未验证 提交 96597a85 编写于 作者: Y yeliang2258 提交者: GitHub

Add tests for PaddleInference Pass (#37676)

* add test for conv_elementwise_add2_act_fuse_pass and conv_elementwise_add_act_fuse_pass

* Add conv_eltwiseadd_bn_fuse_pass test and fix test_conv_elementwise_addX_act_fuse_pass

* add tests for conv_act_mkldnn_fuse_pass

* add test for conv_bias_mkldnn_fuse_pass

* update code

* add conv_act_mkldnn_fuse_pass for relu, relu6, swish, leaky_relu

* update test

* update

* update bug

* update

* update pattern_detector

* fix test_conv_eltwiseadd_bn_fuse_pass

* add diff display notest;test=windows_ci_inference

* fix

* remove test_conv_act_mkldnn_fuse_pass.py

* ifix
上级 e02537f9
......@@ -173,7 +173,7 @@ if(NOT WIN32)
endif()
if (WITH_MKLDNN)
cc_test(test_depthwise_conv_mkldnn_pass SRCS mkldnn/depthwise_conv_mkldnn_pass_tester.cc DEPS depthwise_conv_mkldnn_pass)
cc_test(test_conv_bias_mkldnn_fuse_pass SRCS mkldnn/conv_bias_mkldnn_fuse_pass_tester.cc DEPS conv_bias_mkldnn_fuse_pass naive_executor)
cc_test(test_conv_bias_mkldnn_fuse_pass_cc SRCS mkldnn/conv_bias_mkldnn_fuse_pass_tester.cc DEPS conv_bias_mkldnn_fuse_pass naive_executor)
cc_test(test_conv_activation_mkldnn_fuse_pass SRCS mkldnn/conv_activation_mkldnn_fuse_pass_tester.cc DEPS conv_activation_mkldnn_fuse_pass)
cc_test(test_conv_concat_relu_mkldnn_fuse_pass SRCS mkldnn/conv_concat_relu_mkldnn_fuse_pass_tester.cc DEPS conv_concat_relu_mkldnn_fuse_pass)
cc_test(test_conv_elementwise_add_mkldnn_fuse_pass SRCS mkldnn/conv_elementwise_add_mkldnn_fuse_pass_tester.cc DEPS conv_elementwise_add_mkldnn_fuse_pass pass_test_util)
......
......@@ -807,6 +807,7 @@ PDNode *patterns::ConvBN::operator()(paddle::framework::ir::PDNode *conv_input,
// Bias
eltwise_y_in_var = pattern->NewNode(eltwise_y_in_repr())
->assert_is_op_input("elementwise_add", "Y")
->assert_is_persistable_var()
->AsInput();
eltwise_out_var = pattern->NewNode(eltwise_out_repr())
->AsIntermediate()
......
......@@ -201,6 +201,9 @@ Conv2DSwishFusePass::Conv2DSwishFusePass() {
.End()
.AddOutput("Out")
.IsTensor()
.End()
.AddAttr("beta")
.IsType<float>()
.End();
}
Conv2DHardSwishFusePass::Conv2DHardSwishFusePass() {
......
......@@ -239,6 +239,7 @@ void ConvBiasFusePass::ApplyImpl(ir::Graph* graph) const {
auto input_names = conv->Op()->InputNames();
bool has_bias = std::find(input_names.begin(), input_names.end(), "Bias") !=
input_names.end();
if (has_bias && conv->Op()->Input("Bias").size() > 0) {
auto conv_bias_names = conv->Op()->Input("Bias");
// add eltwise bias to existing conv bias
......
......@@ -80,7 +80,9 @@ if (WITH_MKLDNN AND TENSORRT_FOUND AND WITH_GPU)
set_tests_properties(test_fc_fuse_pass PROPERTIES TIMEOUT 240)
set_tests_properties(test_simplify_with_basic_ops_pass_autoscan PROPERTIES TIMEOUT 60)
set_tests_properties(test_adaptive_pool2d_convert_global_pass_autoscan PROPERTIES TIMEOUT 60)
set_tests_properties(test_conv_eltwiseadd_affine_channel_fuse_pass PROPERTIES TIMEOUT 100)
set_tests_properties(test_conv_eltwiseadd_bn_fuse_pass PROPERTIES TIMEOUT 120)
set_tests_properties(test_conv_elementwise_add2_act_fuse_pass PROPERTIES TIMEOUT 120)
set_tests_properties(test_conv_elementwise_add_act_fuse_pass PROPERTIES TIMEOUT 120)
endif()
if (WITH_MKLDNN)
......
......@@ -144,10 +144,12 @@ class AutoScanTest(unittest.TestCase):
baseline[key].shape == arr.shape,
"The output shapes are not equal, the baseline shape is " +
str(baseline[key].shape) + ', but got ' + str(arr.shape))
diff = abs(baseline[key] - arr)
self.assertTrue(
np.allclose(
baseline[key], arr, atol=atol, rtol=rtol),
