File: frozen_graph_optimizations.cpp

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 (31 lines) | stat: -rw-r--r-- 1,010 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
#include <c10/util/irange.h>
#include <torch/csrc/jit/ir/alias_analysis.h>
#include <torch/csrc/jit/ir/ir_views.h>
#include <torch/csrc/jit/passes/frozen_concat_linear.h>
#include <torch/csrc/jit/passes/frozen_conv_folding.h>
#include <torch/csrc/jit/passes/frozen_graph_optimizations.h>
#include <torch/csrc/jit/passes/frozen_linear_folding.h>
#include <torch/csrc/jit/passes/remove_dropout.h>
#include <torch/csrc/jit/runtime/graph_executor.h>

namespace torch::jit {

void OptimizeFrozenGraph(
    std::shared_ptr<Graph>& graph,
    bool optimize_numerics) {
  removeDropout(graph);
  FrozenConcatLinear(graph);
  // run a couple times to capture Conv -> Mul -> Add etc
  if (optimize_numerics) {
    bool changed = false;
    do {
      changed = false;
      changed |= FoldFrozenConvBatchnorm(graph);
      changed |= FoldFrozenConvAddOrSub(graph);
      changed |= FoldFrozenConvMulOrDiv(graph);
      changed |= FoldFrozenLinearBatchnorm(graph);
    } while (changed);
  }
}

} // namespace torch::jit