xref: /aosp_15_r20/external/pytorch/binaries/dump_operator_names.cc (revision da0073e96a02ea20f0ac840b70461e3646d07c45)
1 /**
2  * Copyright (c) 2016-present, Facebook, Inc.
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  *     http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16 
17 #include <torch/csrc/jit/api/module.h>
18 #include <torch/csrc/jit/mobile/module.h>
19 #include <torch/csrc/jit/serialization/import.h>
20 #include <torch/csrc/jit/runtime/instruction.h>
21 #include <c10/util/Flags.h>
22 
23 #include <fstream>
24 
25 namespace torch {
26 namespace jit {
dump_opnames(const Module & m,std::unordered_set<std::string> & opnames)27 void dump_opnames(const Module& m, std::unordered_set<std::string>& opnames) {
28   auto methods = m.get_methods();
29   for (const auto& method : methods) {
30     const auto& func = method.function();
31     std::cout << "function name: " << func.name() << std::endl;
32     auto graph = toGraphFunction(func).graph()->copy();
33     torch::jit::Code code(graph, "");
34     for (size_t i = 0; i < code.instructions().size(); ++i) {
35       auto ins = code.instructions()[i];
36       auto node = code.instructions_source()[i];
37       if (ins.op == OpCode::OP) {
38         auto opname = node->schema().operator_name();
39         std::string namestr = opname.name;
40         if (!opname.overload_name.empty())
41           namestr += "." + opname.overload_name;
42         std::cout << "    " << namestr << std::endl;
43         opnames.emplace(namestr);
44       }
45     }
46   }
47   for (const auto& sub_m : m.children()) {
48     std::cout << "sub module name: " << sub_m.type()->name()->qualifiedName() << std::endl;
49     dump_opnames(sub_m, opnames);
50   }
51 }
52 }
53 }
54 
55 C10_DEFINE_string(model, "", "The given torch script model.");
56 C10_DEFINE_string(output, "", "The output yaml file of operator list.");
57 
main(int argc,char ** argv)58 int main(int argc, char** argv) {
59   c10::SetUsageMessage(
60     "Dump operators in a script module and its sub modules.\n"
61     "Example usage:\n"
62     "./dump_operator_names"
63     " --model=<model_file>"
64     " --output=<output.yaml>");
65 
66   if (!c10::ParseCommandLineFlags(&argc, &argv)) {
67     std::cerr << "Failed to parse command line flags!" << std::endl;
68     return 1;
69   }
70 
71   CAFFE_ENFORCE_GE(FLAGS_model.size(), 0, "Model file must be specified.");
72   CAFFE_ENFORCE_GE(FLAGS_output.size(), 0, "Output yaml file must be specified.");
73 
74   auto m = torch::jit::load(FLAGS_model);
75   std::unordered_set<std::string> opnames;
76   torch::jit::dump_opnames(m, opnames);
77   std::ofstream ofile(FLAGS_output);
78   std::cout << "-- Final List --" << std::endl;
79   for (const auto& name : opnames) {
80     std::cout << name << std::endl;
81     ofile << "- " << name << std::endl;
82   }
83   ofile.close();
84 }
85