feat(imperative/src): name operators automatically when tracing

GitOrigin-RevId: ff8eb003c5

imperative/python/megengine/core/tensor/megbrain_graph.py

@@ -96,7 +96,7 @@ class Graph(_imperative_rt.ComputingGraph):
         data = data.numpy()
         return self._wrap(_imperative_rt.make_const(self, data, device, data.dtype))
 
-    def make_const(self, data, dtype=None, device=None):
+    def make_const(self, data, dtype=None, device=None, name=None):
         if isinstance(data, _imperative_rt.DeviceTensorND):
             assert dtype is None and device is None
             return self._wrap(_imperative_rt.make_shared(self, data))
@@ -107,7 +107,9 @@ class Graph(_imperative_rt.ComputingGraph):
             elif data.dtype == np.int64:
                 data = data.astype(np.int32)
             device = as_device(device).to_c()
-            return self._wrap(_imperative_rt.make_const(self, data, device, dtype))
+            return self._wrap(
+                _imperative_rt.make_const(self, data, device, dtype, name)
+            )
 
     def make_input(self, *args: "VarNode", device=None, dtype=None, shape=None):
         opnode = InputNode(*args, device=device, dtype=dtype, shape=shape, graph=self)
@@ -305,7 +307,7 @@ def dump_graph(
     output_vars: Union[Dict[str, VarNode], List[VarNode]],
     *,
     keep_var_name: int = 1,
-    keep_op_name: bool = True,
+    keep_opr_name: bool = False,
     keep_param_name: bool = False,
     keep_opr_priority: bool = False,
     strip_info_file=None,
@@ -326,7 +328,7 @@ def dump_graph(
         * 0: none of the names are kept
         * 1: (default)keep names of output vars
         * 2: keep names of all (output and internal) vars
-    :param keep_op_name: whether to keep operator names.
+    :param keep_opr_name: whether to keep operator names.
     :param keep_param_name: whether to keep param names, so param values can be
         easily manipulated after loading model
     :param keep_opr_priority: whether to keep priority setting for operators
@@ -370,7 +372,7 @@ def dump_graph(
     dump_content = _imperative_rt.dump_graph(
         ov,
         keep_var_name,
-        keep_op_name,
+        keep_opr_name,
         keep_param_name,
         keep_opr_priority,
         stat,

imperative/python/megengine/jit/tracing.py

@@ -36,6 +36,7 @@ from ..core.ops.builtin import BackwardGraph, OpDef
 from ..core.ops.special import Const
 from ..core.tensor import megbrain_graph as G
 from ..core.tensor.utils import setscalar
+from ..utils.naming import auto_naming
 from .sublinear_memory_config import SublinearMemoryConfig
 
 
@@ -77,6 +78,7 @@ def exclude_from_trace():
 class TensorInfo:
     __slots__ = (
         # collected attributes
+        "name",
         "external",
         "data_read",
         "shape_read",
@@ -96,6 +98,7 @@ class TensorInfo:
     )
 
     def __init__(self):
+        self.name = None
         self.exported = None
         self.data_read = None
         self.shape_read = None
@@ -290,12 +293,16 @@ class trace:
             h = getattr(x, "_mixin_handle", -1)
             if h < 0 or (not self._capture_as_const and self._tinfo[h].exported):
                 h, info = self._new_handle()
+                name = auto_naming.get_scope() + "." + x.c_name if x.c_name else x._name
+                info.name = name
                 info.external = True
                 info.device = x.device
                 info.dtype = x.dtype
                 info.shape = x.shape
                 if self._capture_as_const:
-                    info.bound_data = RawTensor(x.numpy(), x.dtype, x.device, False)
+                    info.bound_data = RawTensor(
+                        x.numpy(), x.dtype, x.device, False, name
+                    )
 
             ihandles.append(h)
 
@@ -669,6 +676,12 @@ class trace:
         arg_names=None,
         output_names=None,
         append=False,
+        keep_var_name: int = 1,
+        keep_opr_name: bool = False,
+        keep_param_name: bool = False,
+        keep_opr_priority: bool = False,
+        strip_info_file=None,
+        append_json=False,
         optimize_for_inference=True,
         **kwargs
     ):
@@ -681,6 +694,20 @@ class trace:
             use the default name if not specified.
         :param append: whether output is appended to ``file``.
             Only works when ``file`` is str.
+        :param keep_var_name: level for keeping variable names:
+
+            * 0: none of the names are kept
+            * 1: (default)keep names of output vars
+            * 2: keep names of all (output and internal) vars
+        :param keep_opr_name: whether to keep operator names.
+        :param keep_param_name: whether to keep param names, so param values can be
+            easily manipulated after loading model
+        :param keep_opr_priority: whether to keep priority setting for operators
+        :param strip_info_file: a string for path or a file handler. if is not None,
+            then the dump information for code strip would be written to ``strip_info_file``
+        :param append_json: will be check when `strip_info_file` is not None. if set
+            true, the information for code strip will be append to strip_info_file.
+            if set false, will rewrite strip_info_file
         :param optimize_for_inference: enbale optmizations,
             will skip all optimize options if this is False. Default: True
 
@@ -785,7 +812,10 @@ class trace:
                     assert info.external
                     assert info.bound_data
                     h2v[h] = graph.make_const(
-                        info.bound_data.numpy(), dtype=info.dtype, device=info.device,
+                        info.bound_data.numpy(),
+                        dtype=info.dtype,
+                        device=info.device,
+                        name=info.name,
                     )
                 continue
             ivars = []
@@ -795,13 +825,26 @@ class trace:
                     assert info.external
                     assert info.bound_data
                     h2v[h] = graph.make_const(
-                        info.bound_data.numpy(), dtype=info.dtype, device=dumped_device
+                        info.bound_data.numpy(),
+                        dtype=info.dtype,
+                        device=dumped_device,
+                        name=info.name,
                     )
                 ivars.append(h2v[h])
             ovars = G.apply_normal_varnode(op, *ivars)
+
+            auto_naming.record_opnode(ovars[0].op)
+
             assert len(ovars) == len(ohandles)
             h2v.update(zip(ohandles, ovars))
 
+            for i in ohandles:
+                name = auto_naming.get_var_name(i)
+                if name is not None:
+                    h2v[i].name = name
+
+        auto_naming.remove_duplicate_names()
+
         dest_vars = []
         for i, h in enumerate(self._output_bindings):
             v = h2v[h]
@@ -815,7 +858,15 @@ class trace:
         if isinstance(file, str):
             permission = "wb" if append == False else "ab"
             file = open(file, permission)
-        dump_content, dump_info = G.dump_graph(dest_vars)
+        dump_content, dump_info = G.dump_graph(
+            dest_vars,
+            keep_var_name=keep_var_name,
+            keep_opr_name=keep_opr_name,
+            keep_param_name=keep_param_name,
+            keep_opr_priority=keep_opr_priority,
+            strip_info_file=strip_info_file,
+            append_json=append_json,
+        )
         file.write(dump_content)
         return dump_info
 
@@ -1095,20 +1146,22 @@ def apply_compiled_mode(op: OpDef, *args: RawTensor):
     return active_trace._apply_op(op, args)
 
 
-def apply_const_compiled_mode(value, dtype, device, is_const, no_cache):
+def apply_const_compiled_mode(value, dtype, device, is_const, no_cache, name):
     if skip_tracing:
         args = [
             RawTensor(x._dev_tensor()) if x.__class__ is CompiledTensorProxy else x
             for x in args
         ]
         unset_tracing()
-        ret = RawTensor(value, dtype, device, False)
+        ret = RawTensor(value, dtype, device, False, name)
         set_tracing()
         return ret
     return active_trace._apply_const(value, dtype, device)
 
 
 def apply_with_tracing(op: OpDef, *args: RawTensor):
+    if hasattr(op, "scope"):
+        op.scope = auto_naming.get_scope()
     if active_trace._symbolic:
         outputs = apply_symbolic_mode(op, *args)
     else:
@@ -1120,12 +1173,12 @@ def apply_with_tracing(op: OpDef, *args: RawTensor):
     return list(outputs)
 
 
-def apply_const_with_tracing(value, dtype, device, is_const, no_cache):
+def apply_const_with_tracing(value, dtype, device, is_const, no_cache, name):
     if active_trace._symbolic:
         outputs = apply_const_symbolic_mode(value, dtype, device)
     else:
         unset_tracing()
-        outputs = (RawTensor(value, dtype, device, False),)
+        outputs = (RawTensor(value, dtype, device, False, name),)
         set_tracing()
     active_trace._record_const(outputs)
     return list(outputs)

imperative/python/megengine/module/module.py

@@ -12,12 +12,12 @@ from typing import Any, Callable, Iterable, Optional, Set, Tuple, Union
 
 import numpy as np
 
-from ..core._imperative_rt.core2 import pop_scope, push_scope
 from ..core.tensor.utils import make_shape_tuple
 from ..logger import get_logger
 from ..tensor import Parameter, Tensor
 from ..utils.deprecation import deprecated
 from ..utils.hook import HookHandler
+from ..utils.naming import auto_naming
 
 logger = get_logger(__name__)
 