"Output has diff. ")
"Output has diff, Maximum absolute error: {}".format(
np.amax(diff)))
@abc.abstractmethod
def run_test(self, quant=False):
......
# 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 auto_scan_test import PassAutoScanTest, SkipReasons
from program_config import TensorConfig, ProgramConfig, OpConfig
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
import hypothesis
from hypothesis import given, settings, seed, example, assume, reproduce_failure
import hypothesis.strategies as st
class TestConvBiasMkldnnFusePass(PassAutoScanTest):
"""
x_var f_var(persistable)
\ /
conv2d
|
conv2d_var bias_var(persistable)
\ /
elementwise_add
|
elementwise_add_var
"""
def sample_predictor_configs(self, program_config):
# MKLDNN
config = self.create_inference_config(use_gpu=False)
config.enable_mkldnn()
yield config, ["conv2d"], (1e-4, 1e-5)
def is_program_valid(self, prog_config):
paddings = prog_config.ops[0].attrs["paddings"]
strides = prog_config.ops[0].attrs["strides"]
groups = prog_config.ops[0].attrs["groups"]
padding_algorithm = prog_config.ops[0].attrs["padding_algorithm"]
dilations = prog_config.ops[0].attrs["dilations"]
data_format = prog_config.ops[0].attrs["data_format"]
filter_shape = prog_config.weights["filter"].shape
input_shape = prog_config.inputs["input_x"].shape
if data_format != "NCHW":
return False
if padding_algorithm == "VALID":
if ((input_shape[2] - (dilations[0] * (filter_shape[2] - 1) + 1)) / strides[0] + 1) <= 1 or \
((input_shape[3] - (dilations[1] * (filter_shape[3] - 1) + 1)) / strides[1] + 1) <= 1:
return False
if padding_algorithm == "EXPLICIT":
if ((input_shape[2] + paddings[0] + paddings[1] - (dilations[0] * (filter_shape[2] - 1) + 1)) / strides[0] + 1) <= 1 or \
((input_shape[3] + paddings[2] + paddings[3] - (dilations[1] * (filter_shape[3] - 1) + 1)) / strides[1] + 1) <= 1:
return False
if data_format == "NCHW":
if input_shape[1] != filter_shape[1] * groups:
return False
if filter_shape[0] % groups != 0:
return False
else:
if input_shape[3] != filter_shape[1] * groups:
return False
if filter_shape[0] % groups != 0:
return False
return True
def sample_program_config(self, draw):
# 1. Generate shape of input:X of conv2d
x_shape = draw(
st.lists(
st.integers(
min_value=1, max_value=100), min_size=4, max_size=4))
x_shape[1] = draw(st.integers(min_value=1, max_value=10))
# 2. Generate legal attr:data_format of conv2d
data_format = draw(st.sampled_from(["NCHW", "NHWC"]))
# 3. Generate legal shape of input:Y of conv2d
f_shape = draw(
st.lists(
st.integers(
min_value=1, max_value=7), min_size=4, max_size=4))
if data_format == "NCHW":
f_shape[1] = x_shape[1]
else:
f_shape[1] = x_shape[3]
# 4. Generate legal attr:strides of conv2d
strides = draw(
st.lists(
st.integers(
min_value=1, max_value=5), min_size=2, max_size=2))
# 5. Generate legal attr:padding_algorithm of conv2d
padding_algorithm = draw(st.sampled_from(["EXPLICIT", "SAME", "VALID"]))
# 6. Generate legal attr:padding of conv2d
padding = draw(
st.lists(
st.integers(
min_value=1, max_value=5), min_size=4, max_size=4))
# 7. Generate legal attr:groups of conv2d
groups = draw(st.integers(min_value=1, max_value=3))
# 8. Generate legal attr:dilations of conv2d
dilations = draw(
st.lists(
st.integers(
min_value=1, max_value=5), min_size=2, max_size=2))
# 9. Generate legal shape of input:bias of elementwise_add
bias_shape = [f_shape[0]]
# 10. Generate legal shape of attr:axis of elementwise_add
axis = 1
if data_format == "NCHW":
axis = 1
else:
axis = 3
# 11. Generate legal shape of input:bias of conv2d
conv_bias_shape = []
