feat(mge): add opdef serialization and apply_module_trace

GitOrigin-RevId: 5b45bded1d

imperative/python/megengine/experimental/traced_module/__init__.py

@@ -0,0 +1,7 @@
+# MegEngine is Licensed under the Apache License, Version 2.0 (the "License")
+#
+# Copyright (c) 2014-2021 Megvii Inc. All rights reserved.
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.

imperative/python/megengine/experimental/traced_module/serialization.py

@@ -0,0 +1,34 @@
+# MegEngine is Licensed under the Apache License, Version 2.0 (the "License")
+#
+# Copyright (c) 2014-2021 Megvii Inc. All rights reserved.
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+from typing import Dict
+
+from ...core._imperative_rt import OpDef
+from ...core.ops import builtin
+from ...version import __version__
+
+OPDEF_PARAM_LOADER = {}
+
+
+def get_opdef_state(obj: OpDef) -> Dict:
+    state = obj.__getstate__()
+    state["type"] = type(obj)
+    state["version"] = __version__
+    return state
+
+
+def load_opdef_from_state(state: Dict) -> OpDef:
+    assert "type" in state and issubclass(state["type"], OpDef)
+    assert "version" in state
+    opdef_type = state.pop("type")
+    if opdef_type in OPDEF_PARAM_LOADER:
+        loader = OPDEF_PARAM_LOADER[opdef_type]
+        state = loader(state)
+    state.pop("version")
+    opdef_obj = opdef_type()
+    opdef_obj.__setstate__(state)
+    return opdef_obj

imperative/python/src/module_trace.cpp

@@ -0,0 +1,41 @@
+/**
+ * \file imperative/python/src/module_trace.cpp
+ * MegEngine is Licensed under the Apache License, Version 2.0 (the "License")
+ *
+ * Copyright (c) 2014-2021 Megvii Inc. All rights reserved.
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or
+ * implied.
+ */
+
+#include "./module_trace.h"
+#include "./helper.h" // include op pybind11 caster
+
+namespace py = pybind11;
+
+namespace mgb::imperative::python {
+
+apply_result_t apply_module_trace(ApplyContext& ctx) {
+    apply_result_t outputs;
+
+    auto args = py::tuple(ctx.nargs + 1);
+    args[0] = py::cast(ctx.op);
+    for (size_t i = 0; i < ctx.nargs; i++) {
+        args[i + 1] = TensorWrapper::make(ctx.args[i]->shared_from_this());
+    }
+    auto pyout = PyObject_Call(cpp_apply_module_trace, args.ptr(), nullptr);
+    if (!pyout) throw py::error_already_set();
+    auto ret = py::reinterpret_steal<py::object>(pyout);
+
+    // assumption: python function always returns PyList
+    auto tup = py::reinterpret_borrow<py::list>(ret);
+    for (auto i = 0; i < tup.size(); i++) {
+        auto tw = TensorWrapper::try_cast(tup[i].ptr());
+        outputs.emplace_back(tw->m_tensor);
+    }
+    return outputs;
+}
+
+} // namespace mgb::imperative::python

imperative/python/src/module_trace.h

@@ -0,0 +1,20 @@
+/**
+ * \file imperative/python/src/module_trace.h
+ * MegEngine is Licensed under the Apache License, Version 2.0 (the "License")
+ *
+ * Copyright (c) 2014-2021 Megvii Inc. All rights reserved.
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ */
+
+#pragma once
+
+#include "./tensor.h"
+
+namespace mgb::imperative::python {
+
+apply_result_t apply_module_trace(ApplyContext& ctx);
+
+} // namespace mgb::imperative::python

imperative/python/src/ops.cpp

@@ -88,6 +88,19 @@ PyObject* py_new_generic(PyTypeObject* type, PyObject*, PyObject*) {
     return obj;
 }
 