@@ -69,7 +69,9 @@ class Module(metaclass=ABCMeta):
     Base Module class.
     """
 
-    def __init__(self):
+    def __init__(self, name=""):
+        self.name = name
+
         # runtime attributes
         self.training = True
         self.quantize_disabled = False
@@ -79,6 +81,8 @@ class Module(metaclass=ABCMeta):
         self._forward_hooks = OrderedDict()
 
         self._modules = []
+
+        # used for profiler and automatic naming
         self._name = "{anonymous}"
 
     @abstractmethod
@@ -105,7 +109,7 @@ class Module(metaclass=ABCMeta):
         return HookHandler(self._forward_hooks, hook)
 
     def __call__(self, *inputs, **kwargs):
-        push_scope(self._name)
+        auto_naming.push_scope(self.name if self.name else self._name)
         for hook in self._forward_pre_hooks.values():
             modified_inputs = hook(self, inputs)
             if modified_inputs is not None:
@@ -119,7 +123,7 @@ class Module(metaclass=ABCMeta):
             modified_outputs = hook(self, inputs, outputs)
             if modified_outputs is not None:
                 outputs = modified_outputs
-        pop_scope(self._name)
+        auto_naming.pop_scope()
         return outputs
 
     def _flatten(
@@ -579,7 +583,7 @@ class Module(metaclass=ABCMeta):
         value = super().__getattribute__(name)
         if name == "_name":
             return value
-        if _is_module(value):
+        if isinstance(value, (Tensor, Module)):
             value._name = name
         return value
 

imperative/python/megengine/tensor.py

@@ -20,6 +20,7 @@ from .core.tensor.array_method import ArrayMethodMixin
 from .device import _valid_device, get_default_device
 from .logger import get_logger
 from .utils.deprecation import deprecated
+from .utils.naming import auto_naming
 
 
 class Tensor(_Tensor, ArrayMethodMixin):
@@ -27,7 +28,9 @@ class Tensor(_Tensor, ArrayMethodMixin):
     dmap_callback = None
     _q_dict = None
 
-    def __new__(cls, data, dtype=None, device=None, is_const=False, no_cache=False):
+    def __new__(
+        cls, data, dtype=None, device=None, is_const=False, no_cache=False, name=""
+    ):
         if device is None:
             cn = get_default_device()
         elif isinstance(device, str):
@@ -51,8 +54,7 @@ class Tensor(_Tensor, ArrayMethodMixin):
             if isinstance(data, np.ndarray):
                 if 0 in data.strides:
                     data = data.squeeze().reshape(data.shape)
-
-            obj = _Tensor.__new__(cls, data, dtype, cn, is_const, no_cache)
+            obj = _Tensor.__new__(cls, data, dtype, cn, is_const, no_cache, name)
         return obj
 
     @property
@@ -91,6 +93,15 @@ class Tensor(_Tensor, ArrayMethodMixin):
         piece += ", device={}".format(self.device) + ")"
         return piece
 
+    @property
+    def name(self):
+        return self.c_name
+
+    @name.setter
+    def name(self, name):
+        self.c_name = name
+        auto_naming.record_var_name(self._mixin_handle, name)
+
     @deprecated(version="1.0", reason="no need to reuse an existing tensor since 1.0")
     def set_value(self, value):
         if not isinstance(value, _Tensor):

imperative/python/megengine/utils/naming.py

@@ -0,0 +1,63 @@
+# 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 ..core._imperative_rt.core2 import pop_scope, push_scope
+
+
+class AutoNaming:
+    r"""
+    Name all executed operators automaticlly during tracing and record all tensors
+    renamed by the user.
+    """
+
+    def __init__(self):
+        self.scopes = []
+        self.c_ops = []
+        self.name2ops = {}
+        self.handle2names = {}
+
+    def clear(self):
+        for var in vars(self).values():
+            var.clear()
+
+    def push_scope(self, scope):
+        push_scope(scope)
+        self.scopes.append(scope)
+
+    def pop_scope(self):
+        scope = self.scopes.pop()
+        pop_scope(scope)
+
+    def get_scope(self):
+        return ".".join(self.scopes)
+
+    def record_var_name(self, handle, name):
+        self.handle2names[handle] = name
+
+    def get_var_name(self, handle):
+        return self.handle2names.pop(handle, None)
+
+    def record_opnode(self, op):
+        ops = self.name2ops.get(op.name, [])
+        ops.append(op)
+        self.name2ops[op.name] = ops
+
+    def remove_duplicate_names(self):
+        for key, ops in self.name2ops.items():
+            if len(ops) == 1:
+                continue
+            for i, op in enumerate(ops):
+                op.name = key + "[%s]" % str(i)
+                if len(op.outputs) == 1:
+                    continue
+                for var in op.outputs:
+                    var.name = var.name.replace(key, op.name)
+        self.name2ops.clear()
+
+
+auto_naming = AutoNaming()

imperative/python/src/graph_rt.cpp

@@ -294,7 +294,7 @@ void init_graph_rt(py::module m) {
     m.def("dump_graph", [](
         const std::vector<VarNode*>& dest_vars,
         int keep_var_name,
-        bool keep_op_name,
+        bool keep_opr_name,
         bool keep_param_name,
         bool keep_opr_priority,
         py::list& stat,
@@ -307,7 +307,7 @@ void init_graph_rt(py::module m) {
         SymbolVarArray symvars(dest_vars.begin(), dest_vars.end());
 
         ser::GraphDumper::DumpConfig config{keep_var_name, keep_param_name,
-                                       keep_opr_priority, keep_op_name};
+                                       keep_opr_priority, keep_opr_name};
 
         auto rst = dumper->dump(symvars, config);
         for (auto i : rst.inputs) {
@@ -457,13 +457,17 @@ void init_graph_rt(py::module m) {
             return opr::SharedDeviceTensor::make(*graph, std::make_shared<DeviceTensorND>(data)).node();
         });
 
-    m.def("make_const", [](cg::ComputingGraph* graph, py::array data, CompNode cn, DType dtype) {
+    m.def("make_const", [](cg::ComputingGraph* graph, py::array data, CompNode cn, DType dtype, std::optional<std::string> name) {
             if (!cn.valid()) {
                 cn = CompNode::load(get_default_device());
             }
+            OperatorNodeConfig config(cn);
+            if (name) {
+                config.name(*name);
+            }
             auto hv = npy::np2tensor(data.ptr(), npy::Meth::borrow(cn), dtype);
-            return opr::ImmutableTensor::make(*graph, hv, OperatorNodeConfig(cn)).node();
-        });
+            return opr::ImmutableTensor::make(*graph, hv, config).node();
+        }, py::arg(), py::arg(), py::arg(), py::arg(), py::arg() = py::none());
 
     m.def("make_h2d", [](cg::ComputingGraph& graph, CompNode cn, DType dtype, TensorShape shape, std::optional<std::string> name) {
             if (!cn.valid()) {

imperative/python/src/ops.cpp

@@ -99,6 +99,14 @@ PyObject* py_get_generic_impl(PyObject* obj, void* /* closure */) {
 #define py_get_generic(name, attr) \
     py_get_generic_impl<PyOp(name), decltype(std::declval<name>().attr), &name::attr>
 
+template<typename T>
+PyObject* py_get_scope_impl(PyObject* obj, void* /* closure */) {
+    // T: PyOpXXX  inst(): return XXX in opdef.h.inl
+    auto& op = reinterpret_cast<T*>(obj)->inst();
+    return pyobj_convert_generic<std::string>::to(op.scope());
+}
+#define py_get_scope(class) py_get_scope_impl<PyOp(class)>
+
 template<typename T, typename U, U T::Ty::*attr>
 int py_set_generic_impl(PyObject* obj, PyObject* value, void* /* closure */) {
     if (value == NULL) {
@@ -121,6 +129,27 @@ int py_set_generic_impl(PyObject* obj, PyObject* value, void* /* closure */) {
 #define py_set_generic(name, attr) \
     py_set_generic_impl<PyOp(name), decltype(std::declval<name>().attr), &name::attr>
 
