Merge pull request #718 from DrTimothyAldenDavis/dev2

SuiteSparse 7.5.0.beta5

CHOLMOD/Doc/CHOLMOD_UserGuide.tex

@@ -287,9 +287,16 @@ \section{Single-precision sparse matrix support}
 \item {\tt SuiteSparse\_config}: a single place where all sparse matrix
     packages authored or co-authored by Davis are configured.
 \item {\tt METIS 5.1.0}: a graph partitioning package by George Karypis,
-    Univ. of Minnesota.  A slightly revised copy appears in
-    {\tt CHOLMOD/SuiteSparse\_metis}.  Not needed if {\tt -DNPARTITION} is
-    used.  See http://www-users.cs.umn.edu/$\sim$karypis/metis.
+    Univ. of Minnesota.  A slightly revised copy appears in {\tt
+    CHOLMOD/SuiteSparse\_metis}, but with a revised namespace so that no METIS
+    functions in CHOLMOD conflict with the unmodified METIS.  SuiteSparse
+    cannot use any other METIS library except for {\tt
+    CHOLMOD/SuiteSparse\_metis}.  There is no conflict with the unmodified
+    METIS, so it can be linked into the user's application alongside {\tt
+    CHOLMOD/SuiteSparse\_metis}.  This folder containing the modified METIS is
+    not needed if {\tt -DNPARTITION} is used.  See
+    http://www-users.cs.umn.edu/$\sim$karypis/metis for the original version of
+    METIS.
 \end{itemize}
 
 Three other packages are required for optimal performance:
@@ -303,7 +310,7 @@ \section{Single-precision sparse matrix support}
     Not needed if {\tt -DNSUPERNODAL} is used.
     See http://www.netlib.org.
 \item CUDA BLAS:  CHOLMOD can exploit an NVIDIA GPU by using the CUDA BLAS
-    for large supernodes.  This feature is new to CHOLMOD v2.0.0.
+    for large supernodes.
 \end{itemize}
 
 You must first obtain and install LAPACK, and the BLAS.
@@ -325,7 +332,7 @@ \section{Single-precision sparse matrix support}
 \item {\tt -DNPRINT}:       do not print anything.
 \end{itemize}
 
-These settings are controlled in cmake by the following variables, which 
+These settings are controlled in cmake by the following variables, which
 are all \verb'ON' by default:
 \begin{itemize}
 \item \verb'CHOLMOD_CHECK':      if \verb'OFF': do not use the Check module.
@@ -2016,9 +2023,16 @@ \subsection{{\tt cholmod\_sparse\_xtype}: change sparse xtype}
 %---------------------------------------
 
 \input{_sparse_xtype.tex}
-Changes the {\tt xtype} and/or \verb'dtype' of a sparse matrix, to pattern,
-real, complex, or zomplex.  Changing from complex or zomplex to real discards
-the imaginary part.
+Changes the {\tt xtype} and/or \verb'dtype' of a sparse matrix.  The
+\verb'xtype' can be changed to pattern, real, complex, or zomplex.  The
+\verb'dtype' can be changed to single or double.  Changing from complex or
+zomplex to real discards the imaginary part.
+
+The name of this method derives from an earlier version of CHOLMOD where all
+matrices had an implied \verb'dtype' of \verb'CHOLMOD_DOUBLE'.  The method now
+supports changes to the \verb'dtype'.  For backward compatilibilty, the name of
+this method has not changed, and the \verb'to_xdtype' parameter values of the
+prior versions are upwards compatible with this version of CHOLMOD.
 
 %-------------------------------------------------------------------------------
 \newpage \section{{\tt Utility} Module: {\tt cholmod\_factor} object}
@@ -2168,10 +2182,11 @@ \subsection{{\tt cholmod\_change\_factor}: change factor}
 
 Supports all xtypes, except that there is no supernodal zomplex L.
 
-The {\tt to\_xdtype} parameter is used only when converting from symbolic to
+The \verb'to_xtype' parameter is used only when converting from symbolic to
 numeric or numeric to symbolic.  It cannot be used to convert a numeric xtype
-(real, complex, or zomplex) to a different numeric xtype.  For that conversion,
-use {\tt cholmod\_factor\_xtype} instead.
+(real, complex, or zomplex) to a different numeric {\tt xtype}.  It also cannot
+change the \verb'dtype' of a factor (from double to single, for example).
+For those conversions, use \verb'cholmod_factor_xtype' instead.
 
 %---------------------------------------
 \subsection{{\tt cholmod\_pack\_factor}: pack the columns of a factor}
@@ -2228,10 +2243,20 @@ \subsection{{\tt cholmod\_factor\_xtype}: change factor xtype}
 %---------------------------------------
 
 \input{_factor_xtype.tex}
-Changes the {\tt xtype} of a factor, to pattern, real, complex, or zomplex.
-Changing from complex or zomplex to real discards the imaginary part.
-You cannot change a supernodal factor to the zomplex xtype.
-
+Changes the {\tt xtype} and/or \verb'dtype' of a sparse factorization.  The
+\verb'xtype' can be changed to real, complex, or zomplex (not to pattern).
+The \verb'dtype' can be changed to single or double.  Changing from complex or
+zomplex to real discards the imaginary part.
+A supernodal factor cannot be changed to the zomplex \verb'xtype'.
+A factor cannot be converted to an \verb'xtype' of pattern-only with this method;
+use \verb'cholmod_change_factor' for that operation.
+
+The name of this method derives from an earlier version of CHOLMOD where all
+matrices had an implied \verb'dtype' of \verb'CHOLMOD_DOUBLE'.  The method now
+supports changes to the \verb'dtype'.  For backward compatilibilty, the name of
+this method has not changed, and the \verb'to_xdtype' parameter values of the
+prior versions are upwards compatible with this version of CHOLMOD.
+
 %---------------------------------------
 \subsection{How many entries are in a CHOLMOD sparse Cholesky factorization?}
 %---------------------------------------
@@ -2464,9 +2489,17 @@ \subsection{{\tt cholmod\_dense\_xtype}: change dense matrix xtype}
 %---------------------------------------
 
 \input{_dense_xtype.tex}
-Changes the {\tt xtype} and/or {\tt dtype} of a dense matrix, to real, complex,
-or zomplex.  Changing from complex or zomplex to real discards the imaginary
-part.  A dense matrix cannot be converted to an xtype of pattern-only.
+Changes the {\tt xtype} and/or \verb'dtype' of a dense matrix.
+The \verb'xtype' can change to real, complex, or zomplex (not pattern-only).
+The \verb'dtype' can change to single or double.
+Changing from complex or zomplex to real discards the imaginary part.
+A dense matrix cannot be converted to an \verb'xtype' of pattern-only.
+
+The name of this method derives from an earlier version of CHOLMOD where all
+matrices had an implied \verb'dtype' of \verb'CHOLMOD_DOUBLE'.  The method now
+supports changes to the \verb'dtype'.  For backward compatilibilty, the name of
+this method has not changed, and the \verb'to_xdtype' parameter values of the
+prior versions are upwards compatible with this version of CHOLMOD.
 
 %-------------------------------------------------------------------------------
 \newpage \section{{\tt Utility} Module: {\tt cholmod\_triplet} object}
@@ -2534,9 +2567,17 @@ \subsection{{\tt cholmod\_triplet\_xtype}: change triplet xtype}
 %---------------------------------------
 
 \input{_triplet_xtype.tex}
-Changes the {\tt xtype} of a dense matrix, to real, complex, or zomplex.
+Changes the {\tt xtype} and/or \verb'dtype' of a triplet matrix.
+The \verb'xtype' can change to pattern, real, complex, or zomplex,
+and the \verb'dtype' can change to single or double.
 Changing from complex or zomplex to real discards the imaginary part.
 
+The name of this method derives from an earlier version of CHOLMOD where all
+matrices had an implied \verb'dtype' of \verb'CHOLMOD_DOUBLE'.  The method now
+supports changes to the \verb'dtype'.  For backward compatilibilty, the name of
+this method has not changed, and the \verb'to_xdtype' parameter values of the
+prior versions are upwards compatible with this version of CHOLMOD.
+
 %-------------------------------------------------------------------------------
 \newpage \section{{\tt Utility} Module: memory management}
 %-------------------------------------------------------------------------------
@@ -3630,7 +3671,7 @@ \subsection{{\tt cholmod\_rowadd}: add row to factor}
 matrix on input.  Only real matrices are supported (double or single).  The
 dtypes of all matrices must match, except when \verb'L' is symbolic (with an
 xtype of \verb'CHOLMOD_PATTERN').  In this case, \verb'L' is converted to the
-dtype of \verb'R'.  The algorithm is described in \cite{DavisHager05}.  
+dtype of \verb'R'.  The algorithm is described in \cite{DavisHager05}.
 