inputs = dict()
weights = dict()
use_mkldnn = None
if draw(st.booleans()):
conv_bias_shape = [f_shape[0]]
inputs = {
"Input": ["input_x"],
"Filter": ["filter"],
"Bias": ["conv_bias"],
}
weights = {
"filter": TensorConfig(shape=f_shape),
"bias": TensorConfig(shape=bias_shape),
"conv_bias": TensorConfig(shape=conv_bias_shape)
}
use_mkldnn = True
else:
inputs = {
"Input": ["input_x"],
"Filter": ["filter"],
}
weights = {
"filter": TensorConfig(shape=f_shape),
"bias": TensorConfig(shape=bias_shape)
}
use_mkldnn = False
conv2d_op = OpConfig(
"conv2d",
inputs=inputs,
outputs={"Output": ["conv2d_out"]},
strides=strides,
padding_algorithm=padding_algorithm,
paddings=padding,
groups=groups,
dilations=dilations,
data_format=data_format,
use_mkldnn=use_mkldnn)
add_op = OpConfig(
"elementwise_add",
inputs={"X": ["conv2d_out"],
"Y": ["bias"]},
outputs={"Out": ["add_out"]},
axis=axis)
ops = [conv2d_op, add_op]
program_config = ProgramConfig(
ops=ops,
weights=weights,
inputs={"input_x": TensorConfig(shape=x_shape)},
outputs=ops[-1].outputs["Out"])
return program_config
def test(self):
self.run_and_statis(
quant=False,
max_examples=350,
passes=["conv_bias_mkldnn_fuse_pass"])
if __name__ == "__main__":
unittest.main()
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
# 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.
......@@ -12,44 +12,223 @@
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import print_function
import unittest
from auto_scan_test import PassAutoScanTest, SkipReasons
from program_config import TensorConfig, ProgramConfig, OpConfig
import numpy as np
from inference_pass_test import InferencePassTest
import paddle.fluid as fluid
import paddle.fluid.core as core
from paddle.fluid.core import PassVersionChecker
from paddle.fluid.core import AnalysisConfig
"""Test for fusion of conv, elementwise_add and 2 act."""
class ConvElementwiseAdd2ActFusePassTest(InferencePassTest):
def setUp(self):
with fluid.program_guard(self.main_program, self.startup_program):
data = fluid.data(
name="data", shape=[-1, 3, 100, 100], dtype="float32")
add_y2 = fluid.data(
name="add_y2", shape=[1, 3, 98, 98], dtype="float32")
conv_out = fluid.layers.conv2d(
input=data, num_filters=3, filter_size=3, bias_attr=None)
add1_out = fluid.layers.elementwise_add(
add_y2, conv_out, act="relu")
self.feeds = {
"data": np.random.random((1, 3, 100, 100)).astype("float32"),
"add_y2": np.random.random((1, 3, 98, 98)).astype("float32")
}
self.fetch_list = [add1_out]
self.enable_mkldnn = False
def test_check_output(self):
if core.is_compiled_with_cuda():
use_gpu = True
self.check_output_with_option(use_gpu)
self.assertTrue(
PassVersionChecker.IsCompatible(
'conv_elementwise_add2_act_fuse_pass'))
import paddle.inference as paddle_infer
from functools import partial
from typing import Optional, List, Callable, Dict, Any, Set
import unittest
import hypothesis
from hypothesis import given, settings, seed, example, assume, reproduce_failure
import hypothesis.strategies as st
class TestConvElementwiseAdd2ActPass(PassAutoScanTest):
"""
x_var f_var(persistable)
\ /
conv2d
|
conv2d_var y_var(persistable)
\ /
elementwise_add
|
x1_var elementwise_add_out_var
\ /
elementwise_add
|
act
|
act_var
"""
def sample_predictor_configs(self, program_config):
# for gpu
config = self.create_inference_config(use_gpu=True)
yield config, ["conv2d_fusion"], (1e-4, 1e-5)
def is_program_valid(self, prog_config):
paddings = prog_config.ops[0].attrs["paddings"]
strides = prog_config.ops[0].attrs["strides"]
