使用自定义 C++ 类扩展 TorchScript

创建时间： 2020年1月23日 |上次更新时间：2024 年 12 月 2 日 |上次验证： Nov 05， 2024

警告

TorchScript 不再处于积极开发阶段。

本教程是自定义运算符教程的后续教程，并介绍了我们构建的用于将 C++ 类绑定到 TorchScript 的 API 和 Python 同时进行。该 API 与 pybind11 非常相似，大部分概念都会转移 如果您熟悉该系统，请结束。

在 C++ 中实现和绑定类

在本教程中，我们将定义一个简单的 C++ 类，该类维护持久性 state 的值。

// This header is all you need to do the C++ portions of this
// tutorial
#include <torch/script.h>
// This header is what defines the custom class registration
// behavior specifically. script.h already includes this, but
// we include it here so you know it exists in case you want
// to look at the API or implementation.
#include <torch/custom_class.h>

#include <string>
#include <vector>

template <class T>
struct MyStackClass : torch::CustomClassHolder {
  std::vector<T> stack_;
  MyStackClass(std::vector<T> init) : stack_(init.begin(), init.end()) {}

  void push(T x) {
    stack_.push_back(x);
  }
  T pop() {
    auto val = stack_.back();
    stack_.pop_back();
    return val;
  }

  c10::intrusive_ptr<MyStackClass> clone() const {
    return c10::make_intrusive<MyStackClass>(stack_);
  }

  void merge(const c10::intrusive_ptr<MyStackClass>& c) {
    for (auto& elem : c->stack_) {
      push(elem);
    }
  }
};

有几点需要注意：

• torch/custom_class.h是扩展 TorchScript 时需要包含的标头 替换为您的自定义类。

• 请注意，每当我们使用自定义 类，我们通过 的实例来实现。可以看作一个智能指针，但存储了引用计数 直接存储在对象中，而不是单独的元数据块（如 . 内部使用相同的指针类型; 自定义类也必须使用这个指针类型，以便我们可以 一致地管理不同的对象类型。c10::intrusive_ptr<>intrusive_ptrstd::shared_ptrstd::shared_ptrtorch::Tensor

• 要注意的第二件事是用户定义的类必须继承自 。这可确保自定义类具有 存储引用计数。torch::CustomClassHolder

现在让我们看看如何使这个类对 TorchScript 可见，这个过程称为 绑定类：

// Notice a few things:
// - We pass the class to be registered as a template parameter to
//   `torch::class_`. In this instance, we've passed the
//   specialization of the MyStackClass class ``MyStackClass<std::string>``.
//   In general, you cannot register a non-specialized template
//   class. For non-templated classes, you can just pass the
//   class name directly as the template parameter.
// - The arguments passed to the constructor make up the "qualified name"
//   of the class. In this case, the registered class will appear in
//   Python and C++ as `torch.classes.my_classes.MyStackClass`. We call
//   the first argument the "namespace" and the second argument the
//   actual class name.
TORCH_LIBRARY(my_classes, m) {
  m.class_<MyStackClass<std::string>>("MyStackClass")
    // The following line registers the contructor of our MyStackClass
    // class that takes a single `std::vector<std::string>` argument,
    // i.e. it exposes the C++ method `MyStackClass(std::vector<T> init)`.
    // Currently, we do not support registering overloaded
    // constructors, so for now you can only `def()` one instance of
    // `torch::init`.
    .def(torch::init<std::vector<std::string>>())
    // The next line registers a stateless (i.e. no captures) C++ lambda
    // function as a method. Note that a lambda function must take a
    // `c10::intrusive_ptr<YourClass>` (or some const/ref version of that)
    // as the first argument. Other arguments can be whatever you want.
    .def("top", [](const c10::intrusive_ptr<MyStackClass<std::string>>& self) {
      return self->stack_.back();
    })
    // The following four lines expose methods of the MyStackClass<std::string>
    // class as-is. `torch::class_` will automatically examine the
    // argument and return types of the passed-in method pointers and
    // expose these to Python and TorchScript accordingly. Finally, notice
    // that we must take the *address* of the fully-qualified method name,
    // i.e. use the unary `&` operator, due to C++ typing rules.
    .def("push", &MyStackClass<std::string>::push)
    .def("pop", &MyStackClass<std::string>::pop)
    .def("clone", &MyStackClass<std::string>::clone)
    .def("merge", &MyStackClass<std::string>::merge)
  ;
}

