MGPUMpdRdmeSolver.h

Go to the documentation of this file.

/*
 * University of Illinois Open Source License
 * Copyright 2008-2018 Luthey-Schulten Group,
 * Copyright 2012 Roberts Group,
 * All rights reserved.
 * 
 * Developed by: Luthey-Schulten Group
 *               University of Illinois at Urbana-Champaign
 *               http://www.scs.uiuc.edu/~schulten
 * 
 * Developed by: Roberts Group
 *               Johns Hopkins University
 *               http://biophysics.jhu.edu/roberts/
 *
 * Permission is hereby granted, free of charge, to any person obtaining a copy of
 * this software and associated documentation files (the Software), to deal with 
 * the Software without restriction, including without limitation the rights to 
 * use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies 
 * of the Software, and to permit persons to whom the Software is furnished to 
 * do so, subject to the following conditions:
 * 
 * - Redistributions of source code must retain the above copyright notice, 
 * this list of conditions and the following disclaimers.
 * 
 * - Redistributions in binary form must reproduce the above copyright notice, 
 * this list of conditions and the following disclaimers in the documentation 
 * and/or other materials provided with the distribution.
 * 
 * - Neither the names of the Luthey-Schulten Group, University of Illinois at
 * Urbana-Champaign, the Roberts Group, Johns Hopkins University, nor the names
 * of its contributors may be used to endorse or promote products derived from
 * this Software without specific prior written permission.
 * 
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL 
 * THE CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR 
 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 
 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 
 * OTHER DEALINGS WITH THE SOFTWARE.
 *
 * Author(s): Elijah Roberts, Mike Hallock, Zane Thornburg
 */
 
#ifndef LM_RDME_MGPUMPDRDMESOLVER_H_
#define LM_RDME_MGPUMPDRDMESOLVER_H_
 
#include "cuda/lm_cuda.h"
#include "core/ResourceAllocator.h"
#include "rdme/RDMESolver.h"
#include "rdme/ByteLattice.h"
#include "rdme/GPUMapper/MultiGPUMapper.h"
#include "rdme/GPUMapper/SegmentDescriptor.h"
 
#if defined(MACOSX)
#include "GPUMapper/osx_barrier.h"
#endif
 
#define OVERFLOW_MODE_CLASSIC 0
#define OVERFLOW_MODE_RELAXED 1
 
using lm::main::ResourceAllocator;
using lm::rdme::RDMESolver;
using lm::rdme::Lattice;
 
namespace lm {
 
namespace io {
    class Lattice;
    class SpeciesCounts;
}
 
namespace rdme {
 
struct gpu_worker_thread_params;
 
class MGPUMpdRdmeSolver: public RDMESolver
{
    using CMESolver::hookSimulation;
    using RDMESolver::buildDiffusionModel;
 
public:
    MGPUMpdRdmeSolver();
    virtual ~MGPUMpdRdmeSolver();
 
    virtual void initialize(unsigned int replicate,
                            map<string, string>* parameters,
                            ResourceAllocator::ComputeResources* resources);
 
    virtual bool needsReactionModel()  {return true;}
    virtual bool needsDiffusionModel() {return true;}
 
    virtual void buildModel(const uint numberSpeciesA,
                            const uint numberReactionsA,
                            const uint* initialSpeciesCountsA,
                            const uint* reactionTypeA,
                            const double* kA,
                            const int* SA,
                            const uint* DA,
                            const uint kCols = 1);
 
    virtual void buildDiffusionModel(const uint numberSiteTypesA,
                                     const double* DFA,
                                     const uint* RLA,
                                     lattice_size_t latticeXSize,
                                     lattice_size_t latticeYSize,
                                     lattice_size_t latticeZSize,
                                     site_size_t particlesPerSite,
                                     const unsigned int bytes_per_particle,
                                     si_dist_t latticeSpacing,
                                     const uint8_t* latticeData,
                                     const uint8_t* latticeSitesData,
                                     bool rowMajorData = true);
 
    virtual void generateTrajectory();
    virtual void setReactionRate(unsigned int rxid, float rate);
 
protected:
    uint32_t seed;
    double tau;
    bool reactionModelModified;
 
    uint32_t overflowTimesteps;
    uint32_t overflowListUses;
    int      overflow_handling;
 
    // Stored model parameters for const memory
    unsigned int* model_reactionOrders;
    unsigned int* model_reactionSites;
    float*        model_reactionRates;
 
    unsigned int* model_D1;
    unsigned int* model_D2;
 
    int8_t*  model_S;
    float*   model_T;
    uint8_t* model_RL;
 
    size_t zeroOrderSize, firstOrderSize, secondOrderSize;
    float *zeroOrder, *firstOrder, *secondOrder;
 
    MultiGPUMapper* mapper;
    ResourceAllocator::ComputeResources* resources;
    gpu_worker_thread_params* threads;
    pthread_barrier_t start_barrier, stop_barrier, simulation_barrier, overflow_barrier;
 
    int timesteps_to_run;
    uint32_t current_timestep;
    double printPerfInterval;
 
    bool aggcopy_x_unpack;
    bool aggcopy_r_pack;
    bool use_spin_barrier;
 
    virtual void allocateLattice(lattice_size_t latticeXSize,
                                 lattice_size_t latticeYSize,
                                 lattice_size_t latticeZSize,
                                 site_size_t particlesPerSite,
                                 const unsigned int bytes_per_particle,
                                 si_dist_t latticeSpacing);
 
    virtual void writeLatticeData(double time,
                                  ByteLattice* lattice,
                                  lm::io::Lattice* latticeDataSet);
    virtual void writeLatticeSites(double time, ByteLattice* lattice);
 
    virtual void recordSpeciesCounts(double time,
                                     ByteLattice* lattice,
                                     lm::io::SpeciesCounts* speciesCountsDataSet);
    virtual void writeSpeciesCounts(lm::io::SpeciesCounts* speciesCountsDataSet);
 
    virtual int hookSimulation(double time, ByteLattice* lattice);
    virtual void hookCheckSimulation(double time, ByteLattice* lattice);
 
    virtual int run_next_timestep(int gpu, uint32_t timestep);
    virtual uint64_t getTimestepSeed(uint32_t timestep, uint32_t substep);
 
    virtual void computePropensities();
    virtual void initialize_decomposition();
    virtual void setupModelsOnDevice(int gpu);
    virtual void copyModelsToDevice(int gpu);
 
    virtual void start_threads();
    virtual void stop_threads();
    virtual void* run_thread(int);
    friend void* gpu_worker_thread(void* arg);
    
    virtual int handle_all_overflows();
    virtual int handle_overflows(int gpu, void* hptr, void* dptr, int ts);
 
    virtual void calculateXLaunchParameters(dim3* gridSize, dim3* threadBlockSize,
                                            const unsigned int maxXBlockSize,
                                            const unsigned int latticeXSize,
                                            const unsigned int latticeYSize,
                                            const unsigned int latticeZSize);
 
    virtual void calculateYLaunchParameters(dim3* gridSize, dim3* threadBlockSize,
                                            const unsigned int blockXSize,
                                            const unsigned int blockYSize,
                                            const unsigned int latticeXSize,
                                            const unsigned int latticeYSize,
                                            const unsigned int latticeZSize);
 
    virtual void calculateZLaunchParameters(dim3* gridSize, dim3* threadBlockSize,
                                            const unsigned int blockXSize,
                                            const unsigned int blockZSize,
                                            const unsigned int latticeXSize,
                                            const unsigned int latticeYSize,
                                            const unsigned int latticeZSize);
 
    virtual void calculateReactionLaunchParameters(dim3* gridSize, dim3* threadBlockSize,
                                                   const unsigned int blockXSize,
                                                   const unsigned int blockYSize,
                                                   const unsigned int latticeXSize,
                                                   const unsigned int latticeYSize,
                                                   const unsigned int latticeZSize);
};
 
struct gpu_worker_thread_params
{
    pthread_t thread;
    MGPUMpdRdmeSolver* runner;
    MultiGPUMapper* mapper;
    
    int gpu;
    int ngpus;
    int timesteps_to_run;
 
    // cuda objects
    unsigned int *dLattice, *dLatticeTmp;
    uint8_t* dSites;
    cudaStream_t stream1, stream2;
    unsigned int *h_overflows, *d_overflows;
 
    // kernel launch params
    dim3 grid_x, grid_y, grid_z, grid_r;
    dim3 threads_x, threads_y, threads_z, threads_r;
 
    // lattice segment geometry provided by the mapper
    SegmentDescriptor_s* segment;
 
#ifdef MPD_GLOBAL_S_MATRIX
    uint8_t* RLG; // Device global memory pointer for RL matrix
    int8_t*  SG;  // Device global memory pointer for S matrix
#endif
 
#ifdef MPD_GLOBAL_T_MATRIX
    float* TG;
#endif
 
#ifdef MPD_GLOBAL_R_MATRIX
    float*        reactionRatesG;
    unsigned int* reactionOrdersG;
    unsigned int* reactionSitesG;
 
    unsigned int* D1G;
    unsigned int* D2G;
#endif
 
    float *propZeroOrder, *propFirstOrder, *propSecondOrder;
};
 
 
namespace mgpumpdrdme_dev {
__device__ inline size_t local_to_global(unsigned int x, unsigned int y, unsigned int z);
__device__ inline size_t local_index(unsigned int x, unsigned int y, unsigned int z);
__global__ void correct_overflows_mgpu(unsigned int* lattice, unsigned int* siteOverflowList);
 
__global__ void MGPU_x_kernel_unpack(const unsigned int* inLattice,
                                     const uint8_t* inSites,
                                     unsigned int* outLattice,
                                     const unsigned int z_start,
                                     const unsigned long long timestepHash,
                                     unsigned int* siteOverflowList,
                                     unsigned int* buf_top,
                                     unsigned int* buf_bot);
 
__global__ void MGPU_x_kernel(const unsigned int* inLattice,
                              const uint8_t* inSites,
                              unsigned int* outLattice,
                              const unsigned int z_start,
                              const unsigned long long timestepHash,
                              unsigned int* siteOverflowList);
 
__global__ void MGPU_y_kernel(const unsigned int* inLattice,
                              const uint8_t* inSites,
                              unsigned int* outLattice,
                              const unsigned long long timestepHash,
                              unsigned int* siteOverflowList);
 
__global__ void MGPU_z_kernel(const unsigned int* inLattice,
                              const uint8_t* inSites,
                              unsigned int* outLattice,
                              const unsigned long long timestepHash,
                              unsigned int* siteOverflowList,
                              const unsigned int z_start);
 
#ifdef MPD_GLOBAL_S_MATRIX
    #ifdef MPD_GLOBAL_R_MATRIX
        __global__ void MGPU_precomp_reaction_kernel(const unsigned int* inLattice,
                                                     const uint8_t* inSites,
                                                     unsigned int* outLattice,
                                                     const unsigned long long timestepHash,
                                                     unsigned int* siteOverflowList,
                                                     int z_start,
                                                     const __restrict__ int8_t* SG,
                                                     const __restrict__ uint8_t* RLG,
                                                     const unsigned int* __restrict__ reactionOrderG,
                                                     const unsigned int* __restrict__ reactionSitesG,
                                                     const unsigned int* __restrict__ D1G,
                                                     const unsigned int* __restrict__ D2G,
                                                     const float* __restrict__ reactionRatesG,
                                                     const float* __restrict__ qp0,
                                                     const float* __restrict__ qp1,
                                                     const float* __restrict__ qp2);
 