+template<typename T>
+int py_set_scope_impl(PyObject* obj, PyObject* value, void* /* closure */) {
+    if (value == NULL) {
+        PyErr_SetString(PyExc_TypeError, "Cannot delete the attribute");
+        return -1;
+    }
+    auto& op = reinterpret_cast<T*>(obj)->inst();
+    try {
+        op.set_scope(pyobj_convert_generic<std::string>::from(value));
+        return 0;
+    } catch(py::error_already_set& e) {
+        e.restore();
+    } catch(py::builtin_exception& e) {
+        e.set_error();
+    } catch(...) {
+        PyErr_SetString(PyExc_RuntimeError, "Unknown Error");
+    }
+    return -1;
+}
+#define py_set_scope(class) py_set_scope_impl<PyOp(class)>
+
 struct PyOpDef {
     PyObject_HEAD
     std::shared_ptr<OpDef> op;

imperative/python/src/tensor.cpp

@@ -24,6 +24,7 @@
 #include <pybind11/numpy.h>
 #include <pybind11/operators.h>
 #include <range/v3/all.hpp>
+#include <string>
 
 #include <unordered_map>
 
@@ -222,14 +223,15 @@ TensorWrapper::TensorWrapper(PyObject* args, PyObject* kwargs) {
             }
         } else {
             py::detail::loader_life_support life_sup; // FIXME!!!required to cast DType
-            if (nargs != 4 && nargs != 5) {
-                throw py::type_error("expect 4 or 5 arguments");
+            if (nargs != 5 && nargs != 6) {
+                throw py::type_error("expect 5 or 6 arguments");
             }
             auto data = tup[0].cast<py::array>();
             DType dtype = tup[1].cast<DType>();
             CompNode cn = tup[2].cast<CompNode>();
             bool is_const = tup[3].cast<bool>();
-            bool no_cache = nargs == 5 ? tup[4].cast<bool>() : false;
+            bool no_cache = nargs == 6 ? tup[4].cast<bool>() : false;
+            std::string name = tup[nargs - 1].cast<std::string>();
 
             // const op
             if (is_const && is_tracing) {
@@ -259,6 +261,7 @@ TensorWrapper::TensorWrapper(PyObject* args, PyObject* kwargs) {
             }
 
             m_tensor = std::make_shared<Tensor>(handle);
+            m_tensor->user_custom_name = name;
 
             if (data.ndim() == 0) {
                 m_tensor->m_flags |= Tensor::Flags::SCALAR;
@@ -313,6 +316,19 @@ REGISTE_TENSORWRAPPER_PYOBJECT_FUNC(trace_mixin_info)
 #undef REGISTE_TENSORWRAPPER_PYOBJECT_FUNC
 
 
+#define SET_GET_NAME(member)                                     \
+    PyObject* TensorWrapper::member() {                          \
+        return py::cast(m_tensor->member).release().ptr();       \
+    }                                                            \
+    void TensorWrapper::set_##member(PyObject* dest) {           \
+        auto py_dest = py::reinterpret_borrow<py::object>(dest); \
+        m_tensor->member = py_dest.cast<std::string>();          \
+    }
+SET_GET_NAME(user_custom_name)
+SET_GET_NAME(automatic_name)
+#undef SET_GET_NAME
+
+
 PyObject* TensorWrapper::handle() {
     return py::cast(m_tensor->m_handle).release().ptr();
 }
@@ -453,7 +469,11 @@ void TensorWrapper::reset(PyObject* tensor) {
     if (!t) {
         throw py::type_error("expect Tensor");
     }
+    std::string user_custom_name = m_tensor->user_custom_name;
+    std::string automatic_name = m_tensor->automatic_name;
     m_tensor = t->m_tensor;
+    m_tensor->user_custom_name = user_custom_name;
+    m_tensor->automatic_name = automatic_name;
 }
 
 void TensorWrapper::reset_varnode() {
@@ -785,6 +805,8 @@ void init_tensor(py::module m) {
         .def_getset<&TensorWrapper::handle, &TensorWrapper::set_handle>("_handle")
         .def_getset<&TensorWrapper::compiled_info, &TensorWrapper::set_compiled_info>("_compiled_info")
         .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")
         .finalize();
     if (!tensor_type) throw py::error_already_set();
     py::setattr(m, "Tensor", tensor_type);

imperative/python/src/tensor.h

@@ -15,6 +15,7 @@
 
 #include "megbrain/imperative/interpreter.h"
 #include "pybind11/pybind11.h"
+#include <string>
 
 #include "./pyext17.h"
 
@@ -70,6 +71,8 @@ struct Tensor : std::enable_shared_from_this<Tensor>, NonCopyableObj {
     GradInfo m_grad_info;
     TraceInfo m_trace_info;
     SharedHandle m_handle;
+    std::string user_custom_name;
+    std::string automatic_name;
     cg::VarNode* m_var;
 
     using Handle = interpreter::Interpreter::Handle;
@@ -170,6 +173,10 @@ struct TensorWrapper {
     void set_compiled_info(PyObject *);
     PyObject* trace_mixin_info();
     void set_trace_mixin_info(PyObject *);
+    PyObject* user_custom_name();
+    void set_user_custom_name(PyObject *);
+    PyObject* automatic_name();
+    void set_automatic_name(PyObject *);
     PyObject* _use_cnt() { return PyLong_FromSize_t(m_tensor.use_count()); };
 };
 

imperative/python/test/unit/test_dump_naming.py

@@ -0,0 +1,169 @@
+# -*- coding: utf-8 -*-
+# 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.
+import io
+
+import numpy as np
+import pytest
+
+import megengine.functional as F
+import megengine.module as M
+import megengine.utils.comp_graph_tools as cgtools
+from megengine import Parameter, Tensor
+from megengine.core.tensor import megbrain_graph as G
+from megengine.jit.tracing import trace
+from megengine.utils.naming import auto_naming
+
+
+def _dump_and_load(func, symbolic, keep_opr_name=True):
+    auto_naming.clear()
+    func = trace(func, symbolic=symbolic, capture_as_const=True)
+    x = Tensor(np.ones(shape=(2, 3)))
+    func(x).numpy()
+    file = io.BytesIO()
+    func.dump(
+        file,
+        optimize_for_inference=False,
+        arg_names="x",
+        keep_opr_name=keep_opr_name,
+        keep_var_name=2,
+    )
+    file.seek(0)
+    *_, outputs = G.load_graph(file)
+    op = cgtools.get_oprs_seq(outputs)[-1]
+    return op
+
+
+@pytest.mark.parametrize("symbolic", [False, True])
+def test_auto_naming(symbolic):
+    class Simple(M.Module):
+        def __init__(self, name):
+            super().__init__()
+            self.name = name
+
+        def forward(self, x):
+            return x + x
+
+    m = Simple("simple")
+    op = _dump_and_load(m, symbolic)
+    assert op.name == "simple.ADD"
+    assert op.outputs[0].name == "simple.ADD"
+
+
+@pytest.mark.parametrize("symbolic", [False, True])
+def test_user_named_tensor(symbolic):
+    class Simple(M.Module):
+        def __init__(self, name):
+            super().__init__()
+            self.name = name
+            self.k = Parameter(1.0, name="k")
+
+        def forward(self, x):
+            x = x + x
+            x.name = "o_x"
+            return x
+
+    m = Simple("simple")
+
+    op = _dump_and_load(m, symbolic)
+    assert op.name == "simple.ADD"
+    assert op.outputs[0].name == "o_x"
+
+
+@pytest.mark.parametrize("symbolic", [False, True])
+def test_user_named_param(symbolic):
+    class Simple(M.Module):
+        def __init__(self, name):
+            super().__init__()
+            self.name = name
+            self.k = Parameter(2.0, name="k")
+
+        def forward(self, x):
+            return self.k * x
+
+    m = Simple("simple")
+
+    op = _dump_and_load(m, symbolic)
+    assert op.inputs[0].name == "x"
+    assert op.inputs[1].name == "simple.k"
+
+
+@pytest.mark.parametrize("symbolic", [False, True])
+def test_without_module(symbolic):
+    def f(x):
+        return 2 * x
+
+    op = _dump_and_load(f, symbolic)
+    assert op.name == "MUL"
+
+
+@pytest.mark.parametrize("symbolic", [False, True])
+def test_with_submodule(symbolic):
+    class Simple(M.Module):
+        def __init__(self, name):
+            super().__init__()
+            self.name = name
+            self.linear = M.Linear(3, 3)
+
+        def forward(self, x):
+            x = self.linear(x)
+            return x
+
+    m = Simple("simple")
+
+    op = _dump_and_load(m, symbolic)
+    assert op.name == "simple.linear.ADD"
+    assert op.inputs[0].owner.name == "simple.linear.MatrixMul"
+    assert op.outputs[0].name == "simple.linear.ADD"
+
+
+@pytest.mark.parametrize("symbolic", [False, True])
+def test_named_submodule(symbolic):
+    class Simple(M.Module):
+        def __init__(self, name):
+            super().__init__()
+            self.name = name
+            self.linear = M.Linear(3, 3, name="x")
+
+        def forward(self, x):
+            x = self.linear(x)
+            return x
+
+    m = Simple("simple")
+
+    op = _dump_and_load(m, symbolic)
+    assert op.name == "simple.x.ADD"
+    assert op.inputs[0].owner.name == "simple.x.MatrixMul"
+    assert op.outputs[0].name == "simple.x.ADD"
+
+
+@pytest.mark.parametrize("symbolic", [False, True])
+def test_with_same_operators(symbolic):
+    class Simple(M.Module):
+        def __init__(self, name):
+            super().__init__()
+            self.name = name
+
+        def forward(self, x):
+            x = F.relu(x)
+            x = F.relu(x)
+            return x
+
+    m = Simple("simple")
+
+    op = _dump_and_load(m, symbolic)
+    assert op.name == "simple.RELU[1]"
+    assert op.inputs[0].owner.name == "simple.RELU[0]"
+
+
+def test_not_keep_opr_name():
+    def f(x):
+        return 2 * x
+
+    op = _dump_and_load(f, True, False)
+    assert op.name == "MUL(x,2[2])[4]"

