File: Unroll.h

package info (click to toggle)
pytorch 2.6.0%2Bdfsg-8
  • links: PTS, VCS
  • area: main
  • in suites: forky, sid
  • size: 161,672 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 (30 lines) | stat: -rw-r--r-- 843 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
#pragma once
#include <c10/macros/Macros.h>
#include <type_traits>

// Utility to guarantee complete unrolling of a loop where the bounds are known
// at compile time. Various pragmas achieve similar effects, but are not as
// portable across compilers.

// Example: c10::ForcedUnroll<4>{}(f); is equivalent to f(0); f(1); f(2); f(3);

namespace c10 {

template <int n>
struct ForcedUnroll {
  template <typename Func, typename... Args>
  C10_ALWAYS_INLINE void operator()(const Func& f, Args... args) const {
    ForcedUnroll<n - 1>{}(f, args...);
    f(std::integral_constant<int, n - 1>{}, args...);
  }
};

template <>
struct ForcedUnroll<1> {
  template <typename Func, typename... Args>
  C10_ALWAYS_INLINE void operator()(const Func& f, Args... args) const {
    f(std::integral_constant<int, 0>{}, args...);
  }
};

} // namespace c10