1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
|
/***************************************************************************
neighbor_shared.cpp
-------------------
W. Michael Brown (ORNL)
Class for management of data shared by all neighbor lists
__________________________________________________________________________
This file is part of the LAMMPS Accelerator Library (LAMMPS_AL)
__________________________________________________________________________
begin :
email : brownw@ornl.gov
***************************************************************************/
#include <cmath>
#include "lal_precision.h"
#include "lal_neighbor_shared.h"
#if defined(USE_OPENCL)
#include "neighbor_cpu_cl.h"
#include "neighbor_gpu_cl.h"
#elif defined(USE_CUDART)
const char *neighbor_cpu=0;
const char *neighbor_gpu=0;
#else
#include "neighbor_cpu_cubin.h"
#include "neighbor_gpu_cubin.h"
#endif
using namespace LAMMPS_AL;
void NeighborShared::clear() {
if (_compiled) {
if (_gpu_nbor>0) {
if (_gpu_nbor==1) {
k_cell_id.clear();
k_cell_counts.clear();
}
k_build_nbor.clear();
k_transpose.clear();
k_special.clear();
delete build_program;
} else {
k_nbor.clear();
delete nbor_program;
}
_compiled=false;
}
}
double NeighborShared::best_cell_size(const double subx, const double suby,
const double subz, const int nlocal,
const double cut) {
if (_cached_cell_size && _cut_sort==cut) {
_cached_cell_size=false;
return _cell_size;
}
const double box_density = static_cast<double>(nlocal) / (subx*suby*subz);
const double density=box_density*cut*cut*cut;
if (density >= 4.0 * _simd_size) return cut*0.5;
else if (density >= 0.5 * _simd_size) return cut;
const double iters = 60;
const double inc = cut/(iters-1);
const double iss = 1.0 / _simd_size;
double test_size = cut;
double best_iters = 1e200;
double best_size;
for (int i = 0; i < iters; i++) {
const double i_test_size = 1.0/test_size;
const int ncellx = static_cast<int>(ceil(subx*i_test_size));
const int ncelly = static_cast<int>(ceil(suby*i_test_size));
const int ncellz = static_cast<int>(ceil(subz*i_test_size));
const double density = box_density*test_size*test_size*test_size;
const double iters_per_cell = ceil(iss*density);
const double iters = ncellx*ncelly*ncellz*iters_per_cell*
ceil(density*27.0*iss);
if (iters < best_iters) {
best_iters = iters;
best_size = test_size;
}
test_size += inc;
}
const int cells_in_cutoff=static_cast<int>(ceil(cut/best_size));
if (cells_in_cutoff > 2) best_size=cut*0.5;
return best_size;
}
void NeighborShared::compile_kernels(UCL_Device &dev, const int gpu_nbor,
const std::string &flags) {
if (_compiled)
return;
_gpu_nbor=gpu_nbor;
if (_gpu_nbor==0) {
nbor_program=new UCL_Program(dev);
nbor_program->load_string(neighbor_cpu,flags.c_str(),nullptr,stderr);
k_nbor.set_function(*nbor_program,"kernel_unpack");
} else {
build_program=new UCL_Program(dev);
build_program->load_string(neighbor_gpu,flags.c_str(),nullptr,stderr);
if (_gpu_nbor==1) {
k_cell_id.set_function(*build_program,"calc_cell_id");
k_cell_counts.set_function(*build_program,"kernel_calc_cell_counts");
}
k_build_nbor.set_function(*build_program,"calc_neigh_list_cell");
k_transpose.set_function(*build_program,"transpose");
k_special.set_function(*build_program,"kernel_special");
neigh_tex.get_texture(*build_program,"pos_tex");
}
_compiled=true;
}
|