groups = prog_config.ops[0].attrs["groups"]
padding_algorithm = prog_config.ops[0].attrs["padding_algorithm"]
dilations = prog_config.ops[0].attrs["dilations"]
data_format = prog_config.ops[0].attrs["data_format"]
filter_shape = prog_config.weights["filter"].shape
input_shape = prog_config.inputs["input_x"].shape
if data_format != "NCHW":
return False
if padding_algorithm == "VALID":
if int(((input_shape[2] - (dilations[0] * (filter_shape[2] - 1) + 1)) / strides[0] + 1)) <= 0 or \
int(((input_shape[3] - (dilations[1] * (filter_shape[3] - 1) + 1)) / strides[1] + 1)) <= 0:
return False
if padding_algorithm == "EXPLICIT":
if int(((input_shape[2] + paddings[0] + paddings[1] - (dilations[0] * (filter_shape[2] - 1) + 1)) / strides[0] + 1)) <= 0 or \
int(((input_shape[3] + paddings[2] + paddings[3] - (dilations[1] * (filter_shape[3] - 1) + 1)) / strides[1] + 1)) <= 0:
return False
if padding_algorithm == "SAME":
if int((input_shape[2] + strides[0] - 1) / strides[0]) <= 0 or int(
(input_shape[3] + strides[1] - 1) / strides[1]) <= 0:
return False
if data_format == "NCHW":
if input_shape[1] != filter_shape[1] * groups:
return False
if filter_shape[0] % groups != 0:
return False
else:
if input_shape[3] != filter_shape[1] * groups:
return False
if filter_shape[0] % groups != 0:
return False
return True
def sample_program_config(self, draw):
is_not_valid = True
program_config = None
while is_not_valid:
# 1. Generate shape of input:X of conv2d
x_shape = draw(
st.lists(
st.integers(
min_value=1, max_value=100),
min_size=4,
max_size=4))
x_shape[1] = draw(st.integers(min_value=1, max_value=10))
# 2. Generate legal attr:data_format of conv2d
data_format = draw(st.sampled_from(["NCHW", "NHWC"]))
# 3. Generate legal shape of input:Y of conv2d
f_shape = draw(
st.lists(
st.integers(
min_value=1, max_value=7),
min_size=4,
max_size=4))
if data_format == "NCHW":
f_shape[1] = x_shape[1]
else:
f_shape[1] = x_shape[3]
# 4. Generate legal attr:strides of conv2d
strides = draw(
st.lists(
st.integers(
min_value=1, max_value=5),
min_size=2,
max_size=2))
# 5. Generate legal attr:padding_algorithm of conv2d
padding_algorithm = draw(
st.sampled_from(["EXPLICIT", "SAME", "VALID"]))
# 6. Generate legal attr:padding of conv2d
padding = draw(
st.lists(
st.integers(
min_value=1, max_value=5),
min_size=4,
max_size=4))
# 7. Generate legal attr:groups of conv2d
groups = draw(st.integers(min_value=1, max_value=3))
# 8. Generate legal attr:dilations of conv2d
dilations = draw(
st.lists(
st.integers(
min_value=1, max_value=5),
min_size=2,
max_size=2))
# 9. Generate legal elemntwise_add: X of conv2d
bias_2_dict = dict()
bias_2_dict[1] = [x_shape[0], f_shape[0], \
int(((x_shape[2] + padding[0] + padding[1] - (dilations[0] * (f_shape[2] - 1) + 1)) / strides[0] + 1)), \
int(((x_shape[3] + padding[2] + padding[3] - (dilations[1] * (f_shape[3] - 1) + 1)) / strides[1] + 1))]
bias_2_dict[2] = [x_shape[0], f_shape[0], \
int((x_shape[2] + strides[0] - 1) / strides[0]), \
int((x_shape[3] + strides[1] - 1) / strides[1])]
bias_2_dict[3] = [x_shape[0], f_shape[0], \
int(((x_shape[2] - (dilations[0] * (f_shape[2] - 1) + 1)) / strides[0] + 1)), \
int(((x_shape[3] - (dilations[1] * (f_shape[3] - 1) + 1)) / strides[1] + 1))]
bias_index = 1
if padding_algorithm == "SAME":
bias_index = 2
if padding_algorithm == "VALID":
bias_index = 3
bias_2_shape = bias_2_dict[bias_index]
if np.sum(np.array(bias_2_shape) <= 0) == 0:
is_not_valid = False
else:
continue
# 10. Generate legal shape of input:bias of elementwise_add
bias_shape = [f_shape[0]]
# 11. Generate legal attr:axis of elementwise_add_1
axis_1 = 1
