Update version number to 7.0

xiaoyeli · May 12, 2021 · 1a6d5ba · 1a6d5ba
1 parent cba813d
commit 1a6d5ba
Show file tree

Hide file tree

Showing 15 changed files with 175 additions and 100 deletions.
diff --git a/EXAMPLE/README b/EXAMPLE/README
@@ -1,5 +1,5 @@
-		SuperLU_DIST  EXAMPLES
-		======================
+		SuperLU_DIST EXAMPLES
+		=====================
 
 This directory contains sample programs to illustrate how to use
 various functions provided in SuperLU_DIST. You can modify these

diff --git a/FORTRAN/README b/FORTRAN/README
@@ -1,4 +1,5 @@
 		Fortran 90 Interface
+		====================
 
 This directory contains Fortran-90 wrapper routines for SuperLU_DIST.
 The directory contains the following files:

diff --git a/FORTRAN/superlu_c2f_wrap.c b/FORTRAN/superlu_c2f_wrap.c
@@ -4,10 +4,11 @@
  * \brief C interface functions for the Fortran90 wrapper.
  *
  * <pre>
- * -- Distributed SuperLU routine (version 4.1) --
+ * -- Distributed SuperLU routine (version 7.0) --
  * Lawrence Berkeley National Lab, Univ. of California Berkeley.
  * October 2012
  * April 5, 2015
+ * May 12, 2021
  */
 
 #include "superlu_defs.h"

diff --git a/FORTRAN/superlu_dist_config.fh b/FORTRAN/superlu_dist_config.fh
@@ -1,9 +1,11 @@
 
+#define HAVE_CUDA TRUE
 
 #define HAVE_PARMETIS TRUE
 
 
 
+#define XSDK_INDEX_SIZE 64
 
 #if (XSDK_INDEX_SIZE == 64)
 #define _LONGINT 1

diff --git a/README.md b/README.md
@@ -355,7 +355,7 @@ to obtain these parameters. This function is provided in the file SRC/sp_ienv.c.
 Please consult that file for detailed description of the meanings.
 ```
     export NREL=<...>   // supernode relaxation parameter
-    export NSUP=<...>   // maximum allowable supernode size
+    export NSUP=<...>   // maximum allowable supernode size, not to exceed 512
     export FILL=<...>   // estimated fill ratio of nonzeros(L+U)/nonzeros(A)
     export MAX_BUFFER_SIZE=<...>   // maximum buffer size on GPU for GEMM
 ```

diff --git a/SRC/communication_aux.c b/SRC/communication_aux.c
@@ -9,6 +9,16 @@ The source code is distributed under BSD license, see the file License.txt
 at the top-level directory.
 */
 
+/*! @file
+ * \brief Auxiliary routines to support communication in 3D algorithms
+ *
+ * <pre>
+ * -- Distributed SuperLU routine (version 7.0) --
+ * Lawrence Berkeley National Lab, Oak Ridge National Lab
+ * May 12, 2021
+ * </pre>
+ */
+
 #include "superlu_defs.h"
 #if 0
 #include "sec_structs.h"

diff --git a/SRC/sec_structs.c b/SRC/sec_structs.c
@@ -1,3 +1,25 @@
+/*! \file
+Copyright (c) 2003, The Regents of the University of California, through
+Lawrence Berkeley National Laboratory (subject to receipt of any required
+approvals from U.S. Dept. of Energy)
+
+All rights reserved.
+
+The source code is distributed under BSD license, see the file License.txt
+at the top-level directory.
+*/
+
+/*! @file
+ * \brief Auxiliary routines in 3D algorithms
+ *
+ * <pre>
+ * -- Distributed SuperLU routine (version 7.0) --
+ * Lawrence Berkeley National Lab, Oak Ridge National Lab
+ * May 12, 2021
+ * </pre>
+ */
+
+
 //#include <mm_malloc.h>
 #include "superlu_ddefs.h"
 #if 0

diff --git a/SRC/superlu_FortranCInterface.h b/SRC/superlu_FortranCInterface.h
@@ -2,15 +2,15 @@
 #define FC_HEADER_INCLUDED
 
 /* Mangling for Fortran global symbols without underscores. */
-#define FC_GLOBAL(name,NAME) name##_
+#define FC_GLOBAL(name,NAME) name
 
 /* Mangling for Fortran global symbols with underscores. */
-#define FC_GLOBAL_(name,NAME) name##_
+#define FC_GLOBAL_(name,NAME) name
 
