mindspore 2.4.10__cp39-cp39-win_amd64.whl → 2.6.0rc1__cp39-cp39-win_amd64.whl

mindspore/ops_generate/pyboost/pyboost_functions_cpp_generator.py

@@ -0,0 +1,407 @@
+# Copyright 2024 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""
+This module defines the PyboostFunctionsGenerator class for generating C++ functions for PyBoost operations.
+
+The generator processes operator prototypes and constructs the necessary function definitions, including
+conversions for optional parameters and tensor arguments. It generates the registration code and includes
+the necessary header files for the generated functions.
+"""
+
+import os
+
+import common.template as template
+import common.gen_constants as K
+from common.template import Template
+from common.gen_utils import save_file
+from common.op_proto import OpProto
+from common.base_generator import BaseGenerator
+from pyboost import pyboost_utils
+from pyboost.pyboost_utils import get_convert_type_str, is_optional_param, is_op_multi_output, get_input_args_type_str, is_tensor_list
+
+from .op_template_parser import OpTemplateParser
+
+
+class PyboostFunctionsGenerator(BaseGenerator):
+    """
+    Generates PyBoost functions based on operator prototypes.
+
+    This class processes operator prototypes (`op_protos`) to create the necessary C++ function definitions for
+    PyBoost operations. It constructs function bodies, handles optional value conversions, and generates
+    registration code and header inclusions.
+    """
+
+    def __init__(self):
+        """Initializes the PyboostFunctionsGenerator with the necessary templates."""
+        self.pyboost_func_include_header_template = Template(
+            f'#include "{K.MS_PYBOOST_BASE_PATH}/auto_generate/${{operator_name}}.h"\n'
+        )
+        self.convert_optional_to_value_template = Template(
+            "auto ${output} = PyNativeAlgo::PyBoost::OptionalToValue(${input});\n"
+        )
+        self.convert_to_tensor_template = Template(
+            'auto ${output} = PyNativeAlgo::Common::ConvertStubNodeToTensor(${input}, ${need_contiguous}, '
+            'op_run_info->requires_grad);\n'
+        )
+        self.convert_to_tensor_list_template = Template(
+            'auto ${output} = PyNativeAlgo::Common::ConvertStubNodeToValueTuple(${input}, ${need_contiguous}, '
+            'op_run_info->requires_grad);\n'
+        )
+        self.convert_template = Template("auto $arg_name = converter.${convert_func}(args, $arg_index);\n")
+        self.input_args_template = Template(" const ${arg_type}& ${arg_name},")
+        self.PYBOOST_FUNCTION_TEMPLATE = template.PYBOOST_FUNCTION_TEMPLATE
+        self.PYBOOST_COMM_FUNCTION_TEMPLATE = template.PYBOOST_COMM_FUNCTION_TEMPLATE
+        self.PYBOOST_FUNCTION_DYNAMIC_OUTPUT_TEMPLATE = template.PYBOOST_FUNCTION_DYNAMIC_OUTPUT_TEMPLATE
+        self.REGISTER_DEFINE_TEMPLATE = template.REGISTER_DEFINE_TEMPLATE
+        self.REGISTER_TEMPLATE = template.REGISTER_TEMPLATE
+        self.PYBOOST_HEADER_TEMPLATE = template.PYBOOST_FUNCTIONS_CC_TEMPLATE
+        self.TENSOR_FUNC_CLASS_REG = template.TENSOR_FUNC_CLASS_REG
+        self.OP_DEF_INC_HEAD_TEMPLATE = template.OP_DEF_INC_HEAD_TEMPLATE
+
+    def generate(self, work_path, op_protos):
+        """
+        Generates the C++ PyBoost functions and writes them to the specified files.
+
+        This method processes a list of operator prototypes (`op_protos`), extracting necessary information
+        such as operator names, arguments, and conversion types. It constructs the function definitions, includes,
+        and registration code. The generated content is saved to the specified path as a C++ source file.
+
+        Args:
+            work_path (str): The file path where the generated files will be saved.
