File size: 10,406 Bytes
a6dd040
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
/*
 * Copyright (c) 2020 NVIDIA CORPORATION.
 * Copyright (c) 2018-2020 Chris Choy (chrischoy@ai.stanford.edu)
 *
 * Permission is hereby granted, free of charge, to any person obtaining a copy
 * of this software and associated documentation files (the "Software"), to deal
 * in 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:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * 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
 * AUTHORS 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
 * IN THE SOFTWARE.
 *
 * Please cite "4D Spatio-Temporal ConvNets: Minkowski Convolutional Neural
 * Networks", CVPR'19 (https://arxiv.org/abs/1904.08755) if you use any part
 * of the code.
 */
#include "coordinate_map_functors.cuh"
#include "coordinate_map_gpu.cuh"
#include "gpu.cuh"
#include "kernel_region.hpp"
#include "types.hpp"
#include "utils.hpp"

#include <thrust/device_vector.h>
#include <thrust/for_each.h>
#include <thrust/host_vector.h>

#include <torch/extension.h>

namespace minkowski {

using coordinate_type = int32_t;
using index_type = default_types::index_type;
using size_type = default_types::size_type;

/*
 * The number of threads must be > coordinate_size
 */
__global__ void kernel_region_iterator_test(
    coordinate_type const *__restrict__ p_coordinate,
    size_type number_of_coordinates, //
    gpu_kernel_region<coordinate_type> kernel,
    coordinate_type *__restrict__ p_return_coordinates) {
  extern __shared__ coordinate_type sh_coordinate[];

  auto const tx = threadIdx.x;
  auto const bx = blockIdx.x;
  auto const x = blockDim.x * bx + tx;

  size_type coordinate_size = kernel.coordinate_size();
  size_type volume = kernel.volume();

  coordinate_type *sh_tmp = sh_coordinate + tx * coordinate_size;
  coordinate_type *sh_lb =
      sh_coordinate + (CUDA_NUM_THREADS + tx) * coordinate_size;
  coordinate_type *sh_ub =
      sh_coordinate + (2 * CUDA_NUM_THREADS + tx) * coordinate_size;

  index_type *sh_index = reinterpret_cast<index_type *>(
      sh_coordinate + 3 * CUDA_NUM_THREADS * coordinate_size);
  index_type *sh_tensor_stride = sh_index;
  index_type *sh_kernel_size = sh_index + coordinate_size;
  index_type *sh_dilation = sh_index + 2 * coordinate_size;

  if (tx < coordinate_size - 1) {
    sh_tensor_stride[tx] = kernel.tensor_stride()[tx];
    sh_kernel_size[tx] = kernel.kernel_size()[tx];
    sh_dilation[tx] = kernel.dilation()[tx];
  }
  __syncthreads();
  if (x >= number_of_coordinates)
    return;

  // iterate and copy
  index_type out_index = x * kernel.volume();
  kernel.set_bounds(&p_coordinate[x * coordinate_size], sh_lb, sh_ub, sh_tmp);
  for (auto const &coordinate : kernel) {
    for (index_type i = 0; i < coordinate_size; ++i) {
      p_return_coordinates[out_index * coordinate_size + i] = coordinate[i];
    }
    ++out_index;
  }
}

at::Tensor region_iterator_test(const torch::Tensor &coordinates,
                                const torch::Tensor &th_kernel_size) {
  // Create TensorArgs. These record the names and positions of each tensor as
  // parameters.
  torch::TensorArg arg_coordinates(coordinates, "coordinates", 0);
  torch::TensorArg arg_kernel_size(th_kernel_size, "kernel_size", 1);

  torch::CheckedFrom c = "region_iterator_test";
  torch::checkContiguous(c, arg_coordinates);
  torch::checkContiguous(c, arg_kernel_size);
  // must match coordinate_type
  torch::checkScalarType(c, arg_coordinates, torch::kInt);
  torch::checkScalarType(c, arg_kernel_size, torch::kInt);
  torch::checkBackend(c, arg_coordinates.tensor, torch::Backend::CUDA);
  torch::checkBackend(c, arg_kernel_size.tensor, torch::Backend::CPU);
  torch::checkDim(c, arg_coordinates, 2);
  torch::checkDim(c, arg_kernel_size, 1);

  auto const N = (index_type)coordinates.size(0);
  auto const coordinate_size = (index_type)coordinates.size(1);
  coordinate_type *p_coordinate = coordinates.data_ptr<coordinate_type>();
  coordinate_type *p_kernel_size = th_kernel_size.data_ptr<coordinate_type>();

  default_types::stride_type tensor_stride(coordinate_size - 1);
  default_types::stride_type kernel_size(coordinate_size - 1);
  default_types::stride_type dilation(coordinate_size - 1);

  for (index_type i = 0; i < coordinate_size - 1; ++i) {
    tensor_stride[i] = 1;
    kernel_size[i] = p_kernel_size[i];
    dilation[i] = 1;
  }

  auto cpu_kernel = cpu_kernel_region<coordinate_type>(
      RegionType::HYPER_CUBE, coordinate_size, tensor_stride.data(),
      kernel_size.data(), dilation.data());
  auto kernel = gpu_kernel_region<coordinate_type>(cpu_kernel.to_gpu());

  LOG_DEBUG("initialize vectors");

  torch::Tensor out_coordinates = torch::empty(
      {N * kernel.volume(), coordinate_size}, coordinates.options());

  uint32_t shared_memory_size_in_bytes =
      3 * CUDA_NUM_THREADS * coordinate_size * sizeof(coordinate_type) +
      3 * coordinate_size * sizeof(index_type);

  kernel_region_iterator_test<<<GET_BLOCKS(N, CUDA_NUM_THREADS),
                                CUDA_NUM_THREADS,
                                shared_memory_size_in_bytes>>>(
      p_coordinate, //
      N,            //
      kernel,       //
      out_coordinates.data_ptr<coordinate_type>());

