1 /* Copyright 2017 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 #include <memory>
17 #include <vector>
18
19 #include "tensorflow/compiler/xla/client/global_data.h"
20 #include "tensorflow/compiler/xla/client/local_client.h"
21 #include "tensorflow/compiler/xla/client/xla_builder.h"
22 #include "tensorflow/compiler/xla/client/xla_computation.h"
23 #include "tensorflow/compiler/xla/shape_util.h"
24 #include "tensorflow/compiler/xla/status_macros.h"
25 #include "tensorflow/compiler/xla/statusor.h"
26 #include "tensorflow/compiler/xla/test_helpers.h"
27 #include "tensorflow/compiler/xla/tests/client_library_test_base.h"
28 #include "tensorflow/compiler/xla/tests/literal_test_util.h"
29 #include "tensorflow/compiler/xla/tests/test_macros.h"
30 #include "tensorflow/compiler/xla/tests/test_utils.h"
31 #include "tensorflow/compiler/xla/xla_data.pb.h"
32 #include "tensorflow/core/platform/test.h"
33 #include "tensorflow/core/platform/types.h"
34
35 namespace xla {
36 namespace {
37
38 class ClientTest : public ClientLibraryTestBase {};
39
XLA_TEST_F(ClientTest,ExecuteWithLayout)40 XLA_TEST_F(ClientTest, ExecuteWithLayout) {
41 XlaBuilder b(TestName());
42
43 std::vector<std::vector<int64>> layouts = {{0, 1}, {1, 0}};
44 for (const std::vector<int64>& execute_layout : layouts) {
45 for (const std::vector<int64>& transfer_layout : layouts) {
46 Add(ConstantR2<int32>(&b, {{1, 2}, {3, 4}}),
47 ConstantR2<int32>(&b, {{10, 20}, {30, 40}}));
48 TF_ASSERT_OK_AND_ASSIGN(auto computation, b.Build());
49
50 ExecutionOptions execution_options = execution_options_;
51 *execution_options.mutable_shape_with_output_layout() =
52 ShapeUtil::MakeShapeWithLayout(S32, /*dimensions=*/{2, 2},
53 execute_layout)
54 .ToProto();
55 TF_ASSERT_OK_AND_ASSIGN(
56 std::unique_ptr<GlobalData> data,
57 client_->Execute(computation, {}, &execution_options));
58
59 Literal expected_literal = LiteralUtil::CreateR2WithLayout<int32>(
60 {{11, 22}, {33, 44}}, LayoutUtil::MakeLayout(transfer_layout));
61
62 TF_ASSERT_OK_AND_ASSIGN(
63 auto computed, client_->Transfer(*data, &expected_literal.shape()));
64
65 ASSERT_TRUE(LiteralTestUtil::EqualShapesAndLayouts(
66 expected_literal.shape(), computed.shape()));
67 EXPECT_TRUE(LiteralTestUtil::Equal(expected_literal, computed));
68 }
69 }
70 }
71
XLA_TEST_F(ClientTest,ExecuteWithTupleLayout)72 XLA_TEST_F(ClientTest, ExecuteWithTupleLayout) {
73 XlaBuilder b(TestName());
74
75 Tuple(&b, {ConstantR2<int32>(&b, {{1, 2}, {3, 4}}),
76 ConstantR2<int32>(&b, {{10, 20}, {30, 40}})});
77
78 TF_ASSERT_OK_AND_ASSIGN(auto computation, b.Build());
79
80 ExecutionOptions execution_options = execution_options_;
81 // Create a result shape with one element column major and the other row
82 // major.
83 *execution_options.mutable_shape_with_output_layout() =
84 ShapeUtil::MakeTupleShape(
85 {ShapeUtil::MakeShapeWithLayout(S32, /*dimensions=*/{2, 2},
86 /*minor_to_major=*/{0, 1}),
87 ShapeUtil::MakeShapeWithLayout(S32, /*dimensions=*/{2, 2},
88 /*minor_to_major=*/{1, 0})})
89 .ToProto();
90
91 TF_ASSERT_OK_AND_ASSIGN(
92 auto result,
93 client_->ExecuteAndTransfer(computation, {}, &execution_options));
94 LiteralTestUtil::ExpectR2Equal<int32>({{1, 2}, {3, 4}},
95 LiteralSlice(result, {0}));
96 LiteralTestUtil::ExpectR2Equal<int32>({{10, 20}, {30, 40}},
97 LiteralSlice(result, {1}));
98
99 EXPECT_TRUE(result.shape().IsTuple());
100 EXPECT_EQ(2, ShapeUtil::TupleElementCount(result.shape()));
101
102 EXPECT_TRUE(ShapeUtil::Equal(
103 ShapeUtil::GetTupleElementShape(result.shape(), 0),
104 ShapeUtil::MakeShapeWithLayout(S32, /*dimensions=*/{2, 2},
105 /*minor_to_major=*/{0, 1})));
106 EXPECT_TRUE(ShapeUtil::Equal(
107 ShapeUtil::GetTupleElementShape(result.shape(), 1),
108 ShapeUtil::MakeShapeWithLayout(S32, /*dimensions=*/{2, 2},
109 /*minor_to_major=*/{1, 0})));
110 }
111
112 // Disabled for interpreter since ExecuteAsyncOnStream is not implemented on
113 // interpreter backend.
XLA_TEST_F(ClientTest,DISABLED_ON_INTERPRETER (DISABLED_ON_GPU (ExecuteParallel)))114 XLA_TEST_F(ClientTest,
115 DISABLED_ON_INTERPRETER(DISABLED_ON_GPU(ExecuteParallel))) {
116 XlaComputation add_with_one_arg, mul_with_two_args, dot_with_one_arg;
117 Shape shape = ShapeUtil::MakeShape(S32, {2, 2});
118
119 TF_ASSERT_OK_AND_ASSIGN(std::unique_ptr<GlobalData> const_arg,
120 client_->TransferToServer(
121 LiteralUtil::CreateR2<int32>({{5, 6}, {7, 8}})));
122
123 XlaBuilder b(TestName() + ".add");
124 Add(Parameter(&b, 0, shape, "param_0"),
125 ConstantR2<int32>(&b, {{1, 2}, {3, 4}}));
126 TF_ASSERT_OK_AND_ASSIGN(add_with_one_arg, b.Build());
127
128 // We can't really test parallel execution on CPU since all of the cores in a
129 // CPU are presented as a single device. So for now we test "parallel"
130 // execution on a single device.
131 std::vector<Client::XlaComputationInstance> computation_instances;
132 TF_ASSERT_OK_AND_ASSIGN(std::vector<xla::DeviceHandle> devices,
133 client_->GetDeviceHandles(1));
134 ASSERT_EQ(devices.size(), 1);
135
136 ExecutionOptions options = execution_options_;
137 *options.add_device_handles() = devices[0];
138 computation_instances.push_back(Client::XlaComputationInstance(
139 add_with_one_arg, {const_arg.get()}, options, nullptr));
140
141 TF_ASSERT_OK_AND_ASSIGN(auto results,
142 client_->ExecuteParallel(computation_instances));
143 auto expected_result = LiteralUtil::CreateR2<int32>({{6, 8}, {10, 12}});
144
145 TF_ASSERT_OK_AND_ASSIGN(
146 auto result_literal,
147 client_->Transfer(*results[0], &expected_result.shape()));
148
149 EXPECT_TRUE(LiteralTestUtil::Equal(expected_result, result_literal));
150 }
151
152 } // namespace
153 } // namespace xla
154