+            op_protos (list): A list of operator prototypes containing information about the operators to be processed.
+
+        Returns:
+            None
+        """
+        pyboost_func_str = ''
+        pyboost_func_pybind_def = ''
+        pyboost_func_include_headers_str = ''
+        ops_inc_head_set = set()
+        for op_proto in op_protos:
+            if op_proto.op_dispatch is None or not op_proto.op_dispatch.enable:
+                continue
+            op_parser = OpTemplateParser(op_proto)
+            op_pyboost_func_name = op_parser.get_pyboost_func_name()
+            op_def_name_str = op_parser.get_op_def_name_str()
+            type_num, same_type = op_parser.gen_signature_same_type_table()
+            parser_body_str = self._generate_parser_func(op_proto)
+            op_args_str = [op_arg.arg_name for op_arg in op_proto.op_args]
+            convert_stub_str = self._get_convert_stub_str(op_proto)
+            optional_to_value_str = self._get_optional_to_value_str(op_proto)
+            call_args_str = self._get_call_args_str(op_proto)
+            grad_args_str = self._get_grad_args_str(op_proto)
+            cast_args_str = self._get_cast_to_value_str(op_proto)
+            view_arg_str = self._get_first_str(op_proto.op_view, grad_args_str)
+            op_input_args_str = self._get_input_args_str(op_proto)
+            view_arg_str = ", " + view_arg_str if view_arg_str else ''
+            multi_ouptut_str = 'Multi' if is_op_multi_output(op_proto.op_returns) else ''
+            output_num_str = len(op_proto.op_returns)
+            function_tpl = self._get_function_tpl(op_proto)
+            pyboost_func_str += function_tpl.replace(func_name=op_pyboost_func_name,
+                                                     op_def_name=op_def_name_str,
+                                                     type_num=type_num,
+                                                     same_type=same_type,
+                                                     input_args=op_input_args_str,
+                                                     parser_body=parser_body_str,
+                                                     op_name=op_proto.op_class.name,
+                                                     class_name=op_proto.op_class.name,
+                                                     op_args=op_args_str,
+                                                     convert_stub=convert_stub_str,
+                                                     optional_to_value=optional_to_value_str,
+                                                     call_args=call_args_str,
+                                                     grad_args=grad_args_str,
+                                                     cast_args=cast_args_str,
+                                                     view_arg=view_arg_str,
+                                                     is_multi=multi_ouptut_str,
+                                                     output_num=output_num_str,
+                                                     operator_name=op_proto.op_name)
+            pyboost_func_str = pyboost_func_str + template.NEW_LINE + template.NEW_LINE
+            pyboost_op_name = op_parser.get_pyboost_name()
+            pyboost_func_name = op_parser.get_pyboost_func_name()
+            pyboost_func_pybind_def += self.REGISTER_DEFINE_TEMPLATE.replace(
+                pyboost_op_name=pyboost_op_name,
+                pyboost_cfunc_name=pyboost_func_name,
+                class_name=op_proto.op_class.name)
+            pyboost_func_include_headers_str += self.pyboost_func_include_header_template.replace(
+                operator_name=op_proto.op_name)
+            ops_inc_head_set.add(self.OP_DEF_INC_HEAD_TEMPLATE.replace(prefix_char=op_proto.op_class.name[0].lower()))
+        register_func_str = self.REGISTER_TEMPLATE.replace(register_func=pyboost_func_pybind_def)
+        function_class_register = self._get_function_class_register(op_protos)
+        pyboost_func_file = self.PYBOOST_HEADER_TEMPLATE.replace(ops_inc=list(sorted(ops_inc_head_set)),
+                                                                 include_op_header=pyboost_func_include_headers_str,
+                                                                 function_body=pyboost_func_str,
+                                                                 register_function_body=register_func_str,
+                                                                 function_class_register=function_class_register)
+        save_path = os.path.join(work_path, K.PIPELINE_PYBOOST_FUNC_GEN_PATH)
+        file_name = "pyboost_functions.cc"
+        save_file(save_path, file_name, pyboost_func_file)
+
+    def _get_cast_args_with_type_str(self, op_proto, cast_args_str):
+        args_with_type = []
+        for op_arg, cast_args_name in zip(op_proto.op_args, cast_args_str):
+            input_dtype = get_input_dtype(op_arg.arg_dtype, is_optional_param(op_arg))
+            args_with_type.append("const " + input_dtype + " &" + cast_args_name)
+        return list(args_with_type)
+
+    def _get_function_class_register(self, op_protos) -> str:
+        """
+        Generates a function class registration string for tensor functions.
