Anasazi Version of the Day
TbbTsqr_ExplicitQTask.hpp
00001 // @HEADER
00002 // ***********************************************************************
00003 //
00004 //                 Anasazi: Block Eigensolvers Package
00005 //                 Copyright (2010) Sandia Corporation
00006 //
00007 // Under terms of Contract DE-AC04-94AL85000, there is a non-exclusive
00008 // license for use of this work by or on behalf of the U.S. Government.
00009 //
00010 // This library is free software; you can redistribute it and/or modify
00011 // it under the terms of the GNU Lesser General Public License as
00012 // published by the Free Software Foundation; either version 2.1 of the
00013 // License, or (at your option) any later version.
00014 //
00015 // This library is distributed in the hope that it will be useful, but
00016 // WITHOUT ANY WARRANTY; without even the implied warranty of
00017 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00018 // Lesser General Public License for more details.
00019 //
00020 // You should have received a copy of the GNU Lesser General Public
00021 // License along with this library; if not, write to the Free Software
00022 // Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
00023 // USA
00024 // Questions? Contact Michael A. Heroux (maherou@sandia.gov)
00025 //
00026 // ***********************************************************************
00027 // @HEADER
00028 
00029 #ifndef __TSQR_TBB_ExplicitQTask_hpp
00030 #define __TSQR_TBB_ExplicitQTask_hpp
00031 
00032 #include <tbb/task.h>
00033 #include <TbbTsqr_Partitioner.hpp>
00034 #include <Tsqr_SequentialTsqr.hpp>
00035 
00038 
00039 namespace TSQR {
00040   namespace TBB {
00041     
00042     template< class LocalOrdinal, class Scalar >
00043     class ExplicitQTask : public tbb::task {
00044     private:
00045       typedef MatView< LocalOrdinal, Scalar > mat_view;
00046       typedef ConstMatView< LocalOrdinal, Scalar > const_mat_view;
00047       typedef std::pair< mat_view, mat_view > split_t;
00048       typedef std::pair< const_mat_view, const_mat_view > const_split_t;
00049 
00050     public:
00051       ExplicitQTask (const size_t P_first__, 
00052          const size_t P_last__,
00053          MatView< LocalOrdinal, Scalar > Q_out,
00054          const SequentialTsqr< LocalOrdinal, Scalar >& seq,
00055          const bool contiguous_cache_blocks) :
00056   P_first_ (P_first__), P_last_ (P_last__), Q_out_ (Q_out),
00057   seq_ (seq), contiguous_cache_blocks_ (contiguous_cache_blocks)
00058       {}
00059 
00060       tbb::task* execute () {
00061   if (P_first_ > P_last_ || Q_out_.empty())
00062     return NULL;
00063   else if (P_first_ == P_last_)
00064     {
00065       // Fill my partition with zeros.
00066       seq_.fill_with_zeros (Q_out_.nrows(), Q_out_.ncols(), Q_out_.get(), 
00067           Q_out_.lda(), contiguous_cache_blocks_);
00068 
00069       // If our partition is the first (topmost), fill it with
00070       // the first Q_out.ncols() columns of the identity matrix.
00071       if (P_first_ == 0)
00072         {
00073     // Fetch the topmost cache block of my partition.  Its
00074     // leading dimension should be set correctly by
00075     // top_block().
00076     mat_view Q_out_top = seq_.top_block (Q_out_, contiguous_cache_blocks_);
00077     // Set the top block of Q_out to the first ncols
00078     // columns of the identity matrix.
00079     for (LocalOrdinal j = 0; j < Q_out_top.ncols(); ++j)
00080       Q_out_top(j,j) = Scalar(1);
00081         }
00082       return NULL;
00083     }
00084   else
00085     {
00086       // "c": continuation task
00087       tbb::empty_task& c = *new( allocate_continuation() ) tbb::empty_task;
00088 
00089       // Recurse on two intervals: [P_first, P_mid] and [P_mid+1, P_last]
00090       const size_t P_mid = (P_first_ + P_last_) / 2;
00091       split_t Q_split = 
00092         partitioner_.split (Q_out_, P_first_, P_mid, P_last_,
00093           contiguous_cache_blocks_);
00094 
00095       ExplicitQTask& topTask = *new( c.allocate_child() )
00096         ExplicitQTask (P_first_, P_mid, Q_split.first, seq_, 
00097            contiguous_cache_blocks_);
00098       ExplicitQTask& botTask = *new( c.allocate_child() )
00099         ExplicitQTask (P_mid+1, P_last_, Q_split.second, seq_, 
00100            contiguous_cache_blocks_);
00101 
00102       // Set reference count of parent (in this case, the
00103       // continuation task) to 2 (since 2 children -- no
00104       // additional task since no waiting).
00105       c.set_ref_count (2);
00106       c.spawn (botTask);
00107       return &topTask; // scheduler bypass optimization
00108     }
00109       }
00110 
00111     private:
00112       size_t P_first_, P_last_;
00113       mat_view Q_out_;
00114       SequentialTsqr< LocalOrdinal, Scalar > seq_;
00115       Partitioner< LocalOrdinal, Scalar > partitioner_;
00116       bool contiguous_cache_blocks_;
00117     };
00118 
00119   } // namespace TBB
00120 } // namespace TSQR
00121 
00122 
00123 #endif // __TSQR_TBB_ExplicitQTask_hpp
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends