jianmu
/
lammps
forked from lijiext/lammps
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
lammps/lib/gpu/lal_device.cpp

657 lines
21 KiB
C++
Raw Blame History

/***************************************************************************
device.cpp
-------------------
W. Michael Brown (ORNL)
Class for management of the device where the computations are performed
__________________________________________________________________________
This file is part of the LAMMPS Accelerator Library (LAMMPS_AL)
__________________________________________________________________________
begin :
email : brownw@ornl.gov
***************************************************************************/
#include "lal_device.h"
#include "lal_precision.h"
#include <map>
#include <math.h>
#ifdef _OPENMP
#include <omp.h>
#endif
#if defined(USE_OPENCL)
#include "device_cl.h"
#elif defined(USE_CUDART)
const char *device=0;
#else
#include "device_cubin.h"
#endif
using namespace LAMMPS_AL;
#define DeviceT Device<numtyp, acctyp>
template <class numtyp, class acctyp>
DeviceT::Device() : _init_count(0), _device_init(false),
_gpu_mode(GPU_FORCE), _first_device(0),
_last_device(0), _compiled(false) {
}
template <class numtyp, class acctyp>
DeviceT::~Device() {
clear_device();
}
template <class numtyp, class acctyp>
int DeviceT::init_device(MPI_Comm world, MPI_Comm replica, const int first_gpu,
const int last_gpu, const int gpu_mode,
const double p_split, const int nthreads,
const int t_per_atom, const double cell_size) {
_nthreads=nthreads;
#ifdef _OPENMP
omp_set_num_threads(nthreads);
#endif
_threads_per_atom=t_per_atom;
_threads_per_charge=t_per_atom;
if (_device_init)
return 0;
_device_init=true;
_comm_world=replica; //world;
_comm_replica=replica;
_first_device=first_gpu;
_last_device=last_gpu;
_gpu_mode=gpu_mode;
_particle_split=p_split;
_cell_size=cell_size;
// Get the rank/size within the world
MPI_Comm_rank(_comm_world,&_world_me);
MPI_Comm_size(_comm_world,&_world_size);
// Get the rank/size within the replica
MPI_Comm_rank(_comm_replica,&_replica_me);
MPI_Comm_size(_comm_replica,&_replica_size);
// Get the names of all nodes
int name_length;
char node_name[MPI_MAX_PROCESSOR_NAME];
char node_names[MPI_MAX_PROCESSOR_NAME*_world_size];
MPI_Get_processor_name(node_name,&name_length);
MPI_Allgather(&node_name,MPI_MAX_PROCESSOR_NAME,MPI_CHAR,&node_names,
MPI_MAX_PROCESSOR_NAME,MPI_CHAR,_comm_world);
std::string node_string=std::string(node_name);
// Get the number of procs per node
std::map<std::string,int> name_map;
std::map<std::string,int>::iterator np;
for (int i=0; i<_world_size; i++) {
std::string i_string=std::string(&node_names[i*MPI_MAX_PROCESSOR_NAME]);
np=name_map.find(i_string);
if (np==name_map.end())
name_map[i_string]=1;
else
np->second++;
}
int procs_per_node=name_map.begin()->second;
// Assign a unique id to each node
int split_num=0, split_id=0;
for (np=name_map.begin(); np!=name_map.end(); ++np) {
if (np->first==node_string)
split_id=split_num;
split_num++;
}
// Set up a per node communicator and find rank within
MPI_Comm node_comm;
MPI_Comm_split(_comm_world, split_id, 0, &node_comm);
int node_rank;
MPI_Comm_rank(node_comm,&node_rank);
// set the device ID
_procs_per_gpu=static_cast<int>(ceil(static_cast<double>(procs_per_node)/
(last_gpu-first_gpu+1)));
int my_gpu=node_rank/_procs_per_gpu+first_gpu;
// Time on the device only if 1 proc per gpu
_time_device=true;
if (_procs_per_gpu>1)
_time_device=false;
// Set up a per device communicator
MPI_Comm_split(node_comm,my_gpu,0,&_comm_gpu);
MPI_Comm_rank(_comm_gpu,&_gpu_rank);
gpu=new UCL_Device();
if (my_gpu>=gpu->num_devices())
return -2;
#ifndef CUDA_PROXY
if (_procs_per_gpu>1 && gpu->sharing_supported(my_gpu)==false)
return -7;
#endif
if (gpu->set(my_gpu)!=UCL_SUCCESS)
return -6;
gpu->push_command_queue();
gpu->set_command_queue(1);
_long_range_precompute=0;
int flag=0;
for (int i=0; i<_procs_per_gpu; i++) {
if (_gpu_rank==i)
flag=compile_kernels();
gpu_barrier();
}
return flag;
}
template <class numtyp, class acctyp>
int DeviceT::init(Answer<numtyp,acctyp> &ans, const bool charge,
const bool rot, const int nlocal,
const int host_nlocal, const int nall,
Neighbor *nbor, const int maxspecial,
const int gpu_host, const int max_nbors,
const double cell_size, const bool pre_cut,
const int threads_per_atom) {
if (!_device_init)
return -1;
if (sizeof(acctyp)==sizeof(double) && gpu->double_precision()==false)
return -5;
// Counts of data transfers for timing overhead estimates
_data_in_estimate=0;
_data_out_estimate=1;
// Initial number of local particles
int ef_nlocal=nlocal;
if (_particle_split<1.0 && _particle_split>0.0)
ef_nlocal=static_cast<int>(_particle_split*nlocal);
int gpu_nbor=0;
if (_gpu_mode==Device<numtyp,acctyp>::GPU_NEIGH)
gpu_nbor=1;
else if (_gpu_mode==Device<numtyp,acctyp>::GPU_HYB_NEIGH)
gpu_nbor=2;
#ifndef USE_CUDPP
if (gpu_nbor==1)
gpu_nbor=2;
#endif
if (_init_count==0) {
// Initialize atom and nbor data
if (!atom.init(nall,charge,rot,*gpu,gpu_nbor,gpu_nbor>0 && maxspecial>0))
return -3;
_data_in_estimate++;
if (charge)
_data_in_estimate++;
if (rot)
_data_in_estimate++;
} else {
if (atom.charge()==false && charge)
_data_in_estimate++;
if (atom.quaternion()==false && rot)
_data_in_estimate++;
if (!atom.add_fields(charge,rot,gpu_nbor,gpu_nbor>0 && maxspecial))
return -3;
}
if (!ans.init(ef_nlocal,charge,rot,*gpu))
return -3;
if (!nbor->init(&_neighbor_shared,ef_nlocal,host_nlocal,max_nbors,maxspecial,
*gpu,gpu_nbor,gpu_host,pre_cut, _block_cell_2d,
_block_cell_id, _block_nbor_build, threads_per_atom,
_warp_size, _time_device))
return -3;
if (_cell_size<0.0)
nbor->cell_size(cell_size,cell_size);
else
nbor->cell_size(_cell_size,cell_size);
_init_count++;
return 0;
}
template <class numtyp, class acctyp>
int DeviceT::init(Answer<numtyp,acctyp> &ans, const int nlocal,
const int nall) {
if (!_device_init)
return -1;
if (sizeof(acctyp)==sizeof(double) && gpu->double_precision()==false)
return -5;
if (_init_count==0) {
// Initialize atom and nbor data
if (!atom.init(nall,true,false,*gpu,false,false))
return -3;
} else
if (!atom.add_fields(true,false,false,false))
return -3;
if (!ans.init(nlocal,true,false,*gpu))
return -3;
_init_count++;
return 0;
}
template <class numtyp, class acctyp>
void DeviceT::set_single_precompute
(PPPM<numtyp,acctyp,float,_lgpu_float4> *pppm) {
_long_range_precompute=1;
pppm_single=pppm;
}
template <class numtyp, class acctyp>
void DeviceT::set_double_precompute
(PPPM<numtyp,acctyp,double,_lgpu_double4> *pppm) {
_long_range_precompute=2;
pppm_double=pppm;
}
template <class numtyp, class acctyp>
void DeviceT::init_message(FILE *screen, const char *name,
const int first_gpu, const int last_gpu) {
#if defined(USE_OPENCL)
std::string fs="";
#elif defined(USE_CUDART)
std::string fs="";
#else
std::string fs=toa(gpu->free_gigabytes())+"/";
#endif
if (_replica_me == 0 && screen) {
fprintf(screen,"\n-------------------------------------");
fprintf(screen,"-------------------------------------\n");
fprintf(screen,"- Using GPGPU acceleration for %s:\n",name);
fprintf(screen,"- with %d proc(s) per device.\n",_procs_per_gpu);
#ifdef _OPENMP
fprintf(screen,"- with %d thread(s) per proc.\n",_nthreads);
#endif
#ifdef USE_OPENCL
fprintf(screen,"- with OpenCL Parameters for: %s\n",OCL_VENDOR);
#endif
fprintf(screen,"-------------------------------------");
fprintf(screen,"-------------------------------------\n");
int last=last_gpu+1;
if (last>gpu->num_devices())
last=gpu->num_devices();
for (int i=first_gpu; i<last; i++) {
std::string sname;
if (i==first_gpu)
sname=gpu->name(i)+", "+toa(gpu->cores(i))+" cores, "+fs+
toa(gpu->gigabytes(i))+" GB, "+toa(gpu->clock_rate(i))+" GHZ (";
else
sname=gpu->name(i)+", "+toa(gpu->cores(i))+" cores, "+fs+
toa(gpu->clock_rate(i))+" GHZ (";
if (sizeof(PRECISION)==4) {
if (sizeof(ACC_PRECISION)==4)
sname+="Single Precision)";
else
sname+="Mixed Precision)";
} else
sname+="Double Precision)";
fprintf(screen,"GPU %d: %s\n",i,sname.c_str());
}
fprintf(screen,"-------------------------------------");
fprintf(screen,"-------------------------------------\n\n");
}
}
template <class numtyp, class acctyp>
void DeviceT::estimate_gpu_overhead(const int kernel_calls,
double &gpu_overhead,
double &gpu_driver_overhead) {
UCL_H_Vec<int> *host_data_in=NULL, *host_data_out=NULL;
UCL_D_Vec<int> *dev_data_in=NULL, *dev_data_out=NULL, *kernel_data=NULL;
UCL_Timer *timers_in=NULL, *timers_out=NULL, *timers_kernel=NULL;
UCL_Timer over_timer(*gpu);
if (_data_in_estimate>0) {
host_data_in=new UCL_H_Vec<int>[_data_in_estimate];
dev_data_in=new UCL_D_Vec<int>[_data_in_estimate];
timers_in=new UCL_Timer[_data_in_estimate];
}
if (_data_out_estimate>0) {
host_data_out=new UCL_H_Vec<int>[_data_out_estimate];
dev_data_out=new UCL_D_Vec<int>[_data_out_estimate];
timers_out=new UCL_Timer[_data_out_estimate];
}
if (kernel_calls>0) {
kernel_data=new UCL_D_Vec<int>[kernel_calls];
timers_kernel=new UCL_Timer[kernel_calls];
}
for (int i=0; i<_data_in_estimate; i++) {
host_data_in[i].alloc(1,*gpu);
dev_data_in[i].alloc(1,*gpu);
timers_in[i].init(*gpu);
}
for (int i=0; i<_data_out_estimate; i++) {
host_data_out[i].alloc(1,*gpu);
dev_data_out[i].alloc(1,*gpu);
timers_out[i].init(*gpu);
}
for (int i=0; i<kernel_calls; i++) {
kernel_data[i].alloc(1,*gpu);
timers_kernel[i].init(*gpu);
}
gpu_overhead=0.0;
gpu_driver_overhead=0.0;
for (int i=0; i<10; i++) {
gpu->sync();
gpu_barrier();
over_timer.start();
gpu->sync();
gpu_barrier();
double driver_time=MPI_Wtime();
for (int i=0; i<_data_in_estimate; i++) {
timers_in[i].start();
ucl_copy(dev_data_in[i],host_data_in[i],true);
timers_in[i].stop();
}
for (int i=0; i<kernel_calls; i++) {
timers_kernel[i].start();
zero(kernel_data[i],1);
timers_kernel[i].stop();
}
for (int i=0; i<_data_out_estimate; i++) {
timers_out[i].start();
ucl_copy(host_data_out[i],dev_data_out[i],true);
timers_out[i].stop();
}
over_timer.stop();
double time=over_timer.seconds();
driver_time=MPI_Wtime()-driver_time;
if (time_device()) {
for (int i=0; i<_data_in_estimate; i++)
timers_in[i].add_to_total();
for (int i=0; i<kernel_calls; i++)
timers_kernel[i].add_to_total();
for (int i=0; i<_data_out_estimate; i++)
timers_out[i].add_to_total();
}
double mpi_time, mpi_driver_time;
MPI_Allreduce(&time,&mpi_time,1,MPI_DOUBLE,MPI_MAX,gpu_comm());
MPI_Allreduce(&driver_time,&mpi_driver_time,1,MPI_DOUBLE,MPI_MAX,gpu_comm());
gpu_overhead+=mpi_time;
gpu_driver_overhead+=mpi_driver_time;
}
gpu_overhead/=10.0;
gpu_driver_overhead/=10.0;
if (_data_in_estimate>0) {
delete [] host_data_in;
delete [] dev_data_in;
delete [] timers_in;
}
if (_data_out_estimate>0) {
delete [] host_data_out;
delete [] dev_data_out;
delete [] timers_out;
}
if (kernel_calls>0) {
delete [] kernel_data;
delete [] timers_kernel;
}
}
template <class numtyp, class acctyp>
void DeviceT::output_times(UCL_Timer &time_pair, Answer<numtyp,acctyp> &ans,
Neighbor &nbor, const double avg_split,
const double max_bytes, const double gpu_overhead,
const double driver_overhead,
const int threads_per_atom, FILE *screen) {
double single[9], times[9];
single[0]=atom.transfer_time()+ans.transfer_time();
single[1]=nbor.time_nbor.total_seconds()+nbor.time_hybrid1.total_seconds()+
nbor.time_hybrid2.total_seconds();
single[2]=nbor.time_kernel.total_seconds();
single[3]=time_pair.total_seconds();
single[4]=atom.cast_time()+ans.cast_time();
single[5]=gpu_overhead;
single[6]=driver_overhead;
single[7]=ans.cpu_idle_time();
single[8]=nbor.bin_time();
MPI_Reduce(single,times,9,MPI_DOUBLE,MPI_SUM,0,_comm_replica);
double my_max_bytes=max_bytes+atom.max_gpu_bytes();
double mpi_max_bytes;
MPI_Reduce(&my_max_bytes,&mpi_max_bytes,1,MPI_DOUBLE,MPI_MAX,0,_comm_replica);
double max_mb=mpi_max_bytes/(1024.0*1024.0);
if (replica_me()==0)
if (screen && times[5]>0.0) {
fprintf(screen,"\n\n-------------------------------------");
fprintf(screen,"--------------------------------\n");
fprintf(screen," GPU Time Info (average): ");
fprintf(screen,"\n-------------------------------------");
fprintf(screen,"--------------------------------\n");
if (time_device()) {
fprintf(screen,"Data Transfer: %.4f s.\n",times[0]/_replica_size);
fprintf(screen,"Data Cast/Pack: %.4f s.\n",times[4]/_replica_size);
fprintf(screen,"Neighbor copy: %.4f s.\n",times[1]/_replica_size);
if (nbor.gpu_nbor()>0)
fprintf(screen,"Neighbor build: %.4f s.\n",times[2]/_replica_size);
else
fprintf(screen,"Neighbor unpack: %.4f s.\n",times[2]/_replica_size);
fprintf(screen,"Force calc: %.4f s.\n",times[3]/_replica_size);
}
if (nbor.gpu_nbor()==2)
fprintf(screen,"Neighbor (CPU): %.4f s.\n",times[8]/_replica_size);
fprintf(screen,"GPU Overhead: %.4f s.\n",times[5]/_replica_size);
fprintf(screen,"Average split: %.4f.\n",avg_split);
fprintf(screen,"Threads / atom: %d.\n",threads_per_atom);
fprintf(screen,"Max Mem / Proc: %.2f MB.\n",max_mb);
fprintf(screen,"CPU Driver_Time: %.4f s.\n",times[6]/_replica_size);
fprintf(screen,"CPU Idle_Time: %.4f s.\n",times[7]/_replica_size);
fprintf(screen,"-------------------------------------");
fprintf(screen,"--------------------------------\n\n");
}
}
template <class numtyp, class acctyp>
void DeviceT::output_kspace_times(UCL_Timer &time_in,
UCL_Timer &time_out,
UCL_Timer &time_map,
UCL_Timer &time_rho,
UCL_Timer &time_interp,
Answer<numtyp,acctyp> &ans,
const double max_bytes,
const double cpu_time,
const double idle_time, FILE *screen) {
double single[8], times[8];
single[0]=time_out.total_seconds();
single[1]=time_in.total_seconds()+atom.transfer_time()+atom.cast_time();
single[2]=time_map.total_seconds();
single[3]=time_rho.total_seconds();
single[4]=time_interp.total_seconds();
single[5]=ans.transfer_time()+ans.cast_time();
single[6]=cpu_time;
single[7]=idle_time;
MPI_Reduce(single,times,8,MPI_DOUBLE,MPI_SUM,0,_comm_replica);
double my_max_bytes=max_bytes+atom.max_gpu_bytes();
double mpi_max_bytes;
MPI_Reduce(&my_max_bytes,&mpi_max_bytes,1,MPI_DOUBLE,MPI_MAX,0,_comm_replica);
double max_mb=mpi_max_bytes/(1024.0*1024.0);
if (replica_me()==0)
if (screen && times[6]>0.0) {
fprintf(screen,"\n\n-------------------------------------");
fprintf(screen,"--------------------------------\n");
fprintf(screen," GPU Time Info (average): ");
fprintf(screen,"\n-------------------------------------");
fprintf(screen,"--------------------------------\n");
if (time_device()) {
fprintf(screen,"Data Out: %.4f s.\n",times[0]/_replica_size);
fprintf(screen,"Data In: %.4f s.\n",times[1]/_replica_size);
fprintf(screen,"Kernel (map): %.4f s.\n",times[2]/_replica_size);
fprintf(screen,"Kernel (rho): %.4f s.\n",times[3]/_replica_size);
fprintf(screen,"Force interp: %.4f s.\n",times[4]/_replica_size);
fprintf(screen,"Total rho: %.4f s.\n",
(times[0]+times[2]+times[3])/_replica_size);
fprintf(screen,"Total interp: %.4f s.\n",
(times[1]+times[4])/_replica_size);
fprintf(screen,"Force copy/cast: %.4f s.\n",times[5]/_replica_size);
fprintf(screen,"Total: %.4f s.\n",
(times[0]+times[1]+times[2]+times[3]+times[4]+times[5])/
_replica_size);
}
fprintf(screen,"CPU Poisson: %.4f s.\n",times[6]/_replica_size);
fprintf(screen,"CPU Idle Time: %.4f s.\n",times[7]/_replica_size);
fprintf(screen,"Max Mem / Proc: %.2f MB.\n",max_mb);
fprintf(screen,"-------------------------------------");
fprintf(screen,"--------------------------------\n\n");
}
}
template <class numtyp, class acctyp>
void DeviceT::clear() {
if (_init_count>0) {
_long_range_precompute=0;
_init_count--;
if (_init_count==0) {
atom.clear();
_neighbor_shared.clear();
}
}
}
template <class numtyp, class acctyp>
void DeviceT::clear_device() {
while (_init_count>0)
clear();
if (_compiled) {
k_zero.clear();
k_info.clear();
delete dev_program;
_compiled=false;
}
if (_device_init) {
delete gpu;
_device_init=false;
}
}
template <class numtyp, class acctyp>
int DeviceT::compile_kernels() {
int flag=0;
if (_compiled)
return flag;
std::string flags="-cl-mad-enable -D"+std::string(OCL_VENDOR);
dev_program=new UCL_Program(*gpu);
int success=dev_program->load_string(device,flags.c_str());
if (success!=UCL_SUCCESS)
return -4;
k_zero.set_function(*dev_program,"kernel_zero");
k_info.set_function(*dev_program,"kernel_info");
_compiled=true;
UCL_Vector<int,int> gpu_lib_data(15,*gpu,UCL_NOT_PINNED);
k_info.set_size(1,1);
k_info.run(&gpu_lib_data);
gpu_lib_data.update_host(false);
_ptx_arch=static_cast<double>(gpu_lib_data[0])/100.0;
#ifndef USE_OPENCL
if (_ptx_arch>gpu->arch() || floor(_ptx_arch)<floor(gpu->arch()))
return -4;
#endif
_num_mem_threads=gpu_lib_data[1];
_warp_size=gpu_lib_data[2];
if (_threads_per_atom<1)
_threads_per_atom=gpu_lib_data[3];
if (_threads_per_charge<1)
_threads_per_charge=gpu_lib_data[13];
_pppm_max_spline=gpu_lib_data[4];
_pppm_block=gpu_lib_data[5];
_block_pair=gpu_lib_data[6];
_max_shared_types=gpu_lib_data[7];
_block_cell_2d=gpu_lib_data[8];
_block_cell_id=gpu_lib_data[9];
_block_nbor_build=gpu_lib_data[10];
_block_bio_pair=gpu_lib_data[11];
_max_bio_shared_types=gpu_lib_data[12];
_block_ellipse=gpu_lib_data[14];
if (static_cast<size_t>(_block_pair)>gpu->group_size())
_block_pair=gpu->group_size();
if (static_cast<size_t>(_block_bio_pair)>gpu->group_size())
_block_bio_pair=gpu->group_size();
if (_threads_per_atom>_warp_size)
_threads_per_atom=_warp_size;
if (_warp_size%_threads_per_atom!=0)
_threads_per_atom=1;
if (_threads_per_atom & (_threads_per_atom - 1))
_threads_per_atom=1;
if (_threads_per_charge>_warp_size)
_threads_per_charge=_warp_size;
if (_warp_size%_threads_per_charge!=0)
_threads_per_charge=1;
if (_threads_per_charge & (_threads_per_charge - 1))
_threads_per_charge=1;
return flag;
}
template <class numtyp, class acctyp>
double DeviceT::host_memory_usage() const {
return atom.host_memory_usage()+4*sizeof(numtyp)+
sizeof(Device<numtyp,acctyp>);
}
template class Device<PRECISION,ACC_PRECISION>;
Device<PRECISION,ACC_PRECISION> global_device;
int lmp_init_device(MPI_Comm world, MPI_Comm replica, const int first_gpu,
const int last_gpu, const int gpu_mode,
const double particle_split, const int nthreads,
const int t_per_atom, const double cell_size) {
return global_device.init_device(world,replica,first_gpu,last_gpu,gpu_mode,
particle_split,nthreads,t_per_atom,
cell_size);
}
void lmp_clear_device() {
global_device.clear_device();
}
double lmp_gpu_forces(double **f, double **tor, double *eatom,
double **vatom, double *virial, double &ecoul) {
return global_device.fix_gpu(f,tor,eatom,vatom,virial,ecoul);
}
Reference in New Issue View Git Blame Copy Permalink