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
16 #include "absl/strings/str_cat.h"
17 #include "tensorflow/c/eager/c_api.h"
18 #include "tensorflow/c/eager/c_api_experimental.h"
19 #include "tensorflow/c/eager/c_api_internal.h"
20 #include "tensorflow/c/eager/c_api_remote_test_util.h"
21 #include "tensorflow/c/eager/c_api_test_util.h"
22 #include "tensorflow/c/eager/tfe_tensorhandle_internal.h"
23 #include "tensorflow/core/common_runtime/eager/eager_operation.h"
24 #include "tensorflow/core/common_runtime/function_optimization_registry.h"
25 #include "tensorflow/core/distributed_runtime/rpc/grpc_server_lib.h"
26 #include "tensorflow/core/framework/function.h"
27 #include "tensorflow/core/graph/graph.h"
28 #include "tensorflow/core/platform/casts.h"
29 #include "tensorflow/core/platform/errors.h"
30 #include "tensorflow/core/platform/protobuf.h"
31 #include "tensorflow/core/platform/test.h"
32 #include "tensorflow/core/protobuf/cluster.pb.h"
33 #include "tensorflow/core/protobuf/config.pb.h"
34 #include "tensorflow/core/protobuf/tensorflow_server.pb.h"
35
36 namespace {
37
38 using ::tensorflow::string;
39
TestRemoteExecute(bool async)40 void TestRemoteExecute(bool async) {
41 tensorflow::ServerDef server_def = GetServerDef(2);
42
43 // This server def has the task index set to 0.
44 string serialized = server_def.SerializeAsString();
45
46 server_def.set_task_index(1);
47
48 std::unique_ptr<tensorflow::GrpcServer> worker_server;
49 ASSERT_TRUE(tensorflow::GrpcServer::Create(
50 server_def, tensorflow::Env::Default(), &worker_server)
51 .ok());
52 ASSERT_TRUE(worker_server->Start().ok());
53
54 TF_Status* status = TF_NewStatus();
55 TFE_ContextOptions* opts = TFE_NewContextOptions();
56 TFE_ContextOptionsSetAsync(opts, static_cast<unsigned char>(async));
57 TFE_ContextOptionsSetDevicePlacementPolicy(opts,
58 TFE_DEVICE_PLACEMENT_EXPLICIT);
59 TFE_Context* ctx = TFE_NewContext(opts, status);
60 EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
61 TFE_DeleteContextOptions(opts);
62
63 TFE_ContextSetServerDef(ctx, 0, serialized.data(), serialized.size(), status);
64 EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
65
66 TFE_TensorHandle* h0_task0 = TestMatrixTensorHandle(ctx);
67 TFE_TensorHandle* h1_task0 = TestMatrixTensorHandle(ctx);
68 const char remote_device_name[] =
69 "/job:localhost/replica:0/task:1/device:CPU:0";
70 auto* h0_task1 =
71 TFE_TensorHandleCopyToDevice(h0_task0, ctx, remote_device_name, status);
72 ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
73 auto* h1_task1 =
74 TFE_TensorHandleCopyToDevice(h1_task0, ctx, remote_device_name, status);
75 ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
76
77 TFE_Op* matmul = MatMulOp(ctx, h0_task1, h1_task1);
78 TFE_OpSetDevice(matmul, remote_device_name, status);
79 EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
80
81 TFE_TensorHandle* retvals[1];
82 int num_retvals = 1;
83 TFE_Execute(matmul, &retvals[0], &num_retvals, status);
84 EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
85
86 TF_Tensor* t = TFE_TensorHandleResolve(retvals[0], status);
87 ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
88 float product[4] = {0};
89 EXPECT_EQ(sizeof(product), TF_TensorByteSize(t));
90 memcpy(&product[0], TF_TensorData(t), TF_TensorByteSize(t));
91 TF_DeleteTensor(t);
92 EXPECT_EQ(7, product[0]);
93 EXPECT_EQ(10, product[1]);
94 EXPECT_EQ(15, product[2]);
95 EXPECT_EQ(22, product[3]);
96
97 TFE_DeleteTensorHandle(h0_task0);
98 TFE_DeleteTensorHandle(h1_task0);
99 TFE_DeleteTensorHandle(h0_task1);
100 TFE_DeleteTensorHandle(h1_task1);
101 TFE_DeleteTensorHandle(retvals[0]);
102
103 TFE_DeleteOp(matmul);
104
105 TFE_Executor* executor = TFE_ContextGetExecutorForThread(ctx);
106 TFE_ExecutorWaitForAllPendingNodes(executor, status);
107 ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
108 TFE_DeleteExecutor(executor);
109 TFE_DeleteContext(ctx);
110
111 TF_DeleteStatus(status);
112
113 // TODO(b/136478427): Figure out how to correctly shut the server down.
114 worker_server.release();
115 }
116
TEST(CAPI,RemoteExecute)117 TEST(CAPI, RemoteExecute) { TestRemoteExecute(false); }
TEST(CAPI,RemoteExecuteAsync)118 TEST(CAPI, RemoteExecuteAsync) { TestRemoteExecute(true); }
119
TestRemoteExecuteSilentCopiesOp(bool async,bool remote,bool remote_func_outputs=false)120 void TestRemoteExecuteSilentCopiesOp(bool async, bool remote,
121 bool remote_func_outputs = false) {
122 return TestRemoteExecuteSilentCopies(async, remote, /*func=*/false,
123 /*heavy_load_on_streaming_rpc=*/false,
124 remote_func_outputs);
125 }
126
TEST(CAPI,RemoteExecuteSilentCopies)127 TEST(CAPI, RemoteExecuteSilentCopies) {
128 TestRemoteExecuteSilentCopiesOp(/*async=*/false, /*remote=*/true);
129 }
TEST(CAPI,RemoteExecuteSilentCopiesAsync)130 TEST(CAPI, RemoteExecuteSilentCopiesAsync) {
131 TestRemoteExecuteSilentCopiesOp(/*async=*/true, /*remote=*/true);
132 }
TEST(CAPI,RemoteExecuteSilentCopiesLocal)133 TEST(CAPI, RemoteExecuteSilentCopiesLocal) {
134 TestRemoteExecuteSilentCopiesOp(/*async=*/false, /*remote=*/false);
135 }
TEST(CAPI,RemoteExecuteSilentCopiesLocalAsync)136 TEST(CAPI, RemoteExecuteSilentCopiesLocalAsync) {
137 TestRemoteExecuteSilentCopiesOp(/*async=*/true, /*remote=*/false);
138 }
139
140 } // namespace
141