add skelton for bidir

CMakeLists.txt

@@ -148,7 +148,7 @@ src/pgesv/HPL_pdupdateNT.cpp
 src/pgesv/HPL_pdupdateNN.cpp src/pgesv/HPL_pdupdateTN.cpp src/pgesv/HPL_pdupdateTT.cpp
 src/pgesv/HPL_pdupdateTN.cpp src/pgesv/HPL_pdupdateTT.cpp
 src/pauxil/HPL_pwarn.cpp src/comm/HPL_bcast.cpp src/comm/HPL_blong.cpp
-src/comm/HPL_blonM.cpp src/comm/HPL_1ring.cpp src/comm/HPL_2ring.cpp
+src/comm/HPL_blonM.cpp src/comm/HPL_bidir.cpp src/comm/HPL_1ring.cpp src/comm/HPL_2ring.cpp
 src/comm/HPL_1rinM.cpp src/comm/HPL_2rinM.cpp src/comm/HPL_packL.cpp
 src/comm/HPL_sdrv.cpp src/comm/HPL_send.cpp src/pgesv/HPL_pdlaswp00N.cpp
 src/comm/HPL_recv.cpp src/grid/HPL_reduce.cpp src/comm/HPL_binit.cpp

include/hpl_comm.h

@@ -65,7 +65,8 @@ typedef enum
    HPL_2RING_M       = 404,           /* Increasing 2-ring (modified) */
    HPL_BLONG         = 405,                         /* long broadcast */
    HPL_BLONG_M       = 406,               /* long broadcast (modified) */
-   HPL_IBCAST       = 407               /* default bcast in MPI */
+   HPL_IBCAST        = 407,               /* default bcast in MPI */
+   HPL_BIDIR         = 408                /* bidirectional */
 } HPL_T_TOP;
 /*
  * ---------------------------------------------------------------------
@@ -160,6 +161,10 @@ int HPL_binit_blonM STDC_ARGS( ( HPL_T_panel *        ) );
 int HPL_bcast_blonM STDC_ARGS( ( HPL_T_panel *, int * ) );
 int HPL_bwait_blonM STDC_ARGS( ( HPL_T_panel *        ) );
 
+int HPL_binit_bidir STDC_ARGS( ( HPL_T_panel *        ) );
+int HPL_bcast_bidir STDC_ARGS( ( HPL_T_panel *, int * ) );
+int HPL_bwait_bidir STDC_ARGS( ( HPL_T_panel *        ) );
+
 #endif
 /*
  * End of hpl_comm.h

makes/Make.comm

@@ -56,7 +56,7 @@ INCdep           = \
 #
 HPL_comobj       = \
    HPL_1ring.o            HPL_1rinM.o            HPL_2ring.o            \
-   HPL_2rinM.o            HPL_blong.o            HPL_blonM.o            \
+   HPL_2rinM.o            HPL_blong.o            HPL_bidir.o           HPL_blonM.o            \
    HPL_packL.o            HPL_copyL.o            HPL_binit.o            \
    HPL_bcast.o            HPL_bwait.o            HPL_send.o             \
    HPL_recv.o             HPL_sdrv.o
@@ -84,6 +84,8 @@ HPL_2rinM.o            : ../HPL_2rinM.cpp            $(INCdep)
 	$(CC) -o $@ -c $(CCFLAGS)  ../HPL_2rinM.cpp
 HPL_blong.o            : ../HPL_blong.cpp            $(INCdep)
 	$(CC) -o $@ -c $(CCFLAGS)  ../HPL_blong.cpp
+HPL_bidir.o            : ../HPL_bidir.cpp            $(INCdep)
+	$(CC) -o $@ -c $(CCFLAGS)  ../HPL_bidir.cpp
 HPL_blonM.o            : ../HPL_blonM.cpp            $(INCdep)
 	$(CC) -o $@ -c $(CCFLAGS)  ../HPL_blonM.cpp
 HPL_packL.o            : ../HPL_packL.cpp            $(INCdep)

