File size: 7,777 Bytes
66c9c8a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
/** Copyright (c) 2022 NVIDIA CORPORATION.  All rights reserved.
 * NVIDIA CORPORATION and its licensors retain all intellectual property
 * and proprietary rights in and to this software, related documentation
 * and any modifications thereto.  Any use, reproduction, disclosure or
 * distribution of this software and related documentation without an express
 * license agreement from NVIDIA CORPORATION is strictly prohibited.
 */

#include "warp.h"
#include "cuda_util.h"
#include "hashgrid.h"
#include "sort.h"
#include "string.h"

using namespace wp;

#include <map>

namespace 
{
    // host-side copy of mesh descriptors, maps GPU mesh address (id) to a CPU desc
    std::map<uint64_t, HashGrid> g_hash_grid_descriptors;

} // anonymous namespace


namespace wp
{

bool hash_grid_get_descriptor(uint64_t id, HashGrid& grid)
{
    const auto& iter = g_hash_grid_descriptors.find(id);
    if (iter == g_hash_grid_descriptors.end())
        return false;
    else
        grid = iter->second;
        return true;
}

void hash_grid_add_descriptor(uint64_t id, const HashGrid& grid)
{
    g_hash_grid_descriptors[id] = grid;
}

void hash_grid_rem_descriptor(uint64_t id)
{
    g_hash_grid_descriptors.erase(id);

}

// implemented in hashgrid.cu
void hash_grid_rebuild_device(const HashGrid& grid, const wp::vec3* points, int num_points);

} // namespace wp


// host methods
uint64_t hash_grid_create_host(int dim_x, int dim_y, int dim_z)
{
    HashGrid* grid = new HashGrid();
    memset(grid, 0, sizeof(HashGrid));
    
    grid->dim_x = dim_x;
    grid->dim_y = dim_y;
    grid->dim_z = dim_z;

    const int num_cells = dim_x*dim_y*dim_z;   
    grid->cell_starts = (int*)alloc_host(num_cells*sizeof(int));
    grid->cell_ends = (int*)alloc_host(num_cells*sizeof(int));

    return (uint64_t)(grid);
}

void hash_grid_destroy_host(uint64_t id)
{
    HashGrid* grid = (HashGrid*)(id);

    free_host(grid->point_ids);
    free_host(grid->point_cells);
    free_host(grid->cell_starts);
    free_host(grid->cell_ends);

    delete grid;
}

void hash_grid_reserve_host(uint64_t id, int num_points)
{
    HashGrid* grid = (HashGrid*)(id);

    if (num_points > grid->max_points)
    {
        free_host(grid->point_cells);
        free_host(grid->point_ids);
        
        const int num_to_alloc = num_points*3/2;
        grid->point_cells = (int*)alloc_host(2*num_to_alloc*sizeof(int));  // *2 for auxilliary radix buffers
        grid->point_ids = (int*)alloc_host(2*num_to_alloc*sizeof(int));    // *2 for auxilliary radix buffers

        grid->max_points = num_to_alloc;
    }

    grid->num_points = num_points;
}

void hash_grid_update_host(uint64_t id, float cell_width, const wp::vec3* points, int num_points)
{
    HashGrid* grid = (HashGrid*)(id);

    hash_grid_reserve_host(id, num_points);

    grid->cell_width = cell_width;
    grid->cell_width_inv = 1.0f / cell_width;

    // calculate cell for each position
    for (int i=0; i < num_points; ++i)
    {
        grid->point_cells[i] = hash_grid_index(*grid, points[i]);
        grid->point_ids[i] = i;
    }
    
    // sort indices
    radix_sort_pairs_host(grid->point_cells, grid->point_ids, num_points);

    const int num_cells = grid->dim_x * grid->dim_y * grid->dim_z;
    memset(grid->cell_starts, 0, sizeof(int) * num_cells);
    memset(grid->cell_ends, 0, sizeof(int) * num_cells);

    // compute cell start / end
    for (int i=0; i < num_points; ++i)
    {
		// scan the particle-cell array to find the start and end
		const int c = grid->point_cells[i];
		
		if (i == 0)
			grid->cell_starts[c] = 0;
		else
		{
			const int p = grid->point_cells[i-1];

			if (c != p)
			{
				grid->cell_starts[c] = i;
                grid->cell_ends[p] = i;
			}
		}

        if (i == num_points - 1)
        {
            grid->cell_ends[c] = i + 1;
        }
	}
}

