1# Copyright 2019 The TensorFlow Authors. All Rights Reserved. 2# 3# Licensed under the Apache License, Version 2.0 (the "License"); 4# you may not use this file except in compliance with the License. 5# You may obtain a copy of the License at 6# 7# http://www.apache.org/licenses/LICENSE-2.0 8# 9# Unless required by applicable law or agreed to in writing, software 10# distributed under the License is distributed on an "AS IS" BASIS, 11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12# See the License for the specific language governing permissions and 13# limitations under the License. 14# ============================================================================== 15"""Test configs for fused_batch_norm.""" 16from __future__ import absolute_import 17from __future__ import division 18from __future__ import print_function 19 20import numpy as np 21 22import tensorflow.compat.v1 as tf 23from tensorflow.lite.testing.zip_test_utils import create_tensor_data 24from tensorflow.lite.testing.zip_test_utils import make_zip_of_tests 25from tensorflow.lite.testing.zip_test_utils import register_make_test_function 26 27 28@register_make_test_function() 29def make_fused_batch_norm_tests(options): 30 """Make a set of tests to do fused_batch_norm.""" 31 32 test_parameters = [{ 33 "dtype": [tf.float32], 34 "input_shape": [[1, 1, 6, 2]], 35 "epsilon": [0.001, 0.1], 36 "is_training": [False], 37 }] 38 39 # Training support in MLIR converter. 40 if options.use_experimental_converter: 41 test_parameters = test_parameters + [ 42 { 43 "dtype": [tf.float32], 44 "input_shape": [[1, 1, 6, 2]], 45 "epsilon": [0.001, 0.1], 46 "is_training": [True], 47 }, 48 { 49 "dtype": [tf.float32], 50 "input_shape": [[1, None, 6, 2]], 51 "epsilon": [0.001, 0.1], 52 "is_training": [True, False], 53 }, 54 ] 55 56 def build_graph(parameters): 57 """Build the testing graph for fused batch normalization.""" 58 input_shape = parameters["input_shape"] 59 scale_shape = input_shape[3] 60 61 scale = create_tensor_data(parameters["dtype"], scale_shape) 62 offset = create_tensor_data(parameters["dtype"], scale_shape) 63 mean = create_tensor_data(parameters["dtype"], scale_shape) 64 variance = create_tensor_data(parameters["dtype"], scale_shape) 65 66 x = tf.compat.v1.placeholder( 67 dtype=parameters["dtype"], name="x", shape=parameters["input_shape"]) 68 [x_norm, _, _] = tf.compat.v1.nn.fused_batch_norm( 69 x, 70 scale, 71 offset, 72 mean, 73 variance, 74 parameters["epsilon"], 75 data_format="NHWC", 76 is_training=parameters["is_training"]) 77 78 input_tensor = tf.compat.v1.placeholder( 79 dtype=parameters["dtype"], 80 name="input", 81 shape=parameters["input_shape"]) 82 out = tf.add(input_tensor, x_norm) 83 return [x, input_tensor], [out] 84 85 def build_inputs(parameters, sess, inputs, outputs): 86 # Fill dynamic shape with a random number. 87 input_shape = parameters["input_shape"] 88 input_shape = [ 89 np.random.randint(1, 10) if v is None else v for v in input_shape 90 ] 91 92 input_values = [ 93 create_tensor_data(parameters["dtype"], input_shape), 94 create_tensor_data(parameters["dtype"], input_shape) 95 ] 96 97 return input_values, sess.run( 98 outputs, feed_dict=dict(zip(inputs, input_values))) 99 100 make_zip_of_tests(options, test_parameters, build_graph, build_inputs) 101