diff --git a/Addons/CKernelApi/Src/FScalFMMEngine.hpp b/Addons/CKernelApi/Src/FScalFMMEngine.hpp
index a397fb7dfebc00b758f150a96ce1f3b20cc132fe..39cb818787eeb3669f25bb014210ee40d19c0839 100644
--- a/Addons/CKernelApi/Src/FScalFMMEngine.hpp
+++ b/Addons/CKernelApi/Src/FScalFMMEngine.hpp
@@ -178,37 +178,6 @@ public:
}
- /** Test ... */
- struct RunContainer{
- template< int nbAttributeToInsert,class ContainerClass,class LeafClass, class CellClass>
- static void Run(FOctree<FReal,CellClass,ContainerClass,LeafClass> * octree,
- int NbPartToInsert,int * strideForEachAtt,
- FReal* rawDatas){
- generic_tree_abstract_insert<ContainerClass,LeafClass,CellClass,nbAttributeToInsert>(octree,
- NbPartToInsert,strideForEachAtt,rawDatas);
- }
- };
-
- template<class ContainerClass,class LeafClass, class CellClass, int nbAttributeToInsert>
- void generic_tree_abstract_insert(FOctree<FReal,CellClass,ContainerClass,LeafClass> * octree,
- int NbPartToInsert,int * strideForEachAtt,
- FReal* rawDatas){
- for(FSize idxPart = 0; idxPart<NbPartToInsert ; ++idxPart){
- FPoint<FReal> pos = FPoint<FReal>(rawDatas[0],rawDatas[1],rawDatas[2]);
- MortonIndex index = octree->getMortonFromPosition(pos);
- //Insert with how many attributes ???
- octree->insert(pos,idxPart);
- //Get again the container
- ContainerClass * containerToFill = octree->getLeafSrc(index);//cannot be nullptr
- std::array<FReal,nbAttributeToInsert> arrayOfAttribute;
- for(int idxAtt = 0; idxAtt<nbAttributeToInsert ; ++idxAtt){
- arrayOfAttribute[idxAtt] = rawDatas[3+ strideForEachAtt[idxAtt]];
- }
- int idxToRemove = containerToFill->getNbParticles();
- containerToFill->remove(&idxToRemove,1);
- containerToFill->push(pos,idxPart,arrayOfAttribute);
- }
- }
template<class ContainerClass,class LeafClass,class CellClass>
void generic_get_forces_xyz(FOctree<FReal,CellClass,ContainerClass,LeafClass> * octree,
@@ -820,15 +789,17 @@ public:
* get the time spent in each operator.
*/
virtual void get_timers(FReal * Timers){
- const FTic * timers = algoTimer->getAllTimers();
- int nbTimers = algoTimer->getNbOfTimerRecorded();
- for(int idTimer = 0; idTimer<nbTimers ; ++idTimer){
- Timers[idTimer] = timers[idTimer].elapsed();
- }
+ Timers[0] = algoTimer->Timers["P2M"].elapsed();
+ Timers[1] = algoTimer->Timers["M2M"].elapsed();
+ Timers[2] = algoTimer->Timers["M2L"].elapsed();
+ Timers[3] = algoTimer->Timers["L2L"].elapsed();
+ Timers[4] = algoTimer->Timers["L2P"].elapsed();
+ Timers[5] = algoTimer->Timers["P2P"].elapsed();
+ Timers[6] = algoTimer->Timers["NearField"].elapsed();
}
virtual int get_nb_timers(){
- return algoTimer->getNbOfTimerRecorded();
+ return FAlgorithmTimers::nbTimers;
}
virtual void set_upper_limit(int upperLimit){
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 5a771b490d74dbcb44b33cc751d48686e2bd3384..6ce3ed3e37524e5a9cf1ad014d276dc6184bb449 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -34,7 +34,7 @@ GetCpuInfos()
# SCALFMM version number. An even minor number corresponds to releases.
set(SCALFMM_MAJOR_VERSION 1)
-set(SCALFMM_MINOR_VERSION 4)
+set(SCALFMM_MINOR_VERSION 5)
set(SCALFMM_PATCH_VERSION 0)
set(SCALFMM_VERSION "${SCALFMM_MAJOR_VERSION}.${SCALFMM_MINOR_VERSION}.${SCALFMM_PATCH_VERSION}" )
@@ -375,35 +375,53 @@ if (MORSE_DISTRIB_DIR OR EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/CMakeModules/morse/
list(APPEND CMAKE_INSTALL_RPATH "${LAPACK_LIBRARY_DIRS}")
endif()
message (STATUS "check BLAS Fortran mangling")
- # check blas and lapack symbols naming
- set(CMAKE_REQUIRED_LIBRARIES "${BLAS_LIBRARIES}")
- check_function_exists(dgemv_ DGEMV_ADD_)
- set (SCALFMM_BLAS_UPCASE OFF)
- set (SCALFMM_BLAS_NOCHANGE OFF)
- message (STATUS "BLAS dgemv_ " ${DGEMV_ADD_} ${SCALFMM_BLAS_UPCASE})
- if (DGEMV_ADD_)
- set (SCALFMM_BLAS_ADD_ ON)
- message (STATUS "BLAS dgemv_ symbol found, SCALFMM_BLAS_ADD_ is ON")
- else (DGEMV_ADD_)
- set (SCALFMM_BLAS_ADD_ OFF)
+ # add options to let the user be able to force a behavior
+ option( SCALFMM_BLAS_ADD_ "Set to ON to force calls to BLAS Fortran symbols with _ (ex: dgemm_)" OFF )
+ option( SCALFMM_BLAS_UPCASE "Set to ON to force calls to BLAS Fortran symbols in capital (ex: DGEMM)" OFF )
+ option( SCALFMM_BLAS_NOCHANGE "Set to ON to force calls to BLAS Fortran symbols with no change (ex: dgemm)" OFF )
+ # if options not changed by user then auto-detection
+ if (NOT SCALFMM_BLAS_ADD_ AND NOT SCALFMM_BLAS_UPCASE AND NOT SCALFMM_BLAS_NOCHANGE)
+ # give blas libraries and check dgemm symbol
+ set(CMAKE_REQUIRED_LIBRARIES "${BLAS_LIBRARIES}")
+ check_function_exists(dgemv_ DGEMV_ADD_)
check_function_exists(DGEMV DGEMV_UPCASE)
- if (DGEMV_UPCASE)
- set (SCALFMM_BLAS_UPCASE ON)
- message (STATUS "BLAS DGEMV symbol found, SCALFMM_BLAS_UPCASE is ON")
- else (DGEMV_UPCASE)
- # set (SCALFMM_BLAS_UPCASE OFF)
- check_function_exists(dgemv DGEMV_NOCHANGE)
- if (DGEMV_NOCHANGE)
- set (SCALFMM_BLAS_NOCHANGE ON)
- message (STATUS "BLAS dgemv symbol found, SCALFMM_BLAS_NOCHANGE is ON")
- # else (DGEMV_NOCHANGE)
- # set (SCALFMM_BLAS_NOCHANGE OFF)
- endif (DGEMV_NOCHANGE)
- endif (DGEMV_UPCASE)
- endif (DGEMV_ADD_)
- if ( (NOT DGEMV_ADD_) AND (NOT DGEMV_UPCASE) AND (NOT DGEMV_NOCHANGE) )
- message(FATAL_ERROR "BLAS Fortran mangling cannot be properly detected")
- endif ()
+ check_function_exists(dgemv DGEMV_NOCHANGE)
+ # here we consider that the first kind of symbol found will be the one used
+ # current order is: ADD_, UPCASE, NOCHANGE
+ if (DGEMV_ADD_)
+ set (SCALFMM_BLAS_ADD_ ON)
+ set (SCALFMM_BLAS_UPCASE OFF)
+ set (SCALFMM_BLAS_NOCHANGE OFF)
+ message (STATUS "BLAS dgemv_ symbol found, SCALFMM_BLAS_ADD_ is ON")
+ else (DGEMV_ADD_)
+ if (DGEMV_UPCASE)
+ set (SCALFMM_BLAS_ADD_ OFF)
+ set (SCALFMM_BLAS_UPCASE ON)
+ set (SCALFMM_BLAS_NOCHANGE OFF)
+ message (STATUS "BLAS DGEMV symbol found, SCALFMM_BLAS_UPCASE is ON")
+ else (DGEMV_UPCASE)
+ if (DGEMV_NOCHANGE)
+ set (SCALFMM_BLAS_ADD_ OFF)
+ set (SCALFMM_BLAS_UPCASE OFF)
+ set (SCALFMM_BLAS_NOCHANGE ON)
+ message (STATUS "BLAS dgemv symbol found, SCALFMM_BLAS_NOCHANGE is ON")
+ endif (DGEMV_NOCHANGE)
+ endif (DGEMV_UPCASE)
+ endif (DGEMV_ADD_)
+ if ( (NOT DGEMV_ADD_) AND (NOT DGEMV_UPCASE) AND (NOT DGEMV_NOCHANGE) )
+ message(FATAL_ERROR "BLAS Fortran mangling is not properly detected - please check your BLAS libraries")
+ endif ()
+ else (NOT SCALFMM_BLAS_ADD_ AND NOT SCALFMM_BLAS_UPCASE AND NOT SCALFMM_BLAS_NOCHANGE)
+ if (SCALFMM_BLAS_ADD_)
+ message (STATUS "SCALFMM_BLAS_ADD_ is set to ON")
+ endif()
+ if (SCALFMM_BLAS_UPCASE)
+ message (STATUS "SCALFMM_BLAS_UPCASE is set to ON")
+ endif()
+ if (SCALFMM_BLAS_NOCHANGE)
+ message (STATUS "SCALFMM_BLAS_NOCHANGE is set to ON")
+ endif()
+ endif (NOT SCALFMM_BLAS_ADD_ AND NOT SCALFMM_BLAS_UPCASE AND NOT SCALFMM_BLAS_NOCHANGE)
else()
message(WARNING "BLAS has not been found, SCALFMM will continue to compile but some applications will be disabled.")
message(WARNING "If you have BLAS set BLAS_LIBDIR, BLAS_INCDIR or BLAS_DIR (CMake variables using -D or environment variables).")
@@ -466,8 +484,12 @@ if (MORSE_DISTRIB_DIR OR EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/CMakeModules/morse/
endif(SCALFMM_USE_FFT)
list(APPEND FUSE_LIST "FFT")
message(STATUS "SCALFMM_USE_FFT = ${SCALFMM_USE_FFT}")
- list(REMOVE_DUPLICATES SCALFMM_INCLUDES)
- list(REMOVE_DUPLICATES SCALFMM_LIBRARIES)
+ if (SCALFMM_INCLUDES)
+ list(REMOVE_DUPLICATES SCALFMM_INCLUDES)
+ endif()
+ if (SCALFMM_LIBRARIES)
+ list(REMOVE_DUPLICATES SCALFMM_LIBRARIES)
+ endif()
#
# C++ 2011
diff --git a/Src/GroupTree/Chebyshev/FChebCellPOD.hpp b/Src/GroupTree/Chebyshev/FChebCellPOD.hpp
index d16972dac0bb4d73e4085d8f9a21c3028ce6b16b..8d801f5c54d169cf2e22212ccb23dddf33d1dbd2 100644
--- a/Src/GroupTree/Chebyshev/FChebCellPOD.hpp
+++ b/Src/GroupTree/Chebyshev/FChebCellPOD.hpp
@@ -51,6 +51,16 @@ public:
symb->mortonIndex = inMortonIndex;
}
+ /** To get the cell level */
+ int getLevel() const {
+ return symb->level;
+ }
+
+ /** To set the cell level */
+ void setLevel(const int level) {
+ symb->level = level;
+ }
+
/** To get the position */
FTreeCoordinate getCoordinate() const {
return FTreeCoordinate(symb->coordinates[0],
@@ -103,4 +113,3 @@ public:
#endif // FCHEBCELLPOD_HPP
-
diff --git a/Src/GroupTree/Core/FBasicCellPOD.hpp b/Src/GroupTree/Core/FBasicCellPOD.hpp
index 1060b846b79b308b66d1b2054a0e912b38e1ff17..d47258c98c7310d0ea584216f1f8aee967caa9ca 100644
--- a/Src/GroupTree/Core/FBasicCellPOD.hpp
+++ b/Src/GroupTree/Core/FBasicCellPOD.hpp
@@ -9,7 +9,7 @@
struct alignas(FStarPUDefaultAlign::StructAlign) FBasicCellPOD {
MortonIndex mortonIndex;
int coordinates[3];
+ int level;
};
#endif // FBASICCELLPOD_HPP
-
diff --git a/Src/GroupTree/Core/FGroupTree.hpp b/Src/GroupTree/Core/FGroupTree.hpp
index 944d3f7c40e4a788ede9aa7fc2f36e90193485b1..1bcb7d7118ac972a75b99b516451a0042b679aeb 100644
--- a/Src/GroupTree/Core/FGroupTree.hpp
+++ b/Src/GroupTree/Core/FGroupTree.hpp
@@ -99,6 +99,7 @@ public:
CompositeCellClass newNode = newBlock->getCompleteCell(cellIdInBlock);
newNode.setMortonIndex(newNodeIndex);
newNode.setCoordinate(newNodeCoordinate);
+ newNode.setLevel(idxLevel);
// Add leaf
nbParticlesOffsetBeforeLeaf = newParticleBlock->newLeaf(newNodeIndex, cellIdInBlock,
@@ -149,6 +150,7 @@ public:
CompositeCellClass newNode = newBlock->getCompleteCell(cellIdInBlock);
newNode.setMortonIndex(newNodeIndex);
newNode.setCoordinate(newNodeCoordinate);
+ newNode.setLevel(idxLevel);
cellIdInBlock += 1;
blockIteratorInOctree.moveRight();
diff --git a/Src/GroupTree/Rotation/FRotationCellPOD.hpp b/Src/GroupTree/Rotation/FRotationCellPOD.hpp
index 8b86ae7c11f7bbd6bf7457f7bf2958546da90df5..10e13f04ff4b68f66335ffca8e2b024a7ea1ec0e 100644
--- a/Src/GroupTree/Rotation/FRotationCellPOD.hpp
+++ b/Src/GroupTree/Rotation/FRotationCellPOD.hpp
@@ -53,6 +53,16 @@ public:
symb->mortonIndex = inMortonIndex;
}
+ /** To get the cell level */
+ int getLevel() const {
+ return symb->level;
+ }
+
+ /** To set the cell level */
+ void setLevel(const int level) {
+ symb->level = level;
+ }
+
/** To get the position */
FTreeCoordinate getCoordinate() const {
return FTreeCoordinate(symb->coordinates[0],
@@ -109,4 +119,3 @@ public:
#endif // FROTATIONCELLPOD_HPP
-
diff --git a/Src/GroupTree/Taylor/FTaylorCellPOD.hpp b/Src/GroupTree/Taylor/FTaylorCellPOD.hpp
index 9fba78bf2524c40b16de1f95e97e7e4d77864ac8..3c4ff6d22d77bb9999da2b79850de9045b16cc32 100644
--- a/Src/GroupTree/Taylor/FTaylorCellPOD.hpp
+++ b/Src/GroupTree/Taylor/FTaylorCellPOD.hpp
@@ -51,6 +51,16 @@ public:
symb->mortonIndex = inMortonIndex;
}
+ /** To get the cell level */
+ int getLevel() const {
+ return symb->level;
+ }
+
+ /** To set the cell level */
+ void setLevel(const int level) {
+ symb->level = level;
+ }
+
/** To get the position */
FTreeCoordinate getCoordinate() const {
return FTreeCoordinate(symb->coordinates[0],
@@ -102,4 +112,3 @@ public:
};
#endif // FTAYLORCELLPOD_HPP
-
diff --git a/Src/GroupTree/TestKernel/FTestCellPOD.hpp b/Src/GroupTree/TestKernel/FTestCellPOD.hpp
index 1dd5d2110bf0e46bacfba25cabb02af5a84d77c8..6ac22cd16f432a84c1b3385786608ac7766db604 100644
--- a/Src/GroupTree/TestKernel/FTestCellPOD.hpp
+++ b/Src/GroupTree/TestKernel/FTestCellPOD.hpp
@@ -34,6 +34,16 @@ public:
symb->mortonIndex = inMortonIndex;
}
+ /** To get the cell level */
+ int getLevel() const {
+ return symb->level;
+ }
+
+ /** To set the cell level */
+ void setLevel(const int level) {
+ symb->level = level;
+ }
+
/** To get the position */
FTreeCoordinate getCoordinate() const {
return FTreeCoordinate(symb->coordinates[0],
@@ -81,4 +91,3 @@ public:
#endif // FTESTCELLPOD_HPP
-
diff --git a/Src/GroupTree/Uniform/FUnifCellPOD.hpp b/Src/GroupTree/Uniform/FUnifCellPOD.hpp
index 578ee670c2677df738719093379dd62d2b2da4b1..65cca0d4721c81d11c7bafc73702e6e9a688be33 100644
--- a/Src/GroupTree/Uniform/FUnifCellPOD.hpp
+++ b/Src/GroupTree/Uniform/FUnifCellPOD.hpp
@@ -51,6 +51,16 @@ public:
symb->mortonIndex = inMortonIndex;
}
+ /** To get the cell level */
+ int getLevel() const {
+ return symb->level;
+ }
+
+ /** To set the cell level */
+ void setLevel(const int level) {
+ symb->level = level;
+ }
+
/** To get the position */
FTreeCoordinate getCoordinate() const {
return FTreeCoordinate(symb->coordinates[0],
@@ -129,4 +139,3 @@ public:
};
#endif // FUNIFCELLPOD_HPP
-
diff --git a/Src/Kernels/Interpolation/FInterpMatrixKernel_TensorialInteractions.hpp b/Src/Kernels/Interpolation/FInterpMatrixKernel_TensorialInteractions.hpp
index e05977bfa8a209b4a7b31ddfc8ab9a947d205808..e018112f8655543d0f91a78498712bb313c14d47 100644
--- a/Src/Kernels/Interpolation/FInterpMatrixKernel_TensorialInteractions.hpp
+++ b/Src/Kernels/Interpolation/FInterpMatrixKernel_TensorialInteractions.hpp
@@ -45,16 +45,42 @@
* The table applyTab provides the indices in the reduced storage table
* corresponding to the application scheme depicted earlier.
*
- * PB: BEWARE! Homogeneous matrix kernels do not support cell width extension
+ * \warning BEWARE! Homogeneous matrix kernels do not support cell width extension
* yet. Is it possible to find a reference width and a scale factor such that
* only 1 set of M2L opt can be used for all levels??
- *
+ * The definition of the potential p and force f are extended to the case
+* of tensorial interaction kernels:
+*
+*\f$ p_i(x) = K_{ip}(x,y)w_p(y),\f$ \f$ \forall i=1..NPOT, p=1..NPV\f$
+*
+* \f$f_{ik}= w_p(x)K_{ip,k}(x,y)w_p(y)\f$
+*
+* Since the interpolation scheme is such that
+*
+*\f$ p_i(x) \approx S^m(x) L^{m}_{ip}\f$
+*
+* \f$f_{ik}= w_p(x) \nabla_k S^m(x) L^{m}_{ip}\f$
+*
+* with
+*
+* \f$ L^{m}_{ip} = K^{mn}_{ip} S^n(y) w_p(y)\f$ (local expansion)
+*
+ *\f$ M^{m}_{p} = S^n(y) w_p(y)\f$ (multipole expansion)
+*
+* then the multipole exp have NPV components and the local exp NPOT*NPV.
+*
+* NB1: Only the computation of forces requires that the sum over p is
+* performed at L2P step. It could be done at M2L step for the potential.
+*
+* NB2: An efficient application of the matrix kernel is highly kernel
+* dependent, we recommand overriding the P2M/M2L/L2P function of the kernel
+* you are using in order to have opsimal performances + set your own NRHS/NLHS.*
*/
////////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
-//
+///
// Tensorial Matrix Kernels (NCMP>1)
//
// The definition of the potential p and force f are extended to the case
diff --git a/Src/Utils/FAlgorithmTimers.hpp b/Src/Utils/FAlgorithmTimers.hpp
index f9eb1f2a56a7b70e5cf03258e160a75f8d408eed..dc5563b98a1c9d142fd0b11712c3124954a6668e 100644
--- a/Src/Utils/FAlgorithmTimers.hpp
+++ b/Src/Utils/FAlgorithmTimers.hpp
@@ -56,8 +56,13 @@ public:
/// Elapsed time between last FTic::tic() and FTic::tac() for given timer.
double getTime(std::string TimerName) const{
- //assert to verify size
- return Timers.at(TimerName).elapsed();
+ double res = 0;
+ try {
+ res = Timers.at(TimerName).elapsed();
+ } catch(std::out_of_range) {
+ res = 0;
+ }
+ return res;
}
/// Cumulated time between all FTic::tic() and FTic::tac() for given timer.