libhoomd/computes_gpu/TablePotentialGPU.cc

   1 /*
   2 Highly Optimized Object-oriented Many-particle Dynamics -- Blue Edition
   3 (HOOMD-blue) Open Source Software License Copyright 2009-2014 The Regents of
   4 the University of Michigan All rights reserved.
   5
   6 HOOMD-blue may contain modifications ("Contributions") provided, and to which
   7 copyright is held, by various Contributors who have granted The Regents of the
   8 University of Michigan the right to modify and/or distribute such Contributions.
   9
  10 You may redistribute, use, and create derivate works of HOOMD-blue, in source
  11 and binary forms, provided you abide by the following conditions:
  12
  13 * Redistributions of source code must retain the above copyright notice, this
  14 list of conditions, and the following disclaimer both in the code and
  15 prominently in any materials provided with the distribution.
  16
  17 * Redistributions in binary form must reproduce the above copyright notice, this
  18 list of conditions, and the following disclaimer in the documentation and/or
  19 other materials provided with the distribution.
  20
  21 * All publications and presentations based on HOOMD-blue, including any reports
  22 or published results obtained, in whole or in part, with HOOMD-blue, will
  23 acknowledge its use according to the terms posted at the time of submission on:
  24 http://codeblue.umich.edu/hoomd-blue/citations.html
  25
  26 * Any electronic documents citing HOOMD-Blue will link to the HOOMD-Blue website:
  27 http://codeblue.umich.edu/hoomd-blue/
  28
  29 * Apart from the above required attributions, neither the name of the copyright
  30 holder nor the names of HOOMD-blue's contributors may be used to endorse or
  31 promote products derived from this software without specific prior written
  32 permission.
  33
  34 Disclaimer
  35
  36 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER AND CONTRIBUTORS ``AS IS'' AND
  37 ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
  38 WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, AND/OR ANY
  39 WARRANTIES THAT THIS SOFTWARE IS FREE OF INFRINGEMENT ARE DISCLAIMED.
  40
  41 IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
  42 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
  43 BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  44 DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  45 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
  46 OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
  47 ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  48 */
  49
  50 // Maintainer: joaander
  51
  52 #ifdef WIN32
  53 #pragma warning( push )
  54 #pragma warning( disable : 4103 4244 )
  55 #endif
  56
  57 #include <boost/python.hpp>
  58 using namespace boost::python;
  59 #include <boost/bind.hpp>
  60 using namespace boost;
  61
  62 #include "TablePotentialGPU.h"
  63 #include <stdexcept>
  64
  65 /*! \file TablePotentialGPU.cc
  66     \brief Defines the TablePotentialGPU class
  67 */
  68
  69 using namespace std;
  70
  71 /*! \param sysdef System to compute forces on
  72     \param nlist Neighborlist to use for computing the forces
  73     \param table_width Width the tables will be in memory
  74     \param log_suffix Name given to this instance of the table potential
  75 */
  76 TablePotentialGPU::TablePotentialGPU(boost::shared_ptr<SystemDefinition> sysdef,
  77                                      boost::shared_ptr<NeighborList> nlist,
  78                                      unsigned int table_width,
  79                                      const std::string& log_suffix)
  80     : TablePotential(sysdef, nlist, table_width, log_suffix)
  81     {
  82     // can't run on the GPU if there aren't any GPUs in the execution configuration
  83     if (!exec_conf->isCUDAEnabled())
  84         {
  85         m_exec_conf->msg->error() << "Creating a TableForceComputeGPUwith no GPU in the execution configuration" << endl;
  86         throw std::runtime_error("Error initializing TableForceComputeGPU");
  87         }
  88
  89     m_tuner.reset(new Autotuner(32, 1024, 32, 5, 100000, "pair_table", this->m_exec_conf));
  90     }
  91
  92 /*! \post The table based forces are computed for the given timestep. The neighborlist's
  93 compute method is called to ensure that it is up to date.
  94
  95 \param timestep specifies the current time step of the simulation
  96
  97 Calls gpu_compute_table_forces to do the leg work
  98 */
  99 void TablePotentialGPU::computeForces(unsigned int timestep)
 100     {
 101     // start by updating the neighborlist
 102     m_nlist->compute(timestep);
 103
 104     // start the profile
 105     if (m_prof) m_prof->push(exec_conf, "Table pair");
 106
 107     // The GPU implementation CANNOT handle a half neighborlist, error out now
 108     bool third_law = m_nlist->getStorageMode() == NeighborList::half;
 109     if (third_law)
 110         {
 111         m_exec_conf->msg->error() << "TablePotentialGPU cannot handle a half neighborlist" << endl;
 112         throw runtime_error("Error computing forces in TablePotentialGPU");
 113         }
 114
 115     // access the neighbor list
 116     ArrayHandle<unsigned int> d_n_neigh(this->m_nlist->getNNeighArray(), access_location::device, access_mode::read);
 117     ArrayHandle<unsigned int> d_nlist(this->m_nlist->getNListArray(), access_location::device, access_mode::read);
 118     Index2D nli = this->m_nlist->getNListIndexer();
 119
 120     // access the particle data
 121     ArrayHandle<Scalar4> d_pos(m_pdata->getPositions(), access_location::device, access_mode::read);
 122     BoxDim box = m_pdata->getBox();
 123
 124     // access the table data
 125     ArrayHandle<Scalar2> d_tables(m_tables, access_location::device, access_mode::read);
 126     ArrayHandle<Scalar4> d_params(m_params, access_location::device, access_mode::read);
 127
 128     ArrayHandle<Scalar4> d_force(m_force,access_location::device,access_mode::overwrite);
 129     ArrayHandle<Scalar> d_virial(m_virial,access_location::device,access_mode::overwrite);
 130
 131     // run the kernel on all GPUs in parallel
 132     m_tuner->begin();
 133     gpu_compute_table_forces(d_force.data,
 134                              d_virial.data,
 135                              m_virial.getPitch(),
 136                              m_pdata->getN(),
 137                              m_pdata->getNGhosts(),
 138                              d_pos.data,
 139                              box,
 140                              d_n_neigh.data,
 141                              d_nlist.data,
 142                              nli,
 143                              d_tables.data,
 144                              d_params.data,
 145                              m_ntypes,
 146                              m_table_width,
 147                              m_tuner->getParam(),
 148                              m_exec_conf->getComputeCapability());
 149
 150     if (exec_conf->isCUDAErrorCheckingEnabled())
 151         CHECK_CUDA_ERROR();
 152     m_tuner->end();
 153
 154     if (m_prof) m_prof->pop(exec_conf);
 155     }
 156
 157 void export_TablePotentialGPU()
 158     {
 159     class_<TablePotentialGPU, boost::shared_ptr<TablePotentialGPU>, bases<TablePotential>, boost::noncopyable >
 160     ("TablePotentialGPU",
 161      init< boost::shared_ptr<SystemDefinition>,
 162      boost::shared_ptr<NeighborList>,
 163      unsigned int,
 164      const std::string& >())
 165     ;
 166     }
 167
 168 #ifdef WIN32
 169 #pragma warning( pop )
 170 #endif