Add optional use of contiguous arrays for sparse matrix

The reference sparse matrix data structures in GenerateProblem_ref.cpp
can be allocated with a single “new” statement. This feature is turned
on by using the new compile-time macro HPCG_CONTIGUOUS_ARRAYS.  This
features is off by default, but can turned on in the same ways as
working with other HPCG_* defines.

INSTALL

@@ -175,6 +175,11 @@ options and their meaning are:
 
     -DHPCG_DEBUG
 
+* Compile with sparse matrix arrays allocated contiguously. This option
+* may be helpful on systems with pre-fetch::
+
+    -DHPCG_CONTIGUOUS_ARRAYS
+
 * Compile with voluminous debugging information turned on::
 
     -DHPCG_DETAILED_DEBUG

setup/Make.GCC_OMP

@@ -98,6 +98,7 @@ HPCG_LIBS     =
 #
 # -DHPCG_NO_MPI	        Define to disable MPI
 # -DHPCG_NO_OPENMP	Define to disable OPENMP
+# -DHPCG_CONTIGUOUS_ARRAYS Define to have sparse matrix arrays long and contiguous
 # -DHPCG_DEBUG       	Define to enable debugging output
 # -DHPCG_DETAILED_DEBUG Define to enable very detailed debugging output
 #

setup/Make.ICPC_OMP

@@ -98,6 +98,7 @@ HPCG_LIBS     =
 #
 # -DHPCG_NO_MPI	        Define to disable MPI
 # -DHPCG_NO_OPENMP	Define to disable OPENMP
+# -DHPCG_CONTIGUOUS_ARRAYS Define to have sparse matrix arrays long and contiguous
 # -DHPCG_DEBUG       	Define to enable debugging output
 # -DHPCG_DETAILED_DEBUG Define to enable very detailed debugging output
 #

setup/Make.Linux_MPI

@@ -98,6 +98,7 @@ HPCG_LIBS     =
 #
 # -DHPCG_NO_MPI	        Define to disable MPI
 # -DHPCG_NO_OPENMP	Define to disable OPENMP
+# -DHPCG_CONTIGUOUS_ARRAYS Define to have sparse matrix arrays long and contiguous
 # -DHPCG_DEBUG       	Define to enable debugging output
 # -DHPCG_DETAILED_DEBUG Define to enable very detailed debugging output
 #

setup/Make.Linux_Serial

@@ -98,6 +98,7 @@ HPCG_LIBS     =
 #
 # -DHPCG_NO_MPI	        Define to disable MPI
 # -DHPCG_NO_OPENMP	Define to disable OPENMP
+# -DHPCG_CONTIGUOUS_ARRAYS Define to have sparse matrix arrays long and contiguous
 # -DHPCG_DEBUG       	Define to enable debugging output
 # -DHPCG_DETAILED_DEBUG Define to enable very detailed debugging output
 #

setup/Make.MPIICPC_OMP

@@ -98,6 +98,7 @@ HPCG_LIBS     =
 #
 # -DHPCG_NO_MPI		Define to disable MPI
 # -DHPCG_NO_OPENMP	Define to disable OPENMP
+# -DHPCG_CONTIGUOUS_ARRAYS Define to have sparse matrix arrays long and contiguous
 # -DHPCG_DEBUG       	Define to enable debugging output
 # -DHPCG_DETAILED_DEBUG Define to enable very detailed debugging output
 #

setup/Make.MPI_GCC_OMP

@@ -98,6 +98,7 @@ HPCG_LIBS     =
 #
 # -DHPCG_NO_MPI	        Define to disable MPI
 # -DHPCG_NO_OPENMP	Define to disable OPENMP
+# -DHPCG_CONTIGUOUS_ARRAYS Define to have sparse matrix arrays long and contiguous
 # -DHPCG_DEBUG       	Define to enable debugging output
 # -DHPCG_DETAILED_DEBUG Define to enable very detailed debugging output
 #

setup/Make.MPI_ICPC

@@ -98,6 +98,7 @@ HPCG_LIBS     =
 #
 # -DHPCG_NO_MPI	        Define to disable MPI
 # -DHPCG_NO_OPENMP	Define to disable OPENMP
+# -DHPCG_CONTIGUOUS_ARRAYS Define to have sparse matrix arrays long and contiguous
 # -DHPCG_DEBUG       	Define to enable debugging output
 # -DHPCG_DETAILED_DEBUG Define to enable very detailed debugging output
 #