+template<typename T, typename SNIFAE=void>
+struct serialization {
+    static T load(py::object obj) {
+        return py::cast<T>(obj);
+    }
+    template<typename U,
+        typename = std::enable_if_t<std::is_same_v<T, std::decay_t<U>>>>
+    static py::object dump(U&& t) {
+        return py::cast(std::forward<U>(t));
+    }
+};
+
+
 template<typename T>
 void py_dealloc_generic(PyObject* obj) {
     reinterpret_cast<T*>(obj)->op.reset();
@@ -127,6 +140,13 @@ struct PyOpDef {
     static PyGetSetDef py_getsetters[];
     static Py_hash_t tp_hash(PyObject *obj);
     static PyObject* tp_richcompare(PyObject *self, PyObject *other, int op);
+    static PyObject* py_repr(PyObject* self) {
+        return py::cast(
+                       reinterpret_cast<PyOpDef*>(self)->op->make_name())
+                .release()
+                .ptr();
+    }
+
 };
 PyTypeObject PyOpType(OpDef);
 std::unordered_map<mgb::Typeinfo*, PyTypeObject*> PyOp(OpDef)::ctype2pytype;
@@ -191,6 +211,13 @@ struct EnumWrapper {
             std::string(name) + "." + reinterpret_cast<EnumWrapper*>(self)->to_string())
                 .release().ptr();
     }
+
+    static PyObject* py_dump(PyObject* self) {
+        return py::cast(reinterpret_cast<EnumWrapper*>(self)->to_string())
+                .release()
+                .ptr();
+    }
+
     static PyObject* tp_richcompare(PyObject *self, PyObject *other, int op) {
         if (op == Py_EQ || op == Py_NE) {
             T lhs, rhs;
@@ -279,6 +306,19 @@ struct BitCombinedEnumWrapper {
                 reinterpret_cast<BitCombinedEnumWrapper*>(self)->to_string())
                         .release().ptr();
     }
+
+    static PyObject* py_dump(PyObject* self) {
+        std::vector<std::string> result;
+        auto value = reinterpret_cast<BitCombinedEnumWrapper*>(self)->value;
+        uint32_t value_int = static_cast<uint32_t>(value);
+        for (uint32_t i = 0; i < 32; i++) {
+            if (value_int >> i & 1) {
+                result.push_back(members[i]);
+            }
+        }
+        return py::tuple(py::cast(result)).release().ptr();
+    }
+
     static PyObject* py_or(PyObject* self, PyObject* other) {
         if(!(self->ob_type == other->ob_type)){
             return PyErr_Format(
@@ -326,6 +366,24 @@ struct BitCombinedEnumWrapper {
                 return false;
             }
         }
+        if (py::isinstance<py::tuple>(src)) {
+            auto params = py::cast<std::vector<std::string>>(src); 
+            bool first = true;
+            for (auto s : params){
+                auto&& iter = mem2value.find(normalize_enum(s));
+                if (iter != mem2value.end()) {
+                    if (first) {
+                        value = iter->second;
+                        first = false;
+                    } else {
+                        value |= iter->second;
+                    }
+                } else {
+                    return false;
+                }
+            }
+            return true;
+        }
         if (py::isinstance<py::int_>(obj)) {
             auto v = py::cast<std::underlying_type_t<T>>(src);
             if(v > EnumTrait<T>::max) {
@@ -351,6 +409,25 @@ struct BitCombinedEnumWrapper {
     }
 };
 
+template<typename T>
+struct serialization<T,
+        std::enable_if_t<std::is_enum_v<std::decay_t<T>>>> {
+    static T load(py::object obj) {
+        auto caster = pybind11::detail::type_caster<T>();
+        if (caster.load(obj, true)) {
+            return caster;
+        } else {
+                PyErr_SetString(PyExc_RuntimeError,
+                 "load faild \n");
+                return caster;
+        }
+    }
+    static py::object dump(T t) {
+        return py::cast(t).attr("dump")();
+    }
+};
+
+
 void _init_py_op_def(py::module m) {
     using py_op = PyOp(OpDef);
     auto& py_type = PyOpType(OpDef);
@@ -363,6 +440,7 @@ void _init_py_op_def(py::module m) {
     py_type.tp_hash = PyOp(OpDef)::tp_hash;
     py_type.tp_richcompare = PyOp(OpDef)::tp_richcompare;
     py_type.tp_getset = py_op::py_getsetters;
+    py_type.tp_repr = py_op::py_repr;
     mgb_assert(PyType_Ready(&py_type) >= 0);
     m.add_object("OpDef", reinterpret_cast<PyObject*>(&py_type));
 }

