-
Notifications
You must be signed in to change notification settings - Fork 348
/
interpolation_gpu.cu
151 lines (128 loc) · 5.91 KB
/
interpolation_gpu.cu
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
/*
* Copyright (c) 2020 NVIDIA Corporation.
*
* 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.hpp"
#include "coordinate_map_cpu.hpp"
#include "coordinate_map_key.hpp"
#include "coordinate_map_manager.hpp"
#include "errors.hpp"
#include "types.hpp"
#include "utils.hpp"
#include "interpolation_cpu.cpp"
#include "spmm.cuh"
#include <pybind11/pybind11.h>
#include <torch/extension.h>
namespace minkowski {
template <typename coordinate_type,
template <typename C> class TemplatedAllocator>
std::vector<at::Tensor> InterpolationForwardGPU(
at::Tensor const &in_feat, //
at::Tensor const &tfield, //
CoordinateMapKey *p_in_map_key, //
gpu_manager_type<coordinate_type, TemplatedAllocator> *p_map_manager) {
ASSERT(in_feat.is_contiguous(), "in_feat must be contiguous");
ASSERT(in_feat.is_cuda(), "in_feat must be GPU");
ASSERT(in_feat.dim() == 2, "in_feat.dim():", in_feat.dim());
ASSERT(tfield.is_contiguous(), "tfield must be contiguous");
ASSERT(tfield.is_cuda(), "tfield must be GPU");
ASSERT(tfield.dim() == 2, "tfield.dim():", tfield.dim());
ASSERT(tfield.dtype() == in_feat.dtype(),
"tfield and in_feat must have the same dtype");
ASSERT(at::cuda::check_device({in_feat, tfield}),
"in_feat and tfield must be on the same device");
coordinate_map_key_type in_key = p_in_map_key->get_key();
ASSERT(p_map_manager->exists(in_key), ERROR_MAP_NOT_FOUND);
ASSERT(in_feat.size(0) == p_map_manager->size(in_key), "Invalid in_feat size",
in_feat.size(0), "!=", p_map_manager->size(in_key));
auto map_weight =
p_map_manager->interpolation_map_weight(tfield, p_in_map_key);
auto const &in_maps = map_weight[0];
auto const &out_maps = map_weight[1];
auto const &weights = map_weight[2];
LOG_DEBUG("InterpolationForwardKernelGPU");
auto out_feat = coo_spmm<int>(out_maps, in_maps, weights, tfield.size(0),
in_feat.size(0), in_feat, 1, false);
LOG_DEBUG("out_feat shape: (", out_feat.size(0), ",", out_feat.size(1), ")");
// to out_feats
map_weight.insert(map_weight.begin(), out_feat);
return map_weight;
}
template <typename coordinate_type,
template <typename C> class TemplatedAllocator>
at::Tensor InterpolationBackwardGPU(
at::Tensor &grad_out_feat, //
at::Tensor const &in_maps, //
at::Tensor const &out_maps, //
at::Tensor const &weights, //
CoordinateMapKey *p_in_map_key, //
gpu_manager_type<coordinate_type, TemplatedAllocator> *p_map_manager) {
if (!grad_out_feat.is_contiguous())
grad_out_feat = grad_out_feat.contiguous();
ASSERT(grad_out_feat.is_cuda(), "grad_out_feat must be CUDA");
ASSERT(grad_out_feat.dim() == 2, "grad_out_feat.dim():", grad_out_feat.dim());
coordinate_map_key_type in_key = p_in_map_key->get_key();
ASSERT(p_map_manager->exists(in_key), ERROR_MAP_NOT_FOUND);
uint32_t const in_nrows = p_map_manager->size(in_key);
LOG_DEBUG("InterpolationBackwardKernelGPU");
return coo_spmm<int>(in_maps, out_maps, weights, in_nrows,
grad_out_feat.size(0), grad_out_feat, 1, false);
}
// Forward
template std::vector<at::Tensor>
InterpolationForwardGPU<default_types::dcoordinate_type,
detail::default_allocator>(
at::Tensor const &in_feat, //
at::Tensor const &tfield, //
CoordinateMapKey *p_in_map_key, //
gpu_manager_type<default_types::dcoordinate_type, detail::default_allocator>
*p_map_manager);
template std::vector<at::Tensor>
InterpolationForwardGPU<default_types::dcoordinate_type,
detail::c10_allocator>(
at::Tensor const &in_feat, //
at::Tensor const &tfield, //
CoordinateMapKey *p_in_map_key, //
gpu_manager_type<default_types::dcoordinate_type, detail::c10_allocator>
*p_map_manager);
// Backward
template at::Tensor InterpolationBackwardGPU<default_types::dcoordinate_type,
detail::default_allocator>(
at::Tensor &grad_out_feat, //
at::Tensor const &in_map, //
at::Tensor const &out_map, //
at::Tensor const &weight, //
CoordinateMapKey *p_in_map_key, //
gpu_manager_type<default_types::dcoordinate_type, detail::default_allocator>
*p_map_manager);
template at::Tensor InterpolationBackwardGPU<default_types::dcoordinate_type,
detail::c10_allocator>(
at::Tensor &grad_out_feat, //
at::Tensor const &in_map, //
at::Tensor const &out_map, //
at::Tensor const &weight, //
CoordinateMapKey *p_in_map_key, //
gpu_manager_type<default_types::dcoordinate_type, detail::c10_allocator>
*p_map_manager);
} // end namespace minkowski