// device methods
uint64_t hash_grid_create_device(void* context, int dim_x, int dim_y, int dim_z)
{
    ContextGuard guard(context);

    HashGrid grid;
    memset(&grid, 0, sizeof(HashGrid));

    grid.context = context ? context : cuda_context_get_current();

    grid.dim_x = dim_x;
    grid.dim_y = dim_y;
    grid.dim_z = dim_z;

    const int num_cells = dim_x*dim_y*dim_z;   
    grid.cell_starts = (int*)alloc_device(WP_CURRENT_CONTEXT, num_cells*sizeof(int));
    grid.cell_ends = (int*)alloc_device(WP_CURRENT_CONTEXT, num_cells*sizeof(int));

    // upload to device
    HashGrid* grid_device = (HashGrid*)(alloc_device(WP_CURRENT_CONTEXT, sizeof(HashGrid)));
    memcpy_h2d(WP_CURRENT_CONTEXT, grid_device, &grid, sizeof(HashGrid));

    uint64_t grid_id = (uint64_t)(grid_device);
    hash_grid_add_descriptor(grid_id, grid);

    return grid_id;
}

void hash_grid_destroy_device(uint64_t id)
{
    HashGrid grid;
    if (hash_grid_get_descriptor(id, grid))
    {
        ContextGuard guard(grid.context);

        free_device(WP_CURRENT_CONTEXT, grid.point_ids);
        free_device(WP_CURRENT_CONTEXT, grid.point_cells);
        free_device(WP_CURRENT_CONTEXT, grid.cell_starts);
        free_device(WP_CURRENT_CONTEXT, grid.cell_ends);

        free_device(WP_CURRENT_CONTEXT, (HashGrid*)id);
        
        hash_grid_rem_descriptor(id);
    }
}


void hash_grid_reserve_device(uint64_t id, int num_points)
{
    HashGrid grid;

    if (hash_grid_get_descriptor(id, grid))
    {
        if (num_points > grid.max_points)
        {
            ContextGuard guard(grid.context);

            free_device(WP_CURRENT_CONTEXT, grid.point_cells);
            free_device(WP_CURRENT_CONTEXT, grid.point_ids);
            
            const int num_to_alloc = num_points*3/2;
            grid.point_cells = (int*)alloc_device(WP_CURRENT_CONTEXT, 2*num_to_alloc*sizeof(int));  // *2 for auxilliary radix buffers
            grid.point_ids = (int*)alloc_device(WP_CURRENT_CONTEXT, 2*num_to_alloc*sizeof(int));    // *2 for auxilliary radix buffers
            grid.max_points = num_to_alloc;

            // ensure we pre-size our sort routine to avoid
            // allocations during graph capture
            radix_sort_reserve(WP_CURRENT_CONTEXT, num_to_alloc);

            // update device side grid descriptor, todo: this is
            // slightly redundant since it is performed again
            // inside hash_grid_update_device(), but since
            // reserve can be called from Python we need to make 
            // sure it is consistent
            memcpy_h2d(WP_CURRENT_CONTEXT, (HashGrid*)id, &grid, sizeof(HashGrid));

            // update host side grid descriptor
            hash_grid_add_descriptor(id, grid);
        }
    }
}

void hash_grid_update_device(uint64_t id, float cell_width, const wp::vec3* points, int num_points)
{
    
    // ensure we have enough memory reserved for update
    // this must be done before retrieving the descriptor
    // below since it may update it
    hash_grid_reserve_device(id, num_points);

    // host grid must be static so that we can
    // perform host->device memcpy from this variable
    // and have it safely recorded inside CUDA graphs
    static HashGrid grid;

    if (hash_grid_get_descriptor(id, grid))
    {
        ContextGuard guard(grid.context);

        grid.num_points = num_points;
        grid.cell_width = cell_width;
        grid.cell_width_inv = 1.0f / cell_width;

        hash_grid_rebuild_device(grid, points, num_points);

        // update device side grid descriptor
        memcpy_h2d(WP_CURRENT_CONTEXT, (HashGrid*)id, &grid, sizeof(HashGrid));

        // update host side grid descriptor
        hash_grid_add_descriptor(id, grid);
    }
}

#if !WP_ENABLE_CUDA

namespace wp
{

void hash_grid_rebuild_device(const HashGrid& grid, const wp::vec3* points, int num_points)
{

}

} // namespace wp

#endif // !WP_ENABLE_CUDA