MegEngine/imperative/python/test/unit/traced_module/test_modification.py

# 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 pickle
from collections import defaultdict
from itertools import chain

import numpy as np

import megengine.functional as F
import megengine.module as M
import megengine.module.qat as qat
from megengine.module.identity import Identity
from megengine.traced_module import trace_module
from megengine.traced_module.expr import CallFunction, CallMethod, Expr, GetAttr, Input
from megengine.traced_module.node import ModuleNode, Node, TensorNode


class IdentityMod(M.Module):
    def forward(self, x):
        return x


class MyBlock(M.Module):
    def __init__(self, in_channels=3, channels=3):
        super(MyBlock, self).__init__()
        self.conv1 = M.Conv2d(in_channels, channels, 3, 1, padding=1, bias=False)
        self.bn1 = M.BatchNorm2d(channels)
        self.nothing = IdentityMod()

    def forward(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = F.relu(x) + 1
        x = self.nothing(x)
        return x


class MyModule(M.Module):
    def __init__(self):
        super(MyModule, self).__init__()
        self.block0 = MyBlock()
        self.block1 = MyBlock()
        self.nothing = IdentityMod()

    def forward(self, x):
        x = self.block0(x)
        x = self.block1(x)
        x = self.nothing(x)
        return x


class MyBlock1(M.Module):
    def forward(self, a):
        y = F.concat([a, a])
        return a, y


class MyModule1(M.Module):
    def __init__(self):
        super().__init__()
        self.block0 = MyBlock1()
        self.block1 = MyBlock1()

    def forward(self, a):
        a, y1 = self.block0(a)
        a = a + 1
        a, y2 = self.block1(a)
        return a, y1 + y2


class NewModule(M.Module):
    def __init__(self, traced_module):
        super(NewModule, self).__init__()
        self.module = traced_module

    def forward(self, x):
        x = x - 1
        x = self.module(x)
        x = x + 1
        return x


def _check_expr_users(flattened_module):
    node_user = defaultdict(list)
    for expr in flattened_module.graph._exprs:
        for node in expr.inputs:
            node_user[node].append(expr)
    for node in flattened_module.graph.nodes():
        node.users.sort(key=lambda m: m._id)
        node_user[node].sort(key=lambda m: m._id)
        assert node.users == node_user[node]


def _init_cls(cls):
    module = cls()
    x = F.ones((1, 3, 3, 3))
    y = module(x)
    traced_module = trace_module(module, x)
    return traced_module, x, y


def _init_block():
    return _init_cls(MyBlock)


def _init_module():
    return _init_cls(MyModule)


def test_search():
    traced_module, *_ = _init_block()
    graph = traced_module.graph
    relu_expr = graph.get_function_by_type(F.relu).as_unique()
    assert isinstance(relu_expr, CallFunction) and relu_expr.func == F.relu

    conv_node = graph.get_module_by_type(M.Conv2d).as_unique()
    assert isinstance(conv_node, ModuleNode) and conv_node.module_type == M.Conv2d

    add_expr = graph.get_method_by_type("__add__").as_unique()
    assert isinstance(add_expr, CallMethod) and add_expr.method == "__add__"

    conv_node = graph.get_node_by_name("MyBlock_conv1").as_unique()
    assert isinstance(conv_node, ModuleNode) and conv_node.module_type == M.Conv2d


def test_producer_and_users():
    traced_module, *_ = _init_module()

    def _check(exprs):
        for expr in exprs:
            for n in chain(expr.inputs, expr.outputs):
                if not isinstance(n.expr, Input):
                    assert n.expr in exprs
                for e in n.users:
                    assert e in exprs
                    assert n in e.inputs

    for mod in traced_module.modules():
        if not hasattr(mod, "argdef_graph_map"):
            continue
        for g in mod.argdef_graph_map.values():
            _check(g._exprs)


