File: compute_at.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 (45 lines) | stat: -rw-r--r-- 1,114 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
#pragma once

#include <torch/csrc/jit/codegen/cuda/inline_propagator.h>
#include <torch/csrc/jit/codegen/cuda/root_domain_map.h>
#include <torch/csrc/jit/codegen/cuda/transform_replay.h>

#include <c10/macros/Export.h>
#include <c10/util/Exception.h>

#include <deque>
#include <unordered_map>
#include <unordered_set>
#include <vector>

namespace torch {
namespace jit {
namespace fuser {
namespace cuda {

class TensorDomain;
class TensorView;

struct ComputeAt {
 public:
  // Runs the compute at pass making producer look like consumer, computing
  // producer relative to consumer
  static void runAt(
      TensorView* producer,
      TensorView* consumer,
      int64_t consumer_position,
      ComputeAtMode mode = ComputeAtMode::Standard);

  // Runs the compute with pass making consumer look like producer, computing
  // producer relative to consumer
  static void runWith(
      TensorView* producer,
      TensorView* consumer,
      int64_t producer_position,
      ComputeAtMode mode = ComputeAtMode::Standard);
};

} // namespace cuda
} // namespace fuser
} // namespace jit
} // namespace torch