File: comm.h

package info (click to toggle)
pytorch 1.13.1%2Bdfsg-4
  • links: PTS, VCS
  • area: main
  • in suites: bookworm
  • size: 139,252 kB
  • sloc: cpp: 1,100,274; python: 706,454; ansic: 83,052; asm: 7,618; java: 3,273; sh: 2,841; javascript: 612; makefile: 323; xml: 269; ruby: 185; yacc: 144; objc: 68; lex: 44
file content (47 lines) | stat: -rw-r--r-- 1,257 bytes parent folder | download
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
#pragma once

#include <torch/csrc/Export.h>
#include <torch/csrc/autograd/function.h>
#include <torch/csrc/autograd/variable.h>

#include <ATen/ATen.h>
#include <ATen/cuda/ATenCUDAGeneral.h>
#include <ATen/cuda/CUDAContext.h>

#include <cstddef>
#include <vector>

namespace torch {
namespace autograd {

struct TORCH_CUDA_CU_API Scatter : public Node {
  explicit Scatter(
      std::vector<at::Device> devices,
      const c10::optional<std::vector<int64_t>>& chunk_sizes = c10::nullopt,
      int64_t dim = 0,
      const c10::optional<std::vector<c10::optional<at::cuda::CUDAStream>>>&
          streams = c10::nullopt,
      bool unsqueeze_scalars = false);
  ~Scatter() override;

  variable_list apply(variable_list&& inputs) override;

  std::vector<at::Device> devices_;
  c10::optional<std::vector<int64_t>> chunk_sizes_;
  int64_t dim_;
  c10::optional<std::vector<c10::optional<at::cuda::CUDAStream>>> streams_;
  bool unsqueeze_scalars_;
};

struct TORCH_CUDA_CU_API Gather : public Node {
  explicit Gather(const at::Device& destination_device, int64_t dim = 0);
  ~Gather() override;

  variable_list apply(variable_list&& inputs) override;

  at::Device destination_device_;
  int64_t dim_;
};

} // namespace autograd
} // namespace torch