  LOG_DEBUG("End call");
  CUDA_CHECK(cudaStreamSynchronize(0));

  return out_coordinates;
}

std::tuple<std::pair<cpu_in_maps, cpu_out_maps>, size_type, double>
kernel_map_test(const torch::Tensor &in_coordinates,
                const torch::Tensor &out_coordinates,
                const torch::Tensor &kernel_size,
                uint32_t occupancy, //
                uint32_t thread_dim) {
  // Create TensorArgs. These record the names and positions of each tensor as
  // parameters.
  torch::TensorArg arg_in_coordinates(in_coordinates, "coordinates", 0);
  torch::TensorArg arg_out_coordinates(out_coordinates, "coordinates", 1);
  torch::TensorArg arg_kernel_size(kernel_size, "kernel_size", 2);

  torch::CheckedFrom c = "region_iterator_test";
  torch::checkContiguous(c, arg_in_coordinates);
  torch::checkContiguous(c, arg_out_coordinates);
  torch::checkContiguous(c, arg_kernel_size);
  // must match coordinate_type
  torch::checkScalarType(c, arg_in_coordinates, torch::kInt);
  torch::checkScalarType(c, arg_out_coordinates, torch::kInt);
  torch::checkScalarType(c, arg_kernel_size, torch::kInt);
  torch::checkBackend(c, arg_in_coordinates.tensor, torch::Backend::CUDA);
  torch::checkBackend(c, arg_out_coordinates.tensor, torch::Backend::CUDA);
  torch::checkBackend(c, arg_kernel_size.tensor, torch::Backend::CPU);
  torch::checkDim(c, arg_in_coordinates, 2);
  torch::checkDim(c, arg_out_coordinates, 2);
  torch::checkDim(c, arg_kernel_size, 1);

  auto const N_in = (index_type)in_coordinates.size(0);
  auto const D = (index_type)in_coordinates.size(1);

  auto const N_out = (index_type)out_coordinates.size(0);
  auto const D_out = (index_type)out_coordinates.size(1);

  ASSERT(D == D_out, "dimension mismatch");

  coordinate_type const *ptr_in = in_coordinates.data_ptr<coordinate_type>();
  coordinate_type const *ptr_out = out_coordinates.data_ptr<coordinate_type>();

  CoordinateMapGPU<coordinate_type> in_map{N_in, D, occupancy};
  CoordinateMapGPU<coordinate_type> out_map{N_out, D, occupancy};

  auto in_coordinate_range = coordinate_range<coordinate_type>(N_in, D, ptr_in);
  in_map.insert<false>(in_coordinate_range.begin(), // key begin
                       in_coordinate_range.end());  // key end

  auto out_coordinate_range =
      coordinate_range<coordinate_type>(N_out, D, ptr_out);
  out_map.insert<false>(out_coordinate_range.begin(), // key begin
                        out_coordinate_range.end());  // key end

  LOG_DEBUG("coordinate initialization");

  // Kernel region
  coordinate_type *p_kernel_size = kernel_size.data_ptr<coordinate_type>();
  default_types::stride_type tensor_stride;
  default_types::stride_type s_kernel_size;
  default_types::stride_type dilation;
  for (index_type i = 0; i < D - 1; ++i) {
    tensor_stride.push_back(1);
    s_kernel_size.push_back(p_kernel_size[i]);
    dilation.push_back(1);
  }

  auto region = cpu_kernel_region<coordinate_type>(
      RegionType::HYPER_CUBE, D, tensor_stride.data(), s_kernel_size.data(),
      dilation.data());
  LOG_DEBUG("cpu_kernel_region initialized with volume", region.volume());
  region.to_gpu();
  auto gpu_region = gpu_kernel_region<coordinate_type>(region);
  LOG_DEBUG("gpu_kernel_region initialization");

  timer t;
  t.tic();
  auto kernel_map = in_map.kernel_map(
      out_map, gpu_region, CUDAKernelMapMode::SPEED_OPTIMIZED, thread_dim);
  double k_time = t.toc();

  const auto volume = region.volume();
  LOG_DEBUG("kernel_map done");

  cpu_in_maps in_maps(volume);
  cpu_in_maps out_maps(volume);
  for (index_type i = 0; i < volume; ++i) {
    size_type size = kernel_map.kernels.size(i);
    LOG_DEBUG("kernel index", i, "/", volume, "with size", size);
    in_maps[i].resize(size);
    out_maps[i].resize(size);

    if (size > 0) {
      cudaMemcpy(in_maps[i].data(), //
                 kernel_map.in_maps.begin(i), sizeof(index_type) * size,
                 cudaMemcpyDeviceToHost);
      cudaMemcpy(out_maps[i].data(), //
                 kernel_map.out_maps.begin(i), sizeof(index_type) * size,
                 cudaMemcpyDeviceToHost);
    }
  }
  CUDA_CHECK(cudaStreamSynchronize(0));

  return std::make_tuple(std::make_pair(in_maps, out_maps), out_map.size(),
                         k_time);
}

} // namespace minkowski

PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
  m.def("region_iterator_test", &minkowski::region_iterator_test,
        "Minkowski Engine region iterator test");

  m.def("kernel_map_test", &minkowski::kernel_map_test,
        "Minkowski Engine kernel map test");
}