+
+        Args:
+            op_protos (list): A list of tensor op prototypes.
+
+        Returns:
+            str: A concatenated string representing the registration information for tensor
+                 function classes.
+        """
+        function_class_register = ''
+        for op_proto in op_protos:
+            if op_proto.op_dispatch is None or not op_proto.op_dispatch.enable:
+                continue
+            class_name, op_name = op_proto.op_class.name, op_proto.op_name
+            function_class_register += self.TENSOR_FUNC_CLASS_REG.replace(class_name=class_name,
+                                                                          op_name=op_name)
+        return function_class_register
+
+    def _generate_parser_func(self, op_proto: OpProto) -> str:
+        """
+        Generates the parsing function for the operator's arguments.
+
+        This method constructs the code for converting each argument in the operator prototype to its appropriate
+        type, handling optional parameters as necessary.
+
+        Args:
+            op_proto (OpProto): The operator prototype containing the argument information.
+
+        Returns:
+            str: The generated parsing function code as a string.
+        """
+        parser_func_str = ''
+        for index, op_arg in enumerate(op_proto.op_args):
+            is_optional = is_optional_param(op_arg)
+            if op_arg.is_type_id:
+                convert_type_str = get_convert_type_str('type', is_optional)
+            else:
+                convert_type_str = get_convert_type_str(op_arg.arg_dtype, is_optional)
+
+            parser_func_str += self.convert_template.replace(arg_name=op_arg.arg_name, convert_func=convert_type_str,
+                                                             arg_index=pyboost_utils.get_index(index))
+        return parser_func_str
+
+
+    def _get_input_args_str(self, op_proto: OpProto) -> str:
+        """
+        Generates the input arguments list for the pyboost operator.
+
+        Args:
+            op_proto (OpProto): The operator prototype containing the argument information.
+
+        Returns:
+            str: The generated input arguments list as a string.
+        """
+        parser_func_str = ''
+        for _, op_arg in enumerate(op_proto.op_args):
+            is_optional = is_optional_param(op_arg)
+            if op_arg.is_type_id:
+                arg_type_str = get_input_args_type_str('type', is_optional)
+            else:
+                arg_type_str = get_input_args_type_str(op_arg.arg_dtype, is_optional)
+            parser_func_str += self.input_args_template.replace(arg_name=op_arg.arg_name, arg_type=arg_type_str)
+        return parser_func_str[:-1]
+
+
+    def _get_convert_stub_str(self, op_proto: OpProto):
+        """
+        Generates the conversion stub code for the operator's arguments.
+
+        This method creates code for converting operator arguments to tensor format, depending on whether they
+        are view operations or standard tensor operations.
+
+        Args:
+            op_proto (OpProto): The operator prototype containing the argument information.
+
+        Returns:
+            str: The generated conversion stub code as a string.
+        """
+        convert_stub_str = ''
+        need_contiguous = 'true'
+        if op_proto.op_view:
+            # View/ACLNN op does not need to convert to contiguous tensor.
+            need_contiguous = 'false'
+        for op_arg in op_proto.op_args:
+            if pyboost_utils.is_tensor(op_arg):
+                convert_stub_output_name = op_arg.arg_name + '_optional' if is_optional_param(op_arg) \
+                    else op_arg.arg_name + "_tensor"
+                convert_stub_str += self.convert_to_tensor_template.replace(input=op_arg.arg_name,
+                                                                            output=convert_stub_output_name,
+                                                                            need_contiguous=need_contiguous)
+            elif pyboost_utils.is_tensor_list(op_arg):
+                # To adapt the cases where TensorList is optional.
