updating examples/gpu and doc

docs/src/gettingstarted.tex

@@ -584,15 +584,18 @@ Krylov method. At the end of the code, we close the GPU environment
 \caption{setup of a GPU-enabled test program part three.\label{fig:gpu-ex3}}
 \end{listing}
 
-It is very important to employ solvers that are suited
+It is very important to employ smoothers and coarsest solvers that are suited
-to the GPU, i.e.  solvers that do NOT  employ triangular
+to the GPU, i.e.  methods that do NOT employ triangular
-system solve kernels.  Solvers that satisfy this constraint include:
+system solve kernels. Methods that satisfy this constraint include:
 \begin{itemize}
 \item \verb|JACOBI|
+\item \verb|BJAC| with the following methods on the local blocks:
+\begin{itemize}
 \item \verb|INVK|
 \item \verb|INVT|
 \item \verb|AINV|
 \end{itemize}
+\end{itemize}
 and their $\ell_1$ variants. 
 
 %%% Local Variables:

examples/gpu/amg_dexample_gpu.f90

@@ -39,23 +39,18 @@
 !
 ! This sample program solves a linear system obtained by discretizing a
 ! PDE with Dirichlet BCs. The solver is CG, coupled with one of the
-! following multi-level preconditioner, as explained in Section 4.1 of
+! following multi-level preconditioner, as explained in Section 4.2 of
 ! the AMG4PSBLAS User's and Reference Guide:
 !
-! - choice = 1, the default multi-level preconditioner solver, i.e., 
+! - choice = 1, a V-cycle with decoupled smoothed aggregation, 4 Jacobi
-! V-cycle with decoupled smoothed aggregation, 1 hybrid forward/backward
+! sweeps as pre/post-smoother and 8 Jacobi sweeps as coarsest-level
-! GS sweep as pre/post-smoother and UMFPACK as coarsest-level
+! solver with replicated coarsest matrix
-! solver (Sec. 4.1, Listing 1)
 !
-! - choice = 2, a V-cycle preconditioner with 1 block-Jacobi sweep
+! - choice = 2, a W-cycle based on the coupled aggregation relying on matching,
-! (with ILU(0) on the blocks) as pre- and post-smoother, and 8 block-Jacobi
+! with maximum size of aggregates equal to 8 and smoothed prolongators,
-! sweeps (with ILU(0) on the blocks) as coarsest-level solver (Sec. 4.1, Listing 2)
+! 2 sweeps of Block-Jacobi ipre/post-smoother using approximate inverse INVK and
-!
+! 4 sweeps of Block-Jacobi with INVK as coarsest-level solver on distributed
-! - choice = 3,  W-cycle preconditioner based on the coupled aggregation relying
+! coarsest matrix 
-! on matching, with maximum size of aggregates equal to 8 and smoothed prolongators,
-! 2 hybrid forward/backward GS sweeps as pre/post-smoother, a distributed coarsest
-! matrix, and preconditioned Flexible Conjugate Gradient as coarsest-level solver
-! (Sec. 4.1, Listing 3)
 !
 ! The matrix and the rhs are read from files (if an rhs is not available, the
 ! unit rhs is set).
@@ -183,8 +178,9 @@ program amg_dexample_gpu
 
  case(1)
 
-    ! initialize a V-cycle preconditioner with 4 Jacobi sweep 
+    ! initialize a V-cycle preconditioner, relying on decoupled smoothed aggregation 
-    ! and 8 Jacobi sweeps as coarsest-level solver
+    ! with 4 Jacobi sweeps as pre/post-smoother
+    ! and 8 Jacobi sweeps as coarsest-level solver on replicated coarsest matrix
 
     call P%init(ctxt,'ML',info)
     call P%set('SMOOTHER_TYPE','JACOBI',info)
@@ -195,19 +191,22 @@ program amg_dexample_gpu
 
   case(2)
 
-   ! initialize a V-cycle preconditioner based on the coupled aggregation relying on matching,
+   ! initialize a W-cycle preconditioner based on the coupled aggregation relying on matching,
    ! with maximum size of aggregates equal to 8 and smoothed prolongators,
-   ! Block-Jacobi smoother using approximate inverse INVK and
+   ! 2 sweeps of Block-Jacobi pre/post-smoother using approximate inverse INVK and
-   ! and 4 sweeps of INVK on he coarsest level 
+   ! 4 sweeps of Block-Jacobi with INVK on the coarsest level distributed matrix
 
     call P%init(ctxt,'ML',info)
     call P%set('PAR_AGGR_ALG','COUPLED',info)
     call P%set('AGGR_TYPE','MATCHBOXP',info)
     call P%set('AGGR_SIZE',8,info)
     call P%set('ML_CYCLE','WCYCLE',info)
+    call P%set('SMOOTHER_TYPE','BJAC',info)
     call P%set('SMOOTHER_SWEEPS',2,info)
     call P%set('SUB_SOLVE','INVK',info)  
-    call P%set('COARSE_SOLVE','INVK',info)
+    call P%set('COARSE_SOLVE','BJAC',info)
+    call P%set('COARSE_SUBSOLVE','INVK',info)
+    call P%set('COARSE_SWEEPS',4,info)
     call P%set('COARSE_MAT','DIST',info)
     kmethod = 'CG'