diff --git a/paddle/fluid/inference/tensorrt/op_teller.cc b/paddle/fluid/inference/tensorrt/op_teller.cc index 75f5616f7584faa14c84b2a616e563a53ac5f829..1b0c6c0a71d178dc3fad537c4602acca40d85e44 100644 --- a/paddle/fluid/inference/tensorrt/op_teller.cc +++ b/paddle/fluid/inference/tensorrt/op_teller.cc @@ -1079,6 +1079,16 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8, for (auto x : dim) { if (!x) return false; } + } else { + if (BOOST_GET_CONST(bool, desc.GetAttr("reduce_all")) && + !BOOST_GET_CONST(bool, desc.GetAttr("keep_dim"))) + return false; + } + if (desc.HasAttr("reduce_all")) { + int out_dtype = BOOST_GET_CONST(int32_t, desc.GetAttr("out_dtype")); + if (out_dtype != -1) { + return false; + } } } #if IS_TRT_VERSION_GE(7000) diff --git a/python/paddle/fluid/tests/unittests/ir/inference/test_trt_convert_reduce_mean.py b/python/paddle/fluid/tests/unittests/ir/inference/test_trt_convert_reduce_mean.py new file mode 100644 index 0000000000000000000000000000000000000000..6c4c2ef4e1a14044f2d0056577f3851cb09993b4 --- /dev/null +++ b/python/paddle/fluid/tests/unittests/ir/inference/test_trt_convert_reduce_mean.py @@ -0,0 +1,155 @@ +# 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. + +from trt_layer_auto_scan_test import TrtLayerAutoScanTest, SkipReasons +from program_config import TensorConfig, ProgramConfig +import numpy as np +import paddle.inference as paddle_infer +from functools import partial +from typing import Optional, List, Callable, Dict, Any, Set +import unittest + + +class TrtConvertReduceMeanTest(TrtLayerAutoScanTest): + def is_program_valid(self, program_config: ProgramConfig) -> bool: + inputs = program_config.inputs + attrs = [ + program_config.ops[i].attrs + for i in range(len(program_config.ops)) + ] + + ## dim should be in (-rank, rank), and not NONE + rank = len(inputs['input_data'].shape) + for x in attrs[0]["dim"]: + if x >= rank or x <= -rank: + return False + if len(attrs[0]["dim"]) == 0: + return False + ## skip not use + if attrs[0]["out_dtype"] != -1: + return False + + return True + + def sample_program_configs(self): + def generate_input1(attrs: List[Dict[str, Any]]): + return np.random.random([1, 3, 64, 64]).astype(np.float32) + + for keep_dim in [False, True]: + for dim in [[], [1], [0], [0, 1], [1, 2, 3], [-2, 0, 3], [-3], + [-4, 1], [3, 4, 5]]: + for reduce_all in [False, True]: + for out_dtype in [-1, 0, 1]: + dics = [{ + "keep_dim": keep_dim, + "dim": dim, + "reduce_all": reduce_all, + "out_dtype": out_dtype + }, {}] + + ops_config = [{ + "op_type": "reduce_mean", + "op_inputs": { + "X": ["input_data"] + }, + "op_outputs": { + "Out": ["reduce_output_data"] + }, + "op_attrs": dics[0] + }] + ops = self.generate_op_config(ops_config) + + program_config = ProgramConfig( + ops=ops, + weights={}, + inputs={ + "input_data": TensorConfig(data_gen=partial( + generate_input1, dics)) + }, + outputs=["reduce_output_data"]) + + if not self.is_program_valid(program_config): + continue + + yield program_config + + def sample_predictor_configs( + self, program_config) -> (paddle_infer.Config, List[int], float): + def generate_dynamic_shape(attrs): + self.dynamic_shape.min_input_shape = {"input_data": [1, 3, 32, 32]} + self.dynamic_shape.max_input_shape = {"input_data": [4, 3, 64, 64]} + self.dynamic_shape.opt_input_shape = {"input_data": [1, 3, 64, 64]} + + def clear_dynamic_shape(): + self.dynamic_shape.min_input_shape = {} + self.dynamic_shape.max_input_shape = {} + self.dynamic_shape.opt_input_shape = {} + + def generate_trt_nodes_num(attrs, dynamic_shape): + if dynamic_shape: + if (not attrs[0]['keep_dim']) and attrs[0]['reduce_all']: + return 0, 3 + else: + return 1, 2 + else: + if 0 in attrs[0]['dim'] or attrs[0]['reduce_all']: + return 0, 3 + else: + return 1, 2 + + attrs = [ + program_config.ops[i].attrs + for i in range(len(program_config.ops)) + ] + + # for static_shape + clear_dynamic_shape() + self.trt_param.precision = paddle_infer.PrecisionType.Float32 + yield self.create_inference_config(), generate_trt_nodes_num( + attrs, False), 1e-5 + self.trt_param.precision = paddle_infer.PrecisionType.Half + yield self.create_inference_config(), generate_trt_nodes_num( + attrs, False), (1e-5, 1e-5) + + # for dynamic_shape + generate_dynamic_shape(attrs) + self.trt_param.precision = paddle_infer.PrecisionType.Float32 + yield self.create_inference_config(), generate_trt_nodes_num(attrs, + True), 1e-5 + self.trt_param.precision = paddle_infer.PrecisionType.Half + yield self.create_inference_config(), generate_trt_nodes_num( + attrs, True), (1e-5, 1e-5) + + pass + + def add_skip_trt_case(self): + def