paddlepaddle--paddle/test/ir/inference/test_fc_fuse_pass.py

# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
#
# 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.

import unittest

import hypothesis.strategies as st
import numpy as np
from auto_scan_test import IgnoreReasons, PassAutoScanTest
from program_config import OpConfig, ProgramConfig, TensorConfig


class TestFcFusePass(PassAutoScanTest):
    r"""
    x_var   y_var(persistable)
      \       /
         mul     bias_var(persistable)
          |
      mul_out_var  bias_var(persistable)
            \        /
          elementwise_add
    """

    def sample_predictor_configs(self, program_config):
        # cpu
        before_num_ops = len(program_config.ops) + 2
        config = self.create_inference_config(use_gpu=False)
        yield config, ["fc"], (1e-5, 1e-5)

        # for gpu
        config = self.create_inference_config(use_gpu=True)
        yield config, ["fc"], (1e-5, 1e-5)

        # trt static_shape
        config = self.create_trt_inference_config()
        yield config, ['fc'], (1e-5, 1e-5)

    def add_ignore_pass_case(self):
        # Here we put some skip rules to avoid known bugs
        def teller1(program_config, predictor_config):
            # shape of bias should be [1, mul_y_shape[-1]] or [mul_y_shape[-1]]
            x_shape = list(program_config.inputs["mul_x"].shape)
            y_shape = list(program_config.weights["mul_y"].shape)
            bias_shape = program_config.weights["bias"].shape
            bias_shape = list(program_config.weights["bias"].shape)

            if predictor_config.tensorrt_engine_enabled():
                # TensorRT can't handle all the situation of elementwise_add
                # disable it until this problem fixed
                predictor_config.exp_disable_tensorrt_ops(["elementwise_add"])

            if bias_shape != [y_shape[-1]] and bias_shape != [1, y_shape[-1]]:
                return True
            return False

        def teller2(program_config, predictor_config):
            # TODO fuse has bug while axis != -1
            axis = program_config.ops[1].attrs["axis"]
            if (
                axis != -1
                and axis != program_config.ops[0].attrs["x_num_col_dims"]
            ):
                return True
            return False

        self.add_ignore_check_case(
            teller1,
            IgnoreReasons.PASS_ACCURACY_ERROR,
            "The pass output has diff while shape of bias is not [out_size] or [1, out_size].",
        )
        self.add_ignore_check_case(
            teller2,
            IgnoreReasons.PASS_ACCURACY_ERROR,
            "The pass output has diff while axis of elementwise_add is not -1.",
        )

    def is_program_valid(self, prog_config):
        add_x_rank = prog_config.ops[0].attrs["x_num_col_dims"] + 1
        add_y_rank = len(prog_config.weights["bias"].shape)
        axis = prog_config.ops[1].attrs["axis"]
        if add_x_rank == add_y_rank:
            if axis != -1 or axis != 0:
                return False
        return True

    def sample_program_config(self, draw):
        # 1. Generate shape of input:X of mul
        x_shape = draw(
            st.lists(
                st.integers(min_value=1, max_value=4), min_size=2, max_size=4
            )
        )
        # 2. Generate attr:x_num_col_dims/y_num_col_dims of mul
        x_num_col_dims = draw(
            st.integers(min_value=1, max_value=len(x_shape) - 1)
        )
        y_num_col_dims = 1
        # 3. Generate legal shape of input:Y of mul
        y_shape = draw(
            st.lists(
                st.integers(min_value=1, max_value=8), min_size=2, max_size=2
            )
        )
        y_shape[0] = int(np.prod(x_shape[x_num_col_dims:]))
        # 4. Generate legal attr:axis of elementwise_add
        mul_out_shape = x_shape[:x_num_col_dims] + y_shape[1:]
        axis = draw(st.integers(min_value=-1, max_value=x_num_col_dims))
        # 5. Generate legal shape of input:Y of elementwise_add
        if axis >= 0:
            max_bias_rank = x_num_col_dims + 1 - axis
            bias_rank = draw(st.integers(min_value=1, max_value=max_bias_rank))
            bias_shape = mul_out_shape[axis : axis + bias_rank]
        else:
            max_bias_rank = 1
            bias_rank = draw(
                st.integers(min_value=1, max_value=len(mul_out_shape))
            )
            bias_shape = mul_out_shape[-1 * bias_rank :]
        # 6. Random choose if use broadcast for elementwise_add, e.g [3, 4] -> [1, 4]
        if draw(st.booleans()):
            broadcast_dims = draw(st.integers(min_value=1, max_value=bias_rank))
            for i in range(0, broadcast_dims):
                bias_shape[i] = 1
        # 7. Random choose if add a relu operator
        has_relu = draw(st.booleans())

        # Now we have all the decided parameters to compose a program
        # shape of inputs/weights tensors: x_shape, y_shape, bias_shape...
        # parameters of operators: x_num_col_dims, y_num_col_dims, axis...
        # a random boolean value(has_relu) to decide if program include a relu op

        # Here we will compose a program
        # Still has some risks that the program is invalid or cause bug while running
        # Use function `is_program_valid` to filter the invalid programs before running
        # Use function `add_skip_pass_case` to ignore the programs even if they cause bug while running
        mul_op = OpConfig(
            "mul",
            inputs={"X": ["mul_x"], "Y": ["mul_y"]},
            outputs={"Out": ["mul_out"]},
            x_num_col_dims=x_num_col_dims,
            y_num_col_dims=y_num_col_dims,
        )
        add_op = OpConfig(
            "elementwise_add",
            inputs={"X": ["mul_out"], "Y": ["bias"]},
            outputs={"Out": ["add_out"]},
            axis=axis,
        )
        ops = [mul_op, add_op]
        if has_relu:
            relu_op = OpConfig(
                "relu", inputs={"X": ["add_out"]}, outputs={"Out": ["relu_out"]}
            )
            ops.append(relu_op)
        program_config = ProgramConfig(
            ops=ops,
            weights={
                "mul_y": TensorConfig(shape=y_shape),
                "bias": TensorConfig(shape=bias_shape),
            },
            inputs={
                "mul_x": TensorConfig(shape=x_shape),
            },
            outputs=ops[-1].outputs["Out"],
        )
        return program_config

    def test(self):
        self.run_and_statistics(
            quant=False, max_examples=500, passes=["fc_fuse_pass"]
        )


if __name__ == "__main__":
    unittest.main()