def test_insert():
    traced_module, x, expect = _init_block()
    graph = traced_module.graph
    relu_out = graph.get_function_by_type(F.relu).as_unique().outputs[0]
    with graph.insert_exprs():
        neg_out = F.neg(relu_out)
    graph.replace_node({relu_out: neg_out})
    graph.compile()
    np.testing.assert_allclose(expect - 1, 1 - traced_module(x), atol=1e-6)


def test_insert_module():
    class Neg(M.Module):
        def __init__(self, name):
            super().__init__(name)
            self.identity = M.Identity()
            self.identity_list = [M.Identity(), M.Identity()]
            self.identity_dict = {"0": M.Identity(), "1": M.Identity()}
            self.param = F.zeros((1,))

        def forward(self, x):
            x = self.identity(x)
            for m in self.identity_dict:
                x = self.identity_dict[m](x)
            for m in self.identity_list:
                x = m(x)
            return F.neg(x) + self.param

    traced_module, x, expect = _init_block()
    graph = traced_module.graph
    relu_out = graph.get_function_by_type(F.relu).as_unique().outputs[0]
    self = graph.inputs[0]
    setattr(traced_module, "neg", Neg(name="neg"))
    setattr(traced_module, "neg2", Neg(name="neg"))
    setattr(traced_module, "param", F.zeros((1,)))

    with graph.insert_exprs():
        neg_out = self.neg(relu_out)
        neg_out = self.neg2(relu_out)
        neg_out = neg_out + self.param
    graph.replace_node({relu_out: neg_out})
    graph.compile()

    np.testing.assert_allclose(expect - 1, 1 - traced_module(x), atol=1e-6)
    assert traced_module.neg.graph is not None
    assert traced_module.neg2.graph is not None
    assert traced_module.neg2.param is not None
    assert len(traced_module.neg.graph._exprs) == 13
    for n in traced_module.graph.nodes():
        if isinstance(n, TensorNode):
            assert n.value is None


def test_insert_qat_module():
    class concat(qat.Concat):
        pass

    traced_module, x, expect = _init_block()
    graph = traced_module.graph
    self = graph.inputs[0]
    out = graph.outputs[0]
    setattr(traced_module, "cat_0", qat.Concat())
    setattr(traced_module, "cat_1", concat())

    with graph.insert_exprs():
        x_0 = self.cat_0([out, out])
        x_1 = self.cat_1([out, x_0])
    graph.replace_node({out: x_1})
    graph.compile()

    x = F.copy(x)
    np.testing.assert_allclose(
        F.concat([expect, expect, expect]), traced_module(x), atol=1e-6
    )
    assert not hasattr(traced_module.cat_0, "graph")
    assert traced_module.cat_1.graph is not None


def test_add_input_and_output():
    traced_module, x, y = _init_module()

    data_node = traced_module.graph.add_input_node(shape=(1, 3, 224, 224), name="data")
    traced_module.graph.add_output_node(data_node)

    assert data_node.name == "data"
    assert traced_module.graph.inputs[-1] == data_node
    assert len(traced_module.graph.inputs) == 3
    assert len(traced_module.graph.outputs) == 2

    y1, y2 = traced_module(x, x)
    np.testing.assert_equal(y1.numpy(), y.numpy())
    np.testing.assert_equal(y2.numpy(), x.numpy())

    y1, y2 = traced_module(x, y)
    np.testing.assert_equal(y2.numpy(), y.numpy())

    traced_module.graph.reset_outputs(
        ({"orig_out": traced_module.graph.outputs[0]}, traced_module.graph.outputs[1])
    )

    out = traced_module(x, x)
    assert isinstance(out, tuple)
    assert isinstance(out[0], dict)
    np.testing.assert_equal(out[0]["orig_out"].numpy(), y.numpy())
    np.testing.assert_equal(out[1].numpy(), x.numpy())


def test_delete():
    traced_module, x, expect = _init_block()
    graph = traced_module.graph
    relu_expr = graph.get_function_by_type(F.relu).as_unique()
    node = relu_expr.outputs
    repl_node = relu_expr.inputs
    graph.replace_node({node[0]: repl_node[0]})
    graph.compile()
    np.testing.assert_allclose(expect - 1, F.relu(traced_module(x) - 1), atol=1e-6)