+                convert_stub_output_name = op_arg.arg_name + '_optional' if is_optional_param(op_arg) \
+                    else op_arg.arg_name + "_tensor_list"
+                convert_stub_str += self.convert_to_tensor_list_template.replace(input=op_arg.arg_name,
+                                                                                 output=convert_stub_output_name,
+                                                                                 need_contiguous=need_contiguous)
+        return convert_stub_str
+
+    def _get_optional_to_value_str(self, op_proto: OpProto):
+        """
+        Generates the code for converting optional arguments to their corresponding values.
+
+        This method constructs code to handle optional arguments and converts them to their actual values,
+        ensuring proper handling for tensors and lists.
+
+        Args:
+            op_proto (OpProto): The operator prototype containing the argument information.
+
+        Returns:
+            str: The generated code for converting optional arguments to values as a string.
+        """
+        optional_to_value_str = ''
+        for op_arg in op_proto.op_args:
+            if is_optional_param(op_arg):
+                if pyboost_utils.is_tensor(op_arg) or pyboost_utils.is_tensor_list(op_arg):
+                    convert_stub_output_name = op_arg.arg_name + '_optional'
+                    cast_output = 'cast_' + convert_stub_output_name
+                    convert_optional_to_value_name = op_arg.arg_name + '_value'
+                    optional_to_value_str += \
+                        self.convert_optional_to_value_template.replace(input=cast_output,
+                                                                        output=convert_optional_to_value_name)
+                else:
+                    call_arg = op_arg.arg_name
+                    convert_optional_to_value_name = op_arg.arg_name + '_value'
+                    optional_to_value_str += \
+                        self.convert_optional_to_value_template.replace(input=call_arg,
+                                                                        output=convert_optional_to_value_name)
+        return optional_to_value_str
+
+    def _get_call_args_str(self, op_proto: OpProto):
+        """
+        Generates the list of call arguments for the operator.
+
+        This method constructs a list of argument names for the function call, adapting the names for
+        optional tensors and tensor lists as needed.
+
+        Args:
+            op_proto (OpProto): The operator prototype containing the argument information.
+
+        Returns:
+            list: A list of formatted argument names for the function call.
+        """
+        call_args_str = []
+        for op_arg in op_proto.op_args:
+            if pyboost_utils.is_tensor(op_arg):
+                convert_stub_output_name = op_arg.arg_name + '_optional' if is_optional_param(op_arg) \
+                    else op_arg.arg_name + "_tensor"
+                call_arg = convert_stub_output_name
+            elif pyboost_utils.is_tensor_list(op_arg):
+                convert_stub_output_name = op_arg.arg_name + '_optional' if is_optional_param(op_arg) \
+                    else op_arg.arg_name + "_tensor_list"
+                call_arg = convert_stub_output_name
+            else:
+                call_arg = op_arg.arg_name
+            call_args_str.append(call_arg)
+        return call_args_str
+
+    def _get_grad_args_str(self, op_proto: OpProto):
+        """
+        Generates the list of gradient arguments for the operator.
+
+        This method constructs a list of argument names used for computing gradients, adapting for
+        optional tensors and tensor lists as necessary.
+
+        Args:
+            op_proto (OpProto): The operator prototype containing the argument information.
+
+        Returns:
+            list: A list of formatted gradient argument names.
+        """
+        grad_args_str = []
+        for op_arg in op_proto.op_args:
+            if pyboost_utils.is_tensor(op_arg):
+                grad_arg = op_arg.arg_name + "_value" if is_optional_param(op_arg) else \
+                    f"cast_" + op_arg.arg_name + "_tensor"
+            elif pyboost_utils.is_tensor_list(op_arg):
+                if is_optional_param(op_arg):
+                    # To adapt the cases where TensorList is optional.
