mirror of
https://github.com/PaddlePaddle/FastDeploy.git
synced 2025-10-06 09:07:10 +08:00
6343b0db47
* Add notes for tensors * Optimize some apis * move some warnings * Support build with Paddle2ONNX * Add protobuf support * Fix compile on mac * add clearn package script * Add paddle2onnx code * remove submodule * Add onnx ocde * remove softlink * add onnx code * fix error * Add cmake file * fix patchelf * update paddle2onnx * Delete .gitmodules --------- Co-authored-by: PaddleCI <paddle_ci@example.com> Co-authored-by: pangyoki <pangyoki@126.com> Co-authored-by: jiangjiajun <jiangjiajun@baidu.lcom>
98 lines
3.1 KiB
C++
98 lines
3.1 KiB
C++
/*
|
|
* SPDX-License-Identifier: Apache-2.0
|
|
*/
|
|
|
|
// ATTENTION: The code in this file is highly EXPERIMENTAL.
|
|
// Adventurous users should note that the APIs will probably change.
|
|
|
|
#pragma once
|
|
|
|
#include "onnx/common/ir.h"
|
|
#include "onnx/common/ir_pb_converter.h"
|
|
#include "onnx/common/stl_backports.h"
|
|
#include "onnx/proto_utils.h"
|
|
|
|
#include "onnxoptimizer/pass_manager.h"
|
|
#include "onnxoptimizer/pass_registry.h"
|
|
|
|
#include "vector"
|
|
|
|
namespace ONNX_NAMESPACE {
|
|
namespace optimization {
|
|
|
|
struct Optimizer {
|
|
static GlobalPassRegistry passes;
|
|
|
|
public:
|
|
Optimizer(const std::vector<std::string> &names, const bool fixed_point);
|
|
~Optimizer();
|
|
|
|
ModelProto optimize(const ModelProto &mp_in) {
|
|
ModelProto mp_compatible = AddInitializerToInput(mp_in);
|
|
bool has_initializer_not_in_input =
|
|
(mp_in.graph().input_size() != mp_compatible.graph().input_size());
|
|
std::shared_ptr<Graph> g(ImportModelProto(mp_compatible));
|
|
|
|
if (g.get() == nullptr) {
|
|
std::cerr << "Warning: onnx optimizer is unable to parse input model. "
|
|
<< "(The IR version of the ONNX model may be too old.)"
|
|
<< std::endl;
|
|
// If we can't parse the file, just return the input.
|
|
return mp_in;
|
|
}
|
|
|
|
ModelProto mp_out = PrepareOutput(mp_in);
|
|
this->pass_manager->run(*g);
|
|
ExportModelProto(&mp_out, g);
|
|
// `has_initializer_not_in_input` means the original model prefer
|
|
// initializer to be not in input, so the new initializer introduced by
|
|
// both `AddInitializerToInput` and optimization passes will be remove from
|
|
// input
|
|
if (has_initializer_not_in_input) {
|
|
mp_out.mutable_graph()->mutable_input()->DeleteSubrange(
|
|
mp_in.graph().input_size(),
|
|
mp_out.graph().input_size() - mp_in.graph().input_size());
|
|
}
|
|
return mp_out;
|
|
}
|
|
|
|
private:
|
|
std::shared_ptr<PassManager> pass_manager;
|
|
|
|
ModelProto AddInitializerToInput(const ModelProto &original_model) {
|
|
ModelProto model = original_model;
|
|
std::vector<std::string> input_names;
|
|
for (const auto &x : model.graph().input()) {
|
|
input_names.push_back(x.name());
|
|
}
|
|
for (const auto &x : model.graph().initializer()) {
|
|
if (std::find(input_names.begin(), input_names.end(), x.name()) ==
|
|
input_names.end()) {
|
|
auto *value_info = model.mutable_graph()->add_input();
|
|
value_info->set_name(x.name());
|
|
TypeProto *type = value_info->mutable_type();
|
|
auto *tensor = type->mutable_tensor_type();
|
|
tensor->set_elem_type(x.data_type());
|
|
auto *shape = tensor->mutable_shape();
|
|
for (const auto &dim : x.dims()) {
|
|
TensorShapeProto::Dimension *new_dim = shape->add_dim();
|
|
new_dim->set_dim_value(dim);
|
|
}
|
|
}
|
|
}
|
|
return model;
|
|
}
|
|
};
|
|
|
|
const std::vector<std::string> GetAvailablePasses();
|
|
|
|
const std::vector<std::string> GetFuseAndEliminationPass();
|
|
|
|
ModelProto Optimize(const ModelProto &mp_in,
|
|
const std::vector<std::string> &names);
|
|
|
|
ModelProto OptimizeFixed(const ModelProto &mp_in,
|
|
const std::vector<std::string> &names);
|
|
} // namespace optimization
|
|
} // namespace ONNX_NAMESPACE
|