File: inductor_ops.h

package info (click to toggle)
pytorch-cuda 2.6.0%2Bdfsg-7
  • links: PTS, VCS
  • area: contrib
  • in suites: forky, sid, trixie
  • size: 161,620 kB
  • sloc: python: 1,278,832; cpp: 900,322; ansic: 82,710; asm: 7,754; java: 3,363; sh: 2,811; javascript: 2,443; makefile: 597; ruby: 195; xml: 84; objc: 68
file content (39 lines) | stat: -rw-r--r-- 1,095 bytes parent folder | download | duplicates (3)
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
#pragma once

#include <ATen/Tensor.h>

namespace torch::inductor {

TORCH_API at::Tensor _mm_plus_mm_out(
    at::Tensor& out,
    const at::Tensor& a,
    const at::Tensor& b,
    const at::Tensor& c,
    const at::Tensor& d);

// After adding _mm_plus_mm_out, this should not be exposed and called by model
// code. Keeping it around for backward compatibility. Will be deprecated later.
TORCH_API at::Tensor _mm_plus_mm(
    const at::Tensor& a,
    const at::Tensor& b,
    const at::Tensor& c,
    const at::Tensor& d,
    at::Tensor& out);

TORCH_API at::Tensor _alloc_from_pool(
    const at::Tensor& self,
    int64_t offset_bytes,
    at::ScalarType dtype,
    at::IntArrayRef size,
    at::IntArrayRef stride);

// Similar to as_strided with the following differences
// - offset is added to the existing offset (rather than replacing it)
// - view tracking is disabled similar to unsafe_view
TORCH_API at::Tensor _reinterpret_tensor(
    const at::Tensor& self,
    at::IntArrayRef size,
    at::IntArrayRef stride,
    int64_t offset_increment = 0);

} // namespace torch::inductor