imperative/python/test/unit/test_tracing.py

@@ -148,7 +148,7 @@ def test_dump():
     dump_info = f.dump(file)
     assert dump_info.nr_opr == 3
     np.testing.assert_equal(dump_info.inputs, ["arg_0", "arg_1"])
-    np.testing.assert_equal(dump_info.outputs, ["ADD(arg_0,arg_1)[4]"])
+    np.testing.assert_equal(dump_info.outputs, ["ADD"])
     file.seek(0)
     infer_cg = cgtools.GraphInference(file)
     result = list((infer_cg.run(a, b)).values())[0]

imperative/src/impl/op_def.cpp

@@ -75,10 +75,6 @@ std::vector<std::pair<const char*, std::string>> OpDef::props(
     return def.trait()->props(def);
 }
 
-const char* OpDef::name() const {
-    return trait()->name;
-}
-
 std::string OpDef::to_string() const {
     std::string builder = "{";
     for (auto&& [name, value]: props(*this)) {
@@ -107,6 +103,20 @@ const OpTrait* OpDef::trait() const {
     return m_trait;
 }
 
+const std::string OpDef::scope() const {
+    return m_scope;
+}
+
+void OpDef::set_scope(const std::string& scope) {
+    m_scope = scope;
+}
+
+const std::string OpDef::make_name() const {
+    if (m_scope.empty())
+        return trait()->make_name(*this);
+    return m_scope + "." + trait()->make_name(*this);
+}
+
 } // namespace imperative
 } // namespace mgb
 

imperative/src/impl/op_trait.h

@@ -75,6 +75,7 @@ using GradMaker = detail::OpMeth<
 using Props = detail::OpMeth<decltype(OpDef::props)>;
 using HashFunc = detail::OpMeth<size_t(const OpDef&)>;
 using IsSame = detail::OpMeth<bool(const OpDef&, const OpDef&)>;
+using MakeNameFunc = detail::OpMeth<std::string(const OpDef&)>;
 
 struct OpTrait {
     const char* name;
@@ -88,6 +89,7 @@ struct OpTrait {
     Props props;
     HashFunc hash;
     IsSame is_same_st;
+    MakeNameFunc make_name;
     OpTrait(const char* name);
     static OpTrait* find_by_name(const char* name);
     static OpTrait* find_by_typeinfo(Typeinfo* type);
@@ -104,7 +106,8 @@ struct OpTrait {
     cb(make_backward_graph) \
     cb(props) \
     cb(hash) \
-    cb(is_same_st)
+    cb(is_same_st) \
+    cb(make_name)
 
 struct OpTraitRegistry {
     OpTrait* trait;

imperative/src/impl/ops/batch_norm.cpp

@@ -30,13 +30,14 @@ cg::OperatorNodeBase* apply_on_var_node(
     size_t nr_inp = inputs.size();
     mgb_assert(nr_inp == 3 ||nr_inp == 5,
               "BatchNorm expects 3 or 5 inputs; got %lu actually", nr_inp);
+    OperatorNodeConfig config{bn_opr.make_name()};
     if (nr_inp == 3) {
         return opr::BatchNorm::make(
-            inputs[0], inputs[1], inputs[2], bn_opr.param())[0]
+            inputs[0], inputs[1], inputs[2], bn_opr.param(), config)[0]
             .node()->owner_opr();
     } else {
         return opr::BatchNorm::make(
-            inputs[0], inputs[1], inputs[2], inputs[3], inputs[4], bn_opr.param())[0]
+            inputs[0], inputs[1], inputs[2], inputs[3], inputs[4], bn_opr.param(), config)[0]
             .node()->owner_opr();
     }
 }

imperative/src/impl/ops/broadcast.cpp

@@ -27,10 +27,11 @@ std::shared_ptr<OpDef> make_from_op_node(cg::OperatorNodeBase* node_) {
 cg::OperatorNodeBase* apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
-    def.cast_final_safe<Broadcast>();
+    auto&& op = def.cast_final_safe<Broadcast>();
     size_t nr_inp = inputs.size();
     mgb_assert(nr_inp == 2, "Broadcast expects 2 inputs; got %lu actually", nr_inp);
-    return opr::Broadcast::make(inputs[0], inputs[1]).node()->owner_opr();
+    OperatorNodeConfig config{op.make_name()};
+    return opr::Broadcast::make(inputs[0], inputs[1], config).node()->owner_opr();
 }
 
 bool valid_broadcast(const TensorShape& src_shape,
@@ -96,7 +97,8 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& op = static_cast<const Reshape&>(def);
     mgb_assert(inputs.size() == 2);
-    return opr::Reshape::make(inputs[0], inputs[1], op.param());
+    OperatorNodeConfig config{op.make_name()};
+    return opr::Reshape::make(inputs[0], inputs[1], op.param(), config);
 }
 
 std::tuple<SmallVector<LogicalTensorDesc>, bool> infer_output_attrs_fallible(

imperative/src/impl/ops/collective_comm.cpp

@@ -35,7 +35,7 @@ cg::OperatorNodeBase* apply_on_var_node(
     auto disable = std::make_shared<DTypeScalar>();
     disable->set(0);
 
-    cg::OperatorNodeConfig config;
+    OperatorNodeConfig config{comm.make_name()};
     if (comm.comp_node.size() > 0) {
         config.comp_node(CompNode::load(comm.comp_node));
     }

imperative/src/impl/ops/cond_take.cpp

@@ -23,12 +23,12 @@ namespace {
 cg::OperatorNodeBase* apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
-    def.cast_final_safe<CondTake>();
+    auto&& op = def.cast_final_safe<CondTake>();
     auto&& graph = inputs[0]->owner_graph();
 
     opr::CondTake::Param param;
     param.val = 1;
-    cg::OperatorNodeConfig config;
+    OperatorNodeConfig config{op.make_name()};
     cg::OperatorNodeBase* opr = graph->insert_opr(
             std::make_unique<opr::CondTake>(
                     inputs[0], inputs[1], param, config));

imperative/src/impl/ops/elemwise.cpp

@@ -31,7 +31,8 @@ cg::OperatorNodeBase* apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
     auto&& elemwise_opr = def.cast_final_safe<Elemwise>();
-    return opr::Elemwise::make(inputs, elemwise_opr.mode).node()->owner_opr();
+    OperatorNodeConfig config{elemwise_opr.make_name()};
+    return opr::Elemwise::make(inputs, elemwise_opr.mode, config).node()->owner_opr();
 }
 
 std::tuple<SmallVector<LogicalTensorDesc>, bool> infer_output_attrs_fallible(

imperative/src/impl/ops/img_proc.cpp

@@ -23,7 +23,8 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& op = static_cast<const CvtColor&>(def);
     mgb_assert(inputs.size() == 1);
-    return opr::CvtColor::make(inputs[0], op.param());
+    OperatorNodeConfig config{op.make_name()};
+    return opr::CvtColor::make(inputs[0], op.param(), config);
 }
 OP_TRAIT_REG(CvtColor, CvtColor)
     .apply_on_var_node(apply_on_var_node)

imperative/src/impl/ops/io_remote.cpp

@@ -32,7 +32,7 @@ cg::OperatorNodeBase* apply_on_var_node_remote_send(
             ssprintf("%s:%d", send.addr.data(), send.port));
     auto&& graph = inputs[0]->owner_graph();
 
