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 conv.""" 16from __future__ import absolute_import 17from __future__ import division 18from __future__ import print_function 19 20import numpy as np 21import tensorflow.compat.v1 as tf 22from tensorflow.lite.testing.zip_test_utils import create_tensor_data 23from tensorflow.lite.testing.zip_test_utils import make_zip_of_tests 24from tensorflow.lite.testing.zip_test_utils import register_make_test_function 25 26 27@register_make_test_function() 28def make_conv_tests(options): 29 """Make a set of tests to do convolution.""" 30 31 test_parameters = [ 32 { 33 "input_shape": [[1, 3, 4, 3], [4, 6, 6, 1]], 34 "filter_shape": [[1, 1], [2, 3], [3, 3]], 35 "strides": [[1, 1, 1, 1], [1, 2, 3, 1]], 36 "dilations": [[1, 1, 1, 1], [1, 3, 2, 1], [1, 2, 2, 1]], 37 "padding": ["SAME", "VALID"], 38 "data_format": ["NHWC"], # TODO(aselle): NCHW would be good 39 "constant_filter": [True, False], 40 "channel_multiplier": [1, 2], 41 "fully_quantize": [False], 42 "quant_16x8": [False], 43 "dynamic_range_quantize": [False] 44 }, 45 { 46 "input_shape": [[1, 3, 4, 3]], 47 "filter_shape": [[1, 1], [2, 3]], 48 "strides": [[1, 1, 1, 1]], 49 "dilations": [[1, 1, 1, 1]], 50 "padding": ["SAME"], 51 "data_format": ["NHWC"], 52 "constant_filter": [True], 53 "channel_multiplier": [1, 2], 54 "fully_quantize": [True], 55 "quant_16x8": [True], 56 "dynamic_range_quantize": [False], 57 }, 58 # TODO(b/134702301): The fully_quantize param is just ignored by the MLIR 59 # testing path now, resulting in duplicate tests. Either ignore these 60 # tests or handle it properly in the mlir_convert() function. 61 { 62 "input_shape": [[1, 3, 4, 3], [4, 6, 6, 1]], 63 "filter_shape": [[1, 1], [2, 3], [3, 3]], 64 "strides": [[1, 1, 1, 1], [1, 2, 3, 1]], 65 "dilations": [[1, 1, 1, 1], [1, 3, 2, 1], [1, 2, 2, 1]], 66 "padding": ["SAME", "VALID"], 67 "data_format": ["NHWC"], # TODO(aselle): NCHW would be good 68 "constant_filter": [True], 69 "channel_multiplier": [1, 2], 70 "fully_quantize": [True], 71 "quant_16x8": [False], 72 "dynamic_range_quantize": [False] 73 }, 74 { 75 "input_shape": [[1, 3, 4, 3]], 76 "filter_shape": [[1, 1]], 77 "strides": [[1, 1, 1, 1], [1, 2, 3, 1]], 78 "dilations": [[1, 1, 1, 1]], 79 "padding": ["SAME", "VALID"], 80 "data_format": ["NHWC"], 81 "constant_filter": [True], 82 "channel_multiplier": [2], 83 "fully_quantize": [False], 84 "quant_16x8": [False], 85 "dynamic_range_quantize": [True] 86 }, 87 ] 88 89 def get_tensor_shapes(parameters): 90 input_shape = parameters["input_shape"] 91 filter_size = parameters["filter_shape"] 92 filter_shape = filter_size + [ 93 input_shape[3], parameters["channel_multiplier"] 94 ] 95 return [input_shape, filter_shape] 96 97 def build_graph(parameters): 98 """Build a conv graph given `parameters`.""" 99 input_shape, filter_shape = get_tensor_shapes(parameters) 100 input_tensor = tf.compat.v1.placeholder( 101 dtype=tf.float32, name="input", shape=input_shape) 102 103 # Get filter input either as a placeholder or constants. Also get a list of 104 # the input tensors that are represented as placeholders. 105 if parameters["constant_filter"]: 106 filter_input = create_tensor_data( 107 np.float32, filter_shape, min_value=-10, max_value=10) 108 input_tensors = [input_tensor] 109 else: 110 filter_input = tf.compat.v1.placeholder( 111 dtype=tf.float32, name="filter", shape=filter_shape) 112 input_tensors = [input_tensor, filter_input] 113 114 out = tf.nn.conv2d( 115 input_tensor, 116 filter_input, 117 strides=parameters["strides"], 118 dilations=parameters["dilations"], 119 padding=parameters["padding"], 120 data_format=parameters["data_format"]) 121 return input_tensors, [out] 122 123 def build_inputs(parameters, sess, inputs, outputs): 124 # Build list of input values either containing 1 tensor (input) or 2 tensors 125 # (input, filter) based on whether filter is constant or variable input. 126 input_shape, filter_shape = get_tensor_shapes(parameters) 127 values = [ 128 create_tensor_data(np.float32, input_shape, min_value=-1, max_value=1) 129 ] 130 if not parameters["constant_filter"]: 131 values.append(create_tensor_data(np.float32, filter_shape)) 132 return values, sess.run(outputs, feed_dict=dict(zip(inputs, values))) 133 134 make_zip_of_tests( 135 options, 136 test_parameters, 137 build_graph, 138 build_inputs, 139 expected_tf_failures=60) 140