imperative/python/src/pyext17.h

@@ -451,18 +451,11 @@ public:
     template<typename... Args>
     static PyObject* cnew(Args&&... args) {
         auto* pytype = type().operator->();
-        auto* self = pytype->tp_alloc(pytype, 0);
-        auto* inst = reinterpret_cast<wrap_t*>(self)->inst();
-        if constexpr (has_vectorcall && tp_vectorcall::valid) {
-            reinterpret_cast<wrap_t*>(self)->vectorcall_slot = &tp_vectorcall::template impl<>;
-        }
-        new(inst) T(std::forward<Args>(args)...);
-        return self;
+        return cnew_with_type(pytype, std::forward<Args>(args)...);
     }
 
     template<typename... Args>
     static PyObject* cnew_with_type(PyTypeObject* pytype, Args&&... args) {
-        
         auto* self = pytype->tp_alloc(pytype, 0);
         auto* inst = reinterpret_cast<wrap_t*>(self)->inst();
         if constexpr (has_vectorcall && tp_vectorcall::valid) {

imperative/python/src/tensor.cpp

@@ -20,6 +20,7 @@
 #include "./tensor.h"
 #include "./grad.h"
 #include "./trace.h"
+#include "./module_trace.h"
 #include "./common.h"
 #include "./numpy_dtypes.h"
 #include "./graph_rt.h"
@@ -41,6 +42,7 @@ interpreter::Interpreter::Channel* interpreter_for_py;
 
 PyObject *cpp_apply_with_tracing, *cpp_apply_const_with_tracing;
 PyObject *cpp_apply_backward_varnode;
+PyObject *cpp_apply_module_trace;
 
 std::shared_ptr<Tensor> make_const(imperative::TensorPtr value) {
     if (!(ApplyContext::global_enable & Tensor::Flags::TRACE)) {
@@ -70,6 +72,7 @@ std::shared_ptr<Tensor> make_const(imperative::TensorPtr value) {
 REGISTE_APPLY_FUNC(cpp_apply_with_tracing)
 REGISTE_APPLY_FUNC(cpp_apply_const_with_tracing)
 REGISTE_APPLY_FUNC(cpp_apply_backward_varnode)
+REGISTE_APPLY_FUNC(cpp_apply_module_trace)
 
 #undef REGISTE_APPLY_FUNC
 
@@ -79,6 +82,14 @@ Tensor::flags_t ApplyContext::global_enable = 0;
 void set_tracing() { ApplyContext::global_enable |= Tensor::Flags::TRACE; }
 void unset_tracing() { ApplyContext::global_enable &= ~Tensor::Flags::TRACE; }
 
+void set_module_tracing() { ApplyContext::global_enable |= Tensor::Flags::MODULE_TRACE; }
+void unset_module_tracing() { ApplyContext::global_enable &= ~Tensor::Flags::MODULE_TRACE; }
+bool is_tracing_module() {
+    return ApplyContext::global_enable & Tensor::Flags::MODULE_TRACE;
+}
+
+
+
 bool skip_tracing = false;
 
 apply_result_t apply(ApplyContext& ctx) {
@@ -117,6 +128,11 @@ apply_result_t apply(ApplyContext& ctx) {
         return ret;
     }
 
+    if (flags & Tensor::Flags::MODULE_TRACE) {
+        return apply_module_trace(ctx);
+    }
+
+
     if (flags & Tensor::Flags::TRACE) {
         return apply_trace(ctx);
     } else {
@@ -310,6 +326,21 @@ REGISTE_TENSORWRAPPER_FUNC(bool, recording)
 