+                    convert_optional_to_value_name = op_arg.arg_name + "_value"
+                    grad_arg = convert_optional_to_value_name
+                else:
+                    convert_stub_output_name = op_arg.arg_name + "_tensor_list"
+                    grad_arg = "cast_" + convert_stub_output_name
+            else:
+                grad_arg = "cast_" + op_arg.arg_name
+                if is_optional_param(op_arg):
+                    convert_optional_to_value_name = op_arg.arg_name + "_value"
+                    grad_arg = convert_optional_to_value_name
+            grad_args_str.append(grad_arg)
+        return grad_args_str
+
+    def _get_cast_to_value_str(self, op_proto: OpProto):
+        """
+        Generates the list of cast arguments for the operator.
+
+        This method constructs a list of argument names that need to be cast to their corresponding types.
+
+        Args:
+            op_proto (OpProto): The operator prototype containing the argument information.
+
+        Returns:
+            list: A list of formatted cast argument names.
+        """
+        cast_args_str = []
+        for op_arg in op_proto.op_args:
+            cast_str = 'cast_'
+            if pyboost_utils.is_tensor(op_arg):
+                convert_stub_output_name = op_arg.arg_name + '_optional' if is_optional_param(op_arg) \
+                    else op_arg.arg_name + "_tensor"
+                cast_arg = cast_str + convert_stub_output_name
+            elif pyboost_utils.is_tensor_list(op_arg):
+                # To adapt the cases where TensorList is optional.
+                convert_stub_output_name = op_arg.arg_name + '_optional' if is_optional_param(op_arg) \
+                    else op_arg.arg_name + "_tensor_list"
+                cast_arg = cast_str + convert_stub_output_name
+            else:
+                cast_arg = cast_str + op_arg.arg_name
+            cast_args_str.append(cast_arg)
+        return cast_args_str
+
+    def _get_first_str(self, is_view_or_inplace: bool, grad_args: list):
+        """
+        Generates the view base str of arguments for the operator.
+
+        This method constructs a list of argument names that need to be cast to their corresponding types.
+
+        Args:
+            is_view_or_inplace (bool): Whether the op is view op or inplace op.
+            grad_args (list): grad args
+
+        Returns:
+            str: Formatted view or inplace first argument names.
+        """
+        arg_str = ''
+        for i, grad_arg in enumerate(grad_args):
+            if is_view_or_inplace and i == 0:
+                arg_str = grad_arg
+                break
+        return arg_str
+
+
+    def _get_function_tpl(self, op_proto: OpProto):
+        if len(op_proto.op_returns) == 1 and is_tensor_list(op_proto.op_returns[0]):
+            # op output size is unknown
+            return self.PYBOOST_FUNCTION_DYNAMIC_OUTPUT_TEMPLATE
+        return self.PYBOOST_COMM_FUNCTION_TEMPLATE \
+            if op_proto.op_dispatch.is_comm_op else self.PYBOOST_FUNCTION_TEMPLATE

mindspore/ops_generate/pyboost/pyboost_functions_h_generator.py

@@ -0,0 +1,100 @@
+# Copyright 2024 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""
+This module defines the `PyboostFunctionsHeaderGenerator` class, which is responsible for generating
+the header file (`pyboost_functions.h`) for Pyboost function declarations.
+
+The class uses templates and operation prototypes to create function declarations based on the
+operation's primitive and arguments. The generated file is saved to the specified path.
+"""
+
+import os
+
+import common.template as template
+import common.gen_constants as K
+from common.template import Template
+from common.gen_utils import save_file
+from common.base_generator import BaseGenerator
+from common.op_proto import OpProto
+
+from .op_template_parser import OpTemplateParser
+from .pyboost_utils import is_optional_param, get_input_args_type_str
+
+
+class PyboostFunctionsHeaderGenerator(BaseGenerator):
+    """
+    A class to generate the `pyboost_functions.h` header file, which contains Pyboost function declarations.
+    """
+
+    def __init__(self):
+        """Initializes the PyboostFunctionsHeaderGenerator with the necessary templates."""