# 12. Generate legal attr:axis of elementwise_add_2
axis_2 = -1
conv2d_op = OpConfig(
"conv2d",
inputs={"Input": ["input_x"],
"Filter": ["filter"]},
outputs={"Output": ["conv2d_out"]},
strides=strides,
padding_algorithm=padding_algorithm,
paddings=padding,
groups=groups,
dilations=dilations,
data_format=data_format)
add_1_op = OpConfig(
"elementwise_add",
inputs={"X": ["conv2d_out"],
"Y": ["bias_1"]},
outputs={"Out": ["add_1_out"]},
axis=axis_1)
add_2_op = OpConfig(
"elementwise_add",
inputs={"X": ["bias_2"],
"Y": ["add_1_out"]},
outputs={"Out": ["add_out"]},
axis=axis_2)
relu_op = OpConfig(
"relu",
inputs={"X": ["add_out"]},
outputs={"Out": ["relu_out"]})
ops = [conv2d_op, add_1_op, add_2_op, relu_op]
program_config = ProgramConfig(
ops=ops,
weights={
"filter": TensorConfig(shape=f_shape),
"bias_1": TensorConfig(shape=bias_shape),
},
inputs={
"input_x": TensorConfig(shape=x_shape),
"bias_2": TensorConfig(shape=bias_2_shape)
},
outputs=ops[-1].outputs["Out"], )
return program_config
def test(self):
self.run_and_statis(
quant=False,
max_examples=300,
passes=["conv_elementwise_add2_act_fuse_pass"])
if __name__ == "__main__":
......
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
# 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.
......@@ -12,46 +12,177 @@
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import print_function
import unittest
from auto_scan_test import PassAutoScanTest, IgnoreReasons
from program_config import TensorConfig, ProgramConfig, OpConfig
import numpy as np
from inference_pass_test import InferencePassTest
import paddle.fluid as fluid
import paddle.fluid.core as core
from paddle.fluid.core import PassVersionChecker
from paddle.fluid.core import AnalysisConfig
"""Test for fusion of conv, elementwise_add and act."""
class ConvElementwiseAddActFusePassTest(InferencePassTest):
def setUp(self):
with fluid.program_guard(self.main_program, self.startup_program):
data = fluid.data(
name="data", shape=[-1, 3, 100, 100], dtype="float32")
param_attr = fluid.ParamAttr(
initializer=fluid.initializer.Xavier(uniform=False),
learning_rate=0.001)
conv_out = fluid.layers.conv2d(
input=data,
num_filters=3,
filter_size=3,
bias_attr=param_attr,
act="relu")
self.feeds = {
"data": np.random.random((1, 3, 100, 100)).astype("float32")
}
self.fetch_list = [conv_out]
self.enable_mkldnn = False
def test_check_output(self):
if core.is_compiled_with_cuda():
use_gpu = True
self.check_output_with_option(use_gpu)
self.assertTrue(
PassVersionChecker.IsCompatible(
'conv_elementwise_add_act_fuse_pass'))
import paddle.inference as paddle_infer
from functools import partial
from typing import Optional, List, Callable, Dict, Any, Set
import unittest
import hypothesis
from hypothesis import given, settings, seed, example, assume, reproduce_failure
import hypothesis.strategies as st
class TestConvElementwiseAddActPass(PassAutoScanTest):
"""
x_var f_var(persistable)
\ /
conv2d
|
conv2d_var y_var(persistable)
\ /
elementwise_add
|
elementwise_add_var
|
act
|
act_var
"""
def sample_predictor_configs(self, program_config):
# for gpu
config = self.create_inference_config(use_gpu=True)
yield config, ["conv2d_fusion"], (1e-4, 1e-5)
def is_program_valid(self, prog_config):
paddings = prog_config.ops[0].attrs["paddings"]
strides = prog_config.ops[0].attrs["strides"]
groups = prog_config.ops[0].attrs["groups"]
padding_algorithm = prog_config.ops[0].attrs["padding_algorithm"]
dilations = prog_config.ops[0].attrs["dilations"]
data_format = prog_config.ops[0].attrs["data_format"]
