• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1# Copyright 2021 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"""Tests for Tensorflow -> jitrt compilation."""
16
17import numpy as np
18
19from tensorflow.compiler.mlir.tfrt.jit.python_binding import tf_jitrt
20from tensorflow.python.platform import test
21
22specializations = [
23    tf_jitrt.Specialization.ENABLED,
24    tf_jitrt.Specialization.DISABLED,
25    tf_jitrt.Specialization.ALWAYS,
26]
27
28jitrt = tf_jitrt.TfJitRtExecutor()
29
30
31class TfReshapeTest(test.TestCase):
32
33  def test_reshape_unknown_1d(self):
34    for specialize in specializations:
35      mlir_function = """
36        func.func @test(%arg0: tensor<?xf32>, %arg1: tensor<2xi32>)
37            -> tensor<?x?xf32> {
38          %0 = "tf.Reshape"(%arg0, %arg1)
39              : (tensor<?xf32>, tensor<2xi32>) -> tensor<?x?xf32>
40          func.return %0 : tensor<?x?xf32>
41        }"""
42
43      compiled = jitrt.compile(mlir_function, 'test', specialize)
44
45      d0 = np.random.randint(1, 10) * 2
46
47      arg0 = np.random.uniform(0, 10.0, size=(d0)).astype(np.float32)
48
49      shape = np.array([2, d0 / 2]).astype(np.int32)
50      [res] = jitrt.execute(compiled, [arg0, shape])
51      np.testing.assert_allclose(res, np.reshape(arg0, shape), atol=0.0)
52
53      shape = np.array([2, -1]).astype(np.int32)
54      [res] = jitrt.execute(compiled, [arg0, shape])
55      np.testing.assert_allclose(res, np.reshape(arg0, shape), atol=0.0)
56
57      with self.assertRaises(RuntimeError):
58        shape = np.array([30, -1]).astype(np.int32)
59        [res] = jitrt.execute(compiled, [arg0, shape])
60
61  def test_reshape_unknown_2d(self):
62    for specialize in specializations:
63      mlir_function = """
64        func.func @test(%arg0: tensor<?x?xf32>, %arg1: tensor<1xi32>)
65            -> tensor<?xf32> {
66          %0 = "tf.Reshape"(%arg0, %arg1)
67              : (tensor<?x?xf32>, tensor<1xi32>) -> tensor<?xf32>
68          func.return %0 : tensor<?xf32>
69        }"""
70
71      compiled = jitrt.compile(mlir_function, 'test', specialize)
72
73      d0 = np.random.randint(1, 10) * 2
74      d1 = np.random.randint(1, 10) * 2
75
76      arg0 = np.random.uniform(0, 10.0, size=(d0, d1)).astype(np.float32)
77
78      shape = np.array([d0 * d1]).astype(np.int32)
79      [res] = jitrt.execute(compiled, [arg0, shape])
80      np.testing.assert_allclose(res, np.reshape(arg0, shape), atol=0.0)
81
82      shape = np.array([-1]).astype(np.int32)
83      [res] = jitrt.execute(compiled, [arg0, shape])
84      np.testing.assert_allclose(res, np.reshape(arg0, shape), atol=0.0)
85
86      with self.assertRaises(RuntimeError):
87        shape = np.array([d0]).astype(np.int32)
88        [res] = jitrt.execute(compiled, [arg0, shape])
89
90  def test_reshape_zero_dim(self):
91    for specialize in specializations:
92      mlir_function = """
93        func.func @test(%arg0: tensor<?xf32>, %arg1: tensor<1xi32>)
94            -> tensor<?xf32> {
95          %0 = "tf.Reshape"(%arg0, %arg1)
96              : (tensor<?xf32>, tensor<1xi32>) -> tensor<?xf32>
97          func.return %0 : tensor<?xf32>
98        }"""
99
100      compiled = jitrt.compile(mlir_function, 'test', specialize)
101
102      empty = np.array([]).astype(np.float32)
103
104      zero = np.array([0]).astype(np.int32)
105      [res] = jitrt.execute(compiled, [empty, zero])
106      np.testing.assert_equal(res.shape, [0])
107
108      neg1 = np.array([-1]).astype(np.int32)
109      [res] = jitrt.execute(compiled, [empty, neg1])
110      np.testing.assert_equal(res.shape, [0])
111
112      with self.assertRaises(RuntimeError):
113        neg2 = np.array([-2]).astype(np.int32)
114        [res] = jitrt.execute(compiled, [empty, neg2])
115
116      with self.assertRaises(RuntimeError):
117        one = np.array([1]).astype(np.int32)
118        [res] = jitrt.execute(compiled, [empty, one])
119
120  def test_reshape_zero_dim_3d(self):
121    for specialize in specializations:
122      mlir_function = """
123        func.func @test(%arg0: tensor<?xf32>, %arg1: tensor<3xi32>)
124            -> tensor<?x?x?xf32> {
125          %0 = "tf.Const"() { value = dense<[3, 0, 5]> : tensor<3xi32> }
126              : () -> tensor<3xi32>
127          %1 = "tf.Reshape"(%arg0, %0)
128              : (tensor<?xf32>, tensor<3xi32>) -> tensor<?x?x?xf32>
129          %2 = "tf.Reshape"(%1, %arg1)
130              : (tensor<?x?x?xf32>, tensor<3xi32>) -> tensor<?x?x?xf32>
131          func.return %2 : tensor<?x?x?xf32>
132        }"""
133
134      compiled = jitrt.compile(mlir_function, 'test', specialize)
135
136      empty = np.array([]).astype(np.float32)
137
138      shape = np.array([3, 0, -1]).astype(np.int32)
139      [res] = jitrt.execute(compiled, [empty, shape])
140      # TODO(kramerb): This should be [3, 0, 5]
141      np.testing.assert_equal(res.shape, [3, 0, 0])
142
143      with self.assertRaises(RuntimeError):
144        shape = np.array([3, -1, -1]).astype(np.int32)
145        [res] = jitrt.execute(compiled, [empty, shape])
146
147if __name__ == '__main__':
148  test.main()
149