-    cg::OperatorNodeConfig config;
+    OperatorNodeConfig config{send.make_name()};
     cg::OperatorNodeBase* opr =
             graph->insert_opr(std::make_unique<mgb::opr::RemoteSend>(
                     send.key, inputs[0], group_client, true, config));
@@ -42,11 +42,13 @@ cg::OperatorNodeBase* apply_on_var_node_remote_send(
 cg::OperatorNodeBase* apply_on_var_node_remote_recv(
         const OpDef& def, const VarNodeArray& inputs) {
     auto&& recv = def.cast_final_safe<RemoteRecv>();
+    OperatorNodeConfig config{recv.cn};
+    config.name(recv.make_name());
     auto group_client = std::make_shared<GroupClientProxy>(
             ssprintf("%s:%d", recv.addr.data(), recv.port));
     auto&& graph = inputs[0]->owner_graph();
     return graph->insert_opr(std::make_unique<mgb::opr::RemoteRecv>(
-            recv.key, inputs[0], *graph, group_client, OperatorNodeConfig{recv.cn},
+            recv.key, inputs[0], *graph, group_client, config,
             recv.shape, recv.dtype));
 }
 

imperative/src/impl/ops/matrix_inverse.cpp

@@ -21,8 +21,10 @@ namespace {
 auto apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
+    auto&& op = def.cast_final_safe<MatrixInverse>();
     mgb_assert(inputs.size() == 1);
-    return opr::MatrixInverse::make(inputs[0]);
+    OperatorNodeConfig config{op.make_name()};
+    return opr::MatrixInverse::make(inputs[0], {}, config);
 }
 OP_TRAIT_REG(MatrixInverse, MatrixInverse)
     .apply_on_var_node(apply_on_var_node)

imperative/src/impl/ops/nms.cpp

@@ -29,7 +29,9 @@ cg::OperatorNodeBase* apply_on_var_node(
     param.iou_thresh = nms_keep.iou_thresh;
     param.max_output = nms_keep.max_output;
 
-    return NMSKeepOpr::make(inputs[0], param).node()->owner_opr();
+    OperatorNodeConfig config{nms_keep.make_name()};
+
+    return NMSKeepOpr::make(inputs[0], param, config).node()->owner_opr();
 }
 
 OP_TRAIT_REG(NMSKeep, NMSKeep, NMSKeepOpr)

imperative/src/impl/ops/opr_attr.cpp

@@ -79,11 +79,13 @@ public:
 cg::OperatorNodeBase* apply_on_var_node(
         const OpDef& def, const VarNodeArray& inputs) {
     auto&& attr = def.cast_final_safe<OprAttr>();
+    auto config = attr.config;
+    config.name(attr.make_name());
     mgb_assert(!inputs.empty());
     auto registry = serialization::OprRegistry::find_by_name(attr.type);
     mgb_assert(registry, "operator %s not found", attr.type.c_str());
     OprParamsLoadContext ctx{attr.param, inputs[0]->owner_graph()};
-    return registry->loader(ctx, inputs, attr.config);
+    return registry->loader(ctx, inputs, config);
 }
 
 std::shared_ptr<OpDef> make_from_op_node(cg::OperatorNodeBase* opr) {
@@ -99,10 +101,15 @@ std::vector<std::pair<const char*, std::string>> props(const OpDef& def) {
     return {};
 }
 
+std::string make_name(const OpDef& def) {
+    return "OprAttr";
+}
+
 OP_TRAIT_REG(OprAttr, OprAttr)
     .make_from_op_node(make_from_op_node)
     .apply_on_var_node(apply_on_var_node)
     .props(props)
+    .make_name(make_name)
     .fallback();
 
 } // anonymous namespace

imperative/src/impl/ops/resize.cpp

@@ -24,7 +24,8 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& op = static_cast<const Resize&>(def);
     mgb_assert(inputs.size() == 2);
-    return opr::Resize::make(inputs[0], inputs[1], op.param());
+    OperatorNodeConfig config{op.make_name()};
+    return opr::Resize::make(inputs[0], inputs[1], op.param(), config);
 }
 
 OP_TRAIT_REG(Resize, Resize)

imperative/src/impl/ops/specializations.cpp

@@ -46,7 +46,8 @@ auto apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
     auto&& conv = static_cast<const Convolution&>(def);
-    return opr::Convolution::make(inputs[0], inputs[1], conv.param(), conv.policy());
+    OperatorNodeConfig config{conv.make_name()};
+    return opr::Convolution::make(inputs[0], inputs[1], conv.param(), conv.policy(), config);
 }
 
 OP_TRAIT_REG(Convolution, Convolution, opr::Convolution)
@@ -60,7 +61,7 @@ auto apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
     auto&& conv = static_cast<const ConvolutionBackwardData&>(def);
-    cg::OperatorNodeConfig config;
+    OperatorNodeConfig config{conv.make_name()};
     if (inputs.size() == 2) {
         return opr::ConvolutionBackwardData::make(inputs[0], inputs[1], conv.param(), conv.policy(), config);
     } else {
@@ -88,7 +89,8 @@ auto apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
     auto&& ds = static_cast<const Dimshuffle&>(def);
-    return opr::Dimshuffle::make(inputs[0], ds.pattern);
+    OperatorNodeConfig config{ds.make_name()};
+    return opr::Dimshuffle::make(inputs[0], ds.pattern, 0UL, config);
 }
 
 OP_TRAIT_REG(Dimshuffle, Dimshuffle, opr::Dimshuffle)
@@ -107,7 +109,8 @@ auto apply_on_var_node(
     for (auto&& i : add_axis.axis) {
         param.push_back(Desc::make_add(i));
     }
-    return opr::AxisAddRemove::make(inputs[0], param);
+    OperatorNodeConfig config{add_axis.make_name()};
+    return opr::AxisAddRemove::make(inputs[0], param, config);
 }
 
 OP_TRAIT_REG(AddAxis, AddAxis)
@@ -125,7 +128,8 @@ auto apply_on_var_node(
     for (auto&& i : remove_axis.axis) {
         param.push_back(Desc::make_remove(i));
     }
-    return opr::AxisAddRemove::make(inputs[0], param);
+    OperatorNodeConfig config{remove_axis.make_name()};
+    return opr::AxisAddRemove::make(inputs[0], param, config);
 }
 
 OP_TRAIT_REG(RemoveAxis, RemoveAxis)
@@ -138,7 +142,8 @@ auto apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
     auto&& topk = static_cast<const TopK&>(def);
-    return opr::TopK::make(inputs[0], inputs[1], topk.param())[0]
+    OperatorNodeConfig config{topk.make_name()};
+    return opr::TopK::make(inputs[0], inputs[1], topk.param(), config)[0]
             .node()->owner_opr();
 }
 
@@ -152,10 +157,12 @@ auto apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
     auto&& reduce = static_cast<const Reduce&>(def);
+    OperatorNodeConfig config{reduce.make_name()};
     if (inputs.size() > 1) {
-        return opr::Reduce::make(inputs[0], reduce.param(), inputs[1]);
+        return opr::Reduce::make(inputs[0], reduce.param(), inputs[1], config);
     } else {
-        return opr::Reduce::make(inputs[0], reduce.param());
+        return opr::Reduce::make(
+            inputs[0], reduce.param(), (cg::VarNode*)nullptr, config);
     }
 }
 
@@ -175,7 +182,8 @@ auto apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
     auto&& pool = static_cast<const AdaptivePooling&>(def);
-    return opr::AdaptivePooling::make(inputs[0], inputs[1], pool.param());
+    OperatorNodeConfig config{pool.make_name()};
+    return opr::AdaptivePooling::make(inputs[0], inputs[1], pool.param(), config);
 }
 
