MCPcopy Create free account
hub / github.com/DeepRec-AI/DeepRec / Options

Class Options

tensorflow/core/common_runtime/graph_optimizer.h:31–57  ·  view source on GitHub ↗

Source from the content-addressed store, hash-verified

29 using NodePredicate = std::function<bool(const Node*)>;
30
31 struct Options {
32 // If not null it maps from nodes in graph to partially-known
33 // shapes of their outputs, and may be used, e.g., in the constant folding
34 // pass. The use of shape_map implies that the mapping from node name to the
35 // vector of partial shapes of its outputs is stable, i.e., no optimization
36 // pass may replace a node with a different node of the same name that has a
37 // different number of outputs, or outputs with different known shapes.
38 // TODO(b/65453533) introduce a unique way to name nodes in a graph.
39 std::unordered_map<string, std::vector<PartialTensorShape>>* shape_map =
40 nullptr;
41
42 // If not null then only nodes for which cse_consider_fn returns true will
43 // be considered for CSE.
44 NodePredicate cse_consider_fn = nullptr;
45
46 // If not null then only nodes for which cf_consider_fn returns true will be
47 // considered for CF.
48 NodePredicate cf_consider_fn = nullptr;
49
50 // If true, multi-device functions will be inlined if
51 // opts_.do_function_inlining() is true.
52 bool inline_multi_device_functions = false;
53
54 // If true, functions in implementation selection group will be inlined if
55 // opts_.do_function_inlining() is true.
56 bool inline_impl_selection_group_functions = false;
57 };
58
59 explicit GraphOptimizer(const OptimizerOptions& opts);
60 ~GraphOptimizer();

Callers 1

OptimizeGraphFunction · 0.70

Calls

no outgoing calls

Tested by

no test coverage detected