使用 CMake 将示例构建为 C++ 项目

现在，我们将使用 CMake 构建系统构建上述 C++ 代码。首先，获取所有 C++ 代码 到目前为止，我们已经介绍了它并将其放在一个名为 . 然后，编写一个简单的文件并将其放在 同一目录。这应该看起来像这样：class.cppCMakeLists.txtCMakeLists.txt

cmake_minimum_required(VERSION 3.1 FATAL_ERROR)
project(custom_class)

find_package(Torch REQUIRED)

# Define our library target
add_library(custom_class SHARED class.cpp)
set(CMAKE_CXX_STANDARD 14)
# Link against LibTorch
target_link_libraries(custom_class "${TORCH_LIBRARIES}")

此外，创建一个目录。您的文件树应如下所示：build

custom_class_project/
  class.cpp
  CMakeLists.txt
  build/

我们假设您已按照 中所述的相同方式设置环境 上一个教程。 继续调用 cmake，然后调用 make 来构建项目：

$ cd build
$ cmake -DCMAKE_PREFIX_PATH="$(python -c 'import torch.utils; print(torch.utils.cmake_prefix_path)')" ..
  -- The C compiler identification is GNU 7.3.1
  -- The CXX compiler identification is GNU 7.3.1
  -- Check for working C compiler: /opt/rh/devtoolset-7/root/usr/bin/cc
  -- Check for working C compiler: /opt/rh/devtoolset-7/root/usr/bin/cc -- works
  -- Detecting C compiler ABI info
  -- Detecting C compiler ABI info - done
  -- Detecting C compile features
  -- Detecting C compile features - done
  -- Check for working CXX compiler: /opt/rh/devtoolset-7/root/usr/bin/c++
  -- Check for working CXX compiler: /opt/rh/devtoolset-7/root/usr/bin/c++ -- works
  -- Detecting CXX compiler ABI info
  -- Detecting CXX compiler ABI info - done
  -- Detecting CXX compile features
  -- Detecting CXX compile features - done
  -- Looking for pthread.h
  -- Looking for pthread.h - found
  -- Looking for pthread_create
  -- Looking for pthread_create - not found
  -- Looking for pthread_create in pthreads
  -- Looking for pthread_create in pthreads - not found
  -- Looking for pthread_create in pthread
  -- Looking for pthread_create in pthread - found
  -- Found Threads: TRUE
  -- Found torch: /torchbind_tutorial/libtorch/lib/libtorch.so
  -- Configuring done
  -- Generating done
  -- Build files have been written to: /torchbind_tutorial/build
$ make -j
  Scanning dependencies of target custom_class
  [ 50%] Building CXX object CMakeFiles/custom_class.dir/class.cpp.o
  [100%] Linking CXX shared library libcustom_class.so
  [100%] Built target custom_class

您会发现现在 （除其他外） 有一个动态库 文件。在 Linux 上，这可能被命名为 。所以文件树应该看起来像这样：libcustom_class.so

custom_class_project/
  class.cpp
  CMakeLists.txt
  build/
    libcustom_class.so

从 Python 和 TorchScript 使用 C++ 类

现在我们已经将类及其注册编译成了一个文件， 我们可以将该 .so 加载到 Python 中并尝试一下。下面是一个脚本 证明：.so

import torch

# `torch.classes.load_library()` allows you to pass the path to your .so file
# to load it in and make the custom C++ classes available to both Python and
# TorchScript
torch.classes.load_library("build/libcustom_class.so")
# You can query the loaded libraries like this:
print(torch.classes.loaded_libraries)
# prints {'/custom_class_project/build/libcustom_class.so'}

# We can find and instantiate our custom C++ class in python by using the
# `torch.classes` namespace:
#
# This instantiation will invoke the MyStackClass(std::vector<T> init)
# constructor we registered earlier
s = torch.classes.my_classes.MyStackClass(["foo", "bar"])