 OP_TRAIT_REG(AdaptivePooling, AdaptivePooling)
@@ -189,6 +197,7 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& conv = static_cast<const ConvBias&>(def);
     cg::OperatorNodeConfig config{conv.dtype};
+    config.name(conv.make_name());
     if (inputs.size() == 2) {
         return opr::ConvBias::make(inputs[0], inputs[1], conv.param(), conv.policy(), config);
     } else if (inputs.size() == 3) {
@@ -210,6 +219,7 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& conv = static_cast<const BatchConvBias&>(def);
     cg::OperatorNodeConfig config{conv.dtype};
+    config.name(conv.make_name());
     if (inputs.size() == 2) {
         return opr::BatchConvBias::make(inputs[0], inputs[1], conv.param(), conv.policy(), config);
     } else if (inputs.size() == 3) {
@@ -230,7 +240,8 @@ auto apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
     auto&& pool = static_cast<const Pooling&>(def);
-    return opr::Pooling::make(inputs[0], pool.param());
+    OperatorNodeConfig config{pool.make_name()};
+    return opr::Pooling::make(inputs[0], pool.param(), config);
 }
 OP_TRAIT_REG(Pooling, Pooling)
     .apply_on_var_node(apply_on_var_node)
@@ -243,8 +254,9 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& matmul = static_cast<const MatrixMul&>(def);
     mgb_assert(inputs.size() == 2);
+    OperatorNodeConfig config{matmul.make_name()};
     return opr::MatrixMul::make(inputs[0], inputs[1], matmul.param(),
-                                matmul.policy());
+                                matmul.policy(), config);
 }
 OP_TRAIT_REG(MatrixMul, MatrixMul)
     .apply_on_var_node(apply_on_var_node)
@@ -257,8 +269,9 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& matmul = static_cast<const BatchedMatrixMul&>(def);
     mgb_assert(inputs.size() == 2);
+    OperatorNodeConfig config{matmul.make_name()};
     return opr::BatchedMatrixMul::make(inputs[0], inputs[1], matmul.param(),
-                                       matmul.policy());
+                                       matmul.policy(), config);
 }
 OP_TRAIT_REG(BatchedMatrixMul, BatchedMatrixMul)
     .apply_on_var_node(apply_on_var_node)
@@ -267,10 +280,12 @@ OP_TRAIT_REG(BatchedMatrixMul, BatchedMatrixMul)
 
 namespace { namespace dot {
 auto apply_on_var_node(
-        const OpDef&,
+        const OpDef& def,
         const VarNodeArray& inputs) {
+    auto&& op = def.cast_final_safe<Dot>();
     mgb_assert(inputs.size() == 2);
-    return opr::Dot::make(inputs[0], inputs[1]);
+    OperatorNodeConfig config{op.make_name()};
+    return opr::Dot::make(inputs[0], inputs[1], config);
 }
 OP_TRAIT_REG(Dot, Dot)
     .apply_on_var_node(apply_on_var_node)
@@ -282,7 +297,8 @@ auto apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
     auto&& argsort = static_cast<const Argsort&>(def);
-    return opr::Argsort::make(inputs[0], argsort.param());
+    OperatorNodeConfig config{argsort.make_name()};
+    return opr::Argsort::make(inputs[0], argsort.param(), config);
 }
 OP_TRAIT_REG(Argsort, Argsort)
     .apply_on_var_node(apply_on_var_node)
@@ -294,7 +310,8 @@ auto apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
     auto&& argmax = static_cast<const Argmax&>(def);
-    return opr::Argmax::make(inputs[0], argmax.param());
+    OperatorNodeConfig config{argmax.make_name()};
+    return opr::Argmax::make(inputs[0], argmax.param(), config);
 }
 OP_TRAIT_REG(Argmax, Argmax)
     .apply_on_var_node(apply_on_var_node)
@@ -306,7 +323,8 @@ auto apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
     auto&& argmin = static_cast<const Argmin&>(def);
-    return opr::Argmin::make(inputs[0], argmin.param());
+    OperatorNodeConfig config{argmin.make_name()};
+    return opr::Argmin::make(inputs[0], argmin.param(), config);
 }
 OP_TRAIT_REG(Argmin, Argmin)
     .apply_on_var_node(apply_on_var_node)
@@ -318,11 +336,13 @@ auto apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
     auto&& warp = static_cast<const WarpPerspective&>(def);
+    OperatorNodeConfig config{warp.make_name()};
     if (inputs.size() == 3) {
-        return opr::WarpPerspective::make(inputs[0], inputs[1], inputs[2], warp.param());
+        return opr::WarpPerspective::make(inputs[0], inputs[1], inputs[2], warp.param(), config);
     } else {
         mgb_assert(inputs.size() == 4);
-        return opr::WarpPerspective::make(inputs[0], inputs[1], inputs[2], inputs[3], warp.param());
+        return opr::WarpPerspective::make(
+            inputs[0], inputs[1], inputs[2], inputs[3], warp.param(), config);
     }
 }
 OP_TRAIT_REG(WarpPerspective, WarpPerspective)
@@ -336,7 +356,8 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& local = static_cast<const GroupLocal&>(def);
     mgb_assert(inputs.size() == 2);
-    return opr::GroupLocal::make(inputs[0], inputs[1], local.param());
+    OperatorNodeConfig config{local.make_name()};
+    return opr::GroupLocal::make(inputs[0], inputs[1], local.param(), config);
 }
 OP_TRAIT_REG(GroupLocal, GroupLocal)
     .apply_on_var_node(apply_on_var_node)
@@ -349,7 +370,8 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& op = static_cast<const IndexingOneHot&>(def);
     mgb_assert(inputs.size() == 2);
-    return opr::IndexingOneHot::make(inputs[0], inputs[1], op.param());
+    OperatorNodeConfig config{op.make_name()};
+    return opr::IndexingOneHot::make(inputs[0], inputs[1], op.param(), config);
 }
 OP_TRAIT_REG(IndexingOneHot, IndexingOneHot)
     .apply_on_var_node(apply_on_var_node)
@@ -362,7 +384,8 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& op = static_cast<const IndexingSetOneHot&>(def);
     mgb_assert(inputs.size() == 3);
-    return opr::IndexingSetOneHot::make(inputs[0], inputs[1], inputs[2], op.param());
+    OperatorNodeConfig config{op.make_name()};
+    return opr::IndexingSetOneHot::make(inputs[0], inputs[1], inputs[2], op.param(), config);
 }
 OP_TRAIT_REG(IndexingSetOneHot, IndexingSetOneHot)
     .apply_on_var_node(apply_on_var_node)
@@ -375,7 +398,8 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& op = static_cast<const TypeCvt&>(def);
     mgb_assert(inputs.size() == 1);
-    return opr::TypeCvt::make(inputs[0], op.dtype);
+    OperatorNodeConfig config{op.make_name()};
+    return opr::TypeCvt::make(inputs[0], op.dtype, config);
 }
 OP_TRAIT_REG(TypeCvt, TypeCvt)
     .apply_on_var_node(apply_on_var_node)
@@ -388,6 +412,7 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& op = static_cast<const Concat&>(def);
     cg::OperatorNodeConfig config{op.comp_node};
+    config.name(op.make_name());
     return opr::Concat::make(inputs, op.axis, config);
 }
 OP_TRAIT_REG(Concat, Concat)
@@ -402,6 +427,7 @@ auto apply_on_var_node(
     auto&& op = static_cast<const Copy&>(def);
     mgb_assert(inputs.size() == 1);
     cg::OperatorNodeConfig config{op.comp_node};
+    config.name(op.make_name());
     return opr::Copy::make(inputs[0], config);
 }
 OP_TRAIT_REG(Copy, Copy)
@@ -411,10 +437,12 @@ OP_TRAIT_REG(Copy, Copy)
 
 namespace { namespace identity {
 auto apply_on_var_node(
-        const OpDef&,
+        const OpDef& def,
         const VarNodeArray& inputs) {
+    auto&& op = def.cast_final_safe<Identity>();
     mgb_assert(inputs.size() == 1);
-    return opr::Identity::make(inputs[0]);
+    OperatorNodeConfig config{op.make_name()};
+    return opr::Identity::make(inputs[0], config);
 }
 OP_TRAIT_REG(Identity, Identity)
     .apply_on_var_node(apply_on_var_node)
@@ -427,7 +455,8 @@ auto apply_on_var_node(
     const VarNodeArray& inputs) {
         auto&& op = static_cast<const AssertEqual&>(def);
         mgb_assert(inputs.size() == 2);
-        return opr::AssertEqual::make(inputs[0],inputs[1],op.param());
+        OperatorNodeConfig config{op.make_name()};
+        return opr::AssertEqual::make(inputs[0], inputs[1], op.param(), config);
 