filter_shape = prog_config.weights["filter"].shape
input_shape = prog_config.inputs["input_x"].shape
if data_format != "NCHW":
return False
if padding_algorithm == "VALID":
if ((input_shape[2] - (dilations[0] * (filter_shape[2] - 1) + 1)) / strides[0] + 1) <= 1 or \
((input_shape[3] - (dilations[1] * (filter_shape[3] - 1) + 1)) / strides[1] + 1) <= 1:
return False
if padding_algorithm == "EXPLICIT":
if ((input_shape[2] + paddings[0] + paddings[1] - (dilations[0] * (filter_shape[2] - 1) + 1)) / strides[0] + 1) <= 1 or \
((input_shape[3] + paddings[2] + paddings[3] - (dilations[1] * (filter_shape[3] - 1) + 1)) / strides[1] + 1) <= 1:
return False
if data_format == "NCHW":
if input_shape[1] != filter_shape[1] * groups:
return False
if filter_shape[0] % groups != 0:
return False
else:
if input_shape[3] != filter_shape[1] * groups:
return False
if filter_shape[0] % groups != 0:
return False
return True
def sample_program_config(self, draw):
# 1. Generate shape of input:X of conv2d
x_shape = draw(
st.lists(
st.integers(
min_value=1, max_value=100), min_size=4, max_size=4))
x_shape[1] = draw(st.integers(min_value=1, max_value=10))
# 2. Generate legal attr:data_format of conv2d
data_format = draw(st.sampled_from(["NCHW", "NHWC"]))
# 3. Generate legal shape of input:Y of conv2d
f_shape = draw(
st.lists(
st.integers(
min_value=1, max_value=7), min_size=4, max_size=4))
if data_format == "NCHW":
f_shape[1] = x_shape[1]
else:
f_shape[1] = x_shape[3]
# 4. Generate legal attr:strides of conv2d
strides = draw(
st.lists(
st.integers(
min_value=1, max_value=5), min_size=2, max_size=2))
# 5. Generate legal attr:padding_algorithm of conv2d
padding_algorithm = draw(st.sampled_from(["EXPLICIT", "SAME", "VALID"]))
# 6. Generate legal attr:padding of conv2d
padding = draw(
st.lists(
st.integers(
min_value=1, max_value=5), min_size=4, max_size=4))
# 7. Generate legal attr:groups of conv2d
groups = draw(st.integers(min_value=1, max_value=3))
# 8. Generate legal attr:dilations of conv2d
dilations = draw(
st.lists(
st.integers(
min_value=1, max_value=5), min_size=2, max_size=2))
# 9. Generate legal input:ResidualData of conv2d
res_shape = []
if draw(st.booleans()):
res_shape = draw(
st.lists(
st.integers(
min_value=1, max_value=100),
min_size=4,
max_size=4))
# 10. Generate legal shape of input:bias of elementwise_add
bias_shape = [f_shape[0]]
# 11. Generate legal attr:axis of elementwise_add
axis = 1
conv2d_op = OpConfig(
"conv2d",
inputs={
"Input": ["input_x"],
"Filter": ["filter"],
"ResidualData": ["residualdata"]
},
outputs={"Output": ["conv2d_out"]},
strides=strides,
padding_algorithm=padding_algorithm,
paddings=padding,
groups=groups,
dilations=dilations,
data_format=data_format)
add_op = OpConfig(
"elementwise_add",
inputs={"X": ["conv2d_out"],
"Y": ["bias"]},
outputs={"Out": ["add_out"]},
axis=axis)
relu_op = OpConfig(
"relu", inputs={"X": ["add_out"]}, outputs={"Out": ["relu_out"]})
ops = [conv2d_op, add_op, relu_op]
program_config = ProgramConfig(
ops=ops,
weights={
"filter": TensorConfig(shape=f_shape),
"bias": TensorConfig(shape=bias_shape),
},
inputs={
"input_x": TensorConfig(shape=x_shape),
"residualdata": TensorConfig(shape=res_shape)
},
outputs=ops[-1].outputs["Out"], )
return program_config
def test(self):
self.run_and_statis(
quant=False,
max_examples=400,
passes=["conv_elementwise_add_act_fuse_pass"])
if __name__ == "__main__":
......
# 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 auto_scan_test import PassAutoScanTest, SkipReasons
from program_config import TensorConfig, ProgramConfig, OpConfig