 /* Mangling for Fortran module symbols without underscores. */
-#define FC_MODULE(mod_name,name, mod_NAME,NAME) __##mod_name##_MOD_##name
+#define FC_MODULE(mod_name,name, mod_NAME,NAME) __##mod_name##_NMOD_##name
 
 /* Mangling for Fortran module symbols with underscores. */
-#define FC_MODULE_(mod_name,name, mod_NAME,NAME) __##mod_name##_MOD_##name
+#define FC_MODULE_(mod_name,name, mod_NAME,NAME) __##mod_name##_NMOD_##name
 
 #endif
diff --git a/SRC/superlu_defs.h b/SRC/superlu_defs.h
@@ -23,6 +23,7 @@ at the top-level directory.
  *     February 8, 2019    version 6.1.1
  *     November 12, 2019   version 6.2.0
  *     October 23, 2020    version 6.4.0
+ *     May 12, 2021        version 7.0.0
  * </pre>
  */
 
@@ -72,10 +73,10 @@ at the top-level directory.
  *   #endif
  * Versions 4.x and earlier do not include a #define'd version numbers.
  */
-#define SUPERLU_DIST_MAJOR_VERSION     6
-#define SUPERLU_DIST_MINOR_VERSION     4
+#define SUPERLU_DIST_MAJOR_VERSION     7
+#define SUPERLU_DIST_MINOR_VERSION     0
 #define SUPERLU_DIST_PATCH_VERSION     0
-#define SUPERLU_DIST_RELEASE_DATE      "October 23, 2020"
+#define SUPERLU_DIST_RELEASE_DATE      "May 12, 2021"
 
 #include "superlu_dist_config.h"
 /* Define my integer size int_t */

diff --git a/SRC/superlu_dist_config.h b/SRC/superlu_dist_config.h
@@ -1,7 +1,7 @@
 /* superlu_dist_config.h.in */
 
 /* Enable CUDA */
-/* #undef HAVE_CUDA */
+#define HAVE_CUDA TRUE
 
 /* Enable parmetis */
 #define HAVE_PARMETIS TRUE
@@ -13,7 +13,7 @@
 /* #undef HAVE_COMBBLAS */
 
 /* enable 64bit index mode */
-/* #undef XSDK_INDEX_SIZE */
+#define XSDK_INDEX_SIZE 64
 
 #if (XSDK_INDEX_SIZE == 64)
 #define _LONGINT 1

diff --git a/SRC/superlu_grid3d.c b/SRC/superlu_grid3d.c
@@ -3,8 +3,8 @@
  *
  * <pre>
  * -- Distributed SuperLU routine (version 7.0.0) --
- * Lawrence Berkeley National Lab, Univ. of California Berkeley.
- * March 30, 2019
+ * Lawrence Berkeley National Lab, Oak Ridge National Lab
+ * May 12, 2021
  * </pre>
  */
 

diff --git a/SRC/supernodalForest.c b/SRC/supernodalForest.c
@@ -1,3 +1,12 @@
+/*! @file
+ * \brief SuperLU utilities
+ *
+ * <pre>
+ * -- Distributed SuperLU routine (version 7.0) --
+ * Lawrence Berkeley National Lab, Oak Ridge National Lab
+ * May 12, 2021
+ * </pre>
+ */
 #include <stdio.h>
 #include <assert.h>
 #include "superlu_ddefs.h"

diff --git a/SRC/supernodal_etree.c b/SRC/supernodal_etree.c
@@ -1,4 +1,13 @@
-/*function to generate supernodal etree*/
+/*! @file
+ * \brief function to generate supernodal etree
+*
+ * <pre>
+ * -- Distributed SuperLU routine (version 7.0) --
+ * Lawrence Berkeley National Lab, Oak Ridge National Lab
+ * May 12, 2021
+ * </pre>
+ */
+
 #include <stdio.h>
 #include <assert.h>
 #include "superlu_ddefs.h"

diff --git a/SRC/treeFactorization.c b/SRC/treeFactorization.c
@@ -9,90 +9,100 @@ The source code is distributed under BSD license, see the file License.txt
 at the top-level directory.
 */
 