# We can call methods in Python
s.push("pushed")
assert s.pop() == "pushed"

# Test custom operator
s.push("pushed")
torch.ops.my_classes.manipulate_instance(s)  # acting as s.pop()
assert s.top() == "bar" 

# Returning and passing instances of custom classes works as you'd expect
s2 = s.clone()
s.merge(s2)
for expected in ["bar", "foo", "bar", "foo"]:
    assert s.pop() == expected

# We can also use the class in TorchScript
# For now, we need to assign the class's type to a local in order to
# annotate the type on the TorchScript function. This may change
# in the future.
MyStackClass = torch.classes.my_classes.MyStackClass


@torch.jit.script
def do_stacks(s: MyStackClass):  # We can pass a custom class instance
    # We can instantiate the class
    s2 = torch.classes.my_classes.MyStackClass(["hi", "mom"])
    s2.merge(s)  # We can call a method on the class
    # We can also return instances of the class
    # from TorchScript function/methods
    return s2.clone(), s2.top()


stack, top = do_stacks(torch.classes.my_classes.MyStackClass(["wow"]))
assert top == "wow"
for expected in ["wow", "mom", "hi"]:
    assert stack.pop() == expected

使用自定义类保存、加载和运行 TorchScript 代码

我们还可以在 C++ 进程中使用自定义注册的 C++ 类，使用 libtorch 中。例如，让我们定义一个简单的 实例化并调用 MyStackClass 类上的方法：nn.Module

import torch

torch.classes.load_library('build/libcustom_class.so')


class Foo(torch.nn.Module):
    def __init__(self):
        super().__init__()

    def forward(self, s: str) -> str:
        stack = torch.classes.my_classes.MyStackClass(["hi", "mom"])
        return stack.pop() + s


scripted_foo = torch.jit.script(Foo())
print(scripted_foo.graph)

scripted_foo.save('foo.pt')

foo.pt现在包含序列化的 TorchScript 程序。

现在，我们将定义一个新的 CMake 项目来展示如何加载 此模型及其所需的 .so 文件。有关如何执行此操作的完整处理， 请查看 在 C++ 中加载 TorchScript 模型教程 。

与之前类似，让我们创建一个包含以下内容的文件结构：

cpp_inference_example/
  infer.cpp
  CMakeLists.txt
  foo.pt
  build/
  custom_class_project/
    class.cpp
    CMakeLists.txt
    build/

请注意，我们已经复制了序列化文件以及源文件 树。我们将 作为依赖项添加到此 C++ 项目中，以便我们可以 将自定义类构建到二进制文件中。foo.ptcustom_class_projectcustom_class_project

让我们填充以下内容：infer.cpp

#include <torch/script.h>

#include <iostream>
#include <memory>

int main(int argc, const char* argv[]) {
  torch::jit::Module module;
  try {
    // Deserialize the ScriptModule from a file using torch::jit::load().
    module = torch::jit::load("foo.pt");
  }
  catch (const c10::Error& e) {
    std::cerr << "error loading the model\n";
    return -1;
  }

  std::vector<c10::IValue> inputs = {"foobarbaz"};
  auto output = module.forward(inputs).toString();
  std::cout << output->string() << std::endl;
}

同样，让我们定义我们的 CMakeLists.txt 文件：

cmake_minimum_required(VERSION 3.1 FATAL_ERROR)
project(infer)

find_package(Torch REQUIRED)

add_subdirectory(custom_class_project)

# Define our library target
add_executable(infer infer.cpp)
set(CMAKE_CXX_STANDARD 14)
# Link against LibTorch
target_link_libraries(infer "${TORCH_LIBRARIES}")
# This is where we link in our libcustom_class code, making our
# custom class available in our binary.
target_link_libraries(infer -Wl,--no-as-needed custom_class)