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
import hypothesis
from hypothesis import given, settings, seed, example, assume, reproduce_failure
import hypothesis.strategies as st
class TestConvEltwiseaddBnFusePass(PassAutoScanTest):
"""
x_var f_var(persistable)
\ /
conv2d
|
conv2d_var bias_var(persistable)
\ /
elementwise_add
|
elementwise_add_var Scale(persistable) Bias(persistable) Mean(persistable) Variance(persistable)
|
batch_norm
|
Y MeanOut VarianceOut SavedMeanSavedVariance
"""
def sample_predictor_configs(self, program_config):
# cpu
config = self.create_inference_config(use_gpu=False)
yield config, ["conv2d", "elementwise_add"], (1e-4, 1e-5)
# MKLDNN
config = self.create_inference_config(use_gpu=False)
config.enable_mkldnn()
yield config, ["conv2d", "elementwise_add"], (1e-4, 1e-5)
# for gpu
config = self.create_inference_config(use_gpu=True)
yield config, ["conv2d", "elementwise_add"], (1e-4, 1e-5)
def is_program_valid(self, prog_config):
paddings = prog_config.ops[0].attrs["paddings"]
strides = prog_config.ops[0].attrs["strides"]
groups = prog_config.ops[0].attrs["groups"]
padding_algorithm = prog_config.ops[0].attrs["padding_algorithm"]
dilations = prog_config.ops[0].attrs["dilations"]
data_format = prog_config.ops[0].attrs["data_format"]
filter_shape = prog_config.weights["filter"].shape
input_shape = prog_config.inputs["input_x"].shape
if data_format != "NCHW":
return False
if padding_algorithm == "VALID":
if ((input_shape[2] - (dilations[0] * (filter_shape[2] - 1) + 1)) / strides[0] + 1) <= 1 or \
((input_shape[3] - (dilations[1] * (filter_shape[3] - 1) + 1)) / strides[1] + 1) <= 1:
return False
if padding_algorithm == "EXPLICIT":
if ((input_shape[2] + paddings[0] + paddings[1] - (dilations[0] * (filter_shape[2] - 1) + 1)) / strides[0] + 1) <= 1 or \
((input_shape[3] + paddings[2] + paddings[3] - (dilations[1] * (filter_shape[3] - 1) + 1)) / strides[1] + 1) <= 1:
return False
if data_format == "NCHW":
if input_shape[1] != filter_shape[1] * groups:
return False
if filter_shape[0] % groups != 0:
return False
else:
if input_shape[3] != filter_shape[1] * groups:
return False
if filter_shape[0] % groups != 0:
return False
bn_scale = np.array(prog_config.weights["scale_in"].data)
bn_bias = np.array(prog_config.weights["bias_in"].data)
bn_mean = np.array(prog_config.weights["mean_in"].data)
bn_variance = np.array(prog_config.weights["variance_in"].data)
epsilon = np.array(prog_config.ops[-1].attrs["epsilon"])
bn_variance = bn_variance + epsilon
if np.isnan(bn_variance).any():
return False
bn_variance = np.sqrt(bn_variance)
if np.sum(bn_variance == 0.0) > 0:
return False
bn_variance = bn_scale / bn_variance
if np.isnan(bn_variance).any():
return False
return True
def sample_program_config(self, draw):
# 1. Generate shape of input:X of conv2d
x_shape = draw(
st.lists(
st.integers(
min_value=10, max_value=100),
min_size=4,
max_size=4))
x_shape[1] = draw(st.integers(min_value=1, max_value=10))
# 2. Generate legal attr:data_format of conv2d
data_format = draw(st.sampled_from(["NCHW", "NHWC"]))
# 2. Generate legal shape of input:Y of conv2d
f_shape = draw(
st.lists(
st.integers(
min_value=1, max_value=7), min_size=4, max_size=4))
if data_format == "NCHW":
f_shape[1] = x_shape[1]
else:
f_shape[1] = x_shape[3]
# 3. Generate legal attr:strides of conv2d
strides = draw(
st.lists(
st.integers(
min_value=1, max_value=5), min_size=2, max_size=2))
# 4. Generate legal attr:padding_algorithm of conv2d
padding_algorithm = draw(st.sampled_from(["EXPLICIT", "SAME", "VALID"]))