 #undef REGISTE_TENSORWRAPPER_FUNC
 
+PyObject* TensorWrapper::module_trace_info() {
+    if (!m_tensor->m_module_trace_info.ptr()) {
+        PyErr_SetString(PyExc_AttributeError,
+                        "Has no attribute named \'_NodeMixin__node\', please "
+                        "set it first");
+        return nullptr;
+    }
+    return m_tensor->m_module_trace_info.inc_ref().ptr();
+}
+
+void TensorWrapper::set_module_trace_info(PyObject* obj) {
+    m_tensor->m_module_trace_info = py::reinterpret_borrow<py::object>(obj);
+}
+
+
 
 #define REGISTE_TENSORWRAPPER_PYOBJECT_FUNC(member)                                 \
         PyObject* TensorWrapper::member() {                                         \
@@ -495,7 +526,9 @@ void TensorWrapper::reset(PyObject* tensor) {
     }
     std::string user_custom_name = m_tensor->user_custom_name;
     std::string automatic_name = m_tensor->automatic_name;
+    auto module_trace_info = m_tensor->m_module_trace_info;
     m_tensor = t->m_tensor;
+    m_tensor->m_module_trace_info = module_trace_info;
     m_tensor->user_custom_name = user_custom_name;
     m_tensor->automatic_name = automatic_name;
 }
@@ -856,6 +889,7 @@ void init_tensor(py::module m) {
         .def_getset<&TensorWrapper::trace_mixin_info, &TensorWrapper::set_trace_mixin_info>("_trace_mixin_info")
         .def_getset<&TensorWrapper::user_custom_name, &TensorWrapper::set_user_custom_name>("c_name")
         .def_getset<&TensorWrapper::automatic_name, &TensorWrapper::set_automatic_name>("_name")
+        .def_getset<&TensorWrapper::module_trace_info, &TensorWrapper::set_module_trace_info>("_NodeMixin__node")
         .finalize();
     if (!tensor_type) throw py::error_already_set();
     py::setattr(m, "Tensor", tensor_type);
@@ -998,7 +1032,7 @@ void init_tensor(py::module m) {
     m.def("set_cpp_apply_with_tracing", &set_cpp_apply_with_tracing);
     m.def("set_cpp_apply_const_with_tracing", &set_cpp_apply_const_with_tracing);
     m.def("set_cpp_apply_backward_varnode", &set_cpp_apply_backward_varnode);
-
+    m.def("set_cpp_apply_module_trace", &set_cpp_apply_module_trace);
     m.attr("skip_tracing") = &skip_tracing;
 
     py::class_<SharedHandle>(m, "SharedHandle")
@@ -1016,6 +1050,9 @@ void init_tensor(py::module m) {
     m.def("set_allow_higher_order_directive", [](bool value){
         GradKey::allow_higher_order_directive = value;
     });
+    m.def("set_module_tracing", &set_module_tracing);
+    m.def("unset_module_tracing", &unset_module_tracing);
+    m.def("is_tracing_module", &is_tracing_module);
 }
 
 #undef MGE_PY_INTERFACE

imperative/python/src/tensor.h

@@ -96,6 +96,7 @@ struct Tensor : std::enable_shared_from_this<Tensor>, NonCopyableObj {
         static constexpr flags_t SCALAR = 1;
         static constexpr flags_t GRAD = 1 << 1;
         static constexpr flags_t TRACE = 1 << 2;
+        static constexpr flags_t MODULE_TRACE = 1 << 3;
     };
 
     flags_t m_flags = 0;
@@ -106,6 +107,7 @@ struct Tensor : std::enable_shared_from_this<Tensor>, NonCopyableObj {
     std::string user_custom_name;
     std::string automatic_name;
     cg::VarNode* m_var;
+    pybind11::object m_module_trace_info;
 
     using Handle = interpreter::Interpreter::Handle;
 
@@ -158,10 +160,10 @@ struct TensorWrapper {
     using wrap_t = pyext17::wrap<TensorWrapper>;
     friend wrap_t;
 