scripts/config/HPL_8GPU.dat

@@ -3,7 +3,7 @@ Innovative Computing Laboratory, University of Tennessee
 HPL.out      output file name (if any)
 0            device out (6=stdout,7=stderr,file)
 1            # of problems sizes (N)
-256512       N
+255360 256000 256512       N
 1            # of NBs
 384          NBs
 0            PMAP process mapping (0=Row-,1=Column-major)
@@ -19,8 +19,8 @@ HPL.out      output file name (if any)
 2            NDIVs
 1            # of recursive panel fact.
 2            RFACTs (0=left, 1=Crout, 2=Right)
-1            # of broadcast
-6            BCASTs (0=1rg,1=1rM,2=2rg,3=2rM,4=Lng,5=LnM,6=ibcast)
+8            # of broadcast
+7            BCASTs (0=1rg,1=1rM,2=2rg,3=2rM,4=Lng,5=LnM,6=ibcast)
 1            # of lookahead depth
 1            DEPTHs (>=0)
 1            SWAP (0=bin-exch,1=long,2=mix)

src/Makefile.am

@@ -8,7 +8,7 @@ blas/HPL_dscal.c blas/HPL_dtrsm.c blas/HPL_dtrsv.c blas/HPL_idamax.c \
 blas/HPL_dgemv.c blas/HPL_dscal.c blas/HPL_daxpy.c \
 blas/HPL_dcopy.c blas/HPL_dgemm.c blas/HPL_dgemv.c blas/HPL_dger.c \
 comm/HPL_sdrv.c comm/HPL_send.c comm/HPL_recv.c comm/HPL_bcast.c \
-comm/HPL_binit.c comm/HPL_bwait.c comm/HPL_blong.c comm/HPL_1ring.c \
+comm/HPL_binit.c comm/HPL_bwait.c comm/HPL_blong.c comm/HPL_bidir.c comm/HPL_1ring.c \
 comm/HPL_1rinM.c comm/HPL_2rinM.c comm/HPL_2ring.c comm/HPL_blonM.c comm/HPL_packL.c \
 grid/HPL_reduce.c grid/HPL_sum.c grid/HPL_grid_info.c grid/HPL_grid_init.c \
 grid/HPL_all_reduce.c grid/HPL_broadcast.c grid/HPL_grid_exit.c grid/HPL_max.c \

