Myia 流程分析

[leozp]

parser.py

block

attributes

• parser : Parser
• matured : bool
• variable : Dict[str, ANFNode]
• preds: List[Block]
• phi_nodes : Dict[Parameter, str]
• jumps : Dict[Block, Apply]
• graph : Graph

  method

• set_phi_arguments
• mature
• _resolve_ast_type
• make_resolve
• operation
• force_bool
• read
• write
• jump

• cond

  parser

  attributes

• function
• line_offset
• filename
• global_namespace
• closure_namespace
• graph

  method

• make_location
• make_condition_blocks
• parse
• process_FunctionDef
• _process_function
• process_node
• process_BinOp
• process_UnaryOp
• process_Compare
• process_BoolOp
• process_IfExp
• process_Lambda
• process_Name

[leozp]

wrap_primitives

Pipeline step to wrap primitives in non-call positions into graphs.

在trainsforms.py中定义。 从处理过程上看类似于剪枝，将Graph中与结果计算无关的constant node进行封装，使其无法调用

    step_wrap_primitives = WrapPrimitives.partial()

    Inputs:
        graph: A graph

    Outputs:
        graph: The transformed graph

• 功能：独立功能模块 WrapPrimitives
• 流程：遍历常量Node，将使用常量Node的Graph进行封装，并set_edge设置

[leozp]

compile

将图集转换为指令列表

    step_compile = CompileGraphs.partial(linear_impl='nnvm', target='cpu', dev_id=0)

    Inputs:
        graph: A graph

    Outputs:
        mapping: map each graph to its starting position in the code list.
        uinstrs: list of unlinked instructions for all the graphs in
                 the cluster, starting with the passed-in graph.

graph_transform转换具体参数和操作配置如下，原形实现中调用TVM进行图编译。

    graph_transform = PipelineDefinition(
        resources=dict(
            lin_convert=nnvm_convert,
            target='cpu',
            dev_id=0,
        ),
        steps=dict(
            split=SplitGraph.partial(),
            compile=CompileGraph.partial(),
            optimize=OptimizeInstrs.partial(),
        )
    )

编译操作过程：

• 1、SplitGraph 将图切分为线性部分和控制流
• 2、CompileGraph 将切分后的图转换为线性指令序列
• 3、OptimizeInstrs 进行指令优化，目前暂未实现

SplitGraph

class SplitGraph(PipelineStep):
    """Pipeline step to cut the graph into linear portions and control flow.

    Inputs:
        graph: A graph

    Outputs:
        splits: list of graph portions

    """

处理流程：

• 只处理apply节点
• apply节点且input不为常量，或者input为return，partial，switch，make_tuple之一，进行切分
• 切分点集，保存到splits列表

CompileGraph

class CompileGraph(PipelineStep):
    """Step to convert splits into linear instruction flow.

    Inputs:
        graph: A graph
        splits: list of graph portions

    Outputs:
        uinstrs: list of instructions for the graph (unlinked)

    """

[leozp]

link

完成指令序列的链接

    step_link = LinkInstrs.partial()

LinkInstrs

class LinkInstrs(PipelineStep):
    """Link unlinked instructions.

    Inputs:
        mapping: graph map
        uinstrs: unlinked instructions

    Outputs:
        instrs: linked instructions

    """

    def step(self, mapping, uinstrs):
        """Link instructions."""
        for i in range(len(uinstrs)):
            instr = uinstrs[i]
            if instr[0] == 'push_graph':
                uinstrs[i] = ('push', mapping[instr[1]])

        return {'instrs': uinstrs}

处理流程：

• 依次遍历uinstrs指令序列
• 若第一个指令为"push_graph"，则将随后一个指令替换为"push"及mapping中对应指令
• 遍历完成后实现指令链接

[leozp]

export

生成可调用的指令集，并运行

    step_export = VMExporter.partial()

[Olivert007]

博主，您好，咨询一下，https://github.com/leozp/Myia-Issues/wiki/Code-Review-%E2%80%94%E2%80%94-Myia-Compile，文章中的流程图是怎么生成的