        __global__ void MGPU_precomp_reaction_kernel_packing(const unsigned int* inLattice,
                                                             const uint8_t* inSites,
                                                             unsigned int* outLattice,
                                                             const unsigned long long timestepHash,
                                                             unsigned int* siteOverflowList,
                                                             int z_start,
                                                             const __restrict__ int8_t* SG,
                                                             const __restrict__ uint8_t* RLG,
                                                             const unsigned int* __restrict__ reactionOrdersG,
                                                             const unsigned int* __restrict__ reactionSitesG,
                                                             const unsigned int* __restrict__ D1G,
                                                             const unsigned int* __restrict__ D2G,
                                                             const float* __restrict__ reactionRatesG,
                                                             const float* __restrict__ qp0,
                                                             const float* __restrict__ qp1,
                                                             const float* __restrict__ qp2,
                                                             unsigned int* buf_top,
                                                             unsigned int* buf_bot);
 
        __global__ void MGPU_reaction_kernel(const unsigned int* inLattice,
                                             const uint8_t* inSites,
                                             unsigned int* outLattice,
                                             const unsigned long long timestepHash,
                                             unsigned int* siteOverflowList,
                                             const unsigned int z_start,
                                             const int8_t* __restrict__ SG,
                                             const uint8_t* __restrict__ RLG,
                                             const unsigned int* __restrict__ reactionOrdersG,
                                             const unsigned int* __restrict__ reactionSitesG,
                                             const unsigned int* __restrict__ D1G,
                                             const unsigned int* __restrict__ D2G,
                                             const float* __restrict__ reactionRatesG);
    #else
        __global__ void MGPU_precomp_reaction_kernel(const unsigned int* inLattice,
                                                     const uint8_t* inSites,
                                                     unsigned int* outLattice,
                                                     const unsigned long long timestepHash,
                                                     unsigned int* siteOverflowList,
                                                     int z_start,
                                                     const __restrict__ int8_t* SG,
                                                     const __restrict__ uint8_t* RLG,
                                                     const float* __restrict__ qp0,
                                                     const float* __restrict__ qp1,
                                                     const float* __restrict__ qp2);
 
        __global__ void MGPU_precomp_reaction_kernel_packing(const unsigned int* inLattice,
                                                             const uint8_t* inSites,
                                                             unsigned int* outLattice,
                                                             const unsigned long long timestepHash,
                                                             unsigned int* siteOverflowList,
                                                             int z_start,
                                                             const __restrict__ int8_t* SG,
                                                             const __restrict__ uint8_t* RLG,
                                                             const float* __restrict__ qp0,
                                                             const float* __restrict__ qp1,
                                                             const float* __restrict__ qp2,
                                                             unsigned int* buf_top,
                                                             unsigned int* buf_bot);
 
        __global__ void MGPU_reaction_kernel(const unsigned int* inLattice,
                                             const uint8_t* inSites,
                                             unsigned int* outLattice,
                                             const unsigned long long timestepHash,
                                             unsigned int* siteOverflowList,
                                             const unsigned int z_start,
                                             const int8_t* __restrict__ SG,
                                             const uint8_t* __restrict__ RLG);
    #endif
#else
    __global__ void MGPU_precomp_reaction_kernel(const unsigned int* inLattice,
                                                 const uint8_t* inSites,
                                                 unsigned int* outLattice,
                                                 const unsigned long long timestepHash,
                                                 unsigned int* siteOverflowList,
                                                 int z_start,
                                                 const float* __restrict__ qp0,
                                                 const float* __restrict__ qp1,
                                                 const float* __restrict__ qp2);
 
    __global__ void MGPU_precomp_reaction_kernel_packing(const unsigned int* inLattice,
                                                         const uint8_t* inSites,
                                                         unsigned int* outLattice,
                                                         const unsigned long long timestepHash,
                                                         unsigned int* siteOverflowList,
                                                         int z_start,
                                                         const float* __restrict__ qp0,
                                                         const float* __restrict__ qp1,
                                                         const float* __restrict__ qp2,
                                                         unsigned int* buf_top,
                                                         unsigned int* buf_bot);
 
    __global__ void MGPU_reaction_kernel(const unsigned int* inLattice,
                                         const uint8_t* inSites,
                                         unsigned int* outLattice,
                                         const unsigned long long timestepHash,
                                         unsigned int* siteOverflowList,
                                         const unsigned int z_start);
#endif
}
}
}
 
#endif