src/Makefile.in

@@ -147,6 +147,7 @@ am_libhpl_a_OBJECTS = auxil/HPL_dlatcpy.$(OBJEXT) \
 	comm/HPL_recv.$(OBJEXT) comm/HPL_bcast.$(OBJEXT) \
 	comm/HPL_binit.$(OBJEXT) comm/HPL_bwait.$(OBJEXT) \
 	comm/HPL_blong.$(OBJEXT) comm/HPL_1ring.$(OBJEXT) \
+	comm/HPL_bidir.$(OBJEXT) comm/HPL_1ring.$(OBJEXT) \
 	comm/HPL_1rinM.$(OBJEXT) comm/HPL_2rinM.$(OBJEXT) \
 	comm/HPL_2ring.$(OBJEXT) comm/HPL_blonM.$(OBJEXT) \
 	comm/HPL_packL.$(OBJEXT) grid/HPL_reduce.$(OBJEXT) \
@@ -229,6 +230,7 @@ am__depfiles_remade = auxil/$(DEPDIR)/HPL_dlacpy.Po \
 	comm/$(DEPDIR)/HPL_2rinM.Po comm/$(DEPDIR)/HPL_2ring.Po \
 	comm/$(DEPDIR)/HPL_bcast.Po comm/$(DEPDIR)/HPL_binit.Po \
 	comm/$(DEPDIR)/HPL_blonM.Po comm/$(DEPDIR)/HPL_blong.Po \
+	comm/$(DEPDIR)/HPL_bidir.Po \
 	comm/$(DEPDIR)/HPL_bwait.Po comm/$(DEPDIR)/HPL_packL.Po \
 	comm/$(DEPDIR)/HPL_recv.Po comm/$(DEPDIR)/HPL_sdrv.Po \
 	comm/$(DEPDIR)/HPL_send.Po grid/$(DEPDIR)/HPL_all_reduce.Po \
@@ -428,7 +430,7 @@ blas/HPL_dscal.c blas/HPL_dtrsm.c blas/HPL_dtrsv.c blas/HPL_idamax.c \
 blas/HPL_dgemv.c blas/HPL_dscal.c blas/HPL_daxpy.c \
 blas/HPL_dcopy.c blas/HPL_dgemm.c blas/HPL_dgemv.c blas/HPL_dger.c \
 comm/HPL_sdrv.c comm/HPL_send.c comm/HPL_recv.c comm/HPL_bcast.c \
-comm/HPL_binit.c comm/HPL_bwait.c comm/HPL_blong.c comm/HPL_1ring.c \
+comm/HPL_binit.c comm/HPL_bwait.c comm/HPL_blong.c comm/HPL_bidir.c comm/HPL_1ring.c \
 comm/HPL_1rinM.c comm/HPL_2rinM.c comm/HPL_2ring.c comm/HPL_blonM.c comm/HPL_packL.c \
 grid/HPL_reduce.c grid/HPL_sum.c grid/HPL_grid_info.c grid/HPL_grid_init.c \
 grid/HPL_all_reduce.c grid/HPL_broadcast.c grid/HPL_grid_exit.c grid/HPL_max.c \
@@ -583,6 +585,8 @@ comm/HPL_bwait.$(OBJEXT): comm/$(am__dirstamp) \
 	comm/$(DEPDIR)/$(am__dirstamp)
 comm/HPL_blong.$(OBJEXT): comm/$(am__dirstamp) \
 	comm/$(DEPDIR)/$(am__dirstamp)
+comm/HPL_bidir.$(OBJEXT): comm/$(am__dirstamp) \
+	comm/$(DEPDIR)/$(am__dirstamp)
 comm/HPL_1ring.$(OBJEXT): comm/$(am__dirstamp) \
 	comm/$(DEPDIR)/$(am__dirstamp)
 comm/HPL_1rinM.$(OBJEXT): comm/$(am__dirstamp) \
@@ -818,6 +822,7 @@ distclean-compile:
 @AMDEP_TRUE@@am__include@ @am__quote@comm/$(DEPDIR)/HPL_binit.Po@am__quote@ # am--include-marker
 @AMDEP_TRUE@@am__include@ @am__quote@comm/$(DEPDIR)/HPL_blonM.Po@am__quote@ # am--include-marker
 @AMDEP_TRUE@@am__include@ @am__quote@comm/$(DEPDIR)/HPL_blong.Po@am__quote@ # am--include-marker
+@AMDEP_TRUE@@am__include@ @am__quote@comm/$(DEPDIR)/HPL_bidir.Po@am__quote@ # am--include-marker
 @AMDEP_TRUE@@am__include@ @am__quote@comm/$(DEPDIR)/HPL_bwait.Po@am__quote@ # am--include-marker
 @AMDEP_TRUE@@am__include@ @am__quote@comm/$(DEPDIR)/HPL_packL.Po@am__quote@ # am--include-marker
 @AMDEP_TRUE@@am__include@ @am__quote@comm/$(DEPDIR)/HPL_recv.Po@am__quote@ # am--include-marker
@@ -1086,6 +1091,7 @@ distclean: distclean-am
 	-rm -f comm/$(DEPDIR)/HPL_binit.Po
 	-rm -f comm/$(DEPDIR)/HPL_blonM.Po
 	-rm -f comm/$(DEPDIR)/HPL_blong.Po
+	-rm -f comm/$(DEPDIR)/HPL_bidir.Po
 	-rm -f comm/$(DEPDIR)/HPL_bwait.Po
 	-rm -f comm/$(DEPDIR)/HPL_packL.Po
 	-rm -f comm/$(DEPDIR)/HPL_recv.Po
@@ -1233,6 +1239,7 @@ maintainer-clean: maintainer-clean-am
 	-rm -f comm/$(DEPDIR)/HPL_binit.Po
 	-rm -f comm/$(DEPDIR)/HPL_blonM.Po
 	-rm -f comm/$(DEPDIR)/HPL_blong.Po
+	-rm -f comm/$(DEPDIR)/HPL_bidir.Po
 	-rm -f comm/$(DEPDIR)/HPL_bwait.Po
 	-rm -f comm/$(DEPDIR)/HPL_packL.Po
 	-rm -f comm/$(DEPDIR)/HPL_recv.Po

src/comm/HPL_bcast.cpp

@@ -109,6 +109,7 @@ int HPL_bcast
       case HPL_BLONG_M : ierr = HPL_bcast_blonM( PANEL, IFLAG ); break;
       case HPL_BLONG   : ierr = HPL_bcast_blong( PANEL, IFLAG ); break;
       case HPL_IBCAST  : ierr = HPL_BE_bcast_ibcast( PANEL, IFLAG, HPL_TR); break;
+      case HPL_BIDIR   : ierr = HPL_bcast_bidir( PANEL, IFLAG ); break;
       default          : ierr = HPL_SUCCESS;
    }
 