teller1(program_config, predictor_config): + if program_config.ops[0].attrs['out_dtype'] != -1: + return True + return False + + self.add_skip_case( + teller1, SkipReasons.TRT_NOT_IMPLEMENTED, + "NOT Implemented: we will add out_dtype not equal to -1 in the future" + ) + + pass + + def test(self): + self.add_skip_trt_case() + self.run_test() + + +if __name__ == "__main__": + unittest.main() diff --git a/python/paddle/fluid/tests/unittests/ir/inference/test_trt_convert_reduce_sum.py b/python/paddle/fluid/tests/unittests/ir/inference/test_trt_convert_reduce_sum.py new file mode 100644 index 0000000000000000000000000000000000000000..91e1c0677ac481023134fc8c587705f7f07457f5 --- /dev/null +++ b/python/paddle/fluid/tests/unittests/ir/inference/test_trt_convert_reduce_sum.py @@ -0,0 +1,155 @@ +# 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. + +from trt_layer_auto_scan_test import TrtLayerAutoScanTest, SkipReasons +from program_config import TensorConfig, ProgramConfig +import numpy as np +import paddle.inference as paddle_infer +from functools import partial +from typing import Optional, List, Callable, Dict, Any, Set +import unittest + + +class TrtConvertReduceSumTest(TrtLayerAutoScanTest): + def is_program_valid(self, program_config: ProgramConfig) -> bool: + inputs = program_config.inputs + attrs = [ + program_config.ops[i].attrs + for i in range(len(program_config.ops)) + ] + + ## dim should be in (-rank, rank), and not NONE + rank = len(inputs['input_data'].shape) + for x in attrs[0]["dim"]: + if x >= rank or x <= -rank: + return False + if len(attrs[0]["dim"]) == 0: + return False + ## skip not use + if attrs[0]["out_dtype"] != -1: + return False + + return True + + def sample_program_configs(self): + def generate_input1(attrs: List[Dict[str, Any]]): + return np.random.random([1, 3, 64, 64]).astype(np.float32) + + for keep_dim in [False, True]: + for dim in [[], [1], [0], [0, 1], [1, 2, 3], [-2, 0, 3], [-3], + [-4, 1], [3, 4, 5]]: + for reduce_all in [False, True]: + for out_dtype in [-1, 0, 1]: + dics = [{ + "keep_dim": keep_dim, + "dim": dim, + "reduce_all": reduce_all, + "out_dtype": out_dtype + }, {}] + + ops_config = [{ + "op_type": "reduce_sum", + "op_inputs": { + "X": ["input_data"] + }, + "op_outputs": { + "Out": ["reduce_output_data"] + }, + "op_attrs": dics[0] + }] + ops = self.generate_op_config(ops_config) + + program_config = ProgramConfig( + ops=ops, + weights={}, + inputs={ + "input_data": TensorConfig(data_gen=partial( + generate_input1, dics)) + }, + outputs=["reduce_output_data"]) + + if not self.is_program_valid(program_config): + continue + + yield program_config + + def sample_predictor_configs( + self, program_config) -> (paddle_infer.Config, List[int], float): + def generate_dynamic_shape(attrs): + self.dynamic_shape.min_input_shape = {"input_data": [1, 3, 32, 32]} + self.dynamic_shape.max_input_shape = {"input_data": [4, 3, 64, 64]} + self.dynamic_shape.opt_input_shape = {"input_data": [1, 3, 64, 64]} + + def clear_dynamic_shape(): + self.dynamic_shape.min_input_shape = {} + self.dynamic_shape.max_input_shape = {} + self.dynamic_shape.opt_input_shape = {} + + def generate_trt_nodes_num(attrs, dynamic_shape): + if dynamic_shape: + if (not attrs[0]['keep_dim']) and attrs[0]['reduce_all']: + return 0, 3 + else: + return 1, 2 + else: + if 0 in attrs[0]['dim'] or attrs[0]['reduce_all']: + return 0, 3 + else: + return 1, 2 + + attrs = [ + program_config.ops[i].attrs + for i in range(len(program_config.ops)) + ] + + # for static_shape + clear_dynamic_shape() + self.trt_param.precision = paddle_infer.PrecisionType.Float32 + yield self.create_inference_config(), generate_trt_nodes_num( + attrs, False), 1e-5 + self.trt_param.precision = paddle_infer.PrecisionType.Half + yield self.create_inference_config(), generate_trt_nodes_num( + attrs, False), (1e-5, 1e-5) + + # for dynamic_shape + generate_dynamic_shape(attrs) + self.trt_param.precision = paddle_infer.PrecisionType.Float32 + yield self.create_inference_config(), generate_trt_nodes_num(attrs, + True), 1e-5 + self.trt_param.precision = paddle_infer.PrecisionType.Half + yield self.create_inference_config(), generate_trt_nodes_num( + attrs, True), (1e-5, 1e-5) + + pass + + def add_skip_trt_case(self): + def teller1(program_config, predictor_config): + if program_config.ops[0].attrs['out_dtype'] != -1: + return True + return False + + self.add_skip_case( + teller1, SkipReasons.TRT_NOT_IMPLEMENTED, + "NOT Implemented: we will add out_dtype not equal to -1 in the future" + ) + + pass + + def test(self): + self.add_skip_trt_case() + self.run_test() + + +if __name__ == "__main__": + unittest.main()