-#include "superlu_ddefs.h"
-#if 0
-#include "treeFactorization.h"
-#include "trfCommWrapper.h"
-#endif
-
-#if 0 /******** Sherry: Remove extra layer of function calls.  *******/
-int_t sDiagFactIBCast(int_t k,  diagFactBufs_t *dFBuf,
-                      factStat_t *factStat,
-                      commRequests_t *comReqs,
-                      gridinfo_t *grid,
-                      superlu_dist_options_t *options,
-                      double thresh,
-                      LUstruct_t *LUstruct,
-                      SuperLUStat_t *stat, int *info,
-                      SCT_t *SCT,
-		      int tag_ub
-                     )
-{
-    MPI_Request * U_diag_blk_recv_req  = comReqs->U_diag_blk_recv_req;
-    MPI_Request * L_diag_blk_recv_req  = comReqs->L_diag_blk_recv_req;
-    MPI_Request * U_diag_blk_send_req  = comReqs->U_diag_blk_send_req;
-    MPI_Request * L_diag_blk_send_req  = comReqs->L_diag_blk_send_req;
-    int_t * IrecvPlcd_D = factStat->IrecvPlcd_D;
-
-    double * BlockUFactor =  dFBuf->BlockUFactor;
-    double * BlockLFactor =  dFBuf->BlockLFactor;
-    dDiagFactIBCast(k,  k,   BlockUFactor, BlockLFactor,
-                   IrecvPlcd_D,
-                   U_diag_blk_recv_req, L_diag_blk_recv_req,
-                   U_diag_blk_send_req, L_diag_blk_send_req,
-                   grid, options, thresh, LUstruct, stat, info, SCT, tag_ub);
-    return 0;
-}
-int_t sLPanelUpdate( int_t k,  diagFactBufs_t *dFBuf,
-                     factStat_t *factStat,
-                     commRequests_t *comReqs,
-                     gridinfo_t *grid,
-                     LUstruct_t *LUstruct, SCT_t *SCT)
-{
-    MPI_Request * U_diag_blk_recv_req  = comReqs->U_diag_blk_recv_req;
-    int_t * IrecvPlcd_D = factStat->IrecvPlcd_D;
-    int_t * factored_L = factStat->factored_L;
-    double * BlockUFactor =  dFBuf->BlockUFactor;
-
-    dLPanelUpdate( k,  IrecvPlcd_D, factored_L,
-		   U_diag_blk_recv_req, BlockUFactor, grid, LUstruct, SCT);
-    return 0;
-}
-
-int_t sUPanelUpdate( int_t k,
-                     int_t ldt,
-                     diagFactBufs_t *dFBuf,
-                     factStat_t *factStat,
-                     commRequests_t *comReqs,
-                     scuBufs_t* scuBufs,
-                     packLUInfo_t* packLUInfo,
-                     gridinfo_t *grid,
-                     LUstruct_t *LUstruct,
-                     SuperLUStat_t *stat, SCT_t *SCT)
-{
-    double* bigV = scuBufs->bigV;
-    Ublock_info_t* Ublock_info = packLUInfo->Ublock_info;
-
-    MPI_Request * L_diag_blk_recv_req  = comReqs->L_diag_blk_recv_req;
-
-    int_t * factored_U = factStat->factored_U;
-
-    double * BlockLFactor =  dFBuf->BlockLFactor;
-    dUPanelUpdate(k, factored_U, L_diag_blk_recv_req, BlockLFactor, bigV, ldt,
-		  Ublock_info, grid, LUstruct, stat, SCT);
-    return 0;
-}
-int_t sIBcastRecvLPanel(
-    int_t k,
-    commRequests_t *comReqs,
-    LUValSubBuf_t* LUvsb,
-    msgs_t* msgs,
-    factStat_t *factStat,
-    gridinfo_t *grid,
-    LUstruct_t *LUstruct, SCT_t *SCT, int tag_ub)
-
-{
-    int* msgcnt  = msgs->msgcnt;
+/*! @file
+ * \brief factorization routines in 3D algorithms
+ *
+ * <pre>
+ * -- Distributed SuperLU routine (version 7.0) --
+ * Lawrence Berkeley National Lab, Oak Ridge National Lab
+ * May 12, 2021
+ * </pre>
+ */
+
+ #include "superlu_ddefs.h"
