File: clear_undefinedness.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 (24 lines) | stat: -rw-r--r-- 875 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
#pragma once

#include <ATen/ATen.h>
#include <ATen/core/ivalue.h>
#include <ATen/core/jit_type.h>
#include <torch/csrc/Export.h>
#include <torch/csrc/jit/ir/ir.h>

namespace torch {
namespace jit {

// Undefinedness makes argument matching fail for regular tensor operations
// if 1+ arguments are undefined or possibly undefined tensors.
// Technically, undefined tensors are **not** tensors as the regular tensor
// operations do not know how to handle them.
// However, in practice, there are guards and conversion operators that
// **always** gate regular operations if undefined tensors may be present
// Eventually, we would love to move to the world where we use optionals
// in lieu of undefined tensors.
// When this happens, this pass will be removed
TORCH_API void ClearUndefinedness(const std::shared_ptr<Graph>& graph);

} // namespace jit
} // namespace torch