-    inline static TensorWrapper* cast(PyObject* op) {return reinterpret_cast<wrap_t*>(op)->inst();}
-    inline static TensorWrapper* try_cast(PyObject* op) {
-        if (!wrap_t::type().isinstance(op)) return nullptr;
-        return cast(op);
+    inline static TensorWrapper* cast(PyObject* obj) {return reinterpret_cast<wrap_t*>(obj)->inst();}
+    inline static TensorWrapper* try_cast(PyObject* obj) {
+        if (!wrap_t::type().isinstance(obj)) return nullptr;
+        return cast(obj);
     }
     inline ObjectPtr<TensorWrapper, pybind11::handle> self() {return wrap_t::pycast(this);}
 
@@ -206,6 +208,8 @@ struct TensorWrapper {
     void set_compiled_info(PyObject *);
     PyObject* trace_mixin_info();
     void set_trace_mixin_info(PyObject *);
+    PyObject* module_trace_info();
+    void set_module_trace_info(PyObject *);
     PyObject* user_custom_name();
     void set_user_custom_name(PyObject *);
     PyObject* automatic_name();
@@ -331,6 +335,7 @@ void init_tensor(pybind11::module);
 
 extern PyObject *cpp_apply_with_tracing;
 extern PyObject *cpp_apply_backward_varnode;
+extern PyObject *cpp_apply_module_trace;
 
 } // namespace mgb::imperative::python
 

imperative/python/test/unit/core/test_serialization.py

@@ -14,6 +14,11 @@ import numpy as np
 
 import megengine as mge
 from megengine import Parameter, Tensor
+from megengine.core.ops import builtin
+from megengine.experimental.traced_module.serialization import (
+    get_opdef_state,
+    load_opdef_from_state,
+)
 
 
 def test_tensor_serialization():
@@ -86,3 +91,25 @@ def test_compatibility():
 
     test_old_tensor("tensor_v1_1.mge")
     test_old_tensor("tensor_v1_2.mge")
+
+
+def test_opdef_serialization():
+    with TemporaryFile() as f:
+        x = builtin.Elemwise(mode="Add")
+        pickle.dump(get_opdef_state(x), f)
+        f.seek(0)
+        load_x = load_opdef_from_state(pickle.load(f))
+        assert x == load_x
+
+    with TemporaryFile() as f:
+        x = builtin.Convolution(stride_h=9, compute_mode="float32")
+        x.strategy = (
+            builtin.Convolution.Strategy.PROFILE
+            | builtin.Convolution.Strategy.HEURISTIC
+            | builtin.Convolution.Strategy.REPRODUCIBLE
+        )
+        pickle.dump(get_opdef_state(x), f)
+        f.seek(0)
+        load_x = load_opdef_from_state(pickle.load(f))
+        assert x.strategy == load_x.strategy
+        assert x == load_x

imperative/tablegen/targets/python_c_extension.cpp

@@ -34,6 +34,7 @@ private:
     void emit_class();
     void emit_py_init();
     void emit_py_getsetters();
+    void emit_py_methods();
     Initproc emit_initproc();
 
     MgbOp& op;
@@ -133,9 +134,16 @@ void $0(PyTypeObject& py_type) {
 
     if (firstOccur) {
         os << tgfmt(R"(
+    static PyMethodDef tp_methods[] = {
+        {const_cast<char*>("dump"), (PyCFunction)$enumTpl<$opClass::$enumClass>::py_dump, METH_NOARGS, NULL},
+        {NULL}  /* Sentinel */
+        };
+    )", &ctx);
+        os << tgfmt(R"(
     static PyType_Slot slots[] = {
         {Py_tp_repr, (void*)$enumTpl<$opClass::$enumClass>::py_repr},
         {Py_tp_richcompare, (void*)$enumTpl<$opClass::$enumClass>::tp_richcompare},
+        {Py_tp_methods, tp_methods},
 )", &ctx);
         if (attr->getEnumCombinedFlag()) {
             // only bit combined enum could new instance because bitwise operation,
@@ -212,17 +220,62 @@ Initproc OpDefEmitter::emit() {
     emit_class();
     emit_py_init();
     emit_py_getsetters();
+    emit_py_methods();
     return emit_initproc();
 }
 
