFBulk.zOffCF = para->getParH(level)->offCF.zOffCF + para->getParH(level)->intCFBorder.kCF;
// copy the created vectors to the memory addresses of the old arrays
+ // this is inefficient :(
for (uint i = 0; i < (uint)iCellCfcBorderVector.size(); i++) {
para->getParH(level)->intCFBorder.ICellCFC[i] = iCellCfcBorderVector[i];
para->getParH(level)->intCFBorder.ICellCFF[i] = iCellCffBorderVector[i];
diff --git a/src/gpu/VirtualFluids_GPU/DataStructureInitializer/GridReaderGenerator/IndexRearrangementForStreams.h b/src/gpu/VirtualFluids_GPU/DataStructureInitializer/GridReaderGenerator/IndexRearrangementForStreams.h
index c1430232b..4b55ff587 100644
--- a/src/gpu/VirtualFluids_GPU/DataStructureInitializer/GridReaderGenerator/IndexRearrangementForStreams.h
+++ b/src/gpu/VirtualFluids_GPU/DataStructureInitializer/GridReaderGenerator/IndexRearrangementForStreams.h
@@ -1,3 +1,9 @@
+//! \file IndexRearrangementForStreams.h
+//! \ingroup GPU
+//! \author Anna Wellmann
+//! \ref master thesis of Anna Wellmann
+
+
#ifndef IndexRearrangementForStreams_H
#define IndexRearrangementForStreams_H
@@ -26,20 +32,45 @@ private:
std::shared_ptr<Parameter> para;
public:
+ //! \brief construct IndexRearrangementForStreams object
IndexRearrangementForStreams(std::shared_ptr<Parameter> para, std::shared_ptr<GridBuilder> builder);
+ //////////////////////////////////////////////////////////////////////////
// communication after coarse to fine
+ //////////////////////////////////////////////////////////////////////////
void initCommunicationArraysForCommAfterFinetoCoarseX(const uint &level, int j, int direction);
void initCommunicationArraysForCommAfterFinetoCoarseY(const uint &level, int j, int direction);
void initCommunicationArraysForCommAfterFinetoCoarseZ(const uint &level, int j, int direction);
+public:
+ //////////////////////////////////////////////////////////////////////////
// split interpolation cells
+ //////////////////////////////////////////////////////////////////////////
+
+ //! \brief split the interpolation cells from coarse to fine into border an bulk
+ //! \details For communication hiding, the interpolation cells from the coarse to the fine grid need to be split into two groups:
+ //!
+ //! - cells which are at the border between two gpus --> "border"
+ //!
+ //! - the other cells which are not directly related to the communication betweeen the two gpus --> "bulk"
+ //! \ref see master thesis of Anna Wellmann (p. 62-68)
void splitCoarseToFineIntoBorderAndBulk(const uint &level);
+
+ //! \brief split the interpolation cells from fine to coarse into border an bulk
+ //! \details For communication hiding, the interpolation cells from the fine to the coarse grid need to be split into two groups:
+ //!
+ //! - cells which are at the border between two gpus --> "border"
+ //!
+ //! - the other cells which are not directly related to the communication betweeen the two gpus --> "bulk"
+ //!
+ //! \ref see master thesis of Anna Wellmann (p. 62-68)
void splitFineToCoarseIntoBorderAndBulk(const uint &level);
private:
+ //////////////////////////////////////////////////////////////////////////
// communication after coarse to fine
+ //////////////////////////////////////////////////////////////////////////
void copyProcessNeighborToCommAfterFtoCX(const uint &level, int indexOfProcessNeighbor);
void copyProcessNeighborToCommAfterFtoCY(const uint &level, int indexOfProcessNeighbor);
void copyProcessNeighborToCommAfterFtoCZ(const uint &level, int indexOfProcessNeighbor);
@@ -73,12 +104,22 @@ private:
std::vector<uint> &sendIndicesForCommAfterFtoCPositions);
void reorderRecvIndicesForCommAfterFtoC(int *recvIndices, int &numberOfRecvNeighborsAfterFtoC, int direction,
int level, std::vector<uint> &sendIndicesForCommAfterFtoCPositions);
-
+
+ private:
+ //////////////////////////////////////////////////////////////////////////
// split interpolation cells
+ //////////////////////////////////////////////////////////////////////////
+
+ //! \brief This function reorders the arrays of CFC/CFF indices and sets the pointers and sizes of the new subarrays:
+ //! \details The coarse cells for interpolation from coarse to fine (iCellCFC) are divided into two subgroups: border and
+ //! bulk. The fine cells (iCellCFF) are reordered accordingly. The offset cells (xOffCF, yOffCF, zOffCF) must be
+ //! reordered in the same way.
void getGridInterfaceIndicesBorderBulkCF(int level);
- void getGridInterfaceIndicesBorderBulkFC(int level);
- friend class IndexRearrangementForStreamsTest;
+ //! \brief This function reorders the arrays of FCC/FCF indices and return pointers and sizes of the new subarrays:
+ //! \details The coarse cells for interpolation from fine to coarse (iCellFCC) are divided into two subgroups: border and
+ //! bulk. The fine cells (iCellFCF) are reordered accordingly.
+ void getGridInterfaceIndicesBorderBulkFC(int level);
};
#endif
diff --git a/src/gpu/VirtualFluids_GPU/Kernel/Kernels/BasicKernels/FluidFlow/Compressible/CumulantK17chimStream/CumulantK17CompChimStreamDevice.cu b/src/gpu/VirtualFluids_GPU/Kernel/Kernels/BasicKernels/FluidFlow/Compressible/CumulantK17chimStream/CumulantK17CompChimStreamDevice.cu
index da37bb1e2..f57fd9dd9 100644
--- a/src/gpu/VirtualFluids_GPU/Kernel/Kernels/BasicKernels/FluidFlow/Compressible/CumulantK17chimStream/CumulantK17CompChimStreamDevice.cu
+++ b/src/gpu/VirtualFluids_GPU/Kernel/Kernels/BasicKernels/FluidFlow/Compressible/CumulantK17chimStream/CumulantK17CompChimStreamDevice.cu
@@ -26,9 +26,9 @@
// You should have received a copy of the GNU General Public License along
// with VirtualFluids (see COPYING.txt). If not, see <http://www.gnu.org/licenses/>.
//
-//! \file Cumulant27chim.cu
+//! \file Cumulant27chimStream.cu
//! \ingroup GPU
-//! \author Martin Schoenherr
+//! \author Martin Schoenherr, Anna Wellmann
//=======================================================================================
/* Device code */
#include "LBM/LB.h"
diff --git a/src/gpu/VirtualFluids_GPU/Parameter/Parameter.h b/src/gpu/VirtualFluids_GPU/Parameter/Parameter.h
index db648344f..940b2493f 100644
--- a/src/gpu/VirtualFluids_GPU/Parameter/Parameter.h
+++ b/src/gpu/VirtualFluids_GPU/Parameter/Parameter.h
@@ -899,8 +899,8 @@ private:
std::unique_ptr<CudaStreamManager> cudaStreamManager;
public:
- //! sets whether streams and thus communication hiding should be used
- /*! This function is only useful for simulations on multiple GPUs. If there is only one MPI process, the passed value is automatically overwritten with false. */
+ //! \brief sets whether streams and thus communication hiding should be used
+ //! \details This function is only useful for simulations on multiple GPUs. If there is only one MPI process, the passed value is automatically overwritten with false.
void setUseStreams(bool useStreams);
bool getUseStreams();
std::unique_ptr<CudaStreamManager> &getStreamManager();
--
GitLab