+ #if 0
+ #include "treeFactorization.h"
+ #include "trfCommWrapper.h"
+ #endif
+
+ #if 0 /******** Sherry: Remove extra layer of function calls.  *******/
+ int_t sDiagFactIBCast(int_t k,  diagFactBufs_t *dFBuf,
+		       factStat_t *factStat,
+		       commRequests_t *comReqs,
+		       gridinfo_t *grid,
+		       superlu_dist_options_t *options,
+		       double thresh,
+		       LUstruct_t *LUstruct,
+		       SuperLUStat_t *stat, int *info,
+		       SCT_t *SCT,
+		       int tag_ub
+		      )
+ {
+     MPI_Request * U_diag_blk_recv_req  = comReqs->U_diag_blk_recv_req;
+     MPI_Request * L_diag_blk_recv_req  = comReqs->L_diag_blk_recv_req;
+     MPI_Request * U_diag_blk_send_req  = comReqs->U_diag_blk_send_req;
+     MPI_Request * L_diag_blk_send_req  = comReqs->L_diag_blk_send_req;
+     int_t * IrecvPlcd_D = factStat->IrecvPlcd_D;
+
+     double * BlockUFactor =  dFBuf->BlockUFactor;
+     double * BlockLFactor =  dFBuf->BlockLFactor;
+     dDiagFactIBCast(k,  k,   BlockUFactor, BlockLFactor,
+		    IrecvPlcd_D,
+		    U_diag_blk_recv_req, L_diag_blk_recv_req,
+		    U_diag_blk_send_req, L_diag_blk_send_req,
+		    grid, options, thresh, LUstruct, stat, info, SCT, tag_ub);
+     return 0;
+ }
+ int_t sLPanelUpdate( int_t k,  diagFactBufs_t *dFBuf,
+		      factStat_t *factStat,
+		      commRequests_t *comReqs,
+		      gridinfo_t *grid,
+		      LUstruct_t *LUstruct, SCT_t *SCT)
+ {
+     MPI_Request * U_diag_blk_recv_req  = comReqs->U_diag_blk_recv_req;
+     int_t * IrecvPlcd_D = factStat->IrecvPlcd_D;
+     int_t * factored_L = factStat->factored_L;
+     double * BlockUFactor =  dFBuf->BlockUFactor;
+
+     dLPanelUpdate( k,  IrecvPlcd_D, factored_L,
+		    U_diag_blk_recv_req, BlockUFactor, grid, LUstruct, SCT);
+     return 0;
+ }
+
+ int_t sUPanelUpdate( int_t k,
+		      int_t ldt,
+		      diagFactBufs_t *dFBuf,
+		      factStat_t *factStat,
+		      commRequests_t *comReqs,
+		      scuBufs_t* scuBufs,
+		      packLUInfo_t* packLUInfo,
+		      gridinfo_t *grid,
+		      LUstruct_t *LUstruct,
+		      SuperLUStat_t *stat, SCT_t *SCT)
+ {
+     double* bigV = scuBufs->bigV;
+     Ublock_info_t* Ublock_info = packLUInfo->Ublock_info;
+
+     MPI_Request * L_diag_blk_recv_req  = comReqs->L_diag_blk_recv_req;
+
+     int_t * factored_U = factStat->factored_U;
+
+     double * BlockLFactor =  dFBuf->BlockLFactor;
+     dUPanelUpdate(k, factored_U, L_diag_blk_recv_req, BlockLFactor, bigV, ldt,
+		   Ublock_info, grid, LUstruct, stat, SCT);
+     return 0;
+ }
+ int_t sIBcastRecvLPanel(
+     int_t k,
+     commRequests_t *comReqs,
+     LUValSubBuf_t* LUvsb,
+     msgs_t* msgs,
+     factStat_t *factStat,
+     gridinfo_t *grid,
+     LUstruct_t *LUstruct, SCT_t *SCT, int tag_ub)
+
+ {
+     int* msgcnt  = msgs->msgcnt;
     MPI_Request *send_req  = comReqs->send_req;
     MPI_Request *recv_req  = comReqs->recv_req;
     int_t * Lsub_buf  = LUvsb->Lsub_buf;

diff --git a/SRC/trfAux.c b/SRC/trfAux.c
@@ -9,6 +9,16 @@ The source code is distributed under BSD license, see the file License.txt
 at the top-level directory.
 */
 
+/*! @file
+ * \brief Auxiliary routines to support 3D algorithms
+ *
+ * <pre>
+ * -- Distributed SuperLU routine (version 7.0) --
+ * Lawrence Berkeley National Lab, Oak Ridge National Lab
+ * May 12, 2021
+ * </pre>
+ */
+
 #include "superlu_ddefs.h"
 
 #if 0