# 5. Generate legal attr:padding of conv2d
padding = draw(
st.lists(
st.integers(
min_value=1, max_value=5), min_size=4, max_size=4))
# 6. Generate legal attr:groups of conv2d
groups = draw(st.integers(min_value=1, max_value=3))
# 7. Generate legal attr:dilations of conv2d
dilations = draw(
st.lists(
st.integers(
min_value=1, max_value=5), min_size=2, max_size=2))
# 9. Generate legal input:ResidualData of conv2d
res_shape = []
if draw(st.booleans()):
res_shape = draw(
st.lists(
st.integers(
min_value=1, max_value=100),
min_size=4,
max_size=4))
# 10. Generate legal shape of input:bias of elementwise_add
bias_shape = [f_shape[0]]
# 11. Generate legal attr:axis of elementwise_add
axis = 1
# 12. Generate legal input:Scale of batch_norm
bn_scale_shape = [f_shape[0]]
# 13. Generate legal input:Bias of batch_norm
bn_bias_shape = [f_shape[0]]
# 14. Generate legal input:Mean of batch_norm
bn_mean_shape = [f_shape[0]]
# 15. Generate legal input:Variance of batch_norm
bn_variance_shape = [f_shape[0]]
# 16. Generate legal attr:epsilon of batch_norm
epsilon = draw(st.floats(min_value=0.00001, max_value=0.001))
def generate_batch_variance():
return (0.1 + (1.0 - 0.1) * np.random.random(bn_variance_shape)
).astype(np.float32)
conv2d_op = OpConfig(
"conv2d",
inputs={
"Input": ["input_x"],
"Filter": ["filter"],
"ResidualData": ["residualdata"]
},
outputs={"Output": ["conv2d_out"]},
strides=strides,
padding_algorithm=padding_algorithm,
paddings=padding,
groups=groups,
dilations=dilations,
data_format=data_format)
add_op = OpConfig(
"elementwise_add",
inputs={"X": ["conv2d_out"],
"Y": ["bias"]},
outputs={"Out": ["add_out"]},
axis=axis)
bn_op = OpConfig(
"batch_norm",
inputs={
"X": ["add_out"],
"Scale": ["scale_in"],
"Bias": ["bias_in"],
"Mean": ["mean_in"],
"Variance": ["variance_in"]
},
outputs={
"Y": ["y_out"],
"MeanOut": ["mean_in"],
"VarianceOut": ["variance_in"],
"SavedMean": ["SavedMean_out"],
"SavedVariance": ["SavedVariance_out"],
"ReserveSpace": ["ReserveSpace_out"]
},
epsilon=epsilon,
is_test=True,
trainable_statistics=False,
data_layout=data_format)
ops = [conv2d_op, add_op, bn_op]
# 17. if the output of bias is more than one
if draw(st.booleans()):
outputs = ops[-1].outputs["Y"]
else:
outputs = ops[-1].outputs["Y"] + ["bias"]
program_config = ProgramConfig(
ops=ops,
weights={
"filter": TensorConfig(shape=f_shape),
"bias": TensorConfig(shape=bias_shape),
"scale_in": TensorConfig(shape=bn_scale_shape),
"bias_in": TensorConfig(shape=bn_bias_shape),
"mean_in": TensorConfig(shape=bn_mean_shape),
"variance_in": TensorConfig(data_gen=generate_batch_variance),
},
inputs={
"input_x": TensorConfig(shape=x_shape),
"residualdata": TensorConfig(shape=res_shape)
},
outputs=outputs)
return program_config
def test(self):
self.run_and_statis(
quant=False,
max_examples=300,
passes=["conv_eltwiseadd_bn_fuse_pass"])
if __name__ == "__main__":
unittest.main()
......@@ -431,7 +431,7 @@ HIGH_PARALLEL_JOB_NEW = [
'test_memory_usage',
'test_sysconfig',
'reader_test',
'test_conv_bias_mkldnn_fuse_pass',
'test_conv_bias_mkldnn_fuse_pass_cc',
'math_function_test',
'beam_search_decode_op_test',
'save_quant2_model_resnet50',
......@@ -1469,7 +1469,7 @@ CPU_PARALLEL_JOB = [
'test_cpu_bfloat16_placement_pass',
'test_cpu_bfloat16_pass',
'test_conv_concat_relu_mkldnn_fuse_pass',
'test_conv_bias_mkldnn_fuse_pass',
'test_conv_bias_mkldnn_fuse_pass_cc',
'test_conv_batch_norm_mkldnn_fuse_pass',
'test_conv3d_transpose_layer',
'test_conv3d_mkldnn_op',
......
文件模式从 100644 更改为 100755
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册