您知道练习： ， ， 和 ：cd buildcmakemake

$ cd build
$ cmake -DCMAKE_PREFIX_PATH="$(python -c 'import torch.utils; print(torch.utils.cmake_prefix_path)')" ..
  -- The C compiler identification is GNU 7.3.1
  -- The CXX compiler identification is GNU 7.3.1
  -- Check for working C compiler: /opt/rh/devtoolset-7/root/usr/bin/cc
  -- Check for working C compiler: /opt/rh/devtoolset-7/root/usr/bin/cc -- works
  -- Detecting C compiler ABI info
  -- Detecting C compiler ABI info - done
  -- Detecting C compile features
  -- Detecting C compile features - done
  -- Check for working CXX compiler: /opt/rh/devtoolset-7/root/usr/bin/c++
  -- Check for working CXX compiler: /opt/rh/devtoolset-7/root/usr/bin/c++ -- works
  -- Detecting CXX compiler ABI info
  -- Detecting CXX compiler ABI info - done
  -- Detecting CXX compile features
  -- Detecting CXX compile features - done
  -- Looking for pthread.h
  -- Looking for pthread.h - found
  -- Looking for pthread_create
  -- Looking for pthread_create - not found
  -- Looking for pthread_create in pthreads
  -- Looking for pthread_create in pthreads - not found
  -- Looking for pthread_create in pthread
  -- Looking for pthread_create in pthread - found
  -- Found Threads: TRUE
  -- Found torch: /local/miniconda3/lib/python3.7/site-packages/torch/lib/libtorch.so
  -- Configuring done
  -- Generating done
  -- Build files have been written to: /cpp_inference_example/build
$ make -j
  Scanning dependencies of target custom_class
  [ 25%] Building CXX object custom_class_project/CMakeFiles/custom_class.dir/class.cpp.o
  [ 50%] Linking CXX shared library libcustom_class.so
  [ 50%] Built target custom_class
  Scanning dependencies of target infer
  [ 75%] Building CXX object CMakeFiles/infer.dir/infer.cpp.o
  [100%] Linking CXX executable infer
  [100%] Built target infer

现在我们可以运行我们令人兴奋的 C++ 二进制文件了：

$ ./infer
  momfoobarbaz

不可思议！

将自定义类移入/移出 IValues

您也可能需要将自定义类移入或移出自定义 C++ 类实例：IValue``s, such as when you take or return ``IValue``s from TorchScript methods or you want to instantiate a custom class attribute in C++. For creating an ``IValue

• torch::make_custom_class<T>()提供类似于 c10：：intrusive_ptr<T 的 API> 因为它将接受你提供给它的任何参数集，请调用构造函数 of T 匹配该参数集，并将该实例包装起来并返回它。 但是，它不是仅返回指向自定义类对象的指针，而是返回 包装对象。然后，您可以直接将其传递给 TorchScript 的 TorchScript 中。IValueIValue

• 如果您已经指向您的类，则 可以使用构造函数 .intrusive_ptrIValue(intrusive_ptr<T>)

要转换回自定义类：IValue

• IValue::toCustomClass<T>()将返回指向 自定义类。在内部，这个函数正在检查 它注册为自定义类，并且 does 实际上包含 自定义类。您可以通过以下方式手动检查 是否包含自定义类 叫。intrusive_ptr<T>IValueTIValueIValueisCustomClass()

为自定义 C++ 类定义序列化/反序列化方法

如果您尝试将具有自定义绑定 C++ 类的 另存为 属性，则会收到以下错误：ScriptModule

# export_attr.py
import torch

torch.classes.load_library('build/libcustom_class.so')


class Foo(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.stack = torch.classes.my_classes.MyStackClass(["just", "testing"])

    def forward(self, s: str) -> str:
        return self.stack.pop() + s


scripted_foo = torch.jit.script(Foo())

scripted_foo.save('foo.pt')
loaded = torch.jit.load('foo.pt')

print(loaded.stack.pop())

$ python export_attr.py
RuntimeError: Cannot serialize custom bound C++ class __torch__.torch.classes.my_classes.MyStackClass. Please define serialization methods via def_pickle for this class. (pushIValueImpl at ../torch/csrc/jit/pickler.cpp:128)

这是因为 TorchScript 无法自动找出哪些信息 save 从 C++ 类中保存。您必须手动指定。实现目标的方法 是定义 和 类的方法 上的特殊方法 。__getstate____setstate__def_pickleclass_