 %---------------------------------------
 \subsection{{\tt cholmod\_rowadd\_solve}: add row to factor and update a

SPQR/Demo/go4.m

@@ -14,7 +14,7 @@
 nmat = length (f) ;
 
 keep = ones (nmat,1) ;
-kinds = UFkinds ;
+kinds = sskinds ;
 for k = 1:nmat
     id = f (k) ;
     if (~isempty (strfind (kinds {id}, 'subsequent')))

SPQR/Demo/go5.m

@@ -14,7 +14,7 @@
 nmat = length (f) ;
 
 keep = ones (nmat,1) ;
-kinds = UFkinds ;
+kinds = sskinds ;
 for k = 1:nmat
     id = f (k) ;
     if (~isempty (strfind (kinds {id}, 'subsequent')))

SPQR/GPUQREngine/CMakeLists.txt

@@ -66,7 +66,12 @@ configure_file ( "Config/GPUQREngine.hpp.in"
 
 #-------------------------------------------------------------------------------
 
-file ( GLOB GPUQRENGINE_SOURCES "Source/*.cpp" "Source/*.cu" "Source/*/*.cpp" )
+file ( GLOB GPUQRENGINE_SOURCES "Source/*.cpp"
+    "Source/*.cu"
+    "Source/BucketList/BucketList.cpp"
+    "Source/LLBundle/LLBundle.cpp"
+    "Source/Scheduler/Scheduler.cpp"
+    "Source/TaskDescriptor/*.cpp" )
 
 set ( GPUQRENGINE_INCLUDES Include Include/Kernel
     Include/Kernel/Apply Include/Kernel/Assemble Include/Kernel/Factorize )

SPQR/GPUQREngine/Include/GPUQREngine_BucketList.hpp

@@ -21,6 +21,7 @@
 #define GPUQRENGINE_BUCKETLIST_HPP
 
 #include <cstddef>
+#include <new>
 
 #include "GPUQREngine_Common.hpp"
 #include "GPUQREngine_TaskDescriptor.hpp"
@@ -74,10 +75,6 @@ template <typename Int = int64_t> class BucketList
     int VThead;              // Index of the first available entry in VTlist
 
     // Constructors
-    void *operator new(std::size_t, BucketList <Int>* p)
-    {
-        return p;
-    }
     BucketList(Front <Int> *f, Int minApplyGranularity);
     ~BucketList();
 
@@ -190,7 +187,9 @@ template <typename Int = int64_t> class BucketList
     );
 };
 
+#if ! defined (GPUQRENGINE_NO_EXTERN_BUCKETLIST)
 extern template class BucketList<int32_t>;
 extern template class BucketList<int64_t>;
+#endif
 
 #endif

SPQR/GPUQREngine/Include/GPUQREngine_Front.hpp

@@ -20,6 +20,7 @@
 #define GPUQRENGINE_FRONT_HPP
 
 #include <cstddef>
+#include <new>
 
 #include "GPUQREngine_Common.hpp"
 #include "GPUQREngine_SparseMeta.hpp"
@@ -55,8 +56,6 @@ template <typename Int = int64_t> class Front
     /* Debug Fields */
     bool printMe;
 
-    void* operator new(std::size_t reqMem, Front* ptr){ return ptr; }
-
     Front(
         Int fids_arg,                   // the front identifier
         Int pids_arg,                   // the parent identifier

SPQR/GPUQREngine/Include/GPUQREngine_LLBundle.hpp

@@ -25,6 +25,7 @@
 #define GPUQRENGINE_LLBUNDLE_HPP
 
 #include <cstddef>
+#include <new>
 
 #include "GPUQREngine_Common.hpp"
 #include "GPUQREngine_TaskDescriptor.hpp"
@@ -78,7 +79,6 @@ template <typename Int = int64_t> class LLBundle
 
     TaskType CurrentTask;
 
-    void *operator new(std::size_t, LLBundle <Int>* p){ return p; }
     //------------------------------------------------------------------------------
     //
     // This file contains the constructor and destructor for the LLBundle class.
@@ -127,6 +127,9 @@ template <typename Int = int64_t> class LLBundle
     void gpuPack(TaskDescriptor *cpuTask);
 };
 
+#if ! defined (GPUQRENGINE_NO_EXTERN_LLBUNDLE)
 extern template class LLBundle<int32_t>;
 extern template class LLBundle<int64_t>;
 #endif
+
+#endif

SPQR/GPUQREngine/Include/GPUQREngine_Scheduler.hpp

@@ -20,6 +20,7 @@
 #define GPUQRENGINE_SCHEDULER_HPP
 
 #include <cstddef>
+#include <new>
 
 #include "GPUQREngine_Common.hpp"
 #include "GPUQREngine_FrontState.hpp"
@@ -99,7 +100,6 @@ class Scheduler
     cudaStream_t memoryStreamD2H;
 