     }
 
@@ -443,7 +472,8 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& op = static_cast<const UniformRNG&>(def);
     mgb_assert(inputs.size() == 1);
-    return opr::UniformRNG::make(inputs[0], op.param());
+    OperatorNodeConfig config{op.make_name()};
+    return opr::UniformRNG::make(inputs[0], op.param(), config);
 }
 OP_TRAIT_REG(UniformRNG, UniformRNG)
     .apply_on_var_node(apply_on_var_node)
@@ -456,7 +486,8 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& op = static_cast<const GaussianRNG&>(def);
     mgb_assert(inputs.size() == 1);
-    return opr::GaussianRNG::make(inputs[0], op.param());
+    OperatorNodeConfig config{op.make_name()};
+    return opr::GaussianRNG::make(inputs[0], op.param(), config);
 }
 OP_TRAIT_REG(GaussianRNG, GaussianRNG)
     .apply_on_var_node(apply_on_var_node)
@@ -469,7 +500,9 @@ VarNodeArray apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& op = static_cast<const ROIAlign&>(def);
     mgb_assert(inputs.size() == 2);
-    auto* opr = opr::ROIAlign::make(inputs[0], inputs[1], op.param()).node()->owner_opr();
+    OperatorNodeConfig config{op.make_name()};
+    auto* opr = opr::ROIAlign::make(
+        inputs[0], inputs[1], op.param(), config).node()->owner_opr();
     return {opr->output(0), opr->output(1)};
 }
 OP_TRAIT_REG(ROIAlign, ROIAlign)
@@ -484,7 +517,8 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& op = static_cast<const NvOf&>(def);
     mgb_assert(inputs.size() == 1);
-    return opr::NvOf::make(inputs[0], op.param());
+    OperatorNodeConfig config{op.make_name()};
+    return opr::NvOf::make(inputs[0], op.param(), config);
 }
 OP_TRAIT_REG(NvOf, NvOf)
     .apply_on_var_node(apply_on_var_node)
@@ -499,6 +533,7 @@ auto apply_on_var_node(
     auto&& op = static_cast<const Linspace&>(def);
     mgb_assert(inputs.size() == 3);
     cg::OperatorNodeConfig config{op.comp_node};
+    config.name(op.make_name());
     return opr::Linspace::make(inputs[0], inputs[1], inputs[2], op.param(), config);
 }
 OP_TRAIT_REG(Linspace, Linspace)
@@ -513,6 +548,7 @@ auto apply_on_var_node(
     auto&& op = static_cast<const Eye&>(def);
     mgb_assert(inputs.size() == 1);
     cg::OperatorNodeConfig config{op.comp_node};
+    config.name(op.make_name());
     opr::Eye::Param param{op.k, op.dtype.enumv()};
     return opr::Eye::make(inputs[0], param, config);
 }
@@ -527,7 +563,10 @@ VarNodeArray apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& op = static_cast<const ROIPooling&>(def);
     mgb_assert(inputs.size() == 3);
-    auto* opr = opr::ROIPooling::make(inputs[0], inputs[1], inputs[2], op.param()).node()->owner_opr();
+    OperatorNodeConfig config{op.make_name()};
+    auto* opr = opr::ROIPooling::make(
+        inputs[0], inputs[1], inputs[2], op.param(), config
+    ).node()->owner_opr();
     return {opr->output(0), opr->output(1)};
 }
 OP_TRAIT_REG(ROIPooling, ROIPooling)
@@ -541,7 +580,8 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& op = static_cast<const Remap&>(def);
     mgb_assert(inputs.size() == 2);
-    return opr::Remap::make(inputs[0], inputs[1], op.param());
+    OperatorNodeConfig config{op.make_name()};
+    return opr::Remap::make(inputs[0], inputs[1], op.param(), config);
 }
 OP_TRAIT_REG(Remap, Remap)
     .apply_on_var_node(apply_on_var_node)
@@ -578,7 +618,8 @@ auto apply_on_var_node( \
         const OpDef& def, \
         const VarNodeArray& inputs) { \
     auto&& op = static_cast<const NAME&>(def); \
-    return opr::NAME::make(IN##NR_INPUT, get_index(inputs, NR_INPUT, op.items)); \
+    OperatorNodeConfig config{op.make_name()}; \
+    return opr::NAME::make(IN##NR_INPUT, get_index(inputs, NR_INPUT, op.items), config); \
 } \
 OP_TRAIT_REG(NAME, NAME) \
     .apply_on_var_node(apply_on_var_node) \
@@ -609,30 +650,35 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& op = static_cast<const FakeQuant&>(def);
     mgb_assert(inputs.size() == 3);
-    return opr::FakeQuant::make(inputs[0], inputs[1], inputs[2], op.param());
+    OperatorNodeConfig config{op.make_name()};
+    return opr::FakeQuant::make(inputs[0], inputs[1], inputs[2], op.param(), config);
 }
 OP_TRAIT_REG(FakeQuant, FakeQuant)
     .apply_on_var_node(apply_on_var_node)
     .fallback();
 }} // fake_quant
+
 namespace { namespace tqt {
 auto apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
     auto&& op = static_cast<const TQT&>(def);
     mgb_assert(inputs.size() == 2);
-    return opr::TQT::make(inputs[0], inputs[1], op.param());
+    OperatorNodeConfig config{op.make_name()};
+    return opr::TQT::make(inputs[0], inputs[1], op.param(), config);
 }
 OP_TRAIT_REG(TQT, TQT)
     .apply_on_var_node(apply_on_var_node)
     .fallback();
 }}  // tqt
+
 namespace { namespace elemwise_multi_type {
 auto apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
     auto&& op = static_cast<const ElemwiseMultiType&>(def);
     OperatorNodeConfig config{op.dtype};
+    config.name(op.make_name());
     return opr::ElemwiseMultiType::make(inputs, op.param(), config);
 }
 OP_TRAIT_REG(ElemwiseMultiType, ElemwiseMultiType)
@@ -646,7 +692,9 @@ auto apply_on_var_node(
         const VarNodeArray& inputs) {
     auto&& op = static_cast<const SVD&>(def);
     mgb_assert(inputs.size() == 1);
-    return opr::SVD::make(inputs[0], op.param())[0].node()->owner_opr()->usable_output();
+    OperatorNodeConfig config{op.make_name()};
+    return opr::SVD::make(inputs[0], op.param(), config)[0]
+        .node()->owner_opr()->usable_output();
 }
 OP_TRAIT_REG(SVD, SVD)
     .apply_on_var_node(apply_on_var_node)

imperative/src/impl/ops/tensor_manip.cpp

@@ -21,7 +21,8 @@ cg::OperatorNodeBase* apply_on_var_node(
         const OpDef& def,
         const VarNodeArray& inputs) {
     auto&& op_def = def.cast_final_safe<GetVarShape>();
-    return opr::GetVarShape::make(inputs, op_def.param()).node()->owner_opr();
+    OperatorNodeConfig config{op_def.make_name()};
+    return opr::GetVarShape::make(inputs, op_def.param(), config).node()->owner_opr();
 }
 
 DispatchMode decide_dispatch_mode(
@@ -152,7 +153,7 @@ cg::OperatorNodeBase* param_pack_split_apply_on_var_node(
     auto&& graph = inputs[0]->owner_graph();
 
     auto&& shapes = get_shapes(param.shapes);
-    cg::OperatorNodeConfig config;
+    OperatorNodeConfig config(param.make_name());
     cg::OperatorNodeBase* opr =
             graph->insert_opr(std::make_unique<mgb::opr::ParamPackSplit>(
                     inputs[0], param.offsets, shapes, config));
@@ -189,7 +190,7 @@ cg::OperatorNodeBase* param_pack_concat_apply_on_var_node(
     auto&& graph = inputs[0]->owner_graph();
 
     VarNodeArray inps(inputs.begin(), inputs.end() - 1);
-    cg::OperatorNodeConfig config;
+    OperatorNodeConfig config{param.make_name()};
     cg::OperatorNodeBase* opr =
             graph->insert_opr(std::make_unique<mgb::opr::ParamPackConcat>(
                     inps, inputs.back(), param.offsets, config));

imperative/src/impl/ops/tensorrt_runtime.cpp

@@ -20,8 +20,9 @@ namespace { namespace tensorrt_runtime {
             const OpDef& def,
             const VarNodeArray& inputs) {
         auto&& op = static_cast<const TensorRTRuntime&>(def);
+        OperatorNodeConfig config{op.make_name()};
         SymbolVarArray sinputs(inputs.begin(), inputs.end());
-        return opr::TensorRTRuntimeOpr::make(op.buf.c_str(), op.buf_size, sinputs);
+        return opr::TensorRTRuntimeOpr::make(op.buf.c_str(), op.buf_size, sinputs, config);
     }
 OP_TRAIT_REG(TensorRTRuntime, TensorRTRuntime)
     .apply_on_var_node(apply_on_var_node)

imperative/src/impl/ops/warp_affine.cpp

@@ -21,7 +21,8 @@ namespace { namespace warp_affine {
             const VarNodeArray& inputs) {
         mgb_assert(inputs.size() == 3);
         auto&& op = static_cast<const WarpAffine&>(def);
-        return opr::WarpAffine::make(inputs[0], inputs[1], inputs[2], op.param());
+        OperatorNodeConfig config{op.make_name()};
+        return opr::WarpAffine::make(inputs[0], inputs[1], inputs[2], op.param(), config);
     }
 