 void OpDefEmitter::emit_class() {
+    auto&& className = op.getCppClassName();
+    std::string method_defs;
+    std::vector<std::string> body;
+
+    llvm::for_each(op.getMgbAttributes(), [&](auto&& attr) {
+        body.push_back(formatv(R"(
+            {{"{0}", serialization<decltype(opdef.{0})>::dump(opdef.{0})})"
+            , attr.name));
+    });
+    method_defs += formatv(R"(
+    static PyObject* getstate(PyObject* self, PyObject*) {{
+        auto& opdef = reinterpret_cast<PyOp({0})*>(self)->inst();
+        static_cast<void>(opdef);
+        std::unordered_map<std::string, py::object> state {{
+            {1}
+        };
+        return py::cast(state).release().ptr();
+    })", className, llvm::join(body, ","));
+
+        body.clear();
+    llvm::for_each(op.getMgbAttributes(), [&](auto&& attr) {
+        body.push_back(formatv(R"(
+        {{
+        auto&& iter = state.find("{0}");
+        if (iter != state.end()) {
+            opdef.{0} = serialization<decltype(opdef.{0})>::load(iter->second);
+        }
+        })", attr.name));
+    });
+
+    method_defs += formatv(R"(
+    static PyObject* setstate(PyObject* self, PyObject* args) {{
+        PyObject* dict = PyTuple_GetItem(args, 0);
+        if (!dict) return NULL;
+        auto state = py::cast<std::unordered_map<std::string, py::object>>(dict);
+        auto& opdef = reinterpret_cast<PyOp({0})*>(self)->inst();
+        static_cast<void>(opdef);
+        {1}
+        Py_RETURN_NONE;
+    })", className, llvm::join(body, "\n"));
+
+
     os << tgfmt(R"(
 PyOpDefBegin($_self) // {
     static PyGetSetDef py_getsetters[];
+    static PyMethodDef tp_methods[];
+    $0
     static int py_init(PyObject *self, PyObject *args, PyObject *kwds);
 // };
 PyOpDefEnd($_self)
-)", &ctx);
+)", &ctx, method_defs);
 }
 
 void OpDefEmitter::emit_py_init() {
@@ -302,6 +355,33 @@ PyGetSetDef PyOp($_self)::py_getsetters[] = {
 )", &ctx, llvm::join(llvm::map_range(op.getMgbAttributes(), f), "\n    "));
 }
 
+void OpDefEmitter::emit_py_methods(){
+
+     // generate methods
+    std::string method_defs;
+    std::vector<std::string> method_items;
+    {
+        auto&& className = op.getCppClassName();
+        // generate getstate
+        method_items.push_back(formatv(
+            "{{const_cast<char*>(\"__getstate__\"), PyOp({0})::getstate, METH_NOARGS, \"{0} getstate\"},",
+            className));
+
+        // generate setstate
+        method_items.push_back(formatv(
+            "{{const_cast<char*>(\"__setstate__\"), PyOp({0})::setstate, METH_VARARGS, \"{0} setstate\"},",
+            className));
+    }
+   
+
+    os << tgfmt(R"(
+    PyMethodDef PyOp($_self)::tp_methods[] = {
+        $0
+        {NULL}  /* Sentinel */
+    };
+    )", &ctx, llvm::join(method_items, "\n    "));
+}
+
 Initproc OpDefEmitter::emit_initproc() {
     std::string initproc = formatv("_init_py_{0}", op.getCppClassName());
     std::string subclass_init_call;
@@ -321,6 +401,7 @@ void $0(py::module m) {
     py_type.tp_dealloc = py_dealloc_generic<py_op>;
     py_type.tp_new = py_new_generic<py_op>;
     py_type.tp_init = py_op::py_init;
+    py_type.tp_methods = py_op::tp_methods;
     py_type.tp_getset = py_op::py_getsetters;
     mgb_assert(PyType_Ready(&py_type) >= 0);
     $1