注意

和 在 TorchScript 中的语义是 相当于 Python pickle 模块的 API。您可以详细了解我们如何使用这些方法。__getstate____setstate__

下面是一个调用示例，我们可以添加到 的注册中以包含序列化方法：def_pickleMyStackClass

    // class_<>::def_pickle allows you to define the serialization
    // and deserialization methods for your C++ class.
    // Currently, we only support passing stateless lambda functions
    // as arguments to def_pickle
    .def_pickle(
          // __getstate__
          // This function defines what data structure should be produced
          // when we serialize an instance of this class. The function
          // must take a single `self` argument, which is an intrusive_ptr
          // to the instance of the object. The function can return
          // any type that is supported as a return value of the TorchScript
          // custom operator API. In this instance, we've chosen to return
          // a std::vector<std::string> as the salient data to preserve
          // from the class.
          [](const c10::intrusive_ptr<MyStackClass<std::string>>& self)
              -> std::vector<std::string> {
            return self->stack_;
          },
          // __setstate__
          // This function defines how to create a new instance of the C++
          // class when we are deserializing. The function must take a
          // single argument of the same type as the return value of
          // `__getstate__`. The function must return an intrusive_ptr
          // to a new instance of the C++ class, initialized however
          // you would like given the serialized state.
          [](std::vector<std::string> state)
              -> c10::intrusive_ptr<MyStackClass<std::string>> {
            // A convenient way to instantiate an object and get an
            // intrusive_ptr to it is via `make_intrusive`. We use
            // that here to allocate an instance of MyStackClass<std::string>
            // and call the single-argument std::vector<std::string>
            // constructor with the serialized state.
            return c10::make_intrusive<MyStackClass<std::string>>(std::move(state));
          });

注意

我们在 pickle API 中采用了与 pybind11 不同的方法。而 pybind11 作为特殊函数传递给 ， 为此，我们有一个单独的方法。这是因为 name 已被占用，我们不想造成混淆。pybind11::pickle()class_::def()def_pickletorch::jit::pickle

一旦我们以这种方式定义了（反）序列化行为，我们的脚本就可以 现在运行成功：

$ python ../export_attr.py
testing

定义采用或返回绑定 C++ 类的自定义运算符

定义自定义 C++ 类后，还可以使用该类 作为参数或从自定义运算符（即自由函数）返回。假设 您有以下 free 函数：

c10::intrusive_ptr<MyStackClass<std::string>> manipulate_instance(const c10::intrusive_ptr<MyStackClass<std::string>>& instance) {
  instance->pop();
  return instance;
}

您可以在块中运行以下代码注册它：TORCH_LIBRARY

    m.def(
      "manipulate_instance(__torch__.torch.classes.my_classes.MyStackClass x) -> __torch__.torch.classes.my_classes.MyStackClass Y",
      manipulate_instance
    );

有关注册 API 的更多详细信息，请参阅自定义操作教程。

完成此操作后，您可以使用如下示例所示的运算：

class TryCustomOp(torch.nn.Module):
    def __init__(self):
        super(TryCustomOp, self).__init__()
        self.f = torch.classes.my_classes.MyStackClass(["foo", "bar"])

    def forward(self):
        return torch.ops.my_classes.manipulate_instance(self.f)

注意

注册将 C++ 类作为参数的运算符需要 自定义类已注册。您可以通过以下方式强制执行此操作 确保自定义类注册和免费函数定义 位于同一个块中，并且自定义类 首先注册。将来，我们可能会放宽此要求， 以便可以按任何顺序注册这些内容。TORCH_LIBRARY

结论

本教程向您介绍了如何向 TorchScript 公开 C++ 类 （以及扩展的 Python），如何注册它的方法，如何使用它 类，以及如何使用 类并在独立的 C++ 进程中运行该代码。您现在已经准备好了 C++使用与 第三方 C++ 库或实现任何其他需要 在 Python、TorchScript 和 C++ 之间平滑混合。

与往常一样，如果您遇到任何问题或有疑问，可以使用我们的论坛或 GitHub 问题与我们联系。此外，我们的常见问题 （FAQ） 页面可能包含有用的信息。