setup/Make.MPI_ICPC_OMP

@@ -98,6 +98,7 @@ HPCG_LIBS     =
 #
 # -DHPCG_NO_MPI	        Define to enable MPI
 # -DHPCG_NO_OPENMP	Define to disable OPENMP
+# -DHPCG_CONTIGUOUS_ARRAYS Define to have sparse matrix arrays long and contiguous
 # -DHPCG_DEBUG       	Define to enable debugging output
 # -DHPCG_DETAILED_DEBUG Define to enable very detailed debugging output
 #

setup/Make.Mac_MPI

@@ -98,6 +98,7 @@ HPCG_LIBS     =
 #
 # -DHPCG_NO_MPI		Define to disable MPI
 # -DHPCG_NO_OPENMP	Define to disable OPENMP
+# -DHPCG_CONTIGUOUS_ARRAYS Define to have sparse matrix arrays long and contiguous
 # -DHPCG_DEBUG       	Define to enable debugging output
 # -DHPCG_DETAILED_DEBUG Define to enable very detailed debugging output
 #

setup/Make.Mac_Serial

@@ -98,6 +98,7 @@ HPCG_LIBS     =
 #
 # -DHPCG_NO_MPI		Define to disable MPI
 # -DHPCG_NO_OPENMP	Define to disable OPENMP
+# -DHPCG_CONTIGUOUS_ARRAYS Define to have sparse matrix arrays long and contiguous
 # -DHPCG_DEBUG       	Define to enable debugging output
 # -DHPCG_DETAILED_DEBUG Define to enable very detailed debugging output
 #

setup/Make.Mac_Serial_debug

@@ -98,6 +98,7 @@ HPCG_LIBS     =
 #
 # -DHPCG_NO_MPI		Define to disable MPI
 # -DHPCG_NO_OPENMP	Define to disable OPENMP
+# -DHPCG_CONTIGUOUS_ARRAYS Define to have sparse matrix arrays long and contiguous
 # -DHPCG_DEBUG       	Define to enable debugging output
 # -DHPCG_DETAILED_DEBUG Define to enable very detailed debugging output
 #

src/GenerateProblem_ref.cpp

@@ -103,14 +103,28 @@ void GenerateProblem_ref(SparseMatrix & A, Vector * b, Vector * x, Vector * xexa
     mtxIndG[i] = 0;
     mtxIndL[i] = 0;
   }
+
+#ifndef HPCG_CONTIGUOUS_ARRAYS
   // Now allocate the arrays pointed to
-  for (local_int_t i=0; i< localNumberOfRows; ++i) {
+  for (local_int_t i=0; i< localNumberOfRows; ++i)
     mtxIndL[i] = new local_int_t[numberOfNonzerosPerRow];
+  for (local_int_t i=0; i< localNumberOfRows; ++i)
     matrixValues[i] = new double[numberOfNonzerosPerRow];
-    mtxIndG[i] = new global_int_t[numberOfNonzerosPerRow];
-  }
-
+  for (local_int_t i=0; i< localNumberOfRows; ++i)
+   mtxIndG[i] = new global_int_t[numberOfNonzerosPerRow];
 
+#else
+  // Now allocate the arrays pointed to
+  mtxIndL[0] = new local_int_t[localNumberOfRows * numberOfNonzerosPerRow];
+  matrixValues[0] = new double[localNumberOfRows * numberOfNonzerosPerRow];
+  mtxIndG[0] = new global_int_t[localNumberOfRows * numberOfNonzerosPerRow];
+
+  for (local_int_t i=1; i< localNumberOfRows; ++i) {
+  mtxIndL[i] = mtxIndL[0] + i * numberOfNonzerosPerRow;
+  matrixValues[i] = matrixValues[0] + i * numberOfNonzerosPerRow;
+  mtxIndG[i] = mtxIndG[0] + i * numberOfNonzerosPerRow;
+  }
+#endif
 
   local_int_t localNumberOfNonzeros = 0;
   // TODO:  This triply nested loop could be flattened or use nested parallelism

src/SparseMatrix.hpp

@@ -150,12 +150,17 @@ inline void ReplaceMatrixDiagonal(SparseMatrix & A, Vector & diagonal) {
  */
 inline void DeleteMatrix(SparseMatrix & A) {
 
+#ifndef HPCG_CONTIGUOUS_ARRAYS
   for (local_int_t i = 0; i< A.localNumberOfRows; ++i) {
     delete [] A.matrixValues[i];
     delete [] A.mtxIndG[i];
     delete [] A.mtxIndL[i];
   }
-
+#else
+  delete [] A.matrixValues[0];
+  delete [] A.mtxIndG[0];
+  delete [] A.mtxIndL[0];
+#endif
   if (A.title)                  delete [] A.title;
   if (A.nonzerosInRow)             delete [] A.nonzerosInRow;
   if (A.mtxIndG) delete [] A.mtxIndG;