    # clear graph
    graph.replace_node({graph.outputs[0]: graph.inputs[1]})
    graph.compile()
    np.testing.assert_equal(len(list(graph._exprs)), 0)
    np.testing.assert_equal(traced_module(x).numpy(), x.numpy())


def test_flatten():
    traced_module, x, expect = _init_module()
    traced_module = traced_module.flatten()
    assert len(traced_module.graph._exprs) == 12
    np.testing.assert_equal(expect.numpy(), traced_module(x).numpy())

    traced_module = traced_module.flatten()
    assert len(traced_module.graph._exprs) == 12
    np.testing.assert_equal(expect.numpy(), traced_module(x).numpy())

    traced_module, x, expect = _init_cls(MyModule1)
    traced_module = traced_module.flatten()
    _check_expr_users(traced_module)


def test_id_and_name():
    def _check_id(traced_module):
        _total_ids = traced_module.graph._total_ids
        node_ids = [n._id for n in traced_module.graph.nodes().as_list()]
        assert len(set(node_ids)) == len(node_ids)
        assert max(node_ids) + 1 == _total_ids[0]

        expr_ids = [n._id for n in traced_module.graph.exprs().as_list()]
        assert len(set(expr_ids)) == len(expr_ids)
        assert max(expr_ids) + 1 == _total_ids[1]

    def _check_name(flatened_module):
        node_names = [n._name for n in flatened_module.graph.nodes().as_list()]
        assert len(set(node_names)) == len(node_names)

    traced_module, x, expect = _init_module()
    _check_id(traced_module)

    flattened_module = traced_module.flatten()
    _check_id(flattened_module)
    _check_name(flattened_module)

    # pickle check
    obj = pickle.dumps(traced_module)
    traced_module = pickle.loads(obj)
    Node._set_next_id(159)
    Expr._set_next_id(1024)

    graph = traced_module.graph
    for expr in graph.get_function_by_type(F.relu).as_list():
        relu_out = expr.outputs[0]
        cur_graph = expr.top_graph
        with cur_graph.insert_exprs():
            neg_out = F.neg(relu_out)
        cur_graph.replace_node({relu_out: neg_out})
        cur_graph.compile()
    _check_id(traced_module)

    flattened_module = traced_module.flatten()
    _check_id(flattened_module)
    _check_name(flattened_module)

    # check trace TracedModule
    obj = pickle.dumps(traced_module)
    traced_module = pickle.loads(obj)
    module = NewModule(traced_module)
    traced_module = trace_module(module, x)
    _check_id(traced_module)

    flattened_module = traced_module.flatten()
    _check_id(flattened_module)
    _check_name(flattened_module)


def test_set_node_name():
    traced_module, x, expect = _init_module()
    graph = traced_module.graph
    output_node = graph.outputs[0]

    def rename(name):
        output_node.name = name

    np.testing.assert_raises(AssertionError, rename, "block1_out")
    rename("output")
    np.testing.assert_equal(str(graph.outputs[0]), "output")


def test_set_graph_name():
    traced_module, x, expect = _init_module()
    graph = traced_module.graph
    output_node = graph.outputs[0]

    node_name = output_node.name

    graph.name = "Top"
    node = graph.get_node_by_name("{}_{}".format("Top", node_name)).as_unique()
    assert node is output_node


def test_extra_block():
    class PostProcess(M.Module):
        def forward(self, x):
            return x * 2

    class Net(M.Module):
        def __init__(self, traced_module):
            super().__init__()
            self.post_process = PostProcess()
            self.traced_module = traced_module

        def forward(self, x):
            x = self.traced_module(x)
            x = self.post_process(x)
            return x

    traced_module, x, expect = _init_block()
    module = Net(traced_module)
    np.testing.assert_allclose(2 * expect, module(x), atol=1e-6)
    traced_module = trace_module(module, x)
    np.testing.assert_allclose(2 * expect, traced_module(x), atol=1e-6)