File: frozen_ops_to_mkldnn.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 (13 lines) | stat: -rw-r--r-- 389 bytes parent folder | download | duplicates (3) 
1
2
3
4
5
6
7
8
9
10
11
12
13
 
#pragma once

#include <torch/csrc/jit/ir/ir.h>

namespace torch::jit {

// Converts operators & their parameters to mkldnn if it is profitable
// Currently encompassing Conv2d and Conv3d, and Linear
// Op must be in float32 and mkldnn must be built
// This pass only works on frozen graph
TORCH_API void ConvertFrozenOpsToMKLDNN(std::shared_ptr<Graph>& graph);

} // namespace torch::jit