src/comm/HPL_binit.cpp

@@ -99,6 +99,7 @@ int HPL_binit
       case HPL_BLONG_M : ierr = HPL_binit_blonM( PANEL ); break;
       case HPL_BLONG   : ierr = HPL_binit_blong( PANEL ); break;
       case HPL_IBCAST  : ierr = HPL_BE_binit_ibcast(PANEL, HPL_TR); break;
+      case HPL_BIDIR   : ierr = HPL_binit_bidir( PANEL ); break;
       default          : ierr = HPL_SUCCESS;
    }
 

src/comm/HPL_bwait.cpp

@@ -100,6 +100,7 @@ int HPL_bwait
       case HPL_BLONG_M : ierr = HPL_bwait_blonM( PANEL ); break;
       case HPL_BLONG   : ierr = HPL_bwait_blong( PANEL ); break;
       case HPL_IBCAST  : ierr = HPL_BE_bwait_ibcast(PANEL, HPL_TR); break;
+      case HPL_BIDIR   : ierr = HPL_bwait_bidir( PANEL ); break;
       default          : ierr = HPL_SUCCESS;
    }
 

testing/ptest/HPL_pdinfo.cpp

@@ -550,6 +550,7 @@ void HPL_pdinfo
          else if( j == 4 ) TP[ i ] = HPL_BLONG;
          else if( j == 5 ) TP[ i ] = HPL_BLONG_M;
          else if( j == 6 ) TP[ i ] = HPL_IBCAST;
+         else if( j == 7 ) TP[ i ] = HPL_BIDIR;
          else              TP[ i ] = HPL_1RING_M;
       }
 /*
@@ -714,7 +715,8 @@ void HPL_pdinfo
          else if( TP[i] == HPL_2RING_M ) iwork[j] = 3;
          else if( TP[i] == HPL_BLONG   ) iwork[j] = 4;
          else if( TP[i] == HPL_BLONG_M ) iwork[j] = 5;
-         else if( TP[i] == HPL_IBCAST )  iwork[j] = 6;
+         else if( TP[i] == HPL_IBCAST  ) iwork[j] = 6;
+         else if( TP[i] == HPL_BIDIR   ) iwork[j] = 7;
          j++;
       }
       for( i = 0; i < *NDHS; i++ ) { iwork[j] = DH[i]; j++; }
@@ -759,6 +761,7 @@ void HPL_pdinfo
          else if( iwork[j] == 4 ) TP[i] = HPL_BLONG;
          else if( iwork[j] == 5 ) TP[i] = HPL_BLONG_M;
          else if( iwork[j] == 6 ) TP[i] = HPL_IBCAST;
+         else if( iwork[j] == 7 ) TP[i] = HPL_BIDIR;
          j++;
       }
       for( i = 0; i < *NDHS; i++ ) { DH[i] = iwork[j]; j++; }
@@ -1024,6 +1027,8 @@ void HPL_pdinfo
             HPL_fprintf( TEST->outfp,       "  BlongM " );
          else if( TP[i] == HPL_IBCAST )
             HPL_fprintf( TEST->outfp,       "  IBCAST " );
+         else if( TP[i] == HPL_BIDIR )
+            HPL_fprintf( TEST->outfp,       "  BIDIR " );
       }
       if( *NTPS > 8 )
       {
@@ -1044,6 +1049,8 @@ void HPL_pdinfo
                HPL_fprintf( TEST->outfp,       "  BlongM " );
             else if( TP[i] == HPL_IBCAST )
                HPL_fprintf( TEST->outfp,       "  IBCAST " );
+            else if( TP[i] == HPL_BIDIR )
+               HPL_fprintf( TEST->outfp,       "  BIDIR " );
          }
          if( *NTPS > 16 )
          {
@@ -1064,6 +1071,8 @@ void HPL_pdinfo
                   HPL_fprintf( TEST->outfp,       "  BlongM " );
                else if( TP[i] == HPL_IBCAST )
                   HPL_fprintf( TEST->outfp,       "  IBCAST " );
+               else if( TP[i] == HPL_BIDIR )
+                  HPL_fprintf( TEST->outfp,       "  BIDIR " );
             }
          }
       }

testing/ptest/HPL_pdtest.cpp

@@ -274,6 +274,7 @@ void HPL_pdtest
       else if( ALGO->btopo == HPL_2RING_M ) ctop = '3';
       else if( ALGO->btopo == HPL_BLONG   ) ctop = '4';
       else if( ALGO->btopo == HPL_IBCAST  ) ctop = '6';
+      else if( ALGO->btopo == HPL_BIDIR   ) ctop = '7';
       else /* if( ALGO->btopo == HPL_BLONG_M ) */ ctop = '5';
 
       if( wtime[0] > HPL_rzero ) {