+        self.PYBOOST_FUNCTION_HEADER_TEMPLATE = template.PYBOOST_FUNCTION_HEADER_TEMPLATE
+
+        self.pyboost_func_template = Template(
+            'py::object PYNATIVE_EXPORT ${func_name}_Base(const PrimitivePtr &prim, const py::list &args);'
+        )
+        self.pyboost_op_func_template = Template(
+            'py::object ME_EXPORT ${func_name}_OP(const PrimitivePtr &prim, '
+            'const std::vector<ops::OP_DTYPE>& source_type, ${input_args});'
+        )
+        self.input_args_template = Template(" const ${arg_type}& ${arg_name},")
+
+    def generate(self, work_path, op_protos):
+        """
+        Generates the Pyboost function header file (`pyboost_functions.h`).
+
+        Args:
+            work_path (str): The directory where the generated file will be saved.
+            op_protos (list): A list of operation prototypes to parse and convert into Pyboost function declarations.
+
+        Returns:
+            None: The method writes the generated header file to the specified directory.
+        """
+        prim_func_list = []
+        op_func_list_str = []
+        for op_proto in op_protos:
+            if op_proto.op_dispatch is None or not op_proto.op_dispatch.enable:
+                continue
+            op_parser = OpTemplateParser(op_proto)
+            op_pyboost_func_name = op_parser.get_pyboost_func_name()
+            op_input_args_str = self._get_input_args_str(op_proto)
+            prim_func_list.append(self.pyboost_func_template.replace(func_name=op_pyboost_func_name))
+            op_func_list_str.append(self.pyboost_op_func_template.replace(func_name=op_pyboost_func_name,
+                                                                          input_args=op_input_args_str))
+        pyboost_func_h_str = self.PYBOOST_FUNCTION_HEADER_TEMPLATE.replace(prim_func_list=prim_func_list,
+                                                                           op_func_list=op_func_list_str)
+        save_path = os.path.join(work_path, K.PIPELINE_PYBOOST_FUNC_GEN_PATH)
+        file_name = "pyboost_functions.h"
+        save_file(save_path, file_name, pyboost_func_h_str)
+
+    def _get_input_args_str(self, op_proto: OpProto) -> str:
+        """
+        Generates the input arguments list for the pyboost operator.
+
+        Args:
+            op_proto (OpProto): The operator prototype containing the argument information.
+
+        Returns:
+            str: The generated input arguments list as a string.
+        """
+        parser_func_str = ''
+        for _, op_arg in enumerate(op_proto.op_args):
+            is_optional = is_optional_param(op_arg)
+            if op_arg.is_type_id:
+                arg_type_str = get_input_args_type_str('type', is_optional)
+            else:
+                arg_type_str = get_input_args_type_str(op_arg.arg_dtype, is_optional)
+            parser_func_str += self.input_args_template.replace(arg_name=op_arg.arg_name, arg_type=arg_type_str)
+        return parser_func_str[:-1]

mindspore/ops_generate/pyboost/pyboost_functions_py_generator.py

@@ -0,0 +1,148 @@
+# Copyright 2024 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""
+This module defines the PyboostFunctionsPyGenerator class for generating Python bindings for PyBoost functions.
+
+The PyboostFunctionsPyGenerator class processes operator prototypes and generates Python functions
+that correspond to the PyBoost operations defined in the operator prototypes. It handles the necessary
+argument processing and includes appropriate documentation descriptions.
+"""
+
+import os
+
+import common.template as template
+import common.gen_constants as K
+from common.gen_utils import save_file
+from common.op_proto import OpProto
+from common.base_generator import BaseGenerator
+
+
+class PyboostFunctionsPyGenerator(BaseGenerator):
+    """
+    Generates Python bindings for PyBoost functions.
+
+    This class is responsible for creating Python function definitions that correspond to the PyBoost
+    operations defined in operator prototypes. It generates a Python file that includes necessary function
+    definitions and their descriptions.
+    """
+
+    def __init__(self):
+        """Initializes the PyboostFunctionsPyGenerator with required templates."""