 OP_TRAIT_REG(WarpAffine, WarpAffine)

imperative/src/include/megbrain/imperative/op_def.h

@@ -36,6 +36,7 @@ class OpDef : public Hashable,
               public NonCopyableObj,
               public std::enable_shared_from_this<OpDef> {
     mutable const OpTrait* m_trait = nullptr;
+    std::string m_scope;
 public:
     virtual ~OpDef() = default;
 
@@ -86,10 +87,14 @@ public:
 
     const OpTrait* trait() const;
 
-    const char* name() const;
-
     std::string to_string() const;
 
+    const std::string scope() const;
+
+    const std::string make_name() const;
+
+    void set_scope(const std::string& scope);
+
     virtual size_t hash() const;
 
     virtual bool is_same_st(const Hashable&) const;

imperative/tablegen/autogen.cpp

@@ -113,9 +113,10 @@ static void gen_op_def_c_header_single(raw_ostream &os, MgbOp& op) {
                 "{0}({0}_)", i.name
             ));
         }
+        paramList.push_back("std::string scope_ = {}");
         gen_ctor(llvm::join(paramList, ", "),
                  ": " + llvm::join(initList, ", "),
-                 " {}");
+                 " { set_scope(scope_); }");
     }
 
     auto packedParams = op.getPackedParams();
@@ -236,11 +237,19 @@ static void gen_op_def_c_body_single(raw_ostream &os, MgbOp& op) {
         os << mlir::tblgen::tgfmt(hashable->getPropsFunctionTemplate(), &ctx);
         os << "}\n";
 
+        // generate make_name()
+        os << formatv(
+            "std::string {0}(const OpDef& def_) {{\n", formatMethImpl("make_name")
+        );
+        os << mlir::tblgen::tgfmt(hashable->getNameFunctionTemplate(), &ctx);
+        os << "}\n";
+
         os << "} // anonymous namespace\n";
 
         methods.push_back("hash");
         methods.push_back("is_same_st");
         methods.push_back("props");
+        methods.push_back("make_name");
     }
     if (!methods.empty()) {
         os << formatv(
@@ -327,7 +336,7 @@ static void gen_op_def_pybind11_single(raw_ostream &os, MgbOp& op, EnumContext&
             targs.push_back(i.attr.getReturnType());
         }
         os << llvm::join(targs, ", ");
-        os << ">()";
+        os << ", std::string>()";
         for (auto &&i : op.getMgbAttributes()) {
             os << formatv(", py::arg(\"{0}\")", i.name);
             auto defaultValue = i.attr.getDefaultValue();
@@ -337,7 +346,7 @@ static void gen_op_def_pybind11_single(raw_ostream &os, MgbOp& op, EnumContext&
                 hasDefaultCtor = true;
             }
         }
-        os << ")";
+        os << ", py::arg(\"scope\") = {})";
     }
     if (hasDefaultCtor) {
         os << "\n    .def(py::init<>())";
@@ -442,6 +451,10 @@ EnumWrapper<{0}::{1}>::type2str = {{
             className, i.name));
     }
 
+    getsetters.push_back(formatv(
+        "{{\"scope\", py_get_scope({0}), py_set_scope({0}), \"scope\", NULL},",
+        className));
+
     // generate tp_init
     std::string initBody;
     if (!op.getMgbAttributes().empty()) {
@@ -449,6 +462,7 @@ EnumWrapper<{0}::{1}>::type2str = {{
         llvm::for_each(op.getMgbAttributes(), [&](auto&& attr) {
             initBody += formatv("\"{0}\", ", attr.name);
         });
+        initBody += "\"scope\", ";
         initBody += "NULL};\n";
         initBody += "    PyObject ";
         std::vector<std::string> attrs;
@@ -456,12 +470,15 @@ EnumWrapper<{0}::{1}>::type2str = {{
             attrs.push_back(formatv("*{0} = NULL", attr.name));
         });
         initBody += llvm::join(attrs, ", ") + ";\n";
+        initBody += "    PyObject *scope = NULL;\n";
         initBody += "    if (!PyArg_ParseTupleAndKeywords(args, kwds, \"|";
-        initBody += std::string(op.getMgbAttributes().size(), 'O');
+        // an extra slot created for name
+        initBody += std::string(op.getMgbAttributes().size() + 1, 'O');
         initBody += "\", const_cast<char**>(kwlist)";
         llvm::for_each(op.getMgbAttributes(), [&](auto&& attr) {
-            initBody += formatv(" ,&{0}", attr.name);
+            initBody += formatv(", &{0}", attr.name);
         });
+        initBody += ", &scope";
         initBody += "))\n";
         initBody += "    return -1;\n";
         llvm::for_each(op.getMgbAttributes(), [&](auto&& attr) {
@@ -483,6 +500,25 @@ EnumWrapper<{0}::{1}>::type2str = {{
     }
 )", className, attr.name);
         });
+
+        initBody += formatv(R"(
+    if (scope) {{
+        try {{
+            reinterpret_cast<PyOp({0})*>(self)->inst().set_scope(
+                pyobj_convert_generic<std::string>::from(scope));
+        } catch(py::error_already_set& e) {{
+            e.restore();
+            return -1;
+        } catch(py::builtin_exception& e) {{
+            e.set_error();
+            return -1;
+        } catch(...) {{
+            PyErr_SetString(PyExc_RuntimeError, "Unknown Error");
+            return -1;
+        }
+    }
+)", className);
+
     }
     initBody += "\n    return 0;";
 

imperative/tablegen/helper.h

@@ -241,6 +241,30 @@ private:
         body += "    return props_;\n";
         return body;
     }
+    std::string getModeName() const {
+        std::string body = formatv(
+            "    auto&& op_ = def_.cast_final_safe<{0}>();\n"
+            "    static_cast<void>(op_);\n",
+            getCppClassName()
+        );
+        for (auto&& it : getMgbAttributes()) {
+            if (it.name == "mode") {
+                auto* enumAttr = llvm::dyn_cast<MgbEnumAttrMixin>(&it.attr);
+                body += "    switch (op_.mode){\n";
+                for (auto&& enumMember: enumAttr->getEnumMembers()) {
+                    body += formatv(
+                        "        case {0}::{1}::{2}:\n",
+                        getCppClassName(), enumAttr->getEnumName(), enumMember
+                    );
+                    body += formatv("            return \"{0}\";\n", enumMember);
+                }
+                body += formatv(
+                    "        default: return \"{0}::Unknown\";\n", getCppClassName());
+                body += "    }\n";
+            }
+        }
+        return body;
+    }
 public:
     static bool classof(const Operator* op) {
         return op->getDef().isSubClassOf("MgbHashableOpMixin");
@@ -264,6 +288,12 @@ public:
         }
         return getDefaultPropsFunction();
     }
+    std::string getNameFunctionTemplate() const {
+        if (getDef().getValueAsBit("usingModeName")) {
+            return getModeName();
+        }
+        return formatv("    return \"{0}\";\n", getCppClassName());
+    }
 };
 
 } // namespace tblgen

sdk/load-and-run/dump_with_testcase_mge.py

@@ -476,6 +476,7 @@ def main():
 
     output_mgbvars = feeds["outputs"]
     output_mgbvars = optimize_for_inference(args, output_mgbvars)
+    output_mgbvars = [var._node for var in output_mgbvars]
 
     inputs = cgtools.get_dep_vars(output_mgbvars, "Host2DeviceCopy")
     inputs = sorted((i.name, i.dtype) for i in inputs)

src/core/include/megbrain/ir/base.td

@@ -242,6 +242,7 @@ class MgbPackedParamBase<string className, string accessor>:
 class MgbHashableOpMixin {
   string hashFunction = ?;
   string cmpFunction = ?;
+  bit usingModeName = 0;
 }
 
 class MgbOp<string mnemonic, list<MgbParamBase> params=[], list<OpTrait> traits=[]>:

src/core/include/megbrain/ir/ops.td

@@ -21,6 +21,7 @@ include "mlir/Interfaces/SideEffectInterfaces.td"
 def Elemwise : MgbHashableOp<"Elemwise", [ElemwiseParam], [NoSideEffect]> {
   let inputs = (ins Variadic<AnyType>:$input);
   let results = (outs AnyType);
+  let usingModeName = 1;
 }
 
 def Reduce: MgbHashableOp<"Reduce", [ReduceParam]>;
@@ -247,6 +248,7 @@ def ElemwiseMultiType: MgbHashableOp<"ElemwiseMultiType", [ElemwiseMultiTypePara
   let extraArguments = (ins
     MgbDTypeAttr:$dtype
   );
+  let usingModeName = 1;
 }
 
 def InplaceAdd: MgbHashableOp<"InplaceAdd", [EmptyParam]>;