|
| 1 | +# Copyright 2020 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 | + |
| 16 | + |
| 17 | +import tensorflow as tf |
| 18 | +from tf_helpers import lax |
| 19 | +from tensorflow.python.platform import test |
| 20 | +from absl.testing import parameterized |
| 21 | +import itertools |
| 22 | +import numpy as onp |
| 23 | +from tensorflow.python.ops import numpy_ops as tfnp |
| 24 | +from jax import numpy as jnp |
| 25 | +import jax |
| 26 | +import sys |
| 27 | + |
| 28 | + |
| 29 | +class TFLaxTest(tf.test.TestCase, parameterized.TestCase): |
| 30 | + |
| 31 | + @parameterized.parameters( |
| 32 | + {"lhs_np": onp.ones((5, 3)), "rhs_np": onp.ones((3, 2)), |
| 33 | + "dims": (((1,), (0,)), ((), ()))}, |
| 34 | + {"lhs_np": onp.ones((5, 3)), "rhs_np": onp.ones((5, 3)), |
| 35 | + "dims": (((0, 1), (0, 1)), ((), ()))}, |
| 36 | + {"lhs_np": onp.ones((5, 3, 2)), "rhs_np": onp.ones((2, 3, 2)), |
| 37 | + "dims": (((1, 2), (1, 0)), ((), ()))}, |
| 38 | + {"lhs_np": onp.ones((6, 5, 3)), "rhs_np": onp.ones((6, 3, 2)), |
| 39 | + "dims": (((2,), (1,)), ((0,), (0,)))}, |
| 40 | + {"lhs_np": onp.ones((6, 3, 5)), "rhs_np": onp.ones((6, 3, 2)), |
| 41 | + "dims": (((1,), (1,)), ((0,), (0,)))}, |
| 42 | + {"lhs_np": onp.ones((5, 3, 2, 2)), "rhs_np": onp.ones((5, 2, 2, 6)), |
| 43 | + "dims": (((2, 3), (1, 2)), ((0,), (0,)))}, |
| 44 | + {"lhs_np": onp.ones((2, 2, 5, 3)), "rhs_np": onp.ones((2, 2, 3, 2)), |
| 45 | + "dims": (((3,), (2,)), ((0, 1), (0, 1)))}, |
| 46 | + {"lhs_np": onp.ones((2, 2, 5, 2)), "rhs_np": onp.ones((2, 2, 3, 2)), |
| 47 | + "dims": (((3,), (1,)), ((0,), (0,)))}, |
| 48 | + {"lhs_np": onp.ones((2, 2, 5, 3, 3)), "rhs_np": onp.ones((2, 3, 2, 3, 2)), |
| 49 | + "dims": (((4,), (1,)), ((0,), (0,)))}, |
| 50 | + ) |
| 51 | + def test_tf_dot_general(self, lhs_np, rhs_np, dims): |
| 52 | + ans = jax.lax.dot_general(lhs_np, rhs_np, dims) |
| 53 | + result = lax.dot_general(lhs_np, rhs_np, dims) |
| 54 | + self.assertAllClose(result, tfnp.array(ans)) |
| 55 | + |
| 56 | + @parameterized.named_parameters([ |
| 57 | + ("_lhs_shape={}_rhs_shape={}_strides={}_padding={}" |
| 58 | + "_lhs_dilation={}_rhs_dilation={}" |
| 59 | + "_feature_group_count={}_batch_group_count={}_dims={}" |
| 60 | + "_perms={}".format(lhs_shape, rhs_shape, |
| 61 | + strides, padding, lhs_dilation, rhs_dilation, |
| 62 | + feature_group_count, batch_group_count, ",".join(dimension_numbers), perms), |
| 63 | + lhs_shape, rhs_shape, strides, padding, lhs_dilation, rhs_dilation, |
| 64 | + feature_group_count, batch_group_count, dimension_numbers, perms) |
| 65 | + for batch_group_count, feature_group_count in [(1, 1)] |
| 66 | + for lhs_shape, rhs_shape in [ |
| 67 | + ((b * batch_group_count, i * feature_group_count, 9, w), |
| 68 | + (j * feature_group_count * batch_group_count, i, 4, 5)) |
| 69 | + for w in [0, 10] |
| 70 | + for b, i, j in itertools.product([2, 3], repeat=3)] |
| 71 | + for strides in [(1, 1), (2, 1)] |
| 72 | + for padding in ['SAME'] |
| 73 | + for lhs_dilation, rhs_dilation in [ |
| 74 | + (None, (1, 1)) |
| 75 | + ] |
| 76 | + for dimension_numbers, perms in [ |
| 77 | + (("NHWC", "HWIO", "NHWC"), ([0, 2, 3, 1], [2, 3, 1, 0])) |
| 78 | + ]]) |
| 79 | + def testConvGeneralDilated(self, lhs_shape, rhs_shape, strides, |
| 80 | + padding, lhs_dilation, rhs_dilation, |
| 81 | + feature_group_count, batch_group_count, |
| 82 | + dimension_numbers, perms): |
| 83 | + tf.print("dimension_numbers: {}".format(dimension_numbers), output_stream=sys.stdout) |
| 84 | + lhs_perm, rhs_perm = perms # permute to compatible shapes |
| 85 | + |
| 86 | + lhs_tf = tfnp.transpose(tfnp.ones(lhs_shape), lhs_perm) |
| 87 | + rhs_tf = tfnp.transpose(tfnp.ones(rhs_shape), rhs_perm) |
| 88 | + |
| 89 | + lhs_jax = jnp.transpose(jnp.ones(lhs_shape), lhs_perm) |
| 90 | + rhs_jax = jnp.transpose(jnp.ones(rhs_shape), rhs_perm) |
| 91 | + |
| 92 | + jax_conv = jax.lax.conv_general_dilated(lhs_jax, rhs_jax, strides, padding, lhs_dilation, |
| 93 | + rhs_dilation, dimension_numbers, feature_group_count, batch_group_count) |
| 94 | + |
| 95 | + tf_conv = lax.conv_general_dilated(lhs_tf, rhs_tf, strides, padding, jax_conv.shape, lhs_dilation, |
| 96 | + rhs_dilation, dimension_numbers, feature_group_count, batch_group_count) |
| 97 | + |
| 98 | + self.assertAllEqual(tf_conv, tfnp.asarray(jax_conv)) |
| 99 | + |
| 100 | + |
| 101 | +if __name__ == "__main__": |
| 102 | + test.main() |
0 commit comments