+        self.IMPORT_PYBOOST_FUNC_HEADER = template.IMPORT_PYBOOST_FUNC_HEADER
+        self.PYBOOST_PY_FUNC_TEMPLATE = template.PYBOOST_PY_FUNC_TEMPLATE
+
+    def generate(self, work_path, op_protos, doc_data):
+        """
+        Generates the Python file containing PyBoost function definitions.
+
+        This method processes the provided operator prototypes (`op_protos`), generates Python function
+        definitions for each operator that meets the specified conditions, and saves the generated content
+        to a Python file.
+
+        Args:
+            work_path (str): The directory path where the generated Python file will be saved.
+            op_protos (list): A list of operator prototypes containing information about the operators.
+            doc_data (dict): A dictionary containing documentation data for the operators.
+
+        Returns:
+            None
+        """
+        gen_py = ''
+        op_desc_dict = self._get_op_description_dict(doc_data)
+        for op_proto in op_protos:
+            # check if the operator is in pyboost scenario
+            if op_proto.op_dispatch is None or not op_proto.op_dispatch.enable:
+                continue
+            if op_proto.op_function.disable:
+                continue
+            if not op_proto.op_function.name.endswith("_ext") and not op_proto.op_name.endswith("_ext"):
+                continue
+
+            description = op_desc_dict.get(op_proto.op_name)
+            func_args, input_args = self._process_args(op_proto.op_args)
+            func_name, func_impl_name = self._get_func_impl_name(op_proto)
+            gen_py += self.PYBOOST_PY_FUNC_TEMPLATE.replace(func_name=func_name,
+                                                            description=description,
+                                                            func_args=func_args,
+                                                            input_args=input_args,
+                                                            func_impl_name=func_impl_name)
+        py_header = template.PY_LICENSE_STR + self.IMPORT_PYBOOST_FUNC_HEADER
+        save_file(os.path.join(work_path, K.PY_AUTO_GEN_PATH), "gen_extend_func.py", py_header + gen_py)
+
+    def _get_op_description_dict(self, doc_yaml_data):
+        """
+        Constructs a dictionary mapping operator names to their descriptions.
+
+        Args:
+            doc_yaml_data (dict): A dictionary containing YAML data for operator documentation.
+
+        Returns:
+            dict: A dictionary mapping operator names to their descriptions.
+        """
+        op_description_dict = {}
+        for operator_name, operator_desc in doc_yaml_data.items():
+            desc = operator_desc.get("description")
+            op_description_dict[operator_name] = desc
+        return op_description_dict
+
+    def _process_args(self, op_args):
+        """
+        Processes the operator arguments to generate function argument strings.
+
+        Args:
+            op_args (list): A list of operator arguments to be processed.
+
+        Returns:
+            tuple: A tuple containing:
+                - func_args (list): A list of formatted function argument strings.
+                - input_args (list): A list of corresponding input argument names.
+        """
+        func_args = []
+        input_args = []
+        for op_arg in op_args:
+            arg_handler = op_arg.arg_handler
+            arg_name = op_arg.arg_name
+            input_arg = arg_name
+            if arg_handler != '' and arg_handler != 'dtype_to_type_id':
+                input_arg = 'converted_' + arg_name
+            input_args.append(input_arg)
+            default_value = op_arg.default
+            if default_value is not None:
+                default_value = '=' + str(default_value)
+                func_args.append(arg_name + default_value)
+            else:
+                func_args.append(arg_name)
+        return func_args, input_args
+
+    def _get_func_impl_name(self, op_proto: OpProto):
+        """
+        Retrieves the implementation function name based on the operator prototype.
+
+        Args:
+            op_proto (OpProto): The operator prototype containing function name information.
+
+        Returns:
+            tuple: A tuple containing:
+                - func_name (str): The name of the function.
+                - func_impl_name (str): The implementation name of the function.
+        """
+        func_name = op_proto.op_name if op_proto.op_function.name == '' \
+            else op_proto.op_function.name
+        if func_name.endswith("_ext"):
+            func_name = func_name[:-4]
+        func_impl_name = func_name
+        if func_name.endswith("_"):
+            func_impl_name = func_name[:-1]
+        return func_name, func_impl_name