     /* Scheduler.cpp */
-    void *operator new(std::size_t, Scheduler <Int>* p){ return p; }
     Scheduler(Front <Int> *fronts, Int numFronts, size_t gpuMemorySize);
     ~Scheduler();
 
@@ -174,6 +174,9 @@ class Scheduler
 #endif
 };
 
+#if ! defined (GPUQRENGINE_NO_EXTERN_SCHEDULER)
 extern template class Scheduler<int32_t>;
 extern template class Scheduler<int64_t>;
 #endif
+
+#endif

SPQR/GPUQREngine/Source/BucketList/BucketList.cpp

@@ -31,6 +31,9 @@
     Bundles = (LLBundle <Int> *) SuiteSparse_free(Bundles); \
     gpuVT = (double **) SuiteSparse_free(gpuVT); \
     wsMongoVT = Workspace::destroy(wsMongoVT);
+
+#define GPUQRENGINE_NO_EXTERN_BUCKETLIST
+
 #include "GPUQREngine_BucketList.hpp"
 template <typename Int>
 BucketList<Int>::BucketList
@@ -103,16 +106,6 @@ BucketList<Int>::BucketList
     }
 }
 
-template BucketList<int32_t>::BucketList
-(
-    Front <int32_t> *F,
-    int32_t minApplyGranularity
-) ;
-template BucketList<int64_t>::BucketList
-(
-    Front <int64_t> *F,
-    int64_t minApplyGranularity
-) ;
 
 template <typename Int>
 BucketList<Int>::~BucketList()
@@ -121,9 +114,6 @@ BucketList<Int>::~BucketList()
 
 }
 
-template BucketList<int32_t>::~BucketList() ;
-template BucketList<int64_t>::~BucketList() ;
-
 
 template <typename Int>
 void BucketList<Int>::Initialize()
@@ -149,5 +139,17 @@ void BucketList<Int>::Initialize()
     }
 }
 
-template void BucketList<int32_t>::Initialize() ;
-template void BucketList<int64_t>::Initialize() ;
+
+// Instantiate BucketList class template with types int32_t and int64_t
+// for export from library.
+
+#include "BucketList_AdvanceBundles.cpp"
+#include "BucketList_CreateBundles.cpp"
+#include "BucketList_FillWorkQueue.cpp"
+#include "BucketList_GrowBundles.cpp"
+#include "BucketList_Manage.cpp"
+#include "BucketList_PostProcessing.cpp"
+
+template class BucketList<int32_t>;
+template class BucketList<int64_t>;
+

SPQR/GPUQREngine/Source/BucketList/BucketList_AdvanceBundles.cpp

@@ -43,6 +43,3 @@ void BucketList<Int>::AdvanceBundles()
         }
     }
 }
-
-template void BucketList<int32_t>::AdvanceBundles() ;
-template void BucketList<int64_t>::AdvanceBundles() ;

SPQR/GPUQREngine/Source/BucketList/BucketList_CreateBundles.cpp

@@ -58,15 +58,6 @@ void BucketList<Int>::CreateBundles
     }
 }
 
-template void BucketList<int32_t>::CreateBundles
-(
-    void
-) ;
-template void BucketList<int64_t>::CreateBundles
-(
-    void
-) ;
-
 // SkipBundleCreation determines whether we should skip creating a new
 // bundle for the specified tile in the specified column bucket.
 template <typename Int>
@@ -93,17 +84,6 @@ bool BucketList<Int>::SkipBundleCreation
     return false;
 }
 
-template bool BucketList<int32_t>::SkipBundleCreation
-(
-    int32_t tile,           // The tile to consider
-    int32_t colBucket       // The column bucket it sits in
-) ;
-template bool BucketList<int64_t>::SkipBundleCreation
-(
-    int64_t tile,           // The tile to consider
-    int64_t colBucket       // The column bucket it sits in
-) ;
-
 // IsInternal determines whether a tile is completely within the bounds
 // of the front because if it isn't then we will need to use the special
 // edge case kernels.
@@ -133,14 +113,3 @@ bool BucketList<Int>::IsInternal
     Int iLast = TILESIZE * (iTile+1) - 1;
     return(iLast < front->fm && jLast < front->fn);
 }
-
-template bool BucketList<int32_t>::IsInternal
-(
-    LLBundle <int32_t>& Bundle,
-    int jLast
-) ;
-template bool BucketList<int64_t>::IsInternal
-(
-    LLBundle <